Pagani

From the Grid to the Smart Grid,
Topologically
Giuliano Andrea Pagani

The work is supported by the University of Groningen under the Ubbo Emmius
fellowship program and by IBM under the IBM Ph.D. Fellowship 2013-14.
Printed by NetzoDruk - www.netzodruk.nl - Groningen

RIJKSUNIVERSITEIT GRONINGEN
From the Grid to the Smart Grid, Topologically
Proefschrift
ter verkrijging van het doctoraat in de

Wiskunde en Natuurwetenschappen
aan de Rijksuniversiteit Groningen
op gezag van de
Rector Magnificus, dr. E. Sterken,
in het openbaar te verdedigen op
vrijdag 14 maart 2014
om 12.45 uur
door
Giuliano Andrea Pagani
geboren op 29 maart 1981

te Parma, Italië
Promotores: Prof. dr. M. Aiello
Beoordelingscommissie: Prof. dr. P. Pelacchi

Prof. dr. F. Reed-Tsochas
Prof. dr. M. van Steen
ISBN: 978-90-367-6813-9
ISBN-Electronic: 978-90-367-6812-2
To my family
Contents
Content ix
1 Introduction 1
1.1 The New Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Motivations for the future grid . . . . . . . . . . . . . . . . . . 4
1.1.2 What is the smart grid? . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Thesis Scope and Organization . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2 Related Work 15
2.1 The Visions of New Grids . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Complex Network Analysis and the Power Grid . . . . . . . . . . . . 18
2.2.1 Power grid characteristics . . . . . . . . . . . . . . . . . . . . . 19
2.2.2 Statistical global graph properties . . . . . . . . . . . . . . . . 19
2.2.3 The small-world property . . . . . . . . . . . . . . . . . . . . . 23
2.2.4 Node degree distribution . . . . . . . . . . . . . . . . . . . . . 23
2.2.5 Betweenness distribution . . . . . . . . . . . . . . . . . . . . . 25
2.2.6 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Power Grid, Network Design and Evolution . . . . . . . . . . . . . . 27
2.3.1 Evolution of complex networks . . . . . . . . . . . . . . . . . . 27
2.3.2 Complex network analysis for power grid design . . . . . . . 29
2.3.3 Power engineering approach for power lines design . . . . . . 30
2.3.4 Power lines adaptation . . . . . . . . . . . . . . . . . . . . . . . 31
2.4 Smart Grid Services and Applications . . . . . . . . . . . . . . . . . . 33
2.4.1 Smart grid software applications and architectures . . . . . . 33
2.4.2 Service-oriented architectures for energy . . . . . . . . . . . . 36
vii
Contents
2.4.3 Agent interactions in the energy market and the smart grid . 38
3 Network Models for the Smart Grid 41

3.1 The Dutch Medium and Low Voltage Grid . . . . . . . . . . . . . . . 41
3.2 Unweighted Power Grid Study . . . . . . . . . . . . . . . . . . . . . . 45
3.2.1 Betweenness analysis . . . . . . . . . . . . . . . . . . . . . . . 49
3.2.2 Fault tolerance . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3 Weighted Power Grid Study . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.1 Betweenness analysis . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.2 Fault tolerance . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.3 Unweighted vs. weighted study . . . . . . . . . . . . . . . . . 58
3.4 Relating Grid Topology to Electricity Distribution Costs . . . . . . . . 59
3.4.1 Application to the Dutch distribution grid . . . . . . . . . . . 63
4 Network Evolutions for the Smart Grid 67

4.1 Network Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.2 Building New Distribution Networks . . . . . . . . . . . . . . . . . . 70
4.2.1 Model parameters . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.2.2 Model generation . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.2.3 Economic considerations . . . . . . . . . . . . . . . . . . . . . 88
4.2.4 Topology costs vs. benefits . . . . . . . . . . . . . . . . . . . . 100
4.3 Evolution of Current Distribution Networks . . . . . . . . . . . . . . 101
4.3.1 Network evolution policies . . . . . . . . . . . . . . . . . . . . 101
4.3.2 Comparison of the evolution strategies . . . . . . . . . . . . . 105
4.3.3 Economic considerations . . . . . . . . . . . . . . . . . . . . . 115
4.4 New Topologies for the Smart Grid . . . . . . . . . . . . . . . . . . . . 123
5 ICT Services and Applications for the Smart Grid 125

5.1 The Services: the Future Energy Landscape . . . . . . . . . . . . . . . 125
5.1.1 Smart grid challenges . . . . . . . . . . . . . . . . . . . . . . . 125
5.1.2 Service-oriented architecture supporting the smart grid . . . . 127
5.1.3 Traditional service-oriented architecture vs. energy service-
oriented architecture . . . . . . . . . . . . . . . . . . . . . . . . 128
5.2 Today’s Services for Implementing the Smart Grid . . . . . . . . . . . 130
5.2.1 Interface towards energy price service . . . . . . . . . . . . . . 131
5.2.2 Interface towards energy producing equipment service . . . . 133
5.2.3 Interface towards environmental service . . . . . . . . . . . . 134
5.2.4 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.2.5 Pricing energy from renewables . . . . . . . . . . . . . . . . . 137
5.2.6 Renewable energy forecast . . . . . . . . . . . . . . . . . . . . 138
viii
Contents
5.2.7 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

5.3 Smart Grid Aware Buildings . . . . . . . . . . . . . . . . . . . . . . . . 141
5.3.1 Realization of a smart office . . . . . . . . . . . . . . . . . . . . 142
5.3.2 Benefits of the smart grid-enabled system . . . . . . . . . . . . 143
5.4 Agent-Based Energy Market . . . . . . . . . . . . . . . . . . . . . . . . 146
5.4.1 Roles in the smart grid energy market . . . . . . . . . . . . . . 147
5.4.2 Agent modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.4.3 Agent interactions . . . . . . . . . . . . . . . . . . . . . . . . . 149
5.4.4 Agent behavior in the market . . . . . . . . . . . . . . . . . . . 150
5.5 Bits and Electrons Hand in Hand . . . . . . . . . . . . . . . . . . . . . 152
6 Conclusion 153
6.1 The Emergence of a New Grid . . . . . . . . . . . . . . . . . . . . . . . 153
6.2 Open Issues and Future Directions . . . . . . . . . . . . . . . . . . . . 155
A Graph Theory and Complex Network Fundamentals 159

A.1 Graph Theory and Complex Network Definitions and Properties . . 159
A.2 Graph properties example . . . . . . . . . . . . . . . . . . . . . . . . . 163
B Complex Network Models 167

B.1 Building Synthetic Networks . . . . . . . . . . . . . . . . . . . . . . . 167
B.2 Model Parameters for Synthetic Networks . . . . . . . . . . . . . . . . 175
Bibliography 181
ix
Chapter 1
Introduction
The human body of a man 25 years old at 180 cm and 75 kg, requires just about
2.128 kWh of energy (1830 kcal) a day for his basal metabolic rate [85]. All other
energy consumption that each of us requires is to make our life more productive,
more enjoyable, and more safe-in essence, more rich. Energy is the enabler of mod-
ern societies and cities. If one looks around, everything is powered, moved or lit by
some form of energy. The availability of energy is taken for granted in the modern
society so much, that we perceive its real importance only when outages or black-
out strike. In fact, the effects of prolonged blackouts have threatening consequences
both for population security and safety, as well as for economic production [7, 11].
In developed countries, energy has become such an affordable and easily accessible
commodity that we are reminded of its presence only when the car needs to be filled
or when we receive the bill of the energy retail company.
The scope of the present work is on the electric grid and system, therefore, when
we refer to energy we typically mean it in its electric form. Actually, electricity is not
the only form of energy that was mainly used. This was even more true in the period
before the advent of the Information and Communication Technology (ICT) and the
digital life. Worldwide, in 2011, the total consumption of energy is dominated by
oil and then electricity with 40.8% and 17.7%, respectively; followed by natural gas
(15.5%) [90]. If we look back to 1973, for example, we note that in 2011 we have
an almost doubling in the total consumption (4674 million tons of oil equivalent vs.
8918 million tons of oil equivalent); and looking at the share composition in 1973,
oil was even more important with 48.1% followed by natural gas and coal with 14%
and 13.7% of total consumption, respectively; whereas electricity had just a 9.4%
share [90].
Electricity applications were initially not completely understood and the effects
of electricity were not considered particularly useful outside of the physics lab. The
first application dates back to the incandescent bulb. However, here electricity is
a “carrier” of energy that is transformed in other forms of energy, namely, thermal
energy, and light. The first application of electricity, per se, has to deal with the
telegraph. Again, electricity is used as a carrier this time for the electric signals
encoding of information. Electricity was then used mainly as a carrier of energy
2 1. Introduction
from a source point (i.e., power plant) to the end user. Usually at the source, various
other forms of energy are converted into electricity (e.g., the conversion route may
entail be combustible fuel, to thermal energy, to motion, to electricity). Once in form
of electricity, energy can be transmitted from the source to the destination through
a capillary infrastructure: the power grid.
The power grid is considered one of the greatest achievements of the 19th and
th
20 centuries. Just consider the challenge of bringing the power through a whole
continent, such as Europe or America. A typical simplified scheme of the power
grid is represented in Figure 1.1. One sees an organization of a transmission (extra
high and high voltage) and a distribution grid (medium and low voltage). Energy is
mainly produced in large power plant facilities by a few authorized actors, usually
far from big population or industrial conglomerates where the energy is needed. As
the plants are the components at maximum power in the system, they are connected
to the high voltage level; while end users consume mostly at the medium and low
voltage levels. The technical reason for this voltage stepping is to prevent the dis-
sipation and losses for the transmission of the high amount of power with small
currents, and safety in the surrounding of the end user. In fact, losses are propor-
tional to the square of the current. Therefore, on the one hand, very high voltages
are required to transmit power while having small currents. Whereas, on the other
hand, high voltages are deadly for humans, thus, requiring low voltages in residen-
tial premises. The structure is therefore highly hierarchical. This is also reflected
in its commercial organization, where, at the beginning of the electricity era, there
was a monopoly for regulatory, technical, and economic reasons. Samuell Insull
(1859-1938), initially an assistant to Edison and then an energy entrepreneur him-
self, stated a famous quote that summarizes the approach to the electricity sector in
the early days [131]:
“Every home, every factory, and every transportation line will obtain its
energy from one common source, for the simple reason that that will be the
cheapest way to produce and distribute it.”
The concept is perfectly correct in its historical context and in an electrical system
where energy sources are cheap, cost of transmission are negligible and environ-
mental externalities for energy production are not taken into account. However,
when these factors start to take their matter, the electricity sector might begin to
consider different paradigms and technologies for production and distribution.
1.1. The New Grid 3
Figure 1.1: The physical organization of the power grid. (Source: adapted from
Wikipedia)
1.1 The New Grid

During its evolution, the electrical system has become more and more pervasive,
and ICT and telecontrol have been applied to monitor, control, and actuate; espe-
cially in the high voltage part of the system. The medium and low voltage end of
the grid have received less attention due to their traditional role as a passive com-
ponent of the grid. With the current technology and operation of the grid in the
western world, the power supply is highly reliable; for instance in the Netherlands
in 2012, the electrical system was 99,99486% of the time available with just an aver-
age of 27 minutes downtime per year per customer [149]. The Dutch system has one
the highest availability in Europe. Even though the outlook seems good and with
no issues, the future is not trouble-free.
In a 2003 meeting involving American utilities companies and U.S. governmen-
4 1. Introduction
tal institutions [206], the concluding outlook on the status of the grid was not reas-
suring: the grid was found outdated with limited possibilities of expansion in the
transmission lines close to urban centers, inefficient, and unable to meet the energy
requirements of the information-oriented society and economy. Low investments
were realized especially at the level of distribution, and many projects with the
aim of increasing the capacity were canceled. However, some encouraging points
emerged from the meeting. Examples include the focus on new technologies includ-
ing smart power systems and clean energy systems. In addition, investments on the
order of billions of dollars to replace existing equipment and a plan for new ones
were considered. Another aspect mentioned in the meeting called for clearer rules
for regulation. In essence, the report of the meeting laid the foundation for a smart
grid.
1.1.1 Motivations for the future grid

The motivations that drive the requirement to update and modify the traditional
system lie in several points. The first point of concern is the reduction of green-
house gas emissions: the system has to accommodate large quantities of sustainable
and clean sources of energy. This aspect poses a challenge to the traditional energy
system. Actually, renewable energy poses threats to the electrical system due to their
difficulty in output predictability. In fact, the electric system was designed with the
idea in mind to follow the demand by using resources with a predictable output,
such as fossil fuels, and also having the possibility to dispatch a plant. Dispatching
is the operation of planning the use of a power plant in order to meet an expected fu-
ture demand. Different plants are dispatched at different moments in order to satisfy
demand at any time and given the different inertia of the power plant with different
technologies. Plants with high inertia satisfy the base load (e.g., nuclear and coal),
while plants with smaller inertia satisfy the medium and peak load (e.g., gas and
oil). Renewable energies pose a challenge since they highly depend on natural vari-
ability. Therefore their prediction is not accurate (exceptions are hydro, biomass and
tidal powered plants) and it is clearly impossible to drive nature to follow human
demand. One can see the difficulty in having a system fully based on renewables
with the classic idea of demand following [72]. The system needs, therefore, to be
more flexible to accommodate the energy demand based on the availability of the
renewable resources in a point in time. A second motivation for the future grid is
the technological and economic availability of small generators producing energy
from renewable sources such as photovoltaic (PV) panels and small wind turbines.
These technologies are nowadays often subsidized by governments with the aim
of reducing greenhouse gas emission and to achieve higher penetration of renew-
1.1. The New Grid 5
able energy. A third motivation lies in the transformation that the energy market is
undergoing. From the initial monopoly, the situation has changed in many coun-
tries and now more and more companies are present in the energy business from
energy production, to energy transmission and distribution, to retail and services
provided to the end user. For instance, this is the case of The Netherlands. To en-
able and accelerate this process, governments in the western world have promoted
policies to open the electricity business and facilitate competition with the final aim
to both modernize the energy sector and provide a more convenient service for the
end user. Such tendency of the last decades is termed unbundling. In essence, un-
bundling is the process of dismantling monopolistic and oligarchic energy systems,
by allowing a greater number of parties to operate in a certain role of the energy
sector and market [51, 93]. Considering the vision of combining the previous two
aspects, one notes that it is possible for everyone to become a small producer and en-
ergy provider (of small-scale). Such a small-scale approach is considered beneficial
to the electricity system in many ways: from reduced losses, since source and load
are closer; to system modularity; to smaller investments compared to large-scale en-
ergy solutions [123]. Another motivation towards a modernization of the grid lies
in the necessity of using modern ICT technologies for the operations of the energy
sector. ICT and automation can ease many large scale operations that were (and still
are) done manually such as meter reading, user connection and disconnection, and
network switch operations. In addition, by using ICT in the energy domain provides
new possibilities and scenarios for the end user, one can raise energy awareness and
provide automation for energy efficiency.
1.1.2 What is the smart grid?

Smart grid is a recent term that does not have a unique single meaning. This un-
certainty is apparent in the statement by the The Global Community for Sustainable
Energy Professionals - Leonardo ENERGY1 “A smart grid is neither a clearly defined
single concept, nor a single technology. Rather, it is like a basket containing various combi-
nations of balls. The context and the interpretation depend upon the user.” More specifi-
cally, the Department of Engineering and Public Policy of Carnegie Mellon Univer-
sity describes that the characterization of the smart grid is non-unique and various
aspects are involved in the smart grid concept [139]. According to the authors, there
is not a single meaning for the new energy landscape concept, but rather every ac-
tor involved in the grid has different views of the smart grid according to his related
business; and the various stakeholders of the smart grid landscape can benefit in
different ways from it:
1 http://www.leonardo-energy.org/what-definition-smart-grid
6 1. Introduction
• The customer can benefit from real-time tariffs that reflect the price of electric-
ity, react with his loads to the tariffs and receive information directly from the
meter about his consumption and costs.
• The electricity distribution companies can have a higher level of automation

to manage ordinary operations, critical situations, and a more selective way of
shedding load (e.g., based on the importance of the service provided: hospi-
tals or police stations might be the last to be removed from the network due
to their social and safety importance). For distribution companies, the smart
grid also enables scenarios to manage easier integration of distributed energy
generation facilities in the low level grid (e.g., photovoltaic installations and
wind farms) or enable islanding of subsets of the grid during emergencies.
• The generating and transmission companies can benefit from a more “com-
puterized” grid with more information and data about critical grid measures
(e.g., network’s voltage phase). Having more information enables more auto-
matic and distributed decision-making even far from the control center, thus
optimizing the grid operations.
Now a natural question concerns the actors of this system. The scenario is more
complex when a monopoly is not in place due to higher competition, efficiency re-
quirements, and a different scale of the operations involved. The view of National
Institute of Standards and Technology (NIST) for smart grid described in the NIST
Smart Grid Framework [155] is depicted in Figure 1.2, where the many actors and
their domains in the smart grid landscape are represented. It is thus clear from the
picture that many different actors are present and that many different technologies
need to interact at different levels to properly enable the synergy between the many
actors. In the figure, one notes that what is really new is the information flow be-
tween the various parties involved. Traditional flows of information that involve
the generation and transmission grid with the control operation center and the mar-
ket are still present. These flows of information are essential for the monitoring,
control and operation of the grid; they have been a key component of the grid for
a long time. We now see that the distribution grid is also involved with informa-
tion exchange with the market, the operations, and the customer. The customer
becomes an active player and his/her appliances and equipment will be monitored
by the utility service provider for commercial purposes and at an aggregate level
by the distributor for metering and technical purposes (e.g., power quality). We
also see the interaction directly with the market that enables the user to become
an active participant in selling and buying his own energy. The techniques used
to interact with the end user to modify his load are broadly named demand-side
integration. Demand-side integration incorporates various other terms that stress
1.1. The New Grid 7
Figure 1.2: NIST smart grid framework.
some peculiar aspects, such as demand-side management, where utilities compa-

nies perform activities to influence customer energy behavior (e.g., peak/off-peak
tariffs); demand-response, where the demand is shaped to respond to supply con-
ditions (e.g., real-time pricing); demand-side participation where users participate
in three aspects, providing a more economic, secure, and environmental-friendly
electric system [66]. Depending on the type of customer, utilities implement dif-
ferent methods to control or influence customer’s load, such as direct load control
where the utility controls the customer loads (e.g., air conditioners, water heaters),
or agreeing on a curtailable service during contingencies that give the customer a
discount (usually applied to big industrial or commercial customers), or demand-
side bidding where customers offer curtailment with bids based on the wholesale
energy market. Other operational implementations consider the situation of emer-
gency curtailment and ancillary services for market; in this last one, the customer
bids to stay on a standby situation and curtail if needed. Looking at the energy flow,
we see that it still follows its original route from the generators to the customer.
However, in the last segment of the grid (i.e., distribution), the customer has be-
come a producer (cf. PV panels and wind turbine on top of the customer house in
Figure 1.2) that feeds the energy surplus back into the grid. Actually, a new name
has been coined to indicate the new figure of end user that has both the roles of
producer and consumer of energy: the prosumer.
8 1. Introduction
In summary, smart grid is not a single technology, but rather an approach involv-
ing multiple technologies to improve and change the way the actual power grid is
used and managed. The most important innovative aspect is to have, in parallel
with the energy flow, an information flow that enables more advanced functionali-
ties both for the grid operators and for the end users. A synthesis of the smart grid
approach can be seen in the reports issued by the U.S. Department of Energy [145]
and [146] with the main underlying idea is that the grid should be able to:
• self-diagnose and take appropriate recovery actions in case of faults,
• become more resistant to willing and malicious attacks,
• satisfy the users with an improved power quality meeting their needs and
expectations,
• be ready to integrate different sources of power generation,
• give the end users the possibility to interact and respond to real-time electricity
price signals in a distributed energy market.
1.2 Thesis Scope and Organization

In the vast smart grid domain, the research challenge that we tackle in the present
work concerns the changes necessary to fully enable the smart grid vision. The first
research question deals with the infrastructure. Which development must the phys-
ical power grid infrastructure undergo to support a future scenario where users can
locally produce their own energy and are free to participate as prosumers in an un-
bundled energy market? The second topic of interest deals with the ICT applications
for the smart grid and leads to the following research questions: which architecture
and features of future applications will enable the smart grid? What are the appli-
cation scenarios available for the user in a fully unbundled energy market?
To answer the first question, we focus on the part of the power grid infrastruc-
ture that is close to the end user: the distribution grid. We resort to complex net-
work analysis to analyze the current state of the grid infrastructure and consider
possible evolution in the future. Complex networks are a key framework to better
understand those systems that are composed by many interacting parts in a network
fashion in order to grasp an overall behavior of the system. The study of complex
networks is part of complex systems theory, which is the multidisciplinary scientific
branch that studies systems with many interacting parts (or agents) in which the
overall emerging behavior of the system cannot be deduced by the observation of
the single parts. Other essential features of complex systems are the absence of a
1.2. Thesis Scope and Organization 9
centralized controller and the evolution of the system over time [136]. The choice of
complex systems approach is twofold: first, since the electrical system might evolve
towards a complex system; second, at the present stage, given the major uncertainty
surrounding the smart grid topic, the planning of the new grid is not to be a fully
detailed one, but more of a high level scenario where coarse grain network models
are useful instead of circuits at very precise level of details. Having in mind these
characteristics of complex systems, we envision that with more autonomy in pro-
duction and consumption, the system might reduce the centralization control and
go towards an electrical system that is locally self-organized where users interact
in selling and buying energy. Self-organization and lack of a central controller to-
gether with interaction between multiple entities are some of the fundamental prin-
ciples of complex systems theory. The analysis of networked systems following the
paradigm of complex network analysis has provided interesting results in several
fields (e.g., computer science, health, biology). We consider complex networks as a
useful tool for our goal of realizing a high level decision support system to study
and evolve the smart grid which is a networked system by definition. In particular,
having a weighted representation of the network, where the weights represent phys-
ical properties of the grid, is helpful in approaching the real infrastructure without
all the very low levels details. Network science has its basis in graph theory and in
the mathematical branch that study topology. When referring to networks, topology
is defined as the particular organization (i.e., pattern) of the connections (i.e., links)
between the nodes that compose the network.
To answer the questions more focused on the ICT application side, we focus on
the software approach that is capable of sustaining a set of systems and applications
that are diverse, highly distributed, and with constraints for security and timing.
These characteristics are satisfied by service-oriented architectures (SOAs). Going
more into the application domain of the smart grid, we aim at realizing and testing,
in conditions and environments as close as possible to the real ones, the applications
that will likely to be in place in the future smart grid. The success of the smart grid
will depend on the applications that the end user will be provided to manage his
energy production and consumption. Building and testing these applications in a
living lab environment is the first step to their engineering to a higher scale.
These two topics are closely related: the smart grid is about an improvement of
the infrastructure to make it more efficient, safe, and a rich of source of information,
on the one hand, and enable new applications to promote flexibility in the energy
system, on the other hand. The infrastructure and ICT application domains will
have to interact. Figure 1.3 shows this overall picture. The grid and its components
such as transformers, substations, small-scale and traditional generators, and lines
will have, especially at distribution level, more ICT-based sensing and automation
10 1. Introduction
capabilities compared to the current situation. Each physical piece of equipment

will also be characterized by a service point that it provides to other entities in the
SOA for the smart grid. In the SOA, information layer logical entities provide and
consume information in a standardized (e.g., eXtensible Markup Language-based)
fashion. Logical entities can be supervisory control and data acquisition (SCADA)
systems and energy management systems (EMS) belonging to utilities or to the end
user household. Also the smart meters or building energy management systems,
and the agent technology that will reside in them, will be a part of the SOA infras-
tructure and therefore the gateway to access and provide external services. Using
such an approach, the electricity sector will have a substantial change in its opera-
tions being more ICT and communication intensive and, at the same time, more ef-
ficient and agile (e.g., automated metering reading, load shaping through dynamic
pricing, switch automation). In addition, the smart grid has the objective of raising
user awareness and interest concerning energy. With the smart grid, the energy sec-
tor is not the utility sector anymore where user pays the bill with the same spirit and
will of paying income taxes, but a vibrant sector where new services and opportu-
nities are ongoing towards a more sustainable production and use of energy.
The title of the present thesis mentions the term topologically. In fact, we want to
stress the focus on the new patterns of connections and shapes of the new grid. We
examine in detail how the topological infrastructure that characterizes the grid will
have to change its connectivity in the transition to the smart grid. In addition, we
will look also at the connections between new classes of objects that need to interact,
therefore being connected, from an application point of view. These new objects
that interact digitally are smart meters, SCADA systems, EMS, local energy markets,
distributed power grid sensing equipment, devices and automation actuators inside
the house. The interaction between all of them will build a new panorama for the
energy sector.
The novelty of the thesis resides in the several aspects. First of all, in studying a
new topic such as the smart grid, then in focusing on the distribution network that
has not received much attention in the literature. In addition, the approach used
is based on complex network analysis and our contribution proposes a new frame-
work model of evolution towards the smart grid. Our study takes into account also
the economic factors of the evolution of the grid and the infrastructural elements
that influence the cost of electricity distribution. Our findings lay the foundation for
a decision support system that helps the utilities and governments in considering
evolution scenarios of the distribution infrastructure towards a local energy gen-
eration and exchange. On the software and application side of the smart grid, we
investigate novel scenarios and applications available in the future electrical system;
our main contribution is to have these application tested in real environments with
1.2. Thesis Scope and Organization 11
Figure 1.3: Smart grid physical and information infrastructure.
realistic data in order to prove the feasibility of the smart grid vision and concept.
The thesis, therefore, reflects the two main topics mentioned above: infrastruc-
ture analysis and design, and ICT applications tailored to the smart grid.
The thesis entails three focus areas: visions of the future energy system, complex
network approach to the power grid, and software approach and applications for
the smart grid. Chapter 2 provides the necessary background and state of the art.
First, we focus on the vision and scenarios characterizing the new energy world en-
abled by the smart grid. Second, we review the most relevant work in the field of
complex network analysis applied to the power grid considering different geogra-
phies and the focus of the different studies. We also study the main techniques of
design and improvement of a grid infrastructure. In particular, we look at several
aspects: the complex network analysis work relevant to the power grid design, the
evolution of complex networks where links are missing or need to be added, and
12 1. Introduction
the design approach used in the electrical engineering field. Third, we look at the
literature concerning the software approach for the smart grid services and their
applications.
In Chapter 3, we analyze several samples of the Dutch medium and low voltage
grid by using a complex network analysis approach. Since, in our vision of the smart
grid, the end users are both producers and consumers, we focus on this layer of the
grid close to the end user. We expect the medium and low voltage network to be
crucial and invigorate lots of innovation, also in the topology of the infrastructure.
The study of this layer of the power grid has not received particular attention before,
especially using complex network analysis techniques. Our study takes into account
not only purely topological features, but also physical properties of the lines. In
addition to the topological analysis, we look at economic measures that influence
the price of electricity distribution that are related to the topological aspects of the
grid and apply them to the Dutch case.
Chapter 4 goes further in the complex network approach applied to the power
distribution grid. We explore the evolution of the infrastructure towards the smart
grid. First, we consider the development of a distribution grid infrastructure from
scratch. Therefore, we look at the most famous network models from the literature
and evaluate them against a set of topological metrics that are relevant in the smart
grid context. We also present an evolution of the current Dutch grid samples where
we propose an increase in connectivity of the network compared to the current one.
We explore several strategies of adding new lines between existing nodes and we
evaluate the evolved networks against the set of metrics. In both studies, we empha-
size the economic side in terms of the cost to build such networks (or their evolution
from the current infrastructures) and the benefits that the change in topology reflects
on the electricity distribution costs.
Chapter 5 focuses on ICT applications. We look at the service-oriented approaches
to the smart grid and point out the aspects that make it suitable for this kind of appli-
cation. We then dig into the realization of relevant foreseen services that the smart
grid will need to have. We provide an ideal solution for the future and we also
show how a feasible implementation, close to the ideal, can be realized nowadays
with current technologies and Internet services. An application of these real ser-
vices for the schedule of appliance operations driven by energy dynamic pricing in
an office environment is presented. The last application provides the future energy
market. We present an agent-based platform to perform energy trading between
consumers, prosumers, and traditional energy companies. We show how a bidding
strategy for an agent can be optimized and how software agents can be coupled
with real metering devices such as smart meters.
Chapter 6 concludes this thesis providing some discussion on the topics pre-
1.3. Publications 13
sented and a direction for the future work.

The thesis is concluded with two appendix chapters to provide essential back-
ground. A.1 provides an overview of the most important graph theory and complex
networks concepts. B provides a description of the most known network models
from the complex networks literature that are used in Chapter 4.
1.3 Publications
Part of the work presented in this thesis has been published in or, at the time of
writing, is under consideration for publication by several journals and conferences.
In Table 1.1, we provide an overview of the papers on which this thesis is based and
the chapters they are mostly relevant for. Additionally, the papers [156, 35, 34, 161]
have contributed to shape the smart grid vision reported here, but are not directly
related to any chapter. We stress that the contributions are to be considered a joint
effort with the respective co-authors.
Chapter Venue Citation

Physica A: Statistical Mechanics [166]
and its Applications
2
Service Oriented Computing and [160]
Applications
3 IEEE Transactions on Smart Grid [158]
IEEE PES ISGT 2013 [165]
ISCIS 2012 [162]
4
Physica A: Statistical Mechanics [167]
and its Applications
Submitted to journal Tech. Rep. [163]
IEEE Transactions on Smart Grid [76]
ICAC 2011 Workshop Session [36]
5
ICSOC 2010 Workshop Session [157]
Submitted to journal Tech. Rep. [164]
Table 1.1: Overview of publications and manuscripts under review related to the
topics presented in the chapter of the thesis.
Chapter 2
Related Work
The smart grid is a vision and a trend for the electricity system as described is Chap-
ter 1. It is also a new topic of research that in these past few years has received more
and more attention. The reason is the ingenuity required to provide more flexibility
to the electrical system to accommodate more renewable sources, to design new so-
lutions to include electric vehicles, to consider users that are no longer passive and
just consumers, but that are able to generate their own energy requirements and
feed the surplus to the grid. In the past few years, the academic research devoted to
the smart grid has increased, and a proof is the establishment of scientific journals
on the topic and international conferences and workshop focused to this theme.
Here we provide an overview on the smart grid with special focus on the main
aspects of this thesis: the vision of the future energy interactions; the study of the
topology of the power grid and its evolution following the approach of complex
network analysis (CNA); the status concerning applications involving Information
and Communication Technology (ICT) technologies for the smart grid.
2.1 The Visions of New Grids

In the last decades, the view of considering the electrical system and the power grid
managed and administered by a single company or institution has changed and
questions to this model have appeared. More and more views of involvement of
users in the energy picture are appearing. The idea of a new energy landscape is
highlighted in the book “Power to the people” [207]. In the book, among the vari-
ous solutions to achieve a more sustainable energy future and a viable way for our
planet to avoid the effects of global warming, is a new paradigm where local (i.e., SM
all-scale) production of energy and ICT are the cornerstone of the Internet of energy
that is to come. According to the author it is not only technology driving this change,
but its combination with rising environmental issues and the changes in the politi-
cal/economic landscape fostering liberalization and competitiveness. The vision of
an energy path that can can avoid big power plants and huge investments is also
promoted by the environmentalist Amory Lovins in his book “Small is Profitable:
16 2. Related Work
The Hidden Economic Benefits of Making Electrical Resources the Right Size” [123].
In the book, both the economic and the environmental factors are taken into ac-
count and the viability of solutions based on distributed generation powered by
renewable sources is considered a better alternative than traditional ones character-
ized by pollution and green house gases emissions. Small-scale modular solutions
are better than the traditional oversize power plants from a financial point of view
given the delays and costs that such large projects incur. In addition, with modular
distributed generation plants 207 benefits can be achieved according to Lovins. The
same attitude towards the benefits from economic, environmental and reliability
perspective in adopting micro-power generation (i.e., plants smaller than 10 MW)
is shared by Dunn [62]. In another book titled “Hot, Flat, and Crowded: Why We
Need a Green Revolution - And How It Can Renew America”, Thomas Friedman
sees the Internet of energy as the revolution that can help in reaching a sustainable
future made of green energy together with energy efficiency and energy conserva-
tion [73]. As in the Internet, users create content and organize in communities, in the
Energy Internet users can actively participate in production of energy and organize
in virtual power plants or energy cooperatives. The benefits of an approach to the
electrical system based on micro-grids rather than traditional macro-grids is pro-
vided by Marnay and Venkataramanan [129]. The authors emphasize three benefits
achievable with micro-grids. First, micro-combined heat and power (micro-CHP)
units have the ability to produce heat and electricity close to the location where it is
needed the most. Second, power quality and reliability requirements differ among
the pieces of equipment, performing distinct functions for the end user. Therefore,
the authors speculate on the possibility of realizing different grids to be used to
power equipment with different levels of quality and reliability. This idea is based
on the criticality for society of certain equipment and thus the possibility (or not)
to tolerate more instabilities. The vision of an electrical system more similar to the
Energy Internet is proposed in [205], where the authors propose a system that in or-
der to be self-healing and more efficient should be characterized by agents for every
component or entity (e.g., meters, transformers, generation plants, customers). The
agents can reconfigure and react to the changing conditions in the system. The au-
thors explain three main differences between the data transmission and the electrical
transmission: first, electricity is usually centrally generated and only one or few way
of transmission are available, whereas data are generated and routed everywhere in
the network without a priori fixed paths; second, it is difficult and inefficient to store
energy in considerable quantities while data are easily stored and cached in the In-
ternet; third, the whole Internet is based on the best-effort concept and quality of
service is the exception, whereas the security of supply of the customer is the pri-
mary rule of the power system. The authors propose as a solution for a self healing
2.1. The Visions of New Grids 17
and a more flexible network, to have agents sharing information on the status of
each component, precise customer load forecasting, and virtual electricity storage.
This last concept is achieved through a demand-response approach intervening on
customers’ loads. The idea of a more decentralized grid in the generation and in the
control of the production and the assets is expressed in [19]. The authors analyze
peer-to-peer topologies adequate to the distributed infrastructure where agents (i.e.,
peers in the network) can control equipment and cooperatively decide on the best
action to take when a problem is flooded to the agents. The agents also act on behalf
of users in an online power market and can shed loads on behalf of the users when
the grid is in critical conditions. We see similarities between this approach of consid-
ering the new interactions between agents on the grid and the peer-to-peer concept
of the smart grid that we have described in Section 1. Our approach is more fo-
cused on the high level interaction between prosumers, while other works consider
the peer-to-peer paradigm also for for communication and control purposes [19].
The vision for a future energy scenario where energy services are the core of the
smart grid is proposed in a vision paper about 10 years old by Gellings [75]. The au-
thor envisions a convergence of electricity, communication, and energy services in a
device defined as consumer portal. This portal is the fusion of a smart meter, an In-
ternet access router (the author considers a power line communication access to the
Internet), a Web portal and a communication gateway to the intelligent appliances
at home. Through the portal the energy companies (i) can promote energy-related
services, (ii) can provide the user with tariffs, manage the status (i.e., metering) of
other utilities services (e.g., water, gas), (iii) can monitor the appliances to inspect
for failures, (iv) can monitor the status of distributed energy resources and energy
storage devices. It is surprising how this vision that is on the scene for a decade it is
still very up-to-date and basically today’s ideas concerning the home energy man-
agement systems are unchanged and still far to a massive deployment at customer
side. The idea of a home and building automation platform for energy purposes is
the focus of Moneta et al. [137]. The vision follows those described above, where
an energy market with multiple suppliers and freedom of choice for the customer
are in place. Utilities offer additional services to the customer such as remote man-
agement of the distributed energy resources, the shedding of specific loads, where
the customer can respond to changing tariffs or emergency conditions. The authors
have developed a software platform that enables these functionalities and they have
tested it in a realistic home environment.
In summary, the visions concerning the electricity system of the future are in line
with our own view of the smart grid, as described in Section 1. The future of electric-
ity systems will be much more focused on small-scale generation and distribution
where users will be active participants in generation and balance of the system. ICT
18 2. Related Work
technology will make this change possible through smart meters and home energy
management systems that are able to interact on a free energy market on behalf of
the user. Users will be producers and likely to be energy feeders in local micro-grids.
The energy utilities will be no more utilities, but will evolve towards energy service
companies offering more energy-related added value services (e.g., appliance mon-
itoring, energy efficiency). The grid, especially at medium and low voltage level
will be more remotely managed and controlled with software agents that control
power assets and interact together to decide the best and safest configuration for
the equipment.
2.2 Complex Network Analysis and the Power Grid

The works that we examine consider the power grid networks as graphs following
the mathematical meaning of the term. The main investigation usually performed
when analyzing the power grid concerns reliability. This is almost always the mo-
tivation that drives complex network analysis studies related to high voltage elec-
trical infrastructures. Usually, the investigation involves evaluating the disruption
behavior of the graph when its nodes or edges are removed.
Other terms to compare the various power grid studies involve more general
characteristics of the network. In particular, the geographical location of the an-
alyzed grid is responsible for topological properties due to the different morpho-
logical characteristics of different countries. Another relevant aspect deals with the
layer of the power grid under investigation since differences can emerge from a
topological perspective investigating the different ends in which the grid is usually
partitioned: high, medium and low voltage. It is also important to have informa-
tion if the type of power grid graph comes from a real network infrastructure or it
is a synthetic sample extracted from blueprint models for the power grid electrical
engineering literature.
The motivations to include the works are based on the thoroughness and pre-
cision of the research performed, the rigor in the application of complex network
analysis methodologies and the geography of the power grid analyzed in order to
cover a broad spectrum of the infrastructures realized in the different countries and
identify possible differences or similarities.
The power grid is clearly amenable to such studies and a number of these have
been performed on the high voltage grid. Here we describe the most important as-
pects of each work under investigation. The works shown in Table 2.1 have been
chosen based on the following factors: they are specifically about the power grid,
they cover US, European, Chinese and Indian grids or synthetic topologies from
2.2. Complex Network Analysis and the Power Grid 19
electrical engineering literature, they have samples of different sizes and, most im-
portantly, these are the best-known and most representative works from the com-
plex network analysis and power grid community.
2.2.1 Power grid characteristics

Concerning the grids we consider general and non-technical aspects so to give a
global idea as shown in Table 2.1. Several aspects of comparison are taken into
account: the number of nodes and lines composing the grid (second and third col-
umn)1 ; the type of sample considered either a real grid or synthetic samples, for
instance, coming from IEEE literature such as IEEE Bus systems (fourth column);
the type of grid analyzed (fifth column) in belonging either to the transmission part
(high voltage) or to the distribution part (medium and low voltage); another essen-
tial information deals with the geography of the grid (last column).
The data are in most cases extracted from real samples, that is, they represent
real electric infrastructures. Other works in addition to real power grids consider
synthetic grids. Most of these studies investigating synthetic samples use IEEE
blueprint networks such as IEEE Bus systems, while very few concentrate only
on other synthetic networks (e.g., non-IEEE models, small-world models, random
graphs). Almost all samples belong to the high voltage end of the power grid. The
high voltage grid contains the lines used for long range transmission to which big
power plants are attached too; the only exception is our study [158] that is focused
on the distribution part of the grid.
From a geographical perspective the samples are mainly localized in the United
States and in Europe with some studies that consider Chinese high voltage samples,
and only one that considers India; a map of the countries whose grids are analyzed
is represented in Figure 2.1. Another main commonality is to treat the grid as an
undirected graph where each substation or transformer represents a node and each
line transporting electricity is an edge.
2.2.2 Statistical global graph properties

The main characteristics from a graph and complex network analysis perspective
of the grids under analysis are summarized in Table 2.2 where we consider the or-
der (N ) and size (M ) of the graph (second and third column). The average degree,
computed as xk y 2M N , gives a general idea of how many vertexes is an average
1 Note that the numbers in the second and third column are not the exact numbers, but they are an
approximation to give the idea of the importance of the sample.

20 2. Related Work
Work Number of Number of Sample Network Geography

Nodes Lines Type Type
[4] 14000 19600 Real HV North America
[57] 300 500 Real HV Italy
[42] 314000 N.A. Real HV North America
[89] 4800 5500 Real HV Scandinavia
[183] 2700 3300 Real HV Europe
[58] 370 570 Real HV Italy, France
and Spain
[185] 370 570 Real HV Italy, France
and Spain
[221] 4900 6600 Real HV Western US
[217] 8500 13900 Synthetic HV Western US and
and real New York State Area
[158] 4850 5300 Real MV/LV Netherlands
[132]2 210 320 Synthetic HV China
and real
[182] N.A. N.A. Real HV Europe
[28] 300 411 Synthetic HV
[105] 6400 8700 Synthetic HV North America,
and real Scandinavia and Korea
[83] 900 1150 Real HV China
[216] 3200 7000 Synthetic HV New York State Area
and real
[215] 4900 6600 Real HV Western US
[60] 1700 1800 Real HV China
[65] 39 46 Syntethic HV
[64] 39 46 Syntethic HV
[82] 2500 2900 Real HV China
[198] 15400 18400 Real HV North America and China
[178] 14000 19600 Real HV North America
[26] 550 700 Synthetic HV Italy
and real
[87] 29500 50000 Synthetic HV North America
and real
[175] 400 700 Synthetic
[32] 900 1300 Synthetic HV South-East US
and real
[38] 60 110 Real HV India
Table 2.1: Comparison between studies using CNA for the power grid .
vertex connected to (fourth column). Fifth, sixth and seventh column give infor-
mation about the type of statistical analysis performed on the graph, in particular,
the assessment of node degree distribution and betweenness distribution together
with an evaluation of the path length are considered. Another term of comparison
deals with the type of graph analyzed taking into account weights or not. The last
two columns of the table consider the type of aim of the analysis: either an investi-
gation of the disruption behavior of the grid, or the evaluation of the small-world
properties.
Considering Table 2.2, a difference appears: the studies closer to a topological
2 The values for nodes and lines in this table refer only to a snapshot of Shanghai power grid.
Work Sample Sample Average Node Degree Betweenness Path Length Weighted/Unweighted Resilience SM all-world
Order Size degree Distribution Distribution Analysis Analysis Analysis Investigation
Statistics Statistics
[4] 14000 19600 2.80 X X Unweighted X
[57] 300 500 3.33 X X Indirectly through Weighted not based X
efficiency metric on physical properties
[42] 314000 N.A. N.A. X Unweighted X
[89] 4800 5500 2.29 X X Unweighted X X
[183] 2700 3300 2.44 X Unweighted X
[184] 3000 3800 2.53 X X Unweighted X X
[191] 3000 3800 2.53 X Unweighted X
[58] 370 570 3.08 Indirectly through Unweighted X
efficiency metric
[185] 370 570 3.08 X X Unweighted X
[221] 4900 6600 2.69 X Unweighted X
[217] 8500 13900 3.27 X X Unweighted X
and impedance analysis
[158] 4850 5300 2.18 X X X Both X X
[132]2 210 320 3.05 X Both X X
[182] N.A. N.A. X
[28] 300 411 2.74 Both X
[105] 6400 8700 2.72 X Unweighted X X
[86] 300 411 2.74 X(chart only) X Both X
[218] 8500 13900 3.27 X Unweighted X
[83] 900 1150 2.55 X X X Weighted X X
[216] 3200 7000 4.375 X(chart only) X Weighted X
2.2. Complex Network Analysis and the Power Grid
[215] 4900 6600 2.69 Unweighted X

[60] 1700 1800 2.12 X Both X
[65] 39 46 2.36 Weighted X
[64] 150 46 2.36 Weighted X
[82] 2556 2892 2.26 Weighted X
[198] 15400 18368 2.39 X(for one sample only) X Unweighted X X
[168] 550 800 2.91 X Unweighted X
[178] 14000 19600 2.80 Weighted not based on X
on physical properties
[8] 90 120 2.67 Weighted X
[26] 550 700 2.55 Weighted X
[87] 29500 50000 3.39 Weighted X
[175] 400 700 3.5 X Weighted X X
[32] 900 1300 2.89 Unweighted X
[38] 60 110 3.67 Unweighted
Table 2.2: Comparison of the main characteristics of the graphs related to power grids .
21
22 2. Related Work
Figure 2.1: Map of the power grid infrastructure studied using CNA approach.
characterization use unweighted representation of the edges of the grid and con-
sider always the node degree distribution in the analysis, since it is an important
element to define the type of network (e.g., scale-free network). On the other hand,
the studies that apply a weighted representation of the grid do not consider the node
degree distribution statistics: neither considering the unweighted definition of node
degree, nor using a definition that takes into account weights as proposed in [152].
This last aspect might be worth consider since the degree distribution properties of
the network might change the picture of the node degree distribution in comparison
with unweighted studies.
Centrality measures are not often used, exceptions are [132, 216, 28, 65] these
are necessary to identify the statistical distribution of critical nodes. More atten-
tion to centrality measures, especially using weighted representation of the power
grid based on the capacity or energy flows through the grid, might be beneficial in
understanding the most critical nodes or lines in the power system.
Another recurring theme in the complex network analysis involving the power
grid is the reliability analysis, which actually it is the main motivation that drives
these kind of studies. In fact, many works were performed after major blackout
occurred, such as the North American black-out of 20033 or the Italian one of 20034
(e.g., [4, 56, 57, 42, 178]) or anyway mention blackouts as the main motivation for the
work. The fragility and resilience properties of the power grid has been the major
reason of concern that has determined the focus of such complex network analysis
studies on the high voltage network. That is why almost all studies consider the
behavior of the grid to various attacks to its nodes or edges.
2.2.3 The small-world property

The small-world is a name for a property of networks where a path between any
two nodes is usually small. It has received lots of attention starting with sociological
studies [135, 203], but more recently with application of this concept and model to
many more classes of networks [222, 221, 196, 108]. Among the studies analyzed,
the small-world property investigation is performed by ten out of the thirty two
considered. The various studies look for the satisfaction of the small-world property
described by Watts and Strogatz [222, 221].
In general, there is no definitive answer to the question regarding the member-
ship of power grid networks to the small-world group. It is indeed very specific to
the samples analyzed and no general conclusion can be drawn, this seems especially
true for the high voltage grids, while the medium and low voltage networks seem
far from being small-world networks [158]. The path lengths are similar to those of
random graphs satisfying the condition (L LRG ) proposed in [222], while analyz-
ing the clustering coefficient (γ), one sees that only for few samples the condition
of clustering coefficient much higher than a random graph with the same order and
size (γ " γRG ) is satisfied. This justifies the statement made before stating that no
general definitive answer can be given for the power grid network considering the
membership to the small-world network category.
2.2.4 Node degree distribution

The degree of a node is the number of edges incident to it. However, this informa-
tion is not particularly important for big graphs since keeping track of each node
degree may not be manageable nor meaningful. Instead it is better to have a general
idea of the statistics of the node degree for the whole graph to understand the over-
all tendency of the statistical distribution as a synthesis of the degree behavior of
the network. In particular, its probability distribution gives us some insights of the
general properties of the networks such as the likely or unlikely presence of nodes
with very high degree (sometimes also referred as hubs). Table 2.3 shows the main
information about the degree distribution in the works studied. The second column
gives the type of cumulative node degree distribution. The third column provides
the analytical function obtained in the studies by building a statistical distribution
from the empirical data.
3 http://news.bbc.co.uk/2/hi/americas/3152451.stm
4 http://news.bbc.co.uk/2/hi/3146136.stm
24 2. Related Work
Work Cumulative Node Fitted

Degree Distribution Distribution
y pxq e0.5x
Probability Type
p q 2.5e0.55x
[4] exponential
y1 pxq 0.84x3.04
[57] exponential y x
y2 pxq 0.85x3.09
[42] power-law
y1 pxq e0.81x
y2 pxq e0.54x
[183] exponential
[184] exponential y pxq e0.56x

y1 pxq e0.18x
2
[185] exponential or
y2 pxq e0.21x 0.18e0.25px4q
2 2
sum of exponential
y3 pxq 0.96e0.17x 0.25e0.19px3.9q
2 2
terms
[217, 218] Sum of truncated y1 pxq f1 pxq
geometricaland irregular
discrete terms y2 pxq f2 pxq
y1 pxq x1.49
y2 pxq 0.15e21.47x 0.84e0.49x
[158] power-law (unweighted)
and sum of exponential
y1 pxq e0.65x
terms (weighted)
y2 pxq e0.58x
[83] exponential
Table 2.3: Comparison of the node degree cumulative distribution probability func-
tions.
As seen in the table, the results do not completely agree on the type of the dis-
tribution followed by power grid networks, but generally they are close to an ex-
ponential decay. In the works [217, 218] presented in the table the functions f1 pxq
and f2 pxq are not reported in the table for size reason, but in footnote.3 For studies
concerning multiple samples (i.e., [183, 184, 158]) averages between all samples, or
particular significant samples have been chosen among the many available.
In general the various studies focusing on the high voltage grids agree on a statis-
tical distribution for node degree that follows an exponential (or exponential based)
distribution with characteristic parameters of the exponential curve that depend on
the specific grid. While high voltage grid have been quite extensively analyzed, the
medium and low voltage grids have not found much attention so far and a deeper
and wider investigation needs to be performed in different geographies since the
only study (i.e., [158]) is representative of the Northern part of the Netherlands. In
addition, the distribution grid will be the section of the power grid mostly impacted
by innovations in future power systems i.e., smart grid technology [31].
Very few studies in the literature have shown results of an accurate methods
such as the one based on Kolmogorov-Smirnov method, shown by Clauset et al. [47].
3fpxq° °x x 0.2269p0.7731qx t0.4875δp1q, 0.2700δp2q, 0.2425δp3qu xi 1, 2, . . . 34
1 i
f2 pxq x x 0.4085p0.5916qx t0.3545δ p1q, 0.4499δ p2q, 0.1956δ p3qu xi 1, 2, . . . 16

i
i
The symbol is here to be considered as the convolution operator and the δ is the Dirac delta function.
i
As remarked in [47], many distributions that with a naive fitting analysis might be
considered power-laws at first sight do not prove to belong to such a category when
a more rigorous fit test is adopted.
2.2.5 Betweenness distribution

Betweenness is an important measure to assess how a node is central in a network.
This metric in fact computes how many shortest paths traverse a node, therefore
giving an information of the importance of the node in the path efficiency. The main
characteristics of betweenness studies are summarized in Table 2.4 where the sec-
ond column shows the type of followed distribution, while the analytical function
is represented in the third column. This metric is computed by only five studies
(i.e., [4, 57, 158, 83, 216]). Although the studies that perform this type of analysis are
only few, one can see that there is a tendency for the high voltage network to have
a betweenness distribution close to a power-law. For the medium and low voltage
the situation is less clear: some samples analyzed in [158] follow an exponential de-
cay, especially the low voltage ones, while other, usually the bigger belonging to the
medium voltage, follow a power-law.
Work Cumulative Betweenness Fitted

Distribution Distribution
xq0.7
Probability Type
y pxq p2500
p q 10000p785 xq1.44
[4] power-law
y1 pxq x1.18
[57] power-law y x
4 x
[158] power-law and
y2 pxq 0.68e6.810
1.71
exponential
y pxq x1
1.48
[83] power-law
y pxq x2
Table 2.4: Comparison of the betweenness cumulative distribution probability func-

tions.
Power-law seems the dominating rule for betweenness probability distribution

even if few studies consider this statistical property of graphs. To draw a general
definitive conclusion regarding this property for the power grid more studies are
required. Another aspect to be considered which has not received much attention
so far is the study of betweenness statistics in weighted grid models or when power
flows are considered instead of the pure topological analysis.
2.2.6 Remarks
Complex network analysis studies and their comparison show how important prop-
erties of a real system such as the power grid can be studied using graph modeling
26 2. Related Work
tools and which conclusions about the reliability of the infrastructure can be drawn.
CNA proves to be a good set of tools that provide, although without dealing with
the details and complexities of the electrical properties in the case of the power grid,
a comprehensive and general understanding of the properties that characterize a
network. We see an interesting trend in the various works analyzed, that is, the re-
search towards more complex representation of the properties of the network than
a simple graph. In fact, although complex network analysis can help in understand-
ing the foundational properties of the network aspects of the power gird, it is al-
ways worth to remember that the power grid is subject to the law of physics and
the principles of electrotechnics. From the initial studies (e.g., [221, 4, 42, 89, 198])
considering the power grid just as an undirected graph without any property (i.e.,
weight) on edges and with no characterization of the nodes, more recent studies
take into account the electrical properties of the power grid system. Of course
the aim of these later studies is always to provide, anyway, a simplification of the
highly complex power system. However, they add those essential parameter to
better model the grid characteristics: impedance parameters associated to the trans-
mission lines, power limits supported by the substations (i.e., nodes) such as for
instance in [132, 26, 87, 82, 216]. A more detailed description of the power grid
under investigation (i.e., weighted graph representation) enables to better under-
stand the dynamics guiding the power grid with a mixed approach: both preserv-
ing the idea of the complex network analysis of having a general and statistical
behavior of the overall power grid, and, on the other hand, to take into account the
physical/electrical properties essential to characterize the power grid. Latest results
in [87, 26, 8] show a better agreement to real power grid behavior of the models that
take into account physical parameters, compared to the pure topological analysis,
and the observed behaviors and critical points in real power systems thus justifying
this enhanced complex network analysis approach.
We also see another gap in the current scientific work concerning the power grid
and complex network analysis techniques: the applicability to the real power sys-
tems of the results obtained by the network analysis. Especially, the findings of the
studies that consider the vulnerability and cascading effects of the power grid need
to be confirmed by the transmission and distribution operators. We remark here
that basically all the works lack the cross-check of the theoretical results with the
experience on the field.
A noteworthy general aspect is the role that complex network analysis has for
the power grid infrastructure vulnerability analysis: CNA does not want to substi-
tute the traditional approaches to power systems resilience and safety analysis since
they have proved extremely successful in governing and managing the electrical
system with only occasional highly disruptive events. CNA techniques applied to
2.3. Power Grid, Network Design and Evolution 27
the power grid world represent a simplification of all the complexities governing
the power systems, but such approach can anyway be helpful to give a general vi-
sion that can help in identifying quickly and in a simple way critical spots or aspects
of the power grid which then may be further and deeply analyzed with traditional
electrical engineering tools.
2.3 Power Grid, Network Design and Evolution

Network dynamics consider the change over time of a network. It consists basically
on how new nodes and links are added or the current ones are removed. We look
at evolution with specific focus on link formation concerning Complex Networks.
We then draw the attention on complex network analysis studies to evolve the high
voltage power grid and then we focus on traditional electrical engineering power
lines design and adaptation.
2.3.1 Evolution of complex networks

The study of the dynamics evolution of networks by using complex network anal-
ysis techniques has been widely investigated. The focus is mainly to assess and
analyze existing networks that have evolved over time and understand which prin-
ciple is underlying such evolution. More in general, the evolution is considered
in defining abstract algorithms that can be used to build networks that mimic the
same behavior of natural or technical networks. For a general introduction to com-
plex network analysis generation techniques, we refer to the extensive works such
as [39, 151]. The work of Liben-Nowell and Kleinberg [121] discusses link predic-
tion. The authors consider the evolution of the ArXiv library co-authorship network
(i.e., a social network) and define several quantitative metrics to infer the future
links to be formed in the network (i.e., new co-authored publications). The metrics
that the authors use for predictions are based on evaluating the paths, the neigh-
borhood of nodes, rank of the matrix representing the graph, and the clustering.
Compared to a random prediction of new links the methods used in the paper are
far better; in fact the random predictions is correct less than 1% of the times, while
on some topics of the ArXiv library the predictions reach a correct outcome 50%
of the time. The ideas used in the paper and the metrics used in the prediction
of new connections between nodes prove valuable for the co-authorship social net-
work. Even if at the abstract level, networks are the same (i.e., graphs) it is difficult
to apply these metrics and prediction strategies on a real infrastructure such as the
power grid. In fact, many of the metrics considered in the paper are inspired by the
28 2. Related Work
underlying social aspects of the co-author relationship such as neighbors in com-

mon. This aspect is reinforced by the different results that the same metrics used on
different topics of the ArXiv library sometimes provide in the predictions following
the different metrics. Given the small average node degree of the power grid net-
works, the absence of samples at different time stamps, it is difficult to apply the
metrics and methods of the paper to distribution networks and expect meaningful
and interesting results.
The survey of Lü et al. [125] provides a good overview on link prediction. The au-
thors propose several methods and algorithms based on similarity properties of the
nodes at local level (e.g., the neighborhood of a node), global similarity (i.e., over the
whole network), and quasi-local that combines the best of the two approaches. They
also propose other strategies such as probabilistic models to evaluate the likelihood
of the formation of a link. One of the networks that is considered for link predic-
tion in [125] is the Western U.S. power grid (data already used in [221]). In this case
the methods that give more stress on the local interactions between nodes perform
better in the prediction than other approaches. The authors also state that for infras-
tructural networks the geography plays an important role where the long distant
connections are discouraged and very rare. The paper describes three main applica-
tions of link prediction in the field of Complex Network Analysis: reconstruction of
networks, evolution mechanism of networks, and classification of partially labeled
networks. We consider these prediction mechanisms valuable for networks with
rich neighborhood structures where differences and intersection between neighbor-
hoods are meaningful (e.g., social networks) or where the discovery of links is par-
ticularly costly or long in time and efforts (e.g., protein interactions). The current
power grid networks have low average node degree and the structure is usually
close to radial, especially at medium and low voltage level, thus we consider the
predictions less interesting. However, these techniques and link predictions might
be more applicable in the further evolution and prediction of new links when the
networks are more rich in connectivity. In addition, some techniques for link predic-
tion would benefit from additional information available to characterize the nodes,
i.e., the number of prosumers that are connected to a node (i.e., substation) and the
available production power.
The concept of network evolution of a technological infrastructure is considered
in [176] where a model for the evolution of the Internet Autonomous System (AS)
is provided. As most works (e.g., [61, 12]), this one focuses on creating a model of
the existing evolution of the Internet rather than proposing new ways of evolving
networks. The interesting aspect on the modeling of the Internet is the better results
of the proposed model defined as Parallel Addition and Rewiring Growth (PARG)
compared to other models in the literature to capture specific features of the AS
topology. In particular, the model is able to capture the dissortativeness of hubs

(i.e., the most connected nodes in the network) and linear local assortativeness of
nodes. The model proposed proves successful, however one must recognize that
the AS connections are mainly logical and are due more to business and contractual
agreements than physical interconnections, therefore missing part of the problem
that is essential in the case of the power grid topology.
2.3.2 Complex network analysis for power grid design
Few studies in the complex network analysis landscape consider the possibility of
using the insight gained through the analysis to help the design. These few cases
consider the addition of lines in the network to assess the increase in the reliability
of the entire power grid. Examples are the study of the Italian high voltage grid
[58] and the study of improvement by line addition in Italian, French and Span-
ish grids [185]. Also Holmgren [89] uses complex network analysis to understand
which grid improvement strategies are most beneficial. He shows the different im-
provements of typical complex network analysis metrics (e.g., path length, aver-
age degree, clustering coefficient, network connectivity) although in a very simple
small graphs (less than 10 nodes) when different edges and nodes are added to
the network. Broader is the work of Mei et al. [132] where a self-evolution process
of the high voltage grid is studied with complex network analysis methodologies.
The model for power grid expansion considers an evolution of the network where
power plants and substations are connected in a “local-world” topology through
new transmission lines; overall the power grid reaches in its evolution the small-
world topology after few-steps of the expansion process. Wang et al. [217, 219] study
the power grid to understand the kind of communication system needed to support
the decentralized control. The analyses aim at generating samples using random
topologies based on uniform and Poisson probability distributions and a random
topology with small-world network features. The simulation results are compared
to the real samples of U.S. power grid and synthetic reference models belonging
to the IEEE literature. These works also investigate the property of the physical
impedance to assign to the generated grid samples.
Complex network analysis is not generally used as a design tool to propose new
topologies for the future smart grid as we use in this work where we also assess the
benefits in terms of economical improvement.
30 2. Related Work
2.3.3 Power engineering approach for power lines design
Traditionally power system engineers adopt techniques from electrotechnics, which

model deterministically the power lines and their characteristics, these might in-
clude the exploitation of graph theory principles [213, 54]. The traditional tech-
niques applied by power engineers involve the individuation of an objective func-
tion representing the cost of the power flow along a certain line which is then subject
to physical and energy balance. The problem translates into an optimization prob-
lem. Since long time the models are applied both for the high voltage [74, 116] and
the medium and low voltage planning [213, 54]. Not only optimization, but also
expert systems [126] are developed to help in the process of designing grounding
stations based on physical requirements as well as heuristic approaches based on
engineering experience. The substation grounding issue is approached as an op-
timization problem of construction and conductor costs subject to the constraints
of technical and safety parameters, its solution is investigated through a random
walk search algorithm [79]. In [71] a pragmatic approach using sensitivity analysis
is applied to a linear model of load flow related to various overloading situations
and a contingency analysis (N-1 and N-2 redundancy conditions) is performed with
different grades of uncertainty in medium and long term scenarios.
In the practice, the planning and expansion problem is even more complex since
it implies power plants, transmission lines, substations and the distribution grid.
In [81] all these aspects are assessed separately and several challenges appear. For
instance in the planning of a high voltage overhead transmission line, specific clear-
ance code must be followed and not only load is a key element, but also topography
and weather/climatic (above all wind and ice) conditions play an import role in the
planning of the infrastructure. For substation planning the authors of [81] empha-
size, in addition to the need for upgrading the grid (e.g., load growth, system sta-
bility) and budgeting factors, the multidisciplinary aspects that come from environ-
mental, civil, electrical and communications engineering. A more general approach
is proposed in [81] to deal with power system planning; it might be regarded as a
multi-objective (e.g., economics, environment, feasibility, safety) decision problem
requiring the tools typical of decision analysis [100].
The works mentioned so far take into account mainly the high voltage end of
the grid while not least important is the distribution grid, especially in the vision
of the future electrical system where the end user plays a vital role. The integrated
planning of medium and low voltage networks is tackled by Paiva et al. [169] who
emphasize the need of considering the two networks together to obtain a sensible
optimal planning. The problem is modeled as a mixed integer-linear programming
one considering an objective function for investment, maintenance, operation, and
losses costs that need to be minimized satisfying the constraints of energy balance
and equipment physical limits.
Even more challenges to electrical system planning are posed by the change in
the energy landscape with several companies running different aspects of the busi-
ness (generation, transmission, distribution) and by the increasing of prosumers and
renewable energy resources since these players have less control on each other com-
pared to a fully integrated electricity system and they also havedifferent objectives.
In addition, accommodating more players in the wholesale market transmission ex-
pansion should follow (as it is already for generation) a market based approach i.e.,
the demand forces of the market and its forecast should trigger the expansion of
the grid [30]. The same considerations regarding the need for a different approach
in planning in a deregulated market are expressed in [189] where the optimization
of an objective function in the market environment is applied. Another method to
evaluate transmission expansion plan takes into account the probability reliability
criteria of Loss Of Load Expectation (LOLE). In particular, in [45] an objective func-
tion is proposed that takes into account the cost of constructing a transmission line
between all buses involved. The designed line is then subject to constrains in peak
load demand satisfaction and a certain level of LOLE that the line should not outrun.
In the smart grid framework the planning techniques might be revised espe-
cially for the distribution grid which is the segment that is likely to face the greatest
changes due to the presence of Advanced Metering Infrastructure (AMI) (i.e., bidi-
rectional intelligent digital meters at customer location) and Distribution Automa-
tion (DA) (i.e., feeders can be monitored, controlled in automated way through two-
way communication). In addition, the medium and low voltage grid is no longer
a layer where only energy is consumed, but distributed energy generation facilities
(small-scale photovoltaic systems and small-wind turbines) will be attached to this
segment of the grid; altogether these elements are likely to reshape the way plan-
ning for medium and low voltage are realized [31].
2.3.4 Power lines adaptation

A power engineering problem relevant to the scope of this work is that of distribu-
tion grid topological reconfiguration. The early works on the topic of Merlin and
Back considered the problem with heuristic techniques and algorithms (e.g., branch
and bound) to reconfigure the switches of the distribution infrastructure [134]. Later
works of Cinvilar et al. [46] provide computationally tractable methods to modify
the topology of the distribution grid by opening/closing switches to reduce losses.
The authors provide a simplified formula to estimate the reduction in losses be-
tween the situation before and after the reconfiguration in the topology. Baran and
32 2. Related Work
Wu [17] have also proposed the idea of minimizing the losses and guaranteeing the
load balancing. Basically, the problem deals with finding the minimal spanning tree
of the network, since the radial configuration needs to be preserved, that minimizes
the objective function (i.e., system losses) while satisfying the constraints on volt-
age, capacity of lines and transformers, and reliability. For this problem, Baran et
al. use a simplification of the power flow equations to compute the power flow in
the network to be optimized.
Recently, new interest has grown concerning the reconfiguration problem to-
gether with islanded grids and micro-generation plants. The work of Ramesh et
al. focuses on the minimization of losses in the distribution grid [179]. The authors
provide several options that have been simulated and tested on the field to reduce
losses on real and reactive power. The three solutions proposed concern: distributed
generation, capacitor placement and restructuring of feeders. The first proposal con-
sist in placing local energy generation closer to the end users. In certain buses of the
simulated IEEE 37 Bus, they show loss reduction of up to 9 MW. The second solu-
tion proposes the installation at optimal locations of capacitors. The third proposed
technique concerns the restructuring of the topology of the network. The solution
proposed is basically a reconfiguration of the opening and closing of the switches of
the distribution network to maintain the network radial and satisfy its load require-
ments. This approach although valuable from the practical point of view, lacks the
vision in imagining the future energy scenario and there is no investigation of which
topology is best, or according which principles the new pieces of infrastructure (e.g.,
cables) should be interconnected. The aspect of network reconfiguration of power
networks to achieve a minimization of losses is the topic of [144]. The authors state
that the only way to improve efficiency is by altering the topology. They focus on
two aspects: (i) finding an optimal switching scheme of the switches connecting the
lines and minimize losses, (ii) adding lines. The case study used is a IEEE 14 Bus
where three more lines are added. The authors do not give any details why only
three lines are considered and the motivation to choose specific nodes to attach the
new lines. In addition, the only argument provided concerns the economic aspect
of adding only the three lines, but no quantitative evidence is provided. In [190]
the problem of network reconfiguration in a smart grid environment is addressed.
The idea of the authors is to reconfigure the topology of the network by operating
switches in order to minimize the overload in the branches of the network. The
scenario they consider takes also into account the higher penetration rate that dis-
tributed generation will have in the future. The numerical evaluation of the pro-
posed genetic algorithm to reconfigure the network is realized on a simple 33 Bus
network where the system configuration problem is solved to minimize an objective
function (i.e., losses) and without violating the voltage and current constraints on
2.4. Smart Grid Services and Applications 33
the lines. Even in that work, the importance of topology is claimed and the benefits
in terms of reduced losses are shown. However, no additional lines or investment
of the distribution grid are proposed and the benefit come just from the reconfigura-
tion. No motivation or order of magnitude of the avoided investment are provided
and the effects of topological changes available in the switching are anyway lim-
ited. The work of Xiaodan et al. [225] is in the spirit of studying the reconfiguration
in a micro-grid environment with micro-generation plant based on renewables. Ba-
sically, the problem is considered as two optimization sub-problems: one related to
determine the capacity of each island of the grid that has micro-generation capabil-
ities and the other is a problem of reconfiguration of the distribution grid with the
objective of minimizing the power losses. The optimization algorithm is validated
against two test grids: the IEEE 33 Bus and the PG&E 69 node system. The recon-
figuration for the system in an island and micro-generation uses techniques and ap-
proaches similar to traditional reconfiguration of distribution systems. The problem
considered is an optimization problem and the authors constrain the distribution
grid to be always operated in a radial configuration. Usually, the reconfiguration
problem boils down to the definition of an objective function to minimize the losses
of the system and to establish constraints to satisfy the load. The function is then
solved resorting to some heuristic.
2.4 Smart Grid Services and Applications

The smart grid promises to bring changes to the electrical systems by promoting a
massive use of ICT technologies to enable the various actors in the system to interact
and create new services. Therefore, the ICT aspects concerning the smart grid have
received even more interest than changes in the infrastructure so far. We summarize
the main features of the literature concerning service orientation and energy, smart
grid software applications and architectures, and energy market-oriented solutions.
2.4.1 Smart grid software applications and architectures

The smart grid has grown in popularity in the past few years and several approaches
and architectures have been proposed for the information services related to the
smart grid for the smart home. Most of the works are architecture proposal with
interfaces to services that are implemented abstractly or based on historical energy
data, thus, missing to model the real-time dynamics of the system.
In [187], the authors simulate the effects of aggregate households appliances and
electric vehicles to act as balancing devices for solar installations. They show the
feasibility of reducing the amount of spinning reserves and the benefits in terms
34 2. Related Work
of reduced CO2 emissions, considering several types of solar plants and scenarios
of electric vehicles penetration. The work is entirely a simulation effort with no
real services involved. Another simulation tool for demand-response is proposed
in [214]. The goal is to analyze at a very detailed level (bus voltages, power flows,
etc.) a control strategy to accommodate the wind power of a turbine and its fluctua-
tions (0.5 MW of power) in a small/medium-scale community (650 buildings). The
model shows that by applying a control strategy to the heath pumps in the build-
ings one can provide a considerable smoothing in the variations of power in the
slack bus of the simulated IEEE 13 bus system. The presented system is a complex
MATLAB simulation environment and deals with the electrical problem and not
with the ICT or Internet-based services to enable the demand-response in practice.
The importance of services in the future smart grid is emphasized by Karnouskos
who considers the services as the only way to let the various actors in the smart
grid panorama interact with each other [95]. It is more a vision/position paper than
an analysis, and the work does not propose specific services, technologies or im-
plementations to be used. The work of Strobbe et al. [195] provides an architecture
for the future smart grid that is built on JAVA-based standard protocols. The ser-
vices provided are mainly proposed to engage and raise the awareness of the user
to energy conservation and peak reduction by having information of prices. The
architecture presents interesting aspects such as the ease of extension and the en-
ergy price individuation via the Belgian power exchange trading market. However,
there is no mention of the integration of renewable sources and energy forecast in
the architecture, that are essential building blocks of the future smart grid. In ad-
dition, the work does not consider any automation process towards appliances, but
just the user interaction with an home display that provides consumption and tariff
information. Another example of an architecture based on services is given in [96].
Although the amount of services proposed for the future smart grid is considerable
(energy monitoring, prediction, management, optimization, billing, and brokerage)
and well described also in the technical details of the architecture (e.g., representa-
tional state transfer based), there is basically no description of the implementation
of such services. There are no details of data sources in order to implement such ar-
chitecture in practice. Another architecture for the future grid is proposed in [210].
The authors consider a service architecture that can easily plug-in new services. The
architecture can host smart applications that are able to interact and control devices,
and act as a contact point with the services provided by service-providers related to
energy (e.g., real-time pricing, remote device control). The simulation of the smart
grid is obtained through the OMNeT++ framework where the information about
topology, user consumption and scenario parameters are provided. The architec-
ture misses the interaction with real information and data that are the key to provide
more concrete and valuable results to the mere simulation exercise. The model pre-
sented in [33] focuses on simulating a multi-agent system where each agent is an en-
ergy consumer or producer. The simulation is based on historical demand patterns,
while in the energy generation patterns, context information for solar and wind pro-
duction are taken into account through the interaction with Web services. Although
the approach is interesting and uses real data, it lacks an important ingredient of the
smart grid that is dynamic pricing. Moreover, there is no mention of energy forecast
that is essential in the schedule operations of home appliances, and the time granu-
larity considered in the work is quite long (three hours) compared to the dynamics
of the smart grid. A residential energy management system is described in [181].
In that paper, the demand-response is well described, there are several devices and
scenarios taken into account (dishwasher, dryer, electric vehicle, refrigerator, etc.),
and a return on investment analysis is performed. Though, there are some limita-
tions. First, it is only a simulation and there is no interaction with real equipment or
real services providing consumption information; second, there is no tariff differen-
tiation and the demand-response is only available in a time window during the day;
third, distributed energy generation through renewables is not mentioned. Web ser-
vices are the key components to interact in a smart grid-enabled home as suggested
in [103] and [9]. These works, however, are mostly an exercise in Web service in-
teraction and simulation of communication (representing appliances and sensors)
rather than a real set of services or components that can be applied in a smart grid
test solution. The data related to energy consumed in the home environment is the
U.S. average, the energy extractable by solar panels and wind turbine are static and
with no variation according to environmental condition, moreover only peak and
off-peak tariffs are considered. Such approach of simulating the smart grid is quite
simple and lacks in the dynamism of the smart grid components (i.e., energy prices,
renewable energy production) which are all the crucial aspects of the smart grid.
Considering the broader aspect of the whole interactions between the many actors
of the smart grid, service-oriented architectures (SOAs) have been proposed in sev-
eral works to be the good candidates to solve interoperability issues [160, 197, 157].
In summary, we note that effort has been spent in the simulation of smart grid
aspects; while there is a gap in the investigation of how the future smart grid ser-
vices should be and how they should work. Even more problematic is that the data
and information used to simulate the smart grid (e.g., dynamic pricing, usage of
renewable energy) are usually based on limited historical data, often averaged out,
rather than based on real-time actual data.
36 2. Related Work
2.4.2 Service-oriented architectures for energy
The literature on the energy sector is broad and touches many disciplines. Here we
provide a survey of the most relevant research efforts with respect to the vision of
an open grid and its relation with service-oriented architectures.
SOAs are a way of building scalable and interoperable distributed systems. The
SOA paradigm is based on the concept of loosely coupled entities, called services.
Service providers perform operations that can be accessed by service consumers.
Provider entities publish the services together with the requirements for the access;
the service consumers perform discovery to have information on the available ser-
vices and the way to contact them. Once this information is obtained the service
consumer can directly bind to the service provider and exploit it. A successful im-
plementation of the SOA paradigm is through the use of the eXtensible Markup
Language (XML) technology known as Web services. Web services have proven es-
sential in providing interoperability in heterogeneous systems environments. The
energy sector with many different systems is an ideal candidate to implement the
SOA paradigm especially considering the future view of even more heterogeneous
interoperability required by the smart grid.
The need for integration of the various actors and the corresponding information
systems is something that is not new and that has been under consideration since
the first signals and attempts to unbundle the energy sector. Back in 1995 Dahlfords
and Pilling [59], exposed the necessity of having a dynamic and flexible informa-
tion system. They foresaw the need of companies to be able to interact with many
more players than they were traditionally working with. The suggestion that the
authors propose in the paper is to have a two-way communication interaction with
the metering apparatus along the grid. This concept is still valid since it is one of the
key aspects of the modern concept of the smart grid. Notable is the vision stated in
the paper: “The energy market changes from a Producer push-market to a Customer
pull-market”, which today we could update with what we could call a Prosumer pull-
push-market.
The same integration theme is highlighted by Becker et al. [18]. They stress the
need for flexible and shared information systems to let the utilities operate more ef-
ficiently in the new deregulated energy landscape. The integration is not something
new according to the authors, but previous attempts in the energy sector were not at
all handy. In fact, sometimes the integration were made manually, or made in a time
consuming way such as point-to-point techniques that require great efforts. The so-
lution they mention, that has also become part of standards such as IEC-61970, is to
use the Common Information Model (CIM) representation for devices and objects in
the energy domain and use a Generic Interface Definition (GID) to expose the APIs
that can be accessed. The solution proposed for the integration of loosely-coupled
applications (e.g., ERP, SCADA, CRM, EMS, Energy Trading) is based on a message
oriented middleware and a message broker that together enable the creation of a
message bus in which the applications send and receive data; this can be seen as a
predecessor of modern SOAs. An added value is given by self describing messages
for instance those encoded using XML language.
The data integration issue between the various actors and the benefit obtained
by the adoption of SOA are addressed in [101]. An implementation of a Web ser-
vice SOA for the EMS/SCADA systems based on top of IEC standards gives sev-
eral advantages combining the power system oriented aspects (IEC defined) and
the flexibility and the spread of Web services that provide easier integration with
other companies, reuse of existing infrastructure and smooth development of the
Web service environment [133, 127, 78]. Web service technology is also used as the
communication layer to enable the interaction between all the real-time agents that
at different levels are present in the components of a modular and scalable architec-
ture [140]. This design is suitable for the vision of a smart grid composed of a huge
quantity of devices communicating electricity-related data through the Internet.
SOA is seen as the glue for the new smart grid that can enable both intra-enterprise
interactions and can be even and more present in the inter-enterprise interactions
that characterize even more the smart grid domain. This is the idea and the ap-
proach presented in [177] where the inter-enterprise information exchange interac-
tions modeled in the enterprise architecture framework are linked with the inter-
enterprise data exchange (which are based on the IEC standards) by the definition
of an ontology that can map inter and intra-enterprise domains.
Although all these works are interesting and point out specific aspects of the
interaction between energy systems, they tend to miss the vision and the evolution
that might arise with a smart grid infrastructure. In fact SOA in these works tend
to focus on data integration inside a company or enable interoperability between
different companies in the energy business value chain, but no one points out SOA
characteristics for the incoming next generation grid .
Service-oriented architecture and smart grid
As we have mentioned before, the smart grid is also referred to as the Energy Inter-
net [73, 205]. In such a parallelism, to enable the interoperability between a multi-
tude of actors and applications in various works, SOAs are indicated as central to the
smart grid. Especially at the household level, an SOA approach enables easy interac-
tion between heterogeneous devices. Warmer et al. [220] stress how a service based
architecture can be beneficial in a smart home in the new paradigm of smart grid.
38 2. Related Work
They see the Internet and Web services as the key to enable the interaction between
the house with its smart devices and the supply companies and electricity Distri-
bution Systems Operators (DSOs) to exchange supply bids and demand-response
related functionalities. The authors call for an ontology for the smart grid domain
so that the different actors can seamlessly interact with a common language. The
issue related to ontology is addressed by Considine [50] who remarks the necessity
of an ontology for the smart grid, actually referred to as service-oriented grid.
Collaboration between future smart grid objects, appliances and devices in order
to achieve better energy management and efficiency is the idea of [94]. The author
envisions a collaboration between different entities such as energy resources, en-
ergy marketplaces, enterprises and energy providers through Web services, since
they enable flexible integration without the problems due to implementation de-
tails. In Karnouskos’s vision, a device of the smart grid will be SOA-ready exposing
in a standard way the services it can provide, and at the same time it will be able to
dynamically discover services of other devices through Web Service Dynamic Dis-
covery specifications. One of this devices is the smart meter which acts as service
provider for an enhanced business process in which the meter can not only provide
real-time information, but also take decisions related to energy usage and consump-
tion interacting with other services on the Internet [98].
Cox and Considine stress how collaboration is the essence of the smart grid and
only through an interaction between the many actors involved an effective imple-
mentation of the smart grid may be realized [53]. Among the requirements the au-
thors identify as fundamental (that also appear in the NIST and Grid Wise Archi-
tecture Council stack) some aspects such as transparency, composition, extensibility
and loose coupling are presented, which are also basics for SOAs. The authors also
identify the standards for information exchange to be used in the smart grid for
some aspects such as scheduling and time functions, weather information, device
discovery and market interactions. All these elements fit in a SOA framework.
2.4.3 Agent interactions in the energy market and the smart grid
Even before the advent of the smart grid ideas, researchers have started to investi-
gate agent-based platform to deal with the new energy market challenges that the
unbundling process has brought to energy utilities companies and to energy sys-
tem operators. Especially, the financial aspects of an unbundled energy market and
their relation to technology have attracted research attention, e.g., [194]. The Web
has been considered the appropriate solution to accommodate the requirements of
an unbundled market to be efficient, transparent, and equitable [128]. Agents-based
platforms have been proposed initially in the literature to model and test the ap-
proach of a multi-player market and the technology to support it. This is the fla-
vor of the work of Lai et al. [113], where the issues of a modern unbundled market
with several actors can be properly tackled with agents given their characteristics
of autonomy, social ability, reactivity and proactiveness. In [113] the authors al-
ready saw the agents as the entities that would interact on a fully electronic mar-
ket place. A more conservative approach was proposed in [70] where an agent-
based architecture had the purpose of testing the deregulated market in Croatia.
The importance of the market and trading of energy has become so important that
power analysis tools are now enriched with energy market simulation functional-
ity to provide a richer picture so that integrated simulations of both the aspects
can be performed [122]. Another platform based on agents is presented in [92].
The authors consider an evolved energy market where the contracts are negotiated
autonomously, but with the possibility of human intervention. The agents have
strategies in the contract negotiation and can also learn to adapt during the various
rounds. Other studies have focused on the proposal, test and evaluation of business
strategies to apply to this new kind of market to satisfy equilibrium [200]. Some-
times, interoperability requirements in unbundled markets are addresses, but then
the implementation follows agent specific communication languages (e.g., Knowl-
edge Query and Manipulation Language, Protocol Operational Semantics) [113, 92].
To the best of our knowledge a study of where the agents reside, how they interact,
how these architectures would scale are not addressed in significant detail. In [192]
an energy market operation system is proposed. The architecture described, al-
though based on Web services, does not completely clarify what are the services
available for the market participants to interact with. Other solutions have been
realized [10] to simulate different types of commodity markets (e.g., cotton, corn,
electricity), where general services and interoperability requirements for a SOA rep-
resentation are described. In [208] the focus is modeling both the electricity and
gas energy systems in terms of infrastructure and market characterization. The au-
thors focus on realizing an integrated model of these energy systems and markets
with agents representing the main actors involved. The work is more a modeling
effort using agents than a solution where agents can interact and trade energy and
evaluate the market dynamics.
In the smart grid context the ideas of having agents to deal with the interactions
on the market is one of the cornerstone of the approach. The idea of an agent-based
market where agents trade energy is the core of [111]. Agents are characterized by
a demand (and supply) that changes with the varying price. The goal is to have
a bidding system between all parties involved in order to achieve an equilibrium
(demand and supply) at a certain price for energy. To realize this agent mechanism
called PowerMatcher, several agents are involved such as local agents that bid en-
40 2. Related Work
ergy (demand or supply), auctioneer agents that deal with the price formation pro-
cess, agents that aggregate the local demand and supply and bring it to an higher
level of interaction (e.g., cluster). Agents are considered as the key element in the
new paradigm of distributed generation and prosumers [211]. The vision is of a fully
automated trading mechanism where the price varies as a function of the availability
of energy and the congestion on the network. The paper stresses that the proposed
market is different than what happens in the wholesale market since the agents do
not reveal their reserve price and their pattern in generation/consumption. Another
aspect that is emphasized is the use of micro-storage and the possibility of control-
ling the storage and energy sell/purchase on the market. The authors warn that
agents that react with their storage to a real-time price of energy could create prob-
lems in the system, thus proposing for stability and safety purposes that the price of
energy storage is known and fixed for the whole day. Agents have also been used
to realize simulation of smart grid in a smart city environment [97]. In the paper,
agents do not only represent the power stations, but more in general households,
appliances and electric vehicles are simulated. The goal in the work is not to simu-
late the auction process and the negotiations, rather it is to consider the interactions
between entities, and test the real-time response of agents and the possibility of in-
teracting with devices as if they were managed in a demand-response fashion.
Agents are a very good modeling technique to deal with energy markets in gen-
eral and they are even more adequate in the distributed panorama of the smart
grid. The very inner properties of agents such as autonomy, flexibility, reactivity
and proactivity in their behaviors make them perfect to deal with an auction system
where interactions are required and agents have to adapt to the new conditions of
the rapidly changing market pursuing the goal of the user. Also the social ability
of the agents, if properly programmed, can help in achieving solutions that help in
maintaining the balancing and safety in the network and socially acceptable energy
prices.
Chapter 3
Network Models for the Smart Grid
As introduced in Chapter 1, the power grid is one of the core elements of any power
system and it is the infrastructure that enables the transport and distribution of elec-
trical energy from the producing units to the end users. Following the vision and the
changes that the smart grid is likely to enable, we consider and focus our attention
to the distribution grid. The distribution grid is the terminal part of the power grid
infrastructure that reaches the end user; we consider that this part of the power net-
work will experience to a great extent the changes supposed by the implementation
of the smart grid.
The scientific literature exploiting complex network analysis (CNA) has only fo-
cused on the high voltage end of the power grid, though with the new vision of
the smart grid this is no longer enough. Our aim is to study the topological char-
acteristics of the distribution grid, find the affinity to well-known literature models
and propose designs for the new grid. In addition, we consider economic aspects
for the distribution grid, that is the effect of topology on the price of electricity to
understand if current different distribution networks have different parameters in-
fluencing the electricity distribution costs. This investigation of both the topology
and the relationship of the latter with economic aspects is studied considering a set
of samples belonging to the Dutch distribution grid.
3.1 The Dutch Medium and Low Voltage Grid

We focus on the medium and low voltage power grid networks of (Northern) Nether-
lands, the former has voltage 10kV ¤ VM V ¤ 20kV , whereas the latter has voltage
VLV ¤ 10kV . This choice is dictated by the fact that the Netherlands has a modern
infrastructure with one the highest availability in Europe with the electrical system
being 99.99486% of the time available with just an average of 27 minutes downtime
per year per customer. The Dutch high voltage power grid is owned and man-
aged by one player, Tennet, while the lower layers are partitioned geographically
among fourteen companies. The partition of the territory among energy distribu-
tion companies is shown in Figure 3.1. In the figure each zone has an identifier and
42 3. Network Models for the Smart Grid
a distribution provider: 1) RENDO Netwerken, 2) Cogas Infra en Beheer, 3) Liander

(former Continuon Netbeheer), 6) Stedin (former Eneco), 7) Westland Infra, 8) ONS
Netbeheer (now Stedin), 9) DELTA Netwerkbedrijf, 12) NRE Netwerk, 13) Enexis
(former Essent Netwerk), 14) InfraMosane (now Enexis).
Figure 3.1: Distribution companies over the Netherlands. (Source:

www.energieleveranciers.nl)
The grid information used in this study is provided courtesy of Enexis B.V., the
distribution operator of Northern Netherlands and other regions of the country. The
provided data includes information about the transformers in the grid together with
the distribution substations. The data set also provides information about the dis-
tribution lines used to connect substations containing the length of cable and other
interesting physical properties (e.g., resistance, capacity, voltage).
For the sake of precision, we define the notion of a (Weighted) Power Grid graph
given the information available from the network samples.
• All the substations and transformers are considered equal and are represented
as nodes of the graph.
• The cables connecting the substations are considered equal despite the differ-
ences in voltages and current carried and their physical properties, and thus
modeled as unweighted edges in the graph.
• For the data samples that present disconnected components, each component
is treated as a distinct graph.
3.1. The Dutch Medium and Low Voltage Grid 43
P RESENT STUDY R ANDOM G RAPH

ID Order Size Avg. APL CPL γ APL CPL γ
d
1 17 18 2.118 3.398 3.313 0.00000 1.427 1.688 0.13726
2 15 16 2.133 3.086 3.000 0.00000 2.319 2.358 0.00000
3 24 23 2.087 4.499 4.228 0.00000 3.127 3.091 0.05508
4 30 29 1.933 4.545 4.449 0.00000 1.860 2.242 0.05778
5 188 191 2.032 17.726 17.878 0.00000 3.846 4.345 0.00532
6 10 9 1.800 2.423 2.223 0.00000 0.978 1.167 0.26667
7 63 62 1.968 5.204 5.404 0.00000 2.514 2.904 0.03175
8 28 27 1.929 4.784 5.000 0.00000 2.553 2.945 0.04762
9 133 140 2.105 11.543 11.366 0.01112 3.702 4.172 0.01482
10 124 138 2.226 8.053 7.070 0.00869 3.010 3.540 0.02914
11 31 30 1.935 4.353 4.357 0.00000 1.590 1.969 0.07475
Table 3.1: Low voltage samples from the Northern Netherlands power grid com-
pared with Random graphs of the same size.
• The edges are considered undirected.
These assumptions are common in power grid analysis from a graph theoretic per-
spective, see for instance [56, 184, 222, 221, 185, 57, 89] and lead to the following
definition.
3.1. D EFINITION (P OWER G RID GRAPH ). A Power Grid graph is a graph GpV, E q such
that each element vi P V is either a substation, transformer, or consuming unit of a physical
power grid. There is an edge ei,j pvi , vj q P E between two nodes if there is physical cable
connecting directly the elements represented by vi and vj .
The next step is to bring cable properties into the graph definition.
• For each cable connecting elements in the grid a weight is defined based on
the multiplication of the following quantities:
– The principal resistance characterizing the cable (whose value is given in

Ohm/km).
– The length of the cable (whose value is given in km).
• A special kind of connection is defined in the power grid known as a ‘link’.

These are connections, usually very short, with negligible resistance for which
the specific value is not provided in the dataset. For edges representing these
links a conventional weight of 109 is given. This does not affect the over-
all validity of the weighted model since the number of links in a sample is
extremely limited (about 1% of the overall connections are made of links).
Two examples of graphs belonging to the low voltage and medium voltage network
are shown in Figure 3.2.
3.2. D EFINITION (W EIGHTED P OWER G RID GRAPH ). A Weighted Power Grid graph
is a Power Grid graph Gw pV, E q with an additional function f : E Ñ R associating a real
number to an edge representing a physical property (e.g., the resistance, expressed in Ohm),
of the physical cable represented by the edge.
The analysis we perform uses samples from the low voltage and medium voltage
grids. The low voltage samples sum up to a total of 663 nodes and a 683 edges;
while the medium voltage samples sum up to 4185 nodes and a 4574 edges. The
size of the data set, tough being a sample and not the whole network, is about the
same size or larger than those used in other available studies on the (high voltage)
power grid [222, 196, 57, 89, 183, 184, 185]. We begin our analysis by considering the
unweighted model to derive basic topological properties and then proceed with a
richer investigation by introducing graph weights.
P RESENT STUDY R ANDOM G RAPH

ID Order Size Avg. APL CPL γ APL CPL γ
d
1 444 486 2.189 11.033 10.858 0.00537 5.547 6.163 0.00333
2 472 506 2.144 17.095 17.174 0.01360 5.039 5.700 0.00106
3 238 245 2.059 11.715 11.580 0.00000 3.558 4.234 0.00595
4 263 288 2.190 12.775 12.311 0.01118 5.046 5.368 0.01080
5 217 229 2.111 10.321 10.241 0.00140 4.894 5.391 0.00121
6 191 207 2.168 9.288 8.990 0.00296 4.616 5.079 0.00225
7 884 1059 2.396 9.817 9.527 0.00494 5.440 6.010 0.00170
8 366 382 2.087 15.113 14.546 0.00000 4.691 5.249 0.00405
9 218 232 2.128 10.850 10.915 0.00000 5.454 5.856 0.00539
10 201 204 2.030 15.742 15.257 0.00166 4.898 5.503 0.00491
11 202 213 2.109 13.504 12.891 0.00140 4.801 5.217 0.08750
12 464 499 2.151 13.144 12.703 0.00036 4.718 5.390 0.00209
Table 3.2: Medium voltage samples from the Northern Netherlands power grid com-
pared with Random graphs of the same size.
3.2. Unweighted Power Grid Study 45
(a) Physical sample low voltage #5.
(b) Physical sample medium voltage #8.

Figure 3.2: Distribution grid samples.
3.2 Unweighted Power Grid Study

The typical study of the power grid as a complex system considers high voltage
samples for characterizing the network structural properties and identifying how
fragile the infrastructure is. We use similar techniques for the Medium and Low
voltage. Let us begin by recalling the basic complex network quantities.
Basic properties and small-world analysis
We consider the traditional measures and metrics used in the literature of CNA, de-
scribed in Appendix A and apply them to the data of the Dutch distribution grid.
We divide our data set in samples of topologically connected regions. In Table 3.1,
we report the analysis on the network data modeled as unweighted graphs and we
compare each sample belonging to the low voltage network with a random graph of
the same size and order. The analysis for the medium voltage is reported in Table 3.2.
Referring to the table, the first column is the ID of the sample, the second and third
represent the number of vertexes N (order) and edges M (size), respectively. The
average degree (fourth column) is defined as xk y 2M N . The fifth and sixth columns
report the average path length (APL) and characteristic path length (CPL), that is
the average of the minimum distance between any two given nodes and the median
of the average minimum distance between all node pairs, respectively. For more
detailed definitions we refer to Definitions A.12 and A.13 on page 161. The sev-
enth column provides an indication of the clustering coefficient of the graph, that
is, broadly speaking, an average value of the ability of nodes to participate in con-
nected aggregation with other nodes close to it. For a more detailed definition we
refer to Definition A.8 on page 160. In general for more formal definitions related to
graph theory concepts we refer to Appendix A.
We remark that the average node degree has similar values between for the sam-
ples in the low and medium voltage, the values are both around 2. Computing the
mean over all samples’ average node degree gives a value of xk y 2.074 with a
very small variance σxky 0.017. This value appears to be almost constant consid-
ering the low voltage and medium voltage samples since the variance of the two
categories is even smaller (σxkyLV 0.016, σxkyM V 0.012). An almost constant av-
erage degree is also characteristic of the high voltage power grid [184], though with
a slightly higher value xk y 2.8. This limited number of edges a node can manage
can be regarded as a physical limit that each power grid substation has to satisfy.
Considering path measures: average path length and characteristic path length
of the low voltage segment of the network have generally a smaller path length
compared to the medium voltage one. The clustering coefficient is very small es-
pecially for the low voltage network for which many samples have a zero value
(i.e., absence of triangles in the graph). The difference in path length between the
low voltage network and medium voltage network is due to the higher number of
nodes the medium voltage network samples have while holding the same average
node degree as the low voltage, together with the absence of long distance edges.
This implies a longer path to connect any two nodes in a bigger network. In addi-
tion, these values of APL and CPL are in general quite high, if compared to other
networks such as the World Wide Web.
The clustering coefficients for the low voltage segment of the network are gen-
erally small; this is due to the strong hierarchical design of this layer of the physical
network which resemble a tree-like radial structure. Contrarily, the medium volt-
age segment generally presents higher values for the clustering coefficient. This
finding can be justified by the different purpose the medium voltage network has in
which meshed components and connection redundancies are much more likely to
be present for robustness reasons.
To gain a better understanding of the tables just presented, it is useful to compare
the numbers obtained with those of Random Graphs [67] and to identify the possible
presence of small-world properties. Small-world networks (SW), proposed by Watts
and Strogatz in [222], own two important aspects at the same time: characteristic
path length close in value to the one of a random graph (RG) (CP LSW CP LRG )
and a much higher clustering coefficient (CCSW " CCRG ). Small-worlds are a bet-
ter model than random graphs for social networks and other phenomena, and thus
a candidate for modeling the power grid too. To make the comparison fair, random
graphs are generated with the same number of nodes and edges as the real samples,
imposing the resulting graphs not to have disconnected components. The values
are presented on columns eight to ten of Tables 3.1 and 3.2. We note how the CPL of
the grid samples is on average twice as big as the random generated samples, thus
comparable to the definition of small-world graph according to [222]. In addition
the clustering coefficient of the grid samples is almost always smaller than the result
obtained for the random generated samples; this completely contradicts the defini-
tion of small-world graph according to [222]. Watts and Strogatz [222] impose the
following condition to the graphs they study: N " k " lnpN q " 1 where N is the
number of nodes, k is the number of edges per node. Such a condition is not satis-
fied by the Northern Netherlands samples and generally it is not satisfied by power
grid networks as pointed by Wang et al.in [217]. Interestingly, the same condition is
also not satisfied by the Western States high voltage power grid Watts and Strogatz
use in [222] and Watts analyzes in [221], while the results for CC and CPL satisfy
the conditions for a small-world network. Another study (i.e., [184]) considering the
European high voltage power grid shows that the small-world phenomenon is not
shown by all the considered grids, since especially the smaller (in terms of order and
size) grids fail to satisfy CC condition.
In summary, the Northern Netherlands medium voltage and low voltage sam-
ples show a very small value of average node degree. This is mainly independent
from the order and the different purpose of the network, being almost constant de-
spite the different samples considered. In addition, the path length is quite high,
given the order of the graphs, compared with other types of complex networks e.g.,
the World Wide Web. This relative high path length together with very small clus-
tering properties suggests that the networks analyzed do not strictly follow the def-
inition of small-world or, in terms of decentralized energy negotiation, it suggests
that perhaps a structural change to decrease path length (especially the weighted
one) might be necessary to empower delocalization.
Node degree distribution analysis
To have a general understanding of the overall characteristics of a network it is use-

ful to compute certain statistical measures, one of which is the node degree prob-
ability distribution. The shape of the distribution is a salient characteristic of the
network. For the power grid, the shape is typically either exponential or a power-
law. More precisely an exponential node degree (k) distribution has a fast decay in
the probability of having nodes with relative high node degree and follows a rela-
tion:
P pk q αeβk (3.1)
where α and β are parameters of the specific network considered. While a power-
law distribution has a slower decay with higher probability of having nodes with
high node degree:
P pk q αk γ (3.2)
where α and γ are parameters of the specific network considered.

Power-law distributions are very common in many real life networks both cre-
ated by natural processes (e.g., food-webs, protein interactions) and by artificial
ones (e.g., airline travel routes, Internet routing, telephone call graphs), [13]. Having
a power-law distribution in the node degree statistical distribution means that few
nodes have a very high degree and the majority of nodes have very small degree.
The types of networks that follow this property are referred as Scale-free networks
([16, 14, 5]); typical examples of Scale-free networks are the World Wide Web, the
Internet, metabolic networks, airline routes and many others. From the dynamic
point of view, the model that enables to create such networks is a preferential at-
tachment model, that is, when new nodes and arcs are added to a graph, these are
more likely to connect to nodes that already have a high degree, [15, 3]. In addi-
tion, this network structure provides special reliability properties: high degree of
tolerance to random failures and, at the same time, very high sensitivity to targeted
attacks towards hubs [5, 138, 56].
We compute the node degree distribution for every sample both for low volt-
age and medium voltage segments. For the most significant samples i.e., those be-
longing to the medium voltage and the big ones belonging to the low voltage part,
the node degree cumulative distribution seems to follow a power-law: Pk k γ .
A coarse, but straightforward, method to investigate if the node degree follows a
power-law is to plot the cumulative node degree distribution on a log-log scale [6].
If the distribution in a log-log plot follows a straight line the distribution can be con-
sidered a power-law, while if the decay is faster this might indicate an exponential
distribution. It is also possible to apply data fitting techniques (e.g., non-linear least
square method) to identify the γ parameter of a power-law as suggested by Clauset
et al. [47].
The most significant samples for this kind of analysis are the biggest samples be-
longing to the medium voltage network and the most numerous ones from the low
voltage (i.e., samples #5, #9 and #10 from Table 3.1). All these samples tend to follow
a straight line in the log-log plot as in Figure 3.3. In the figure circles represent sam-
ple data, while the straight line represents a power-law with γ 1.977. Power grids
are not dense network and also many of the samples under investigation have small
order, thus providing few data points. Therefore, for many samples it is difficult to
perfectly fit to a distribution for the node degree following the usual node degree
definition.
We thus conclude that only some samples of the medium voltage and low volt-
age tend to be Scale-free networks, although some exponential tail appear due to
physical and economic constraints in the network. This means robustness in terms
of redundancy of paths, but fragility to attacks on the hubs. The hubs tend to be the
few nodes that most likely lead to the high voltage segment in a certain geographical
location.
3.2.1 Betweenness analysis

Betweenness describes the importance of a node with respect to minimal paths in
the graph. This property is important to identify critical components of the power
grid [5, 138, 57]. For a given node, betweenness, sometimes also referred as load, is
defined as the number of shortest paths that traverse that node. For a more technical
definition refer to Definition A.15 on page 162. The betweenness measure allows
to find if there are nodes that are critical for the whole infrastructure. In fact, the
removal of nodes with the highest betweenness can lead to critical effects on the
network connectivity [138].
Figure 3.3: Node degree cumulative probability distribution for medium voltage
sample #7.
Figure 3.4: Betweenness cumulative probability distribution for low voltage sample
#5.
For the most significant samples in the low voltage network (i.e., samples #5 and
#10) betweenness probability distribution follows an exponential decay, that is, the
nodes with very high values of betweenness are less likely to be present in the net-
work, as shown in Figure 3.4. In the figure circles represent sample data, while the
4
continuous line represents an exponential decay y 0.7699e2.22710 x . This aspect
is not surprising since the low voltage network is quite hierarchical and the paths
tend to follow the few ones admissible by the relative simple topology. In fact, the
betweenness probability distribution charts (Figure 3.4) do not fit a straight line in
Figure 3.5: Betweenness cumulative probability distribution for medium voltage

sample #1.
a logarithmic plot, thus exhibiting a fast decay. In addition a fitting procedure, us-
ing the non-linear least square method gives very good results approximating the
betweenness probability distribution samples with an exponential function. On the
other hand, betweenness in medium voltage segment seems to follow a power-law
decay, this is shown in the logarithmic chart in Figure 3.5. In the figure circles repre-
sent sample data, while the straight line represents a power-law with γ 1.075. The
samples from medium voltage network show a distribution of betweenness with a
much fatter tail than the low voltage ones, that is there are several nodes that are
central in many paths. This is due to the more meshed structure the medium volt-
age network has, compared to the low voltage one. This result for medium voltage
network betweenness is closer to the results obtained for this same metric in high
voltage studies, [5, 57]. In summary, a few nodes are extremely critical to enable the
electricity distribution for the whole network.
3.2.2 Fault tolerance

A way to study the reliability of the network is by analyzing its connectivity when
nodes are removed. There are basically two ways to perform this analysis: choosing
the nodes randomly or selecting the nodes to be removed following a specific signif-
icant property for the network. Similar studies concerning the resilience of the high
voltage power grid exist, e.g., [184, 5]. We apply such technique to the medium volt-
age and low voltage ends of the power grid using three policies for node removal:
random, highest degree and highest betweenness driven choices. The measure that
Figure 3.6: Resilience for node random-based removal for medium voltage sample
#7.
is taken into account is the order of the largest connected component of the network
(i.e., the number of nodes composing the biggest cluster in the network) computed
as a fraction of the original order of the network, and its evolution while nodes of
the network are removed, again the latter are considered as a fraction of the original
order of the network.
The random removal simulates casual errors. As shown in [48], networks that
follow a power-law whose characteristic parameter γ 3 tend to have a high value
for the transition threshold at which they disrupt. In the samples here analyzed, it
seems that this is true especially for the small samples that generally have a cluster
that is 10% of the original when almost 90% of the nodes are removed. The situation
is different for samples with higher order that show a cluster that is reduced to 10%
of the original when about 40% of the nodes are removed, as shown in Figure 3.6.
Even if the degree distributions found for samples following a power-law have a
parameter γ 3 the samples show a threshold effect that is more similar, according
to [48], to networks whose characteristic γ ¡ 3.
The situation is radically different when “targeted attacks” are considered. In
particular two kind of attack policies are investigated: node degree-based removal and
betweenness-based removal. The main difference compared to the random-based re-
moval is the presence of very sharp falls that appear when certain nodes are tar-
geted. The removal of selected nodes can cause a drop in the size of the maximal
connected component even of 40%. Node degree-based removal is much more crit-
ical than the random removal: by just removing 10% of the most connected nodes
Figure 3.7: Resilience for node degree-based removal for medium voltage sample #7.
one reduces the network to only 10% of its original size. The same applies for the
biggest samples considered both in the low voltage and medium voltage network,
as shown in Figure 3.7.
The removal of nodes based on the highest betweenness shows generally the
same behavior, as the degree-based removal, with network disruption that appears
much faster than random-based network failures. Considering the general correla-
tion between nodes with a certain degree and their betweenness, it is not surprising
that the two removal policies have very similar results and shape. The only re-
mark that generally differentiates the betweenness-based removal is a little higher
order of the maximal connected component compared to the one obtained with a
degree-based removal when the same fraction of nodes is removed. In addition,
the decrease of the order of the maximal connected component tends to be slightly
smoother than the degree-based one. Figure 3.8 shows the comparison of the two
removal policies for the samples that show some interesting deviations in the corre-
lation of the degree and betweenness.
In summary, the results for the low voltage and medium voltage show disrup-
tion behaviors that are quite immune to random failures to which the networks
present a constant degrading disruption, while they strongly suffer from the re-
moval of specific nodes.
Figure 3.8: Resilience for node degree-based and betweenness-based removal for
medium voltage sample #10.
3.3 Weighted Power Grid Study

The purely topological study of the power grid just presented already gives im-
portant information about the connectivity and robustness of the medium and low
voltage grids, though it does not consider the different physical properties of the
cables. These can vary greatly for different sections of the grid and provide essen-
tial indications to establish the behavior of a link. Next, we perform an analysis of
the same samples of the grid considering the resistance as the weight of the graph
model.
We take the samples analyzed in Tables 3.1 and 3.2, but now consider the weighted
graph definition. The notion of a characteristic path length can be extended to
take the weights into account yielding the values shown in Tables 3.3 and 3.4. In
each table, the second column contains the characteristic path length resulting in
the weighted graph (WCPL). The third column provides the average value of the
weights of all edges; while the fourth column shows a normalized value for the
weighted characteristic path length (NWCPL) obtained by dividing the WCPL by
the average weight of the edge belonging to the same data sample.
Due to the relative short length of the low voltage networks cables, the WCPLs
for this segment of the network are small, as well as the average weight of each edge
(almost all of them are below the unit). The situation is different for the medium
voltage networks that are higher since the cables and paths span across wider ge-
ographical areas. The discrepancy can be explained by the different purpose for
3.3. Weighted Power Grid Study 55
ID Weighted Char- Edge Average Normalized

acteristic Path Weight Weighted Char-
Length acteristic Path
Length
1 2.000 0.698 2.865
2 1.429 0.595 2.402
3 3.066 0.739 4.149
4 3.087 0.699 4.414
5 12.136 0.741 16.378
6 3.889 1.648 2.360
7 4.162 0.348 11.960
8 5.112 0.876 5.836
9 7.872 0.583 13.503
10 6.407 0.785 8.162
11 2.967 0.592 5.012
Table 3.3: Weighted analysis of the low voltage samples.
ID Weighted Char- Edge Average Normalized

acteristic Path Weight Weighted Char-
Length acteristic Path
Length
1 153.402 8.608 17.821
2 163.067 9.217 17.692
3 127.258 7.122 17.868
4 134.661 13.106 10.275
5 187.084 16.382 11.420
6 185.916 12.779 14.549
7 108.011 11.851 9.987
8 148.058 7.193 20.584
9 99.385 7.421 13.392
10 126.845 6.850 18.518
11 92.060 8.764 10.504
12 38.084 6.915 5.507
13 232.475 13.810 16.834
Table 3.4: Weighted analysis of the medium voltage samples.
which these networks are designed: a bridge network from high voltage transmis-
sion lines and end user distribution (medium voltage network) and the final end de-
livery (low voltage network). In fact, both the WCPL and the edge average weight
for medium voltage samples are approximately two order of magnitude greater than
the low voltage ones. This is indeed due to an extension of medium voltage cables
that range from hundred meters to kilometers, while low voltage extend usually
around tens of meters.
Weighted node degree distribution
In the weighted power grid graph that we consider, there are no weights explicitly
associated to the nodes. However, the weights of the incident edges influence the
node properties. To capture this characteristic one can consider a weighted node
degree for a node whose weight is obtained by the sum of the weights of the incident
edges. For a more formal definition we refer to Definition A.7 on page 160. It is then
easily to extend the concept of node degree distribution to a weighted node degree
distribution. This distribution is straightforwardly obtained by using the weighted
degree concept instead of the traditional node degree in the node degree distribution
(cf. Definition A.16 on page 162).
For the most significant sample of the low voltage, as shown in Figure 3.9, the
shape of the distribution is close to an exponential one with a quite fast decay. In
the figure circles represent sample data, while continuous line represents a sum of
exponential decays y 0.8975e0.9289x 0.0904e0.1379x . The situation looks differ-
ent in medium voltage samples. The very first part of the distribution is well fitted
by an exponential shape, while the central part of the distribution, and especially
the tail, fit best a power-law like shape as visible in Figure 3.10. In the figure circles
represent sample data, while straight line represents a power-law with γ 1.374.
An explanation of such behavior among the the most numerous samples of the two
ends of the grid is due to the order and size of the medium voltage samples that are
from two to four times bigger than the low voltage samples, thus having a higher
likelihood of far different values in weighted node degree.
3.3.1 Betweenness analysis

Considering the weighted definition of path it is possible to compute betweenness
in the weighted scenario and again betweenness of a node can be seen as a random
variable thus obtaining the corresponding probability distribution. The shape of the
distribution does not change much compared to the same unweighted samples for
the low voltage network: the distribution is best approximated by an exponential
decay or by a sum of exponential contributions. For medium voltage samples, the
changes between unweighted and weighted paths influence the betweenness proba-
bility distribution whose shape in these conditions seems to be better approximated
by an exponential or sum of exponential components. This change in the distribu-
tion of the number of shortest paths that traverse a node between the weighted and
the unweighted graph is clearly an indication that some property change between
3.3. Weighted Power Grid Study 57
Figure 3.9: Weighted node degree cumulative probability distribution for low voltage
sample #5.
Figure 3.10: Weighted node degree cumulative probability distribution for medium
voltage sample #1.
the two analysis. In fact, the weighted path analysis better approximates the actual
routes the current flows follow.
3.3.2 Fault tolerance

Fault tolerance can be evaluated based on the removal of nodes following strategies
similar to the unweighted case. Since the random removal yields exactly the same
Figure 3.11: Resilience for node degree-based removal for medium voltage sample
#1.
result for the weighted and unweighted case, here we focus on the node degree-
based removal that which considers the weighted node degree definition. The dis-
ruption behavior of the network samples is very similar to the unweighted situ-
ation: the network suffers deeply these targeted attacks; a very small percentage
of removed nodes causes an important loss in the size of the biggest component
left in the network. There is a general correlation between high degree nodes in
the unweighted graph and high degree nodes in the weighted one. If one takes a
closer look at the disruption charts for the same samples, some small differences
can anyway be noted as in Figure 3.11. The horizontal axis represents the fraction
f of the nodes removed from the original sample; the vertical axis represents the
size of the largest connected component S relative to the initial size of the graph.
Red diamonds represent the weighted node degree-based removal, while blue cir-
cles represent traditional node degree-based removal. The nodes with the highest
weighted degree cause a bigger damage to the network when removed in the very
first iteration than nodes with higher degree in unweighted networks. The situation
then changes in the later stages of the removal process when a bigger disruption is
caused by nodes with higher node degree in traditional sense.
3.3.3 Unweighted vs. weighted study

In Chapter 2 and more thoroughly in [166, 87], the superiority of an approach that
considers the physical properties underlying the network is emphasized. In this
3.4. Relating Grid Topology to Electricity Distribution Costs 59
section and in Section 3.2 we have performed the same type of analysis considering
both the unweighted and weighted definition of graphs, respectively. Our motiva-
tion was not to prove the superiority of one approach over the other, but to consider
if important differences in the metrics analyzed were present.
The similarities lie in the fashion the network disrupts when the most connected
nodes (either in the traditional, or in the weighted definition) are attacked: the re-
moval of just few nodes compromise the connectivity of the whole network (cf. Fig-
ure 3.11). Some dissimilarities are evident considering the node degree distribution.
In fact, the weighted analysis tends to reduce the contribution of the tail components
of the distribution, thus being more compact especially for low voltage samples.
This is due to the small variance of the weighted node degree that especially these
low-end samples of the grid show. Samples that in the unweighted analysis show a
power-law distribution, when considered weighted tend to assume an exponential
form or a sum of exponential contributions. The most interesting dissimilarity in the
analysis is in the path-related properties. The number of nodes traversed in the two
cases is different for the medium voltage networks. This is particularly interesting
from the practical point of view, as it indicates the number of transformers and dis-
tribution substations traversed in the power grid. These points are critical in terms
of additional losses that are associated with substations and transformers, and in
turn in the number of potential points of failure that a path traverses. Figure 3.12
shows the results for the low voltage and medium voltage networks. Each bar rep-
resents the average percentage increase in the number of nodes traversed along the
shortest path between any two nodes for the unweighted and the weighted situa-
tion. It is interesting that for several samples of the low voltage network there is no
difference in the number of traversed nodes, thus reinforcing the idea of a highly hi-
erarchical tree-like radial structure whose paths are fixed by the built-in topology of
the grid independently of the associated edge weights. The situation though is quite
different for the medium voltage. In fact, there is an increment of traversed nodes
between the weighted and unweighted models (especially for the meaningful sam-
ples) on average of about 50%. This is a clear indication of a meshed network for
which there are less imposed paths and in which weights have an important role.
3.4 Relating Grid Topology to Electricity Distribution

Costs
The complex network analysis that we provided so far gives a statistical aggregated
view of the current infrastructure for the low and medium voltage grids. The nat-
ural next question that arises concerns the usability of such infrastructure for the
(a) Low voltage samples (b) Medium voltage samples

Figure 3.12: Percentage increase in number of node traversed between weighted and
unweighted graph definition.
delocalized energy exchange. To answer it, we propose to tie statistical properties

of the power grid with energy trading costs. The cost represents a balance frontier
below which the actors are motivated to trade energy. We do not claim of having
identified “the” cost function, but rather we propose that Complex Network Anal-
ysis measures can be used to evaluate the success of a decentralized energy market.
In general, establishing energy pricing is not a simple task since several aspects
influence the overall price at which electricity is sold [84]. There are aspects con-
nected to the supply side such as fuel prices, policy regulations, load losses and
bidding strategies; on the other hand, there are elements connected to the demand
side such as human behaviors, natural phenomena that influence habits and thus
consumption. Recent proposals and methods for price allocation include nodal pric-
ing [119], which is particularly indicated for distributed generation solutions be-
cause of the price benefits it brings to the small producer [193]. It is also important
to notice that the savings deriving from distribution losses can be extremely impor-
tant [43, 104, 115]. A set of factors is most definitely tied to infrastructural proper-
ties of the distribution network, as illustrated for instance in the economic studies
of Harris and Munasinghe [84, 141], most notably:
• losses both in line and at transformer stations,
• security and capacity factors,
• line redundancy, and
• power transfer limits.

The listed technical parameters are naturally associated with a topological parame-
ter, namely:
• Line losses are related to and thus expressed as a function of the weighted
characteristic path length LlineN f pW CP LN q
• Substation losses are expressed as a function of the (average) number of nodes

visited while traveling from source node to destination node along the weighted
shortest path Lsubstationij f p|W SPij |q. The significant dimension is
LsubstationN f p|W SPN |q where |W SPN | is the average number of nodes tra-
versed in a shortest path in the network N .
• Robustness from a topological perspective is expressed as the fraction of the

maximal connected component compared to its original order once a certain
fraction of nodes is removed. RobN f pN, pq where N is the network under
evaluation and p is the removal policy adopted.
• Line redundancy is simply mapped to a topological metric that counts the

number of paths (without cycles) that are available between any two nodes
and the cost associated to this redundancy Redij f p|Pij |, wij q where Pij is
the set of paths between nodes i and j and wij is the weight of the worst case
redundant path. A global metric for network N is RedN f p|Pij |, wij q, @i, j P
N
• Network capacity may be considered as a function of the weighted charac-

teristic path length where the weight in this case is the maximal supported
operating current of the cable CapN f pW CP LN q
These topological ingredients provide two sorts of measures, the first one α gives
an average of the dissipation in the transmission between two nodes
α f pLlineN , LsubstationN q; (3.3)
the second one β is a measure of reliability/redundancy in the paths among any two
nodes
β f pRobN , RedN , CapN q. (3.4)
We argue that these two factors influence the inclination of prosumers (energy con-
sumers/producers) to trade energy on the power grid. In fact a high value of the α
parameter represents a high level of losses experienced for transporting energy in
the network, either in distribution lines or substations. Additionally, the reliability
and ability to bring sufficient energy to the end users plays an important role. In
Figure 3.13: Transport cost of energy based on the low voltage topological properties.
Figure 3.14: Transport cost of energy based on the medium voltage topological prop-
erties.
fact, if proper levels of robustness of the network or resilience to failures are not the
norm, the prosumer inclination to sell energy as well as the end user to buy it will be
limited. Furthermore, if the availability of redundant paths for electricity routing in

case of partial disruption of the network is insufficient, then it leads to a high value
of β, and in turn a disincentive for trading. To better understand the constituents
of α and β, we consider next a possible instantiation of these parameters using the
data of the Dutch power grid.
3.4.1 Application to the Dutch distribution grid

To give an impression of how the parameters can be used to assess the success of
the energy market, we provide an example next.
• Losses on the transmission/distribution line can be expressed by the quotient

of the weighted characteristic path length and the average weight of a line (a
weighted edge in the graph):
LlineN W CP
w
LN
(3.5)
• Losses at substation level are expressed as the number of nodes (on average)
that are traversed when computing the weighted shortest path between all the
nodes in the network:
LsubstationN N odesW CP L N
(3.6)
• Robustness is evaluated with random removal strategy and the weighted node
degree-based removal by computing the average of the order of maximal con-
nected component between the two situations when the 20% of the nodes of
the original graph are removed. It can be written as:
RobN |M CCRandom20% | 2|M CCN odeDegree20% | (3.7)
• Redundancy is evaluated by covering a random sample of the nodes in the

network (40% of the nodes whose half represents source nodes and the other
half represents destination nodes) and computing for each source and destina-
tion pair the first ten shortest paths of increasing length. If there are less than
ten paths available, the worst case path between the two nodes is considered.
To have a measure of how these resilient paths have an increment in trans-
portation cost, a normalization with the weighted characteristic path length is
performed. We formalize it as:
°
RedN P P
i Sources,j Sinks SPwij
(3.8)
W CP L
• Network capacity is considered as the value of the weighted characteristic

path length, whose weights are the maximal operating current supported, nor-
malized by the average weight of the edges in the network (average current
supported by a line). That is:
CapN W CP
w
LcurrentN
(3.9)
current
With these instantiations, equations (3.3) and (3.4) become:
α f pLlineN , LsubstationN q LlineN LsubstationN (3.10)
β f pRedN , RobN , CapN q

RedN
RobN lnpCapN q
(3.11)
The functions to compute α and β are only few of many available possible ones.
The choice made here is to have a simple mechanism to assess the potential distribu-
tion costs of different networks. Equation (3.10) is a basic sum, since both quantities
are pure numbers, over the losses that are experienced both at line and at substation
level. Equation (3.11) takes into account the aspects of reliability and tolerance of the
network: the higher the β the more prone to failures and less reliable the network
is. The cost of increasing the number of paths to provide more redundancy is the
dividend in the fraction, while elements improving reliability act as divisor.
With these quantities, one can form an impression of what the influence of the
cost of transportation is for the decentralized energy exchange. If the cost is too high
because of an infrastructure with high chances of failure (high β) and high resistance
(high α), then the decentralized market will not be incentivized. On the other hand,
for low α and β, it will be economically attractive to have a decentralized energy
market. In Figures 3.13 and 3.14, we show a combination of α, β obtained with
the functions described above and with an hypothesis of a quadratic increase of
energy price with the increment in α and β. We report the position of the analyzed
samples as white circles in Figures 3.13 and 3.14 respectively for low voltage and
medium voltage samples. By performing an economic study, which we stress is
beyond the scope of the present treatment, one then can identify what the threshold
is for the feasibility of a decentralized market (gray rectangles in Figure 3.15) and
then conclude what topological modifications are necessary to the medium voltage
and low voltage infrastructure in order to allow the energy exchange.
Figure 3.15: Transport cost of energy based on the topological properties for medium
voltage with supposed economic convenience threshold (grey thick line).
Chapter 4
Network Evolutions for the Smart Grid
We have seen that complex network analysis (CNA) can provide a useful insight in
understanding the topology of the distribution grid (cf. Chapter 3). In particular, we
have used complex network analysis techniques not only to analyze the reliability
of the network, as the majority of the literature studies concerning the high voltage
grid, but we have also considered how the topology can play a role in influencing
the price of electricity distribution.
Here we want to go one step further and consider possible scenarios of evolution
of the distribution grid in terms of topology, since we believe that the future smart
grid will have to evolve and a change in the distribution infrastructure. Inevitably,
more generation and more electricity consumption at local-scale will have repercus-
sions on the distribution grid infrastructure as well. More and more prosumers will
feed their locally produced energy into the grid and participate in a local market for
energy trade. The distribution grid has then to accommodate a new paradigm that
is no more the top-down approach where the power is generated in big facilities
remotely located, but it is dominated by a bottom-up approach where the produc-
ers are numerous and energy is produced and consumed at the neighborhood level.
The distribution grid has therefore to be efficient in enabling this paradigm without
posing barriers with high cost in energy transportation. In addition, the massive
presence of electric vehicles will also be an important ingredient in shaping the next
generation of distribution grids. In order to look at the evolutions that may char-
acterize the future distribution grid, we consider here two situations to understand
how the new grids could look like. First, we consider the case of the realization
of a completely new infrastructures such as for new settlements whose topology
blueprint can be decided by an infrastructure planner. The blueprint to be used is
based on well-known models of complex network analysis (Section 4.2). Second,
we consider the case of current networks deployed on the ground and we look at
techniques to adapt the distribution grid topologies to better support the local en-
ergy interactions of the smart grid (Section 4.3). The former can be considered as a
solution for developing new distribution infrastructures in new locations e.g., new
urbanization expansions in developing countries where the core is the local energy
production and distribution, while the latter deals more with the changes required
68 4. Network Evolutions for the Smart Grid
in the current distribution grid to become more efficient and better support energy
exchanges that take place locally. The novelty is first in using the complex network
analysis as a design tool for infrastructure planning and evolution, and, second, the
focus that, compared to the existing studies of power grid and complex network
analysis, is devoted to the distribution grid and not on the high voltage grid.
4.1 Network Metrics

The price of electricity distribution is influenced by the physical characteristic of the
network that transports it. We have shown in Chapter 3 how different samples of
the Dutch distribution grid have differences in their topology and the influence on
the distribution price in that network. To measure those differences, we proposed a
number of metrics useful for analyzing power grid topologies having in mind de-
centralized energy trading. Our goal has been to understand their influence on the
electricity distribution price. Here we define a set of additional metrics inspired
by the analysis carried out on the Dutch samples, and we then use them to mea-
sure properties of networks for the evolution of the distribution grid. We set two
main categories of requirements: qualitative and quantitative desiderata that the
networks optimized for the local energy exchange should satisfy.
Qualitative requirements
The main qualitative requirement we envision for the future distribution network
concerns the modularity of the topology. In the power system domain, the modular-
ity is invoked as a solution that provides benefits reducing uncertainties in energy
demand forecasting and costs for energy generation plants as well as risks of tech-
nological and regulatory obsolescence [123, 88]. Modularity is usually required not
only in the energy sector, but more generally in the design and creation of prod-
ucts or organizations [77]. It is also a principle that is promoted in innovation of
complex systems [68] for the benefits it provides in terms of reduced design and
development time, adaptation and recombination. We define the modularity as the
ability of building the network using a self-similar recurrent approach and having a
repetition of a form of pattern in its structure.
Quantitative requirements
As a global statistical tool, quantitative requirements give a precise indication of net-
work properties. Here are the relevant ones when considering efficiency, resilience,
and robustness of a power system.
4.1. Network Metrics 69
• Characteristic Path Length (CPL) lower or equal to the natural logarithm of order
of graph: CP L ¤ lnpN q. This requirement represents having a general short
path when moving from one node to another. In the grid this provides for
a network with limited losses in the paths used to transfer energy from one
node to another.
• Clustering Coefficient (CC) which is 5 times higher than a corresponding Random

Graph (RG) with same order and size: CC ¥ 5 CCRG . Watts and Strogatz [222]
show that small-world networks have clustering coefficient such that CC "
CCRG . Here we require a similar condition, although less strong by putting a
constant value of 5. This requirement is proposed in order to guarantee a local
clustering among nodes. In fact, it is more likely that energy exchanges occur
at a very local-scale (e.g., neighborhood) when small-scale distributed energy
resources are heavily implemented.
• Betweenness-related requirements:
– A low value for average betweenness in terms of order of the graph υ N σ

,
where σ is the average betweenness of the graph and N is the order of the
graph. For the Internet, Vázquez et al. [209] have found for this metric
υ 2.5. Internet has proved successful to tolerate failures and attacks
[48, 5], therefore we require a similar value for this metric for the future
power grid.
– A coefficient of variation for betweenness cv xs 1 where s is the sample
standard deviation and x is the sample mean of betweenness. Usually,
distribution with cv 1 are known as low-variance ones.
The above two requirements are generally considered to provide network re-
silience by limiting the number of critical nodes that have a high number
of minimal paths traversing them. These properties provide distributions of
shortest paths which are more uniform among all nodes.
• An index for robustness such that RobN ¥ 0.45. Robustness is evaluated with a
random removal strategy and a node degree-based removal strategy by com-
puting the average of the order of the maximal connected component (MCC)
of the graph between the two situations (i.e., random and targeted node re-
moval) when the 20% of the nodes of the original graph are removed. It can
be written as RobN
|M CCRandom20% | |M CCN odeDegree20% | . Such a requirement
2
is about double the value observed for current medium voltage networks and
33% more for low voltage network samples.
• A measure of the cost related to the redundancy of paths available in the network:
AP L10th ¤ 2 CP L. As described in Chapter 3, with this metric we consider
the cost of having redundant paths available between nodes. In particular,
we evaluate the 10th shortest path (i.e., the shortest path when the nine best
ones are not considered) by covering a random sample of the nodes in the net-
work (40% of the nodes whose half represents source nodes and the other half
represents destination nodes). The values for the paths considered are then
averaged. In the case where there are less than ten paths available, the worst
case path between the two nodes is considered. This last condition gives not
completely significant values when applied to networks with small connectiv-
ity (i.e., absence of redundant paths).
Metric Efficiency Resilience Robustness

CPL X
CC X
Avg. Betweenness X
Betw. Coeff. of Variation X
RobN X
AP L10th X X
Table 4.1: Metrics classification related to properties delivered to the network.
We categorize the above quantitative metrics into three macro categories with re-
spect to how they affect the power grid and measure its goodness from a topo-
logical point of view: efficiency deals with the losses in the transfer of energy, re-
silience measures the possibility of having alternative paths if part of the network is
compromised/congested, and robustness concerns failures happening to the over-
all network connectivity. Table 4.1 summarizes the property each metric assesses.
Each metric gives a specific contribution and all the metrics together cover all the
properties a smart grid infrastructure should have.
4.2 Building New Distribution Networks

To address the question of what are the best suited topologies to characterize the
medium and low voltage grids, we study models for graph generation proposed
for social and technological complex networks. For each model we evaluate the
properties of the network for several values of the order of the graph. Following our
analysis of the Northern Dutch medium and low voltage in Chapter 3, we categorize
networks as small, medium and large, see Table 4.2. We then analyze the properties of
4.2. Building New Distribution Networks 71
the networks coming from the generated models by applying the relevant complex
network analysis metrics described in Section 4.1. In this way, CNA is not only a
tool for analysis, but it becomes a design tool for the future electrical grid.
Network layer Category Order Reference Dutch sample

low voltage Small 20 #1
low voltage Medium 90 #7
low voltage Large 200 #5
medium voltage Small 250 #3
medium voltage Medium 500 #12
medium voltage Large 1000 #7
Table 4.2: Categories of medium and low voltage network and their order.
Synthetic models
We look at network models that have proven successful in showing salient char-
acteristics of technological networks (i.e, preferential attachment, Copying Model,
power-law networks), social networks (i.e., small-world, Kronecker graph, recur-
sive matrix) and natural phenomena as well (e.g., Random Graph, small-world,
Forest Fire) to investigate which one is best suited for supporting local-scale energy
exchange from a topological point of view. Next, we provide a brief introduction
to all the models used in the present study, while a more in-depth presentation is
available in Appendix B or [39] and [151].
• Random Graph. A Random Graph is built by connecting each pair of nodes

with an edge with probability p. It is due to the pioneering studies of Erdős
and Rényi [67].
• Small-world Graph. The small-world characterization of graphs has been in-

vestigated by Watts and Strogatz [221, 222], who showed the presence of the
small-world property in many types of networks such as actor acquaintances,
the power grid and neural networks in worms.
• Preferential Attachment. The preferential attachment model represents the

phenomenon happening in real networks where a fraction of nodes has a high
connectivity while the majority of nodes has small node degree. This model
is built upon the observation by Barabási and Albert [15] of a typical pattern
characterizing several type of natural and artificial networks.
• R-MAT. R-MAT (Recursive MATrix) is a model that exploits the representation

of a graph through its adjacency matrix [40]. In particular, it applies a recursive
method to create the adjacency matrix of a graph, thus obtaining a self-similar

graph structure. This model captures the community-based pattern appearing
in some real networks.
• Random Graph with power-law. A Random Graph with power-law model

generates networks characterized by a power-law in the node degree proba-
bility distribution (P pk q k γ ). A graph generated in this fashion has the
majority of nodes with a low degree and a small amount of nodes with a very
high degree. Power-law distributions are very common in many real life net-
works both created by natural processes (e.g., food-webs, protein interactions)
and by artificial ones (e.g., airline travel routes, Internet routing, telephone call
graphs) [13].
• Copying Model. Replicating the structure underlying the links of WWW

pages brought to the Copying Model [107] which captures the tendency of
members of communities with same interests to create pages on the web with
a similar structure of links.
• Forest Fire. In order to capture dynamic aspects of the evolution of networks,

Leskoveck et al. [118] proposed the Forest Fire model. The intuition is that net-
works tend to densify in connectivity and shrink in diameter (i.e., the greatest
shortest path in the network) during the growth process. Several technologi-
cal, social and information networks exhibit this property.
• Kronecker Graph. A generating model with a recursive flavor similar to R-

MAT uses the Kronecker product applied to the adjacency matrix of a graph
[117]. If the Kronecker product is applied to the same matrix, therefore mul-
tiplying the matrix with its elements recursively, a self similar structure arises
in the graph. This model creates networks that show a densification in the
connectivity with a shrinking diameter over time.
An example of graphical representation of the topological models obtained by

applying the generation algorithm of each network model just described is
shown in Figures 4.1 and 4.2.
(a) A Random Graph (199 nodes, 400 (b) A small-world graph (200 nodes, 399
edges). edges).
(c) A preferential attachment graph (200 (d) A R-MAT graph (222 nodes, 499
nodes, 397 edges). edges).
Figure 4.1: Graphical representation of the network models considered in the
study.
(a) A Random Graph with power-law (b) A Copying Model graph (200 nodes,
graph (200 nodes, 399 edges). 199 edges).
(c) A Forest Fire graph (200 nodes, 505 (d) A Kronecker graph (167 nodes, 264
edges). edges).
Figure 4.2: Graphical representation of the network models considered in the
study.
Model generation and metrics computation.
The synthetic topologies that we generate come from complex network analysis lit-
erature. These networks are obtained using the Stanford Network Analysis Project
(SNAP)1 library that implements the generation algorithms of the network topolo-
gies described above and more thoroughly in Appendix B. The analysis of the gen-
erated graphs according to the metrics described in Section 4.1 is performed with
ad hoc created software based on the JAVA graph library JGraphT.2 The versions of
SNAP and JGraphT software libraries used are respectively v10.10.01 and v0.8.1.
Analysis and metric satisfaction of current medium and low voltage networks
The baseline network for comparing possible evolutions must be the current power
grid. Therefore, we use actual samples from the medium and low voltage network
of the Northern Netherlands (cf. Chapter 3). The average degree of the medium
and low voltage samples scores almost constantly around xk y 2 independently of
the order of the network. In the low voltage networks we see a tendency towards
the increase of the characteristic path length, with a value of about 18 when the or-
der and size tend to 200 nodes and edges, respectively. The metric does not have
the same clear tendency for the medium voltage samples. Considering the cluster-
ing coefficient there is a general rule: a null value for the low voltage samples and
small, but at least significant, values for the medium voltage samples. These dif-
ferences in both characteristic path length and clustering coefficient come from the
difference in topology of the two networks. The low voltage network is almost a
non-mashed network which resembles for certain samples trees, closed chains or
radial structures with longer paths on average, especially for networks with bigger
order. On the other side, the medium voltage network is more meshed (despite the
same average node degree) with more connections that act as topological “short-
cuts”. It also has some redundancy in the connections between the neighborhood of
a node, which implies a significant clustering coefficient compared to the low volt-
age network. The analysis of the robustness metric shows generally poor scores that
decrease while the samples increase in order, at least for the low voltage networks,
while the tendency is not clear for the medium voltage samples considered. A com-
mon behavior for the medium voltage samples is the problem they experience in the
biggest component connectivity, when the 20% of the nodes with the highest degree
are removed from the network the order of the MCC falls to 4.56%, 3.66% and 3.96%
of its initial value respectively for the small, medium and large samples (cf. reference
sample in Table 4.2). Considering the additional effort required when the first nine
shortest paths are not available, we see a general increase especially for the low volt-
age samples, where the AP L10th increases three times for the large sample analyzed;
the increase is still present in the medium voltage networks, but it is limited when
1 http://snap.stanford.edu/
2 http://www.jgrapht.org/
compared to the low voltage samples. It is again an indication that the medium volt-
age provides more efficient alternative paths to connect nodes. An exception in the
results is the low voltage medium size sample: here the AP L10th is very close to the
traditional characteristic path length. This is due to the absence of alternative paths,
therefore the only paths between nodes are at the same time the best and worst case
too. Such aspect reinforces the idea of a low voltage network with a fixed structure
(sort of chain, tree-like or radial structures) and a limited redundancy.
Considering the betweenness-related metrics (Table 4 in [159]), one notes an in-
crease in the average betweenness as the samples become more numerous in the two
segments of the network. This same tendency is present in the average betweenness
to order ratio: the biggest samples in terms of order both of low voltage and medium
voltage score highest. In particular, the large sample belonging to the low voltage is
almost twice the value of the biggest sample of the medium voltage. Again it can be
justified by the similar-to-tree structure of the low voltage samples for which nodes
responsible for the paths that enable sub-trees or sub-chains to be connected are the
most high scoring for betweenness. This tendency highly increases the average be-
tweenness (while the mode is usually null). The coefficient of variation is above
one for all the big samples and reaches almost three for the biggest sample belong-
ing to the medium voltage network. Such an high value implies an high standard
deviation in the betweenness of the nodes.
4.2.1 Model parameters

To model the future power grid we compare network topologies that evolve in size
(M ) and order (N ). In particular, we consider the increase of average node degree
(xk y 2MN ). The evolution implies new cables and costs. For the Random Graph,
small-world, preferential attachment and R-MAT models, we consider an evolution
in the magnitude of average node degree of xk y 2 then xk y 4 and xk y 6.
For models that do not allow explicit setting of both size and order, we operate on
other parameters available that generate comparable networks. Another essential
assumption is the interpretation of all the graphs generated as undirected graphs.
Some of the models produce directed graph (e.g., Copying Model) since they tend
to model the WWW network because of the unidirectionality of the links in the Web
pages. The power grid graphs are, on the other hand, considered as undirected
graphs (e.g., [4, 221, 217, 158]) since the power can flow in both directions (in princi-
ple) in the power lines that represent the edges, this aspect is even more emphasized
by the vision of a prosumer-based energy exchange. A more through discussion con-
cerning the other parameters required by the synthetic network models is provided
in Appendix B.
Network Model Order Size Avg. CPL CC Removal Redundancy

type deg. robust- cost
ness (AP L10th )
(RobN )
LV-Small Small-world 20 20 2.000 4.053 0.00000 0.330 7.580
LV-Medium Small-world 90 90 2.000 11.820 0.01593 0.167 12.932
LV-Large Small-world 200 201 2.010 17.397 0.01083 0.109 21.544
MV-Small Small-world 250 250 2.000 24.237 0.00000 0.087 24.534
MV-Medium Small-world 500 501 2.004 28.084 0.00000 0.057 35.413
MV-Large Small-world 1000 1001 2.002 47.077 0.00000 0.040 60.074
LV-Small Preferential 20 19 1.900 2.579 0.00000 0.349 2.800
attachment
LV-Medium Preferential 90 89 1.978 4.315 0.00000 0.263 4.471
attachment
LV-Large Preferential 200 199 1.990 6.523 0.00000 0.206 6.375
attachment
MV-Small Preferential 250 249 1.992 5.426 0.00000 0.245 5.570
attachment
MV-Medium Preferential 500 499 1.996 5.705 0.00000 0.231 5.745
attachment
MV-Large Preferential 1000 999 1.998 6.976 0.00000 0.187 6.908
attachment
LV-Small Random 17 21 2.471 2.938 0.07451 0.390 7.472
Graph
LV-Medium Random 78 92 2.359 5.987 0.03547 0.418 10.974
Graph
LV-Large Random 172 207 2.407 6.254 0.00736 0.354 10.796
Graph
MV-Small Random 224 259 2.313 7.269 0.00000 0.322 12.002
Graph
MV-Medium Random 435 516 2.372 8.380 0.00138 0.321 12.818
Graph
MV-Large Random 863 1026 2.378 9.061 0.00070 0.328 13.446
Graph
LV-Small R-MAT 27 31 2.296 3.615 0.00000 0.356 7.830
LV-Medium R-MAT 88 125 2.841 4.115 0.05688 0.369 6.418
LV-Large R-MAT 199 261 2.623 5.495 0.00737 0.364 8.774
MV-Small R-MAT 195 263 2.697 5.629 0.00865 0.378 8.642
MV-Medium R-MAT 365 523 2.866 5.470 0.01360 0.396 7.646
MV-Large R-MAT 728 1056 2.901 5.726 0.00589 0.363 7.887
Table 4.3: Metrics for small-world, preferential attachment, Random Graph and R-
MAT models with average node degree 2.
Network Model Order Size Avg. be- Avg. Coeff.

type tweenness betw/order variation
LV-Small Small-world 20 20 62.300 3.115 0.804
LV-Medium Small-world 90 90 985.956 10.955 1.307
LV-Large Small-world 200 201 3429.720 17.149 1.260
MV-Small Small-world 250 250 5881.296 23.525 1.598
MV-Medium Small-world 500 501 13980.228 27.960 1.745
MV-Large Small-world 1000 1001 47919.616 47.920 2.279
LV-Small Preferential 20 19 31.400 1.570 2.344
attachment
LV-Medium Preferential 90 89 293.400 3.260 3.068
attachment
LV-Large Preferential 200 199 1089.260 5.446 3.288
attachment
MV-Small Preferential 250 249 1096.144 4.385 3.972
attachment
MV-Medium Preferential 500 499 2401.680 4.803 5.049
attachment
MV-Large Preferential 1000 999 6061.288 6.061 6.240
attachment
LV-Small Random 17 21 31.059 1.827 1.157
Graph
LV-Medium Random 78 92 408.308 5.235 1.126
Graph
LV-Large Random 172 207 938.512 5.456 1.276
Graph
MV-Small Random 224 259 1474.143 6.581 1.265
Graph
MV-Medium Random 435 516 3415.890 7.853 1.204
Graph
MV-Large Random 863 1026 7081.119 8.205 1.264
Graph
LV-Small R-MAT 27 31 70.593 2.615 1.320
LV-Medium R-MAT 88 125 282.500 3.210 1.540
LV-Large R-MAT 199 261 937.578 4.711 1.297
MV-Small R-MAT 195 263 959.118 4.919 1.395
MV-Medium R-MAT 365 523 1692.910 4.638 1.581
MV-Large R-MAT 728 1056 3633.473 4.991 2.004
Table 4.4: Betweenness metrics for small-world, preferential attachment, Random
Graph and R-MAT models with average node degree 2.

ness (AP L10th )
(RobN )
attachment
attachment
attachment
attachment
attachment
attachment
LV-Small Random 20 40 4.000 2.079 0.17667 0.733 4.350
Graph
LV-Medium Random 87 180 4.138 3.174 0.03418 0.735 5.368
Graph
LV-Large Random 199 400 4.020 3.869 0.03064 0.734 6.107
Graph
MV-Small Random 247 500 4.049 4.057 0.01681 0.740 6.432
Graph
MV-Medium Random 494 1000 4.049 4.495 0.00823 0.749 6.670
Graph
MV-Large Random 987 2001 4.055 5.062 0.00359 0.738 7.150
Graph
LV-Small R-MAT 30 59 3.933 2.517 0.27360 0.579 4.511
LV-Medium R-MAT 105 250 4.762 3.019 0.13039 0.581 4.490
LV-Large R-MAT 227 504 4.441 3.619 0.04683 0.601 5.302
MV-Small R-MAT 230 496 4.313 3.736 0.02940 0.626 5.381
MV-Medium R-MAT 420 1004 4.781 3.915 0.00450 0.591 5.249
MV-Large R-MAT 932 2039 4.376 4.562 0.00875 0.690 6.251

attachment
attachment
attachment
attachment
attachment
attachment
LV-Small Random 20 40 23.600 1.180 0.807
Graph
LV-Medium Random 87 180 196.345 2.257 0.850
Graph
LV-Large Random 199 400 589.849 2.964 0.889
Graph
MV-Small Random 247 500 766.389 3.103 0.857
Graph
MV-Medium Random 494 1000 1768.757 3.580 0.972
Graph
MV-Large Random 987 2001 4068.393 4.122 0.942
Graph
LV-Small R-MAT 30 59 44.000 1.467 1.342
LV-Medium R-MAT 105 250 223.733 2.131 1.695
LV-Large R-MAT 227 504 609.419 2.685 1.493
MV-Small R-MAT 230 496 650.374 2.828 1.468
MV-Medium R-MAT 420 1004 1285.786 3.061 1.652
MV-Large R-MAT 932 2039 3422.348 3.672 1.506
Table 4.6: Betweenness metrics for small-world, preferential attachment, Random
Graph and R-MAT models with average node degree 4.
4.2.2 Model generation

Each network model described is generated and analyzed according to the signif-
icant power grid metrics described in Section 4.1. We begin with the models for

ness (AP L10th )
(RobN )
attachment
attachment
attachment
attachment
attachment
attachment
LV-Small Random 20 60 6.000 1.684 0.29599 0.775 3.370
Graph
LV-Medium Random 90 270 6.000 2.640 0.06987 0.791 4.298
Graph
LV-Large Random 200 600 6.000 3.141 0.03991 0.777 4.693
Graph
MV-Small Random 249 750 6.024 3.230 0.01934 0.793 4.884
Graph
MV-Medium Random 499 1500 6.012 3.620 0.00976 0.792 5.284
Graph
MV-Large Random 998 3000 6.012 4.022 0.00544 0.791 5.662
Graph
LV-Small R-MAT 32 87 5.438 2.194 0.21179 0.760 3.945
LV-Medium R-MAT 123 374 6.081 2.926 0.08173 0.717 4.377
LV-Large R-MAT 249 759 6.096 3.165 0.04444 0.736 4.622
MV-Small R-MAT 236 747 6.331 3.143 0.04982 0.746 4.389
MV-Medium R-MAT 466 1512 6.489 3.427 0.04365 0.743 4.805
MV-Large R-MAT 925 3035 6.562 3.742 0.02560 0.723 4.925

attachment
attachment
attachment
attachment
attachment
attachment
LV-Small Random 20 40 14.700 0.735 0.662
Graph
LV-Medium Random 87 180 151.489 1.683 0.809
Graph
LV-Large Random 199 400 431.090 2.155 0.835
Graph
MV-Small Random 247 500 563.839 2.264 0.710
Graph
MV-Medium Random 494 1000 1328.405 2.662 0.745
Graph
MV-Large Random 987 2001 3051.922 3.058 0.771
Graph
LV-Small R-MAT 30 59 38.000 1.188 0.989
LV-Medium R-MAT 105 250 247.008 2.008 1.351
LV-Large R-MAT 227 504 550.538 2.211 1.352
MV-Small R-MAT 230 496 530.093 2.246 1.357
MV-Medium R-MAT 420 1004 1169.382 2.509 1.506
MV-Large R-MAT 932 2039 2599.496 2.810 1.731
Table 4.8: Betweenness-related metrics for small-world, preferential attachment,
Random Graph and R-MAT models with average node degree 6.
which it is possible to explicitly assign order and size (or one of these quantities and
the average node degree); we then proceed analyzing the other models that do not
explicitly allow to set the average node degree parameter.
Comparison of models with average node degree xk y 2
The results for the metrics with average degree xk y 2 for the small-world, pref-
erential attachment, Random Graph and R-MAT models score quite poorly, cf. Ta-
ble 4.3. Low values for the metrics are due to the small connectivity the networks
show. Especially, we highlight the low results of the small-world model under these
conditions.
The betweenness analysis, whose results are presented in Table 4.4, shows an
average for each node that increases with the size of the graph.
Table 4.5 shows the results for small-world, preferential attachment, Random Graph
and R-MAT models with an average degree xk y 4. One notices that the scores for
the metrics improve compared to the xk y 2 case. The average over the character-
istic path length of all the samples reduces from around 10 to a value that is slightly
less than 5. The clustering coefficient has values that are significant and all positive.
The small-world model scores best in this specific metric since it relies on the lattice
topology that with an average degree of 4 connects each node with 4 neighbors. In
particular 3 triangle structures emerge in each neighborhood of a node (of course
before the rewiring process takes place). This provides a substantial contribution
to the quite high clustering coefficient. A graphical comparison for the large sam-
ple for medium voltage considering characteristic path length, clustering coefficient
and robustness are shown in Figure 4.3.
(a) Characteristic path length. (b) Clustering coefficient.
(c) Removal robustness.

Figure 4.3: Results for metrics for the large sample of medium voltage network type
with average node degree 4.
Analyzing the betweenness we see a general improvement in the metrics com-

pared to the xk y 2 case, cf. Table 4.6. The most important improvement is for
the small-world model which, with approximately 4 connections per node, substan-
tially reduces the average betweenness by a factor of 10 compared to the xk y 2
case. A graphical comparison for the results of the large sample for medium voltage
type considering average betweenness order ratio and coefficient of variation metrics
are shown in Figure 4.4.
(a) Betweenness to order ratio. (b) Betweenness coefficient of

variation.
Figure 4.4: Results for metrics for the large sample of medium voltage network type
with average node degree 4.
Table 4.7 shows the results for small-world, preferential attachment, Random Graph
and R-MAT models with an average degree xk y 6. The scores for the metrics con-
sidered improve even more with respect to those of Tables 4.3 and 4.5. The charac-
teristic path length of all the samples has reduced to a value that, considering the
average over all the samples with xk y 6, is about 3; yet 2 hops lower than the situ-
ation with xk y 4. The same tendency for clustering coefficient found for samples
in Table 4.5 applies to this situation too. The small-world model scores highest since
the neighbors of a node have 9 connections with each other (before rewiring), thus
contributing to a high coefficient.
Having increased the average degree to 6 brings benefits to the betweenness
statistics too, cf. Table 4.8. The improvement on the average betweenness order ratio
are about 25% higher than the xk y 4 situation; this ratio, therefore, is now very
close to the experimental values that have been found for the Internet (i.e., 2.5).
A more thorough analysis of these results and a detailed comparison are avail-
able in [159].
Models Independent from the Average Node Degree
The Copying, Forest Fire, and Kronecker models are not generated using explicitly
the average node degree, cf. Appendix B. Therefore, we consider the power grid
metrics on them separately. We remark however that, though not explicitly used
as input parameter, the average node degree of the generated graphs has similar
values to those of Random Graphs, small-world and preferential attachment models
generated with the same order. As a general consideration for these models, we see
a reduced performance in the satisfaction of the metrics. The only models that have
overall scores that become closer to the target parameters requirements is the Forest
Fire and the power-law-based graph with small value of the characteristic parameter
(γ). Therefore, we give a detailed analysis of these models only, and refer to [159]
for the analysis of the other models.
For the Forest Fire model, we assign different forward and backward burning
probabilities to obtain values for the average degree to some extent comparable
with the other models. The model with pf wd pbwd 0.2 can be compared to
models with xk y 2. The Forest Fire scores definitely better than all the others in
clustering coefficient. This is not surprising, if one recalls the algorithm behind the
model: an ambassador node is chosen and with a certain probability a certain num-
ber of ambassador’s neighbors nodes are chosen to establish link to. One can see
how many triangle-like structures tend to appear from such a generating method.
The same observations can be done for the Forest Fire with pf wd pbwd 0.3 when
compared to models with xk y 4: the characteristic path length has a score similar
to node degree dependent models, while this model suffers deeply in the robust-
ness metric which for the biggest samples obtain a score which is half compared
to the other generating models with xk y 4. This is due to the very high dam-
ages imposed to network connectivity when high degree nodes are removed: for
the biggest sample (order of about 1000 nodes), when the 20% of nodes with high-
est degree are removed, the biggest connected component is just 2% of the original
graph order. This is typical of heavy-tailed distributions which Forest Fire models
empirically [118]. The metric that scores best is again the clustering coefficient that
is three times higher (for the biggest sample) than the already quite high value of
the small-world model. Even when we consider denser Forest Fire networks (i.e.,
pf wd pbwd 0.35) the comparison with the model with xk y 6 brings to the
same conclusions: far better clustering coefficient, but an important weakness to
node removal. Betweenness for the Forest Fire model shows a trend when varying
the average node degree, the more the networks becomes connected the better the
metrics related to betweenness become. For the samples with a burning probability
of pf wd pbwd 0.35, the betweenness to order ratio stays below 3. The same be-
havior applies to the coefficient of variation, although it generally scores worse than
the samples already analyzed with similar average degree.
Considering the results of Random Graph with power-law models, there is a
difference for the networks generated with smaller γ parameters (i.e., medium and
low voltage Dutch grid γ 2 and social and technological networks γ 2.3) which
score better than the ones with higher γ (i.e., U.S. Eastern Interconnect and West-
ern Grid γ 3 and U.S. Western power grid γ 4). Refer to Appendix B for the
motivations of choosing the specific γ parameters. The samples with γ 2 and
γ 2.3 show a denser network with higher average node degree, almost double
compared to the other set (γ 3 and γ 4); it results in a beneficial behavior for
the metrics computed which present a smaller characteristic path length. This set of
networks with small γ is comparable for the characteristic path length property to
the values obtained for networks generated with xk y 4. The second set of samples
(i.e., higher γ parameter) shows results that are similar to the ones obtained for sam-
ples generated with xk y 2. A general property that applies to all these power-law
based samples is the problem they suffer from targeted attacks involving the nodes
with high degree, which justifies very poor scores for robustness metric. The be-
tweenness analysis for the power-law based models shows an average betweenness
value that is smaller for the networks with a lower value for the γ coefficient so that
they score very good in the betweenness to order ratio. A lower γ implies a higher
probability in the presence of nodes that have higher node degree; usually there is
quite a good positive correlation between the node degree and the betweenness the
nodes have to sustain (high degree implies high betweenness for that node). It is
therefore understandable why the coefficient of variation is higher for the networks
characterized by a low γ than the ones with higher power-law characteristic param-
eter.
Comparing the generated topologies with the physical ones
The analysis of the Northern Netherlands grid shows an average degree almost con-
stant of about xk y 2. In terms of average node degree the situation is similar to the
high voltage grid based on the data of Eastern and Western high voltage U.S. power
grid. Therefore, it is fair to compare the generated models with similar average de-
gree, the Copying Model ones and the Random Graphs with power-law in node
degree distribution with average node degree around xk y 2. Generated mod-
els, except the model based on Random Graph with power-law, score better than
the physical topologies for all the metrics considered; the characteristic path length
scores half for the R-MAT and Copying Model cases in comparison to the real data.
In addition, the synthetic networks are more robust than the real data samples: R-
MAT and Random Graph score constantly above 0.3 for robustness metric while real
data hardly obtain this value. Clustering coefficients are quite similar since in this
configuration with limited connectivity having triangle structures in the network
is rare; however, we see that R-MAT model has almost always significant cluster-
ing coefficient values. An exception is the small-world model which scores almost
always worse than the real data samples. In fact, under this situation of small av-
erage node degree it is actually not fully correct to consider this synthetic topology
a “small-world”. The same sort of considerations apply to betweenness values: ex-
cept the small-world model all the other synthetic ones score better for the average
betweenness to order ratio metric, while for the coefficient of variation the situation
is similar. If one considers the satisfaction of the desiderata for the actual samples
of the Dutch medium and low voltage grid, summarized in Table 4.9, we notice that
none of the parameters are satisfied. However, networks generated according to
the models with almost the same average node degree (networks with xk y 2 in
Tables 4.10 and networks based on Random Graph with power-law based on data
from Eastern and Western high voltage U.S. power grid and the U.S. Western high
voltage power grid in Table 4.14) do not satisfy all the desiderata as well. There-
fore, this highlights that the first ingredient for the next generation of grids suited
to enable local energy exchange according to the metrics defined is an increase con-
nectivity.
Desiderata Northern Netherlands

medium and low voltage samples
Modularity 7
¤ p q
CP L ln N 7
¥
CC 5 CCRG 7
υ
σ
2.5 7
¤
N
cv 1 7
RobN¥ 0.45 7
¤
AP L10th 2 CP L
Table 4.9: Desiderata parameter compliance of physical samples of the Northern
Netherlands grid.
Increasing the average node degree naturally provides for better values for the
network metrics, as shown in Tables 4.10 and 4.11. The case of the small-world
model is emblematic. The xk y 2 case scores extremely poorly as there are not
enough “shortcuts” in the network so that they can not improve much the charac-
teristic path length. Actually, under such small average degree the condition Watts
and Strogatz impose for their small-world model is not completely satisfied (i.e.,
n " k " lnpnq " 1, where k is the average node degree and n is the order of the
graph). When we move closer to satisfying the small-world condition by increasing
the average node degree, the value of the metrics suddenly change and the models
score extremely high. The small-world scores best for the clustering property and
resilience to failures in the xk y 4 situations. Under these conditions also between-
ness values are quite concentrated around the mean with a coefficient of variation
not exceeding the unit.
Comparing the average values of some metrics for the generated models while
increasing node degree, one notices a natural improvement of the metrics, cf. Ta-
ble 4.12. In fact, we have a reduction in characteristic path length of about 60%
and an increase in the clustering coefficient of one order of magnitude; at the same
time the robustness doubles. With xk y 6 the improvement in the metrics is less
prominent, being between 10% and 20% compared to the xk y 4 case. From the
comparison of the metric results in Tables 4.10 and 4.11, one sees that the small-
world model almost always satisfies the desired requirement from a quantitative
point of view when the average node degree is at least 4. From a qualitative point of
view, the small-world model shows some modularity being generated starting from
a regular lattice and then rewiring a certain fraction of the edges.
The models independent form average node degree perform generally worse
than the other models. The adherence to the target values are shown in Tables 4.13
and 4.14. There is a general prevalence of requirement dissatisfaction, especially for
parameters involving betweenness.
From the topological analysis one can see that between the models analyzed
when there is a minimal connectivity (xk y 4 or xk y 6) the small-world stands
out, cf. Tables 4.10 and 4.11. In Table 4.15 the models with explicit dependence
on node degree are once again compared by assigning a “tick” sign (3) for the ful-
fillment of each of the following properties: qualitative topological parameters (i.e.,
modularity), quantitative topological parameters (Tables 4.10 and 4.11) and the thrift
in network realization (e.g., addition of cables which represent a cost). The latter is
just an estimation, a more detailed analysis of cost in realizing a network belonging
to medium and low voltage with a certain size (i.e., small, medium or large) and the
economic benefits in electricity distribution arising from the enhanced connectivity
is provided in Section 4.2.3. From Table 4.15, we conclude that networks generated
with the small-world model with average degree xk y 4 provide the best balance
between modularity, performance and, thrift for the future power grid.
4.2.3 Economic considerations

Traditionally, the problem of evaluating the expansion of an electrical system is a
complex task that involves both the use of modeling, usually based on optimization
techniques and linear programming [74], and the experience and vision of experts
Desiderata Average node degree k
x y2 Average node degree k
x y 4
SW Pref. Attach. Rnd. Graph R-MAT SW Pref. Attach. Rnd. Graph R-MAT
Modularity 7 7 3 7 7 3
CP L ln N
¤ p q 7 3 3 3 3 3
4.2. Building New Distribution Networks
CC 5 CCRG
¥ 7 7 N/A 3 3 N/A 7
σ
υ N
2.5 7 7 7 7 7 7 7
cv
¤ 1 7 7 7 7 3 7 3 7
RobN
¥ 0.45 7 7 7 7 3 3 3 3
AP L10th
¤ 2 CP L 3 3 3 3 3 3 3 3
Table 4.10: Desiderata parameter compliance of the generated models with node degree k x y 2, 4.
89
4. Network Evolutions for the Smart Grid
Desiderata x y
Average node degree k 6
SW Pref. Attach. Rnd. Graph R-MAT
Modularity 7 7 3
¤ p q
CP L ln N 3 3 3 3
¥
CC 5 CCRG 3 3 N/A
υ σ
2.5 3 3
¤
N
cv 1 3 7 3 7
¥
RobN 0.45 3 3 3 3
AP L10th ¤ 2 CP L 3 3 3 3
x y 6.
Table 4.11: Desiderata parameter compliance of the generated models with node degree k
90
Avg. node degree transition Average metric improvement (%)

CPL CC Robustness
xky 2 Ñ xky 4 61.7 941.6 128.5
xky 4 Ñ xky 6 18.0 11.8 19.6
Table 4.12: Comparison of generated topologies for varying average node degree.
in the field supported by computer computation in their decisions. With more dis-
tributed generating facilities of small capacity (e.g., rooftop photovoltaic panels),
traditional methods have limits and need to be modified or updated to take into ac-
count the new scenario the smart grid framework brings into play. The models that
we have so far analyzed as being candidates for the vision of the future smart grid
need also to be evaluated from the economic point of view. How much will it cost
to generate electrical infrastructures according to these models? What is the actual
cost of adding a physical edge to the topology?
The cost of adding edges
One important difference that a physical infrastructure such as the power grid has,
compared to the WWW or social networks, is the physical presence of cables that
have to connect the medium voltage substations or low voltage end users generating
units. If establishing a link from a Web page to another one is free, each increase in
connectivity in the power grid implies costs in order to build or adapt the substation
or end user premise involved and the cables required for the connection. To assess
these costs in the medium and low voltage infrastructure, we consider a simple
relation where the cost of cabling and cost of substations are added:
¸
N ¸
M
Cimp Sscj Cci (4.1)

j 1
i 1
where Cimpl stands for cost for implementation, Sscj is the adaptation cost for the
substation j and Cci is the cost for the cable i. The cost of the cable can be expressed
as a linear function of the distance the cable i covers: Cci Cuc l where Cuc is
the cable cost per unit of length and l is the lengths of the cable. Several types of
cables exist which are used for power transmission and distribution with varying
physical characteristics and costs; in addition, also the cost for installation can vary
significantly [148]. In the present work, to provide an initial estimate, we simply
consider cabling costs and ignore substation ones. While the former are directly
tied to the topology and length of the links, the latter is dependent on other factors
too (e.g., type of equipment). As a source of data for cable type and pricing, we have
4. Network Evolutions for the Smart Grid
Desiderata Copying Forest Fire Forest Fire Forest Fire Kronecker Kronecker
Model (pb=0.2) (pb=0.3) (pb=0.35) (PG params) (social
net params)
Modularity 7 7 7 3 3
¤ p q
CP L ln N 3 7 3 7 3
¥
CC 5 CCRG 7 3 3 3 7 7
υ σ
2.5 7 7 7 7
¤
N
cv 1 7 7 7 7 7 7
RobN ¥ 0.45 7 7 7 7 7
AP L10th ¤ 2 CP L N/A 7 3 3 3 3
Table 4.13: Desiderata parameter compliance of the generated models.
92
Desiderata RG with power-law RG with power-law RG with power-law RG with power-law
Model (social (East-West US (Western US (NL MLV
net params) HV PG params) HV PG params) PG params)
Modularity 7 7 7 7
4.2. Building New Distribution Networks
CP L ln N
¤ p q 3 7 7 3
CC 5 CCRG
¥ 3 7 3
σ
υ N
2.5 7 7 7 7
cv
¤ 1 7 7 7 7
RobN
¥ 0.45 7 7 7 7
AP L10th
¤ 2 CP L 3 3 3 3
Table 4.14: Desiderata parameter compliance of the generated models.
93
Network Model Avg. node deg. Avg. node deg. Avg. node deg.
x y
k 2 x y
k 4 x y
k 6
Small-world 33 333 33
Preferential 3 33 3
Attachment
Random Graph 3 33 3
R-MAT 33 33 33
Table 4.15: Satisfaction of modularity, performance and cabling cost for generated
models.
been provided (courtesy of Enexis B.V. the Netherlands) with cables characteristics
and prices for the 11 network samples belonging to the low voltage network and
12 samples belonging to the medium voltage of the Northern Netherlands, whose
topological properties we have analyzed in Chapter 3.
Statistical consideration over cables’ price
The length of the cables plays an important role for both total resistance and price.
If one considers the correlation between the price and resistance, high values are
found using Spearman’s rank correlation coefficient [102], shown in Table 24 in [159].
Especially, for generating synthetic networks it is important to obtain values for both
the properties of the cables that are similar to the ones used in practice. Plotting the
two variables characterizing each cable one notices that the majority of the samples
concentrates in the lower tails of the joint distribution. Figure 4.5 shows the relation
between the price and resistance where the values concentrate in the lower corner
of price resistance plane.
In the chart in Figure 4.5, two distinct lines deviate from the low-left corner.
They represent the two main types of cables used in that sample of the low voltage
network to cover different distances and result in increasing price and resistance for
longer lines. The problem of extracting cable properties can be however approached
in another way: evaluate for each type of cable used in a certain sample (small, medium,
and large) how the length of the cables used are distributed. In fact, given a certain type
of cable and its length all other interesting properties required for our weighted
topological analysis (cf. Chapter 3) are then available (i.e., cable total resistance,
cable total cost and cable supported current).
A general tendency appears when fitting the distribution of lengths to cable
types belonging to low voltage and medium voltage: a fast decay in lengths’ prob-
ability distribution with the majority of lengths for the low voltage cables types in
the order of tens of meters, and medium voltage cables about hundreds of meters.
Fitting the length to a statistical probability distribution gives a good approxima-
tion for the low voltage cable lengths as exponential distributions (y fX px; µq
x
1 µ
µe ), while for medium voltage cable lengths the distribution that generally fits
best the data is the generalized extreme value distribution (y fX px; k, µ, σ q
σ p1 k xµ 1 k
σ q exptp1 k x µ k
σ q u); these hypothesis are supported in the re-
1 1 1
sults by the tests of Kolmogorov-Smirnov [130]. A graphical representation of the

length probability cumulative distribution function for a cable type of the medium
voltage network is shown in Figure 4.6.
Assume that, statistically speaking, the distribution of the lengths for each type
of cable in the synthetic networks are the same as in the real samples. Therefore,
knowing the probability of using a certain type of cable i (pcablei ° #cablei
where
k #cablek
#cablei is the number of occurrences of cable type i in a certain network sample)
that has a certain cost and resistance per meter and a specific current supported, it is
then possible to estimate the cables that are used in the synthetic samples together
with their properties. The joint probability of the event of having a cable of length
L of cable type T can be obtained by the conditional probability relationship:
P pL X T q P pL|T qP pT q (4.2)
where P pL|T q is the probability of the event of having a cable of length L condi-
tioned to the event of being of cable type T , and P pT q is the probability of the event
of having a cable of type T .
Figure 4.5: Price-Resistance pairs joint plot for the low voltage large size sample (i.e.,
LV Sample#5).
Sample type Size Cost (thousand euro)

low voltage - Small 20 30
low voltage - Medium 90 78
low voltage - Large 200 449
medium voltage - Small 250 32000
medium voltage - Medium 500 42000
medium voltage - Large 1000 43000
x y 2 synthetic samples.
Table 4.16: Cabling cost for k
Figure 4.6: Cumulative distribution function for cable length (meters) for cable type
“3x1x70al” in Northern Netherlands medium voltage sample size medium.
Economic benefits of highly connected topologies
Once the information about cable prices is available, it is possible to estimate the cost
for realizing a network with a certain connectivity and whether such networks are
able to lower the (economic) barrier towards decentralized energy trading in terms
of cost of electricity distribution. The results for low voltage and medium voltage
networks for small, medium and large types with an average node degree xk y 2,
xky 4 and xky 6 are shown in Table 4.16, Table 4.17, and Table 4.18 respectively.
For medium voltage networks, it is important to clarify that the information
available for cables’ prices in this study are only partial and limited to some tech-
nologies (only few cross sections of aluminum and copper cables). Anyway, in order
to have a glimpse of costs for this type of the network, we fitted to the best inter-
polating curve the available prices as a function of the cross section. The relation
between price and cross section for aluminum cables fits best to a cubic polyno-


mial, while for the copper ones is linear; in this way we estimate the prices for all
the types of cables involved knowing their cross section. The small difference in
costs between the medium and large types of networks for medium voltage is related
mainly to the different technologies (i.e., cable types) in the cables that are used for
these types of networks according to the sample data provided.
Price alone is not enough to describe future scenarios. It is important to in-
vestigate how an enhanced connectivity is beneficial to the electricity distribution
costs. We have shown the benefits in topology-related metrics for more connected
networks earlier in this section, however those results consider only the topology
without any parameter related to the properties of the cables (e.g., resistance and
supported current). In order to consider the effects of topology in electricity distri-
bution costs, we resort to the α and β metrics the we have proposed in Chapter 3.
In order to apply these metrics to power grid networks weights are essential, rep-
resenting physical quantities such as resistance of the cable and maximal operating
current supported by the cable. Once we have the statistical information about the
types and the length of the cables used in a specific type of physical network, (i.e.,
medium or low voltage and its small, medium or large size) it is possible to assign
weights to the edges of the generated graphs. This is done under the assumption
that the same type of cables are used and that the distances covered in general (i.e.,
statistically) remain the same.
We consider α and β for networks generated following the small-world model
since it has proven to be the best one in the pure topological analysis. For low
voltage networks we compute the metrics for networks with an increasing average
node degree (xk y 2, xk y 4 and xk y 6). It is not surprising to see that the
samples with xk y 2 score poorer than the other networks. The network with
medium size scores best and the difference between the network with xk y 6 and
the network with xk y 4 is limited. Robustness (i.e., β parameter) for the medium
and large size networks reaches an high value just with a sufficient connectivity (i.e.,
xky 4) and more connectivity (i.e., xky 6) does not improve much this metric.
The samples with small size score better in the α metric and this is quite reasonable
since the paths are limited, of course due to the reduced order of the network.
Considering α and β for the networks generated for the medium voltage, the
same tendency appears: once the network is sufficiently connected (i.e., xk y 4) the
metrics score definitely better than the xk y 2 situation.
Let us compare α and β of the synthetic networks with the values of the cur-
rent power grid samples of the Northern Netherlands. Considering the low voltage
samples and the synthetic networks designed for this purpose, we generally see an
improvement in the metrics especially in the α values for the xk y 4 and xk y 6
networks. In fact, if we do not consider the synthetic networks with xk y 2, be-
cause of the problems of small-world topology with such small connectivity, there
is an improvement on average in the α metric for synthetic samples with xk y 4
of more than 50% compared to the Northern Netherlands samples. In fact, for the
α metric from an average of about 12 for the physical samples, the xk y 4 syn-
thetic ones score about 6. The improvement is more than 60% when considering
the xk y 6 ones where the average for these synthetic networks scores just below
5. There are improvements also in the β metric, although limited. From an average
around 4 for the physical samples, the xk y 4 ones score on average just below
2.75; while a better result is obtained by xk y 6 which on average score 2.30 (about
40% improvement). The graphical comparison between Dutch samples (white dia-
monds) and generated samples (white circles) is shown in Figure 4.7 in which each
symbol represents a sample in the α, β quadratic function envelope that is chosen as
the type of dependence between the topological parameters and electricity transport
prices.
Taking into account the medium voltage, the Dutch samples of the Northern
Netherlands and the small-world synthetic networks, we see an important improve-
ment in the metrics both in the α and β values for the xk y 4 and xk y 6 networks.
As already mentioned, synthetic networks with xk y 2 should not be considered.
The improvement on average in the α metric is more than 65% comparing to the
xky 4 synthetic samples (from an average of α about 33 for the physical samples,
the xk y 4 synthetic ones score about 11), and an improvement of more than 75%
when comparing to the xk y 6 ones (the average α for xk y 6 synthetic networks

scores around 7.3). There are improvements also in the β metric. In particular, from
an average of around 3.55 for the physical samples, the xk y 4 score on average just
below 1.15; a similar result is obtained by xk y 6 which on average score about 1.2
(more than 65% improvement). The graphical comparison is shown in Figure 4.8.
Figure 4.7: Comparison of the transport cost between synthetic small-world net-
works (withe circles) and Northern Netherlands low voltage samples (white dia-
monds).
Figure 4.8: Comparison for transport cost between synthetic small-world networks
(withe circles) and Northern Netherlands medium voltage samples (white dia-
monds).
4.2.4 Topology costs vs. benefits
Watts and Strogatz’s small-world model captures best the requirements for the new
grid compared to the others analyzed being these dependent on the average node
degree (preferential attachment, R-MAT and Random Graph) or not (Copying Model,
Forest Fire, Kronecker and power-laws). The higher clustering that this models ex-
hibits provides efficient local distribution with paths that are locally short; at the
same time the shortcuts between the local clusters are the elements that keep the
(global) average path limited. These two aspects influence the α parameter which
then stays relatively small. At the same time, the small-world model benefits from
a general robustness against failures: the absence of big hubs that keep the network
together (which are present on the other hand in the power-law-based topologies,
for instance) improves the reliability against attacks, which help obtaining good
scores for the β parameter. More quantitatively, one sees the general improvement
in the metrics characterizing both the parameters influencing the losses (i.e., α pa-
rameter) and the reliability of the grid (i.e., β parameter) while the network becomes
more dense, i.e., more edges are added. On average, we see an improvement of at
least 50% when comparing the physical samples of Northern Netherlands with the
small-world networks with an average degree xk y 4, while better results are ob-
tained with more density (i.e., xk y 6) where the improvement are 60% compared
to the physical samples. This is indeed beneficial to the power grid and, accord-
ing to the relationship with the topology, it should translate into a reduction in the
costs for electricity distribution since α and β are composed by essential ingredients
influencing electricity distribution price.
These benefits come literally at a cost. The network needs more connectivity
therefore costs for extra cabling need to be considered in addition to the cost for
upgrading the substations and end users electricity gateways. A comprehensive re-
turn on investment analysis on this aspect is beyond the scope of the present study.
Nevertheless, it is interesting to see how with the α and β metrics it is possible to
consider how a certain physical sample belonging to a certain size category (small,
medium and large) would improve in its performance if its topology is arranged ac-
cording to the principles of a synthetic model and more connections are added ap-
propriately.
The benefits reached for α and β should translate into a reduction in the cost
for electricity transport and distribution since the parameters that influence these
metrics are directly connected to aspects related to costs. However, the significant
investment required to add more connectivity in the network might not immedi-
ately enable cheaper electricity costs, but on the contrary make it more expensive.
4.3. Evolution of Current Distribution Networks 101
4.3 Evolution of Current Distribution Networks

Another important topic to consider is how to evolve or adapt the current distri-
bution grids to support a prosumer-based energy approach. It is therefore essential
to evaluate the benefits of different ways (or strategies) of evolving the distribution
network in terms of additional connections. It is also essential to consider both the
topological benefits together with costs of realization and reduction in cost of the
electricity distribution. As a case study we consider the evolution of the distribu-
tion grid of the Netherlands that we have analyzed in Chapter 3.
4.3.1 Network evolution policies

Next, we consider evolutions starting from the Dutch samples, that is, adding cables
according to several strategies of network growth. We break the evolutions into
four groups of edge growth: increments of 25%, 50%, 75%, and 100%. The choice of
stopping at 100% is performed based on the results of Section 4.2, where we show
that an average node degree of xk y 4 has the right balance of improved network
qualities and costs of network evolution. We consider several strategies for evolving
the graph by adding more links, namely:
• Assortativity. A network is assortative if nodes having similar characteristics

or properties are connected one another [150]. We consider node degree and
then take two strategies:
– High degree nodes are connected one another. The process starts with con-
sidering a set of nodes with the highest equal node degree and connect
them together. The process goes on considering the next set of nodes with
equal high degree in the order of rank and so on.
– Low degree nodes are connected one another. The process goes on as for
the the high degree strategy, but nodes are linked starting from the cou-
ples with lowest degree.
• Dissortativity is the opposite of assortativity, that is, a network is dissortative

if nodes having different characteristics or properties are connected together.
Therefore, nodes with highest node degree are linked to nodes with lowest
node degree.
• Triangle closure is based on the principle of increasing the clustering coeffi-

cient of the network. At each step, a node is selected at random and for each
pair of its neighbors an edge is added between them, if not already present.
• Least distance gives priority to the connection of nodes that are geographi-
cally closer to each other. This strategy can minimize the costs of cabling since
such costs are directly proportional to the length of cables.
• Random is based on the random selection of nodes to attach edges. At each

step, a pair of distinct nodes are randomly selected and an edge between them
is added.
For every strategy, if two nodes already have an edge that connects them the edge
is not added and the evolution strategy continues. In fact, in the graph models we
only allow a single edge between a pair of nodes, if not already present.
We study the adaptation of current physical networks according the strategies

just described and we analyze the obtained graphs according to a set of metrics that
provide a view of efficiency of the whole network and its adequacy for local energy
distribution (cf. Section 4.1).
(a) Physical sample.

(b) 1st stage of evolution (i.e., +25% (c) 2nd stage of evolution (i.e., +50%
edges). edges).
(d) 3rd stage of evolution (i.e., +75% (e) 4th stage of evolution (i.e., +100%
edges). edges).
Figure 4.9: Stages of evolution of medium voltage sample #8 following least distance
strategy.
To determine which strategy is more rewarding and provides the higher benefits,
one needs to compare the metrics obtained with the different evolution strategies.
If determining a “best” strategy is unrealistic, one should aim at identifying trade-
offs between one strategy compared to another one. We evaluate all the strategies
considering average values over all the samples and considering the satisfaction
of the topological metrics defined in Section 4.1. Visual examples of the evolution
strategies are provided in Figures 4.9 and 4.10. Figure 4.9 represents the evolu-
tion following the least distance strategy, one can see that for a part of the network
nodes that are several hops apart in the original network (they belong to two differ-
ent branches) are now connected by new edges providing a substantial topological
shortcut in the network. Figure 4.10 shows the evolution following the triangle
closure strategy, one sees the appearance of several new triangle structures in the
network.
(a) Physical sample.
(b) 1st stage of evolution (i.e., +25% edges). (c) 2nd stage of evolution (i.e., +50% edges).
(d) 3rd stage of evolution (i.e., +75% edges). (e) 4th stage of evolution (i.e., +100% edges).
Figure 4.10: Stages of evolution of medium voltage sample #8 following triangle clo-
sure strategy.
4.3.2 Comparison of the evolution strategies

Comparison of the evolution strategies for the medium voltage grid
It is difficult and perhaps even methodologically wrong to establish “a winner”

among the various evolution strategies. The improvement in the various metric
values during the evolution does not establish a linear pattern. As a general re-
mark, we reinforce that the addition of links is beneficial to the topological metrics
analyzed and one might speculate that the optimal solution in order to minimize
(or maximize) the metrics is to realize a fully connected network (i.e., clique). With
a real infrastructure such as the power grid this is impossible to achieve from the
economic point of view, but also from a technical point of view (e.g., substations
receiving thousand of connections).
The comparison of the averages over all samples in each evolution step for the
various evolution strategies is shown in Table 4.19. Take the triangle closure strat-
egy, for instance, it is the one that in every step of the evolution process leads to
the maximal values of the clustering coefficient, but it is also the one that has the
worst results together with the least distance strategy for the metrics related to path
length (characteristic path length and AP L10th ). Considering the robustness metric,
the triangle closure offers the smallest improvements, while for the betweenness-
related metrics, the triangle closure strategy is the worst compared to all the others
with the second set of highest values for average betweenness and an increasing
coefficient of variation. The same considerations apply to the least distance strat-
Evolution CPL CC Rob. 10th Avg. Avg. Coeff.

step red. bet. bet./order var.
path
+25% 5.757 0.037 0.326 8.550 1682.845 4.744 2.258
Assortative +50% 5.121 0.067 0.344 7.095 1369.399 4.100 2.259
high +75% 4.724 0.098 0.361 6.493 1245.963 3.799 2.147
degree +100% 4.558 0.122 0.367 6.072 1226.062 3.717 1.904
+25% 6.404 0.045 0.358 9.503 1955.047 5.598 2.039
Assortative +50% 6.320 0.090 0.408 8.492 1931.053 5.508 1.651
low degree +75% 6.261 0.109 0.482 8.149 1913.409 5.440 1.504
+100% 6.172 0.121 0.496 8.014 1888.415 5.357 1.441
+25% 10.761 0.165 0.257 14.559 2698.910 8.130 1.786
Triangle +50% 9.504 0.288 0.321 11.680 2377.892 7.123 1.872
closure +75% 8.482 0.388 0.341 10.053 2104.554 6.289 1.927
+100% 7.618 0.483 0.384 8.997 1884.527 5.563 2.034
+25% 5.803 0.019 0.349 8.891 1727.708 4.949 2.433
+50% 5.591 0.028 0.408 7.904 1645.824 4.710 1.953
Dissortative +75% 5.402 0.034 0.463 7.421 1594.361 4.532 1.739
+100% 5.217 0.033 0.468 7.053 1559.650 4.418 1.613
+25% 10.025 0.141 0.272 13.078 3044.817 9.042 1.703
Least +50% 9.066 0.214 0.360 11.696 2789.684 8.127 1.744
distance +75% 8.535 0.261 0.455 10.705 2646.133 7.652 1.788
+100% 8.109 0.304 0.523 10.013 2525.651 7.275 1.807
+25% 6.249 0.011 0.423 9.954 1785.946 5.265 1.194
+50% 4.990 0.016 0.634 7.780 1378.933 3.998 1.074
Random
+75% 4.393 0.019 0.737 6.739 1178.359 3.394 0.964
+100% 4.000 0.027 0.764 6.080 1044.279 2.988 0.881
Table 4.19: Comparison of evolution strategies for medium voltage network.
egy that to a certain extent is very similar to the triangle closure. This strategy is
the worst concerning the path length, having a characteristic path length more than
double than the best evolution strategy (i.e., random); the same also applies for the
AP L10th . Concerning redundancy aspects at the end of the evolution process, the
least distance strategy is the one (except for random) that has the best robustness
results. In terms of metrics related to betweenness, the least distance strategy shows
results that are even worse compared to the triangle closure one. Considering the
high degree assortative evolution strategy the results show that, overall, it is the
best evolution strategy in the first two steps of the evolution; in the first step, it is
even better than the random one (with the exception of robustness). Considering
betweenness metrics, this strategy is second only to the random strategy, with the
exception of coefficient of variation which is worse than all other metrics. In gen-
eral, this is a good strategy for evolving the network especially concerning the path-
related aspects, with just robustness that lacks behind compared to other evolution
strategies. The assortative low degree strategy is not particularly appealing and it
is outrun by the assortative high degree strategy in every metric except robustness
and coefficient of variation of betweenness, therefore we do not find it particularly
interesting for evolution purposes of the medium voltage networks. The dissorta-
tive strategy does not excel particularly in any of the metrics considered, however
its values are quite fair, especially in the initial stage of the evolution. In particular,
since the values of the metric do not improve substantially in the following steps of
the evolution process, such evolution strategy could be used as a slightly more ro-
bust alternative compared to the assortative high degree in those scenario where the
number of edges to be added is minimal and there is no need of special excellence in
one topological parameter. When enough connectivity is added, the evolution strat-
egy that scores best compared to all the others for all the metrics, with the exception
of the clustering coefficient metric, is the random one. It has already been noted by
Casals and Murtra [183] that some randomness in the network is beneficial espe-
cially for aspects related to robustness. In our analysis, we have the same general
results, with characteristic path length about 4 and a AP L10th of just 6. In addition,
the networks evolved according to this strategy are the most robust with a value
higher than 0.7 when the maximal connectivity is reached. The same considerations
apply for betweenness metrics that obtain the best results when the connectivity is
enhanced reaching a betweenness to order ratio smaller than three and a coefficient
of variation below one at the last evolution step. However, it is difficult to propose
for a distribution provider to improve his grid in a random fashion, even if part of
the weaknesses and inefficiencies come from the lack of such randomness. Consid-
ering rational and evolution strategies that come with a motivation we consider the
assortative high degree strategy as the one that, by evaluating the various topolog-
ical metrics, scores best in the evolution tests that we have performed on the Dutch
medium voltage grids.
Comparison of the evolution strategies for the low voltage distribution grid
As for the medium voltage evolution strategies, one cannot declare a winner. Also
in this case some improvements between the various phases of evolution are not
linear, therefore a strategy that scores best in one evolution step could result worse
compared to another one when more edges are added. We point out once again
that the addition of links is beneficial to the topological metrics analyzed and one
might speculate that the optimal solution to minimize (or maximize) the metrics is
to realize a fully connected network. With a real infrastructure such as the power
grid this is impossible to achieve due to economic and technical considerations.
The comparison of the averages over all samples in each evolution step for the
Evolution CPL CC Rob. 10th Avg. Avg. Coeff.

step red. bet. bet./order var.
path
+25% 3.600 0.062 0.408 7.535 207.959 2.515 1.518
Assortative +50% 3.225 0.157 0.432 5.797 191.974 2.227 1.564
high +75% 2.973 0.234 0.457 5.059 182.954 2.043 1.678
degree +100% 2.729 0.277 0.481 4.977 174.593 1.830 1.772
+25% 3.866 0.057 0.405 8.865 226.416 2.869 1.436
Assortative +50% 3.074 0.127 0.446 6.310 188.824 2.093 1.746
low degree +75% 2.827 0.223 0.509 5.187 177.928 1.825 2.113
+100% 2.766 0.277 0.551 4.730 175.312 1.765 1.939
+25% 5.908 0.131 0.346 9.028 488.127 4.599 1.202
Triangle +50% 5.463 0.250 0.383 8.070 461.893 2.044 1.247
closure +75% 5.020 0.355 0.456 6.973 419.114 1.850 1.310
+100% 4.686 0.462 0.497 6.573 392.781 1.732 1.428
+25% 3.396 0.093 0.403 7.923 194.838 2.413 1.849
+50% 2.826 0.177 0.435 6.168 167.339 1.875 2.318
Dissortative +75% 2.616 0.214 0.465 4.924 150.914 0.667 2.459
+100% 2.476 0.232 0.501 4.618 140.275 0.619 2.502
+25% 3.748 0.039 0.440 8.404 219.250 2.650 1.159
+50% 3.171 0.095 0.583 6.323 172.730 2.113 1.054
Random
+75% 2.841 0.141 0.636 5.450 149.836 0.642 1.038
+100% 2.621 0.156 0.689 4.831 132.417 0.567 0.978
Table 4.20: Comparison of evolution strategies for low voltage network.
various evolution strategies is shown in Table 4.20. Once again, the strategy that
scores best in comparison with the others is the random one for robustness and for
metrics related to betweenness. For robustness, in the last stage of evolution the
value reaches on average 0.7. For betweenness-related metrics the two final stages
of evolution are the best ones for average betweenness and average betweenness to
order ratio. This strategy is the only one for which the coefficient of variation results
below one (on average for all the samples) in the final stage of evolution. How-
ever, the random strategy has its weak point in the clustering coefficient metric that
score below all the others evolution strategies. Considering the assortative strate-
gies, one sees that the two strategies have quite similar scores for the metrics. In the
very first stage, the high degree assortative strategy is better, but later in the evo-
lution the assortative low degree slightly outperforms the assortative high degree
one for clustering coefficient and characteristic path length, while for the robustness
the difference is limited. The results between these two metrics for betweenness are
quite similar and here the only interesting difference is in the coefficient of varia-
tion which scores best for the high node degree strategy. These two strategies are
comparable to the random one for characteristic path length matters, while scoring
worse for robustness, but better for clustering. Considering the clustering coefficient
metric alone, the strategy that outperforms the others is the triangle closure strategy
with values that are double compared to the others at the end of the evolution pro-
cess. However, this strategy is worse than all the others for the characteristic path
length. Although it has the highest values for average betweenness, the coefficient
of variation ranks second compared to the other strategies considered. The strategy
that slightly outperforms the others (except the values for robustness of the random
one) is the dissortative strategy. In fact, the path length is just smaller than 2.5 even
better than the random edge addition. The clustering coefficient is in line with the
assortative strategies and scores around 0.5. Concerning betweenness, this strategy
scores best considering the average betweenness, but for the coefficient of varia-
tion this strategy scores worst. In the low voltage distribution grid it is difficult to
propose a strategy to follow for the evolution given the similarity for the values of
the evolution strategies considered. However, considering non-random evolution
strategies we consider the dissortative strategy a good candidate since scores best in
the evolution tests that we have performed on the Dutch low voltage grids.
Parameters satisfaction for medium voltage networks
We now consider the satisfaction of the quantitative metrics to analyze the appro-
priateness of synthetic topologies to improve the grid in reducing losses, facilitate
local energy distribution, and increasing network robustness.
Assortative high degree The assortative high degree evolution strategy satisfies
the desired values for the metrics concerning the characteristic path length already
from the second step of evolution for ten out of twelve samples. All samples satisfy
the requirements over the redundant path. Considering the clustering coefficient,
the majority of samples satisfies the requirement at the third step of evolution where
seven samples have a CC ¥ 5 CCRG ; in the final stage of evolution this metric is
satisfied by all the samples except two. The real drawback of this metric is repre-
sented by the robustness that never reaches the goal of 0.45, but stops around 0.37.
The unsatisfaction of the metric is also present for betweenness ones: all samples
are one unit larger than the target even in the last step of evolution. In addition, the
coefficient of variation never reaches the target for the samples.
Assortative low degree The assortative low degree evolution strategy satisfies the
desired values for the metrics concerning the clustering coefficient already from the
second step of evolution with eleven out of twelve samples. All the samples satisfy
the requirements over the redundant path already from the first step. Considering
the clustering coefficient, the majority of samples satisfies the requirement at the
second step of evolution where seven samples have a CC ¥ 5 CCRG ; in the final
stage of evolution this metric is satisfied by all the samples except one. This strategy
almost provides the satisfaction of the robustness requirement from the third step
on by having six samples fully compliant and three very close to the 0.45 threshold.
The metrics concerning betweenness are not satisfied both for betweenness to order
ratio and for the coefficient of variation which never reach the target for the samples
at any stage of evolution.
Triangle closure The triangle closure evolution strategy focuses on the clustering
coefficient, therefore reaching the target already in the first step of the evolution.
The evolved graphs do not only satisfy the requirement posed of having CC ¥
5 CCRG , but also a more restrictive requirement of CC ¥ 10 CCRG which
can be considered the condition for satisfying the small-world requirement for this
property cf. [222]. Concerning the path length properties, the characteristic path
length is never lower than the logarithm of the order of the graph. The requirement
over the AP L10th is satisfied already from the first addition of edges. On all other
requirements this strategy is weak. For robustness just few samples reach values
around 0.4, while the majority is about 0.3 when the most of the edges are added.
For metrics that involve betweenness, this evolution strategy scores poorly and it
is not close to the target for both average betweenness to order ratio and for the
coefficient of variation for any single sample.
Dissortative node degree The dissortative node degree strategy scores quite poorly.
As mentioned before, this strategy does not excel in one specific metric, but all the
metrics are slightly improved. Such aspects result in a limited crossing of the thresh-
old for the desiderata parameters imposed. Only after the third evolution step the
characteristic path length is almost satisfied: nine of the twelve samples reach the
target, while one more reaches the target when even more edges are added. A simi-
lar condition is true for the robustness metric. In fact, six samples satisfy the target
after the third step and other three are above the 0.4 value. The clustering coefficient
is always below the threshold, having values that often are slightly higher than the
clustering coefficient of a Random Graph with same order and size. The only met-
ric fully satisfied is the one related to redundant paths in the network. While the
metrics related to betweenness are not satisfied even at later stages.
Least distance The results of the least distance strategy are similar to those of tri-
angle closure. The clustering coefficient desiderata is reached already in the first
step of the evolution. From the second step on, the evolved graphs do not only sat-
isfy the requirement posed of having CC ¥ 5 CCRG , but also a more restrictive
requirement of CC ¥ 10 CCRG which can be considered the condition for satis-

fying the small-world requirement for this property cf. [222]. Concerning the path
length properties, the characteristic path length is never lower than the logarithm
of the order of the graph. However, the easy requirement related to the AP L10th
is satisfied already from the first addition of edges. Robustness is another positive
note of this evolution strategy. From the third step on the seven samples reach the
0.45 target and other three are close to 0.4; in the last step basically all the medium
voltage evolved samples satisfy the desiderata. The situation is not so positive for
betweenness related metrics since they score worse compared to the others strate-
gies for both average betweenness to order ratio and for the coefficient of variation.
Random The random evolution strategy is the one that satisfies most of the desider-
ata parameters not only after the first step of evolution, but already after the second
one. The metrics concerning the characteristic path length are satisfied basically al-
ready from the second step of evolution. The same applies to the redundant path
whose goal is met already in the first evolution step. Robustness is really impres-
sively achieved after the addition of 50% of more edges, with all the samples above
the 0.45 threshold and the majority of them scoring even higher, above 0.6. The
metrics related to betweenness that usually fail for the other samples are here met
completely on the coefficient of variation side after the third evolution stage (eight
out the twelve samples meet the target). In addition, this evolution strategy is the
one that also goes closer to the satisfaction of the betweenness to order ratio. The
only problem of such a strategy is on the clustering coefficient side. With such a
strategy the formation or closure of topological triangle structures is very unlikely.
Parameters satisfaction for low voltage networks
We now consider the satisfaction of the quantitative metrics for the low voltage
evolved samples.
Assortative high degree The assortative high degree evolution strategy almost
satisfies the desired values for the metrics concerning the characteristic path length
from the second step of evolution with eight out of eleven samples; while full com-
pliance is in the last evolution step. However, the requirement for the redundant
path are never fully satisfied. Considering the biggest samples, they never satisfy
the condition CC ¥ 5 CCRG . Also robustness requirement are not satisfied, only
six samples in the last two evolution stages reach the goal of 0.45. The between-
ness requirement is only partially satisfied: eight out of the eleven samples have
a betwenness to order ratio around the target of 2.5. Considering the coefficient of
variation, only the two smallest samples satisfy this property.
Assortative low degree The assortative low degree evolution strategy satisfies or
is close to satisfying the metrics in the last stage of the evolution process. For char-
acteristic path length, already from the second step, ten of the eleven samples reach
the target; on the contrary, the target is never reached for the redundant path re-
quirement. The three biggest samples score sufficiently concerning the clustering
coefficient requirement, while the others have high values, but never satisfy the
high-demanding requirements. Robustness is close to satisfaction in the last stage of
evolution where all samples reach the target except two that stop to a value around
0.4. The metrics concerning betweenness show an almost satisfaction of the average
betweenness to order ratio with eight of the samples well below the target; however
the coefficient of variation never reaches the target for the samples at any stage of
evolution.
Triangle closure The triangle closure evolution strategy focuses on the clustering
coefficient, therefore reaching the target already after the second step of the evolu-
tion. Two of the three biggest samples satisfy also the small-world requirement for
this property cf. [222]. Concerning the path length properties, the characteristic path
length is never lower than the logarithm of the order of the graph. However, the re-
quirement over the AP L10th is satisfied already after the second step of evolution.
In fact, the redundant paths are smaller than twice the characteristic path length.
For robustness, seven out of the eleven samples reach the target in the last step of
evolution. For metrics that involve betweenness, only seven samples are below the
target for the betweenness to order metric, actually the smallest in order, while the
three biggest samples are far from the target. The coefficient of variation increases
in the evolution.
Dissortative node degree The dissortative node degree strategy scores quite well
for evolving low voltage networks. Especially in the last stage of evolution half of
the metrics are satisfied or very close to satisfaction. Already in the second step of
evolution, the characteristic path length is satisfied; actually already in the first step
of evolution eight samples satisfy the condition. Robustness is satisfied with nine
out of the eleven samples that fully comply with the desiderata and the other two
that are above 0.4. The real weak point is the clustering coefficient which even for
the biggest samples is never close to the target. Betweenness-related metrics have a
dissimilar behavior: the betweenness to order ratio satisfies the desiderata for nine
samples, while the coefficient of variation has an increasing trend and in the final
step does not comply with the requirements for any sample.
Random The random evolution strategy proves to be one of the strategies that
satisfies most of the parameters. The metrics concerning the characteristic path
length are satisfied basically already from the second step of evolution with ten of
the eleven samples compliant. A different behavior applies to the redundant paths
whose goal is not met, but in the final step eight of the eleven samples reach the
target. Robustness is almost achieved already after the second step: all the samples
score above 0.4 and just three do not reach the target in that step, but later in the
evolution. In the last step, the satisfaction of the redundant path is almost satisfied
with eight of the samples (containing also the three biggest ones). The metrics re-
lated to betweenness that usually fail for the other types of evolution are here almost
entirely met concerning the coefficient of variation and the betweenness to order ra-
tio in the last stage of evolution. The main drawback is on the clustering coefficient
side. With such a strategy the formation or closure of topological triangle structures
is difficult.
Discussion
Considering the medium voltage grid, the random evolution strategy is the one that
satisfies most of the desiderata requirements (three parameters fully satisfied and
one almost satisfied) already at the second stage of evolution and from the third
step on, four requirements are fully satisfied. Concerning the other strategies that
have an evolution with a specific goal, those that satisfy most of the requirements are
the assortative high degree and the least distance one. Both strategies satisfy three
out of the six parameters and the difference being that the assortative one has good
performance for the characteristic path length parameters, while the least distance
strategy reaches the target for the robustness aspects. Therefore, these sub-optimal
strategies could be used where such different requirements are most needed.
From the comparison of the different evolution strategies and their steps in adding
new edges, one notices that there are evolution strategies that tend to satisfy the ma-
jority of the metrics for the smart grid that we have defined (c.f. 4.1). The evolution
strategy that already in the second step fully satisfies three requirements is the ran-
dom evolution. The requirements satisfied become four when 75% of additional
edges are added. This strategy is in line with the finding of Section 4.2 where a
small-world model was the best solution for modeling a smart grid topology. The
addition of random edges goes into that direction: the network with the rational
structure planned by the power engineers is modified by the addition of random
links, thus in something between the rational topology and the fully random net-
work. The other strategies that are quite successful are the assortative high degree
and the least distance, but only when the edges are doubled. This is the situation
considering a pure topological decision. When also the costs of the evolution are
taken into the picture the optimal evolution strategy might change due to economic
constraints. The extended analysis that considers economic aspects is performed in
Section 4.3.3.
Considering the low voltage grid, the random evolution strategy is the one that
satisfies most of the desiderata requirements at the final stage of evolution (two pa-
rameters fully satisfied and three almost satisfied). Concerning the other strategies
that have an evolution with a specific goal, those that satisfy most of the require-
ments are the assortative low degree followed by the dissortative one which almost
satisfy four and three requirements respectively. Therefore these sub-optimal strate-
gies could be used too. In particular, the assortative low degree satisfies the re-
quirements regarding the clustering coefficient, that are not satisfied neither by the
random, nor by the dissortative strategy.
A commonality between the evolution for medium voltage and low voltage is
the best result achieved by the random evolution of network that scores best among
the strategies considered in this study. An interesting difference lies in the sub-
optimal strategies that score high for the medium voltage and low voltage network
evolutions. For the medium voltage network, it is best to provide more connectivity
between those nodes that already hold a high node degree, thus reinforcing their
role as key components of the network. On the other hand, for the low voltage net-
work, it seems that adding more connectivity between the nodes that have a small
connectivity provides better performances. The assortative low degree aims at cre-
ating more connections and (therefore hubs) where is less connectivity in the current
samples. Also the other strategy that is sub-optimal for low voltage networks, the
dissortative, aims at giving a more important role to the nodes in the periphery of
the network by connecting them to the more connected nodes. A synthesis of the
topological performance of the different strategies is provided in Table 4.21. For
the layer of the power grid considered (column one), each strategy (column two)
is assessed with an optimality level (column three) based on the satisfaction of the
topological metrics. The step during the evolution process in which the most of the
metrics are satisfied is also provided in column four. The achievement of good re-
sults depends on the network layer and on the strategy used. The optimality level
is assigned based on the full satisfaction of the topological metrics described in Sec-
tion 4.1. One can see a distinction between the medium voltage and the low voltage
networks: the low voltage achieve less topological optimality (less metrics are satis-
fied) although requiring on average more evolution steps. Another aspect to note is
that some evolution strategies (i.e., assortative high degree and triangle closure) do
not have an improvement between the various steps despite the addition of more
connectivity. On the other hand, other strategies (i.e., dissortative and random) ben-
efit more from additional connectivity by improving the satisfaction of the metrics.
Network Strategy Optimality Optimality

level reached at step
MV Assortative HD 4
MV Assortative LD 2
MV Triangle closure 1
MV Dissortative 1
MV Least distance 4
MV Random 3
LV Assortative HD 3
LV Assortative LD 4
LV Triangle closure 3
LV Dissortative 4
LV Random 4
Table 4.21: Parameter optimal satisfaction.
4.3.3 Economic considerations

We need to characterize the cables used for the upgrading process in terms of their
physical properties and costs, as we did in Section 4.2.3. The information of length
for each cable in the existing sample networks is available. For the majority of the
nodes in the samples belonging to the medium voltage we know the geographi-
cal coordinate information, so with some approximation it is possible to compute
the length of the new cable connecting nodes that do not have a line connecting
them yet. Therefore, each new added line is assigned a length. In order to define
the properties characterizing the new cables, we adopt a k-nearest neighbor (KNN)
classification based on the length of the cable [52]. The classification is performed
using the Java Machine Learning Library (Java-ML) v0.1.53 . For each new cable, we
consider the m cables in the original network used as the training set for the classifier
that have the length closer to the added one and then the type of cable is provided
by the classification algorithm. We choose this method to identify the cable since it
is simple and it has proven successful in several practical contexts [188]. As the k
parameter we choose k 5 being a compromise towards speed of the algorithm ex-
ecution in the trade-off between precision of the result and speed to achieve it. Once
3 http://java-ml.sourceforge.net/
the properties of the new cables are identified, all the information we need for the
economic analysis are then available: resistance per unit of length, cost per unit of
length, maximal supported current. We remark that our proposed analysis does not
aim at being a comprehensive investment analysis, for which other techniques are
well established, simply an economic evaluation of the proposed evolution strate-
gies to confirm their feasibility or unfeasibility.
To assess these costs in the medium and low voltage infrastructure, we consider
the simple relation of Equation 4.1. As stated before, in this work we provide an
initial estimate and simply consider cabling costs and ignore substation ones. In
addition, the economic analysis can be applied only to the medium voltage network
samples since there is no geographical information about the location of the nodes
in the low voltage networks.
For the cost analysis, in the medium voltage case we limit our investigation to
the three strategies that have scored best in the pure topological analysis (i.e., ran-
dom, assortative high node degree and least distance). The results of the cost for the
evolution of the Dutch samples are shown in Tables 4.22, 4.23, and 4.24. The first
column of each table contains the sample ID, while the second provides the infor-
mation about the evolution step considered, the third column has the information
concerning the cost of the evolution of the network according to the specified strat-
egy. The fourth column contains the information on the fraction of the cost that the
evolution of the infrastructure impacts on the whole cost of the infrastructure. The
most interesting result is the difference in the cost that the three methods of evo-
lution require. The costs of network improvement is similar when considering the
assortative and the random strategy: the cost of adding more edges is on average
about 75% of the cost for cabling of the whole grid infrastructure. The situation is
radically different for the least distance strategy whose development impacts only
marginally in the total cost of the infrastructure, by adding just less than 13% of the
cost of the infrastructure. Purely from the point of view of cabling costs, the most
promising evolution strategy in economic and topological terms for the medium
voltage grid is the strategy that connects the nodes that are geographically closer
(i.e., least distance strategy).
One may then wonder if such investments are beneficial for the end users and
the distribution companies in reducing the cost of electricity flows. We resort to a set
of metrics that we have developed and already applied to the Northern Netherlands
distribution grid and to synthetically generated networks. The goal is to consider
those measures that are critical in contributing to the cost of electricity as elements
in the transmission and distribution networks.
Evolution step Evolution cost (euro) Fraction of evolution

cost on the whole infrastructure
+25% 59334032 0.73
+50% 59334032 0.73
Sample #1
+75% 59334032 0.73
+100% 119505699 0.85
+25% 107698129 0.73
+50% 225049864 0.85
Sample #2
+75% 255361286 0.87
+100% 298792505 0.88
+25% 23007544 0.72
+50% 44739934 0.83
Sample #3
+75% 72515143 0.89
+100% 105567690 0.92
+25% 45762304 0.65
+50% 55159897 0.69
Sample #4
+75% 55159897 0.69
+100% 55159897 0.69
+25% 65393631 0.74
+50% 117984090 0.84
Sample #5
+75% 157876214 0.87
+100% 194838593 0.90
+25% 31063470 0.67
+50% 62971658 0.81
Sample #6
+75% 104711687 0.87
+100% 165938342 0.92
+25% 79148152 0.61
+50% 199600692 0.80
Sample #7
+75% 229365781 0.82
+100% 350782953 0.87
+25% 32826377 0.67
+50% 36976365 0.70
Sample #8
+75% 83305090 0.84
+100% 123241632 0.88
+25% 11739091 0.46
+50% 42894090 0.76
Sample #9
+75% 72834999 0.84
+100% 103147797 0.88
+25% 14899834 0.60
+50% 32136672 0.77
Sample #10
+75% 45646927 0.82
+100% 60546613 0.86
+25% 13094364 0.50
+50% 26377193 0.67
Sample #11
+75% 38709007 0.75
+100% 40774034 0.76
+25% 93369553 0.70
+50% 93369553 0.70
Sample #12
+75% 138091611 0.77
+100% 253056714 0.86
Table 4.22: Cost of medium voltage network evolution for assortative high degree
strategy.

+25% 49585634 0.70
+50% 99550100 0.82
Sample #1
+75% 147237698 0.87
+100% 193060613 0.90
+25% 81443232 0.67
+50% 173835139 0.81
Sample #2
+75% 270738228 0.87
+100% 352561226 0.90
+25% 20401026 0.70
+50% 39401799 0.82
Sample #3
+75% 58698139 0.87
+100% 80038042 0.90
+25% 38550574 0.61
+50% 74405969 0.75
Sample #4
+75% 112507292 0.82
+100% 144092688 0.85
+25% 34665986 0.60
+50% 63522559 0.74
Sample #5
+75% 90171964 0.80
+100% 118746816 0.84
+25% 32756917 0.68
+50% 67956839 0.82
Sample #6
+75% 102189138 0.87
+100% 140632238 0.90
+25% 85124154 0.63
+50% 181095750 0.78
Sample #7
+75% 262020773 0.84
+100% 352245673 0.87
+25% 40434688 0.72
+50% 79435946 0.83
Sample #8
+75% 107843650 0.87
+100% 139828749 0.90
+25% 31096020 0.70
+50% 57036433 0.81
Sample #9
+75% 85190089 0.86
+100% 115442294 0.89
+25% 12677249 0.56
+50% 26126828 0.73
Sample #10
+75% 42325971 0.81
+100% 53357492 0.84
+25% 1214608 0.08
+50% 2553383 0.16
Sample #11
+75% 3117824 0.19
+100% 4494600 0.25
+25% 105240341 0.72
+50% 223059810 0.85
Sample #12
+75% 317337397 0.89
+100% 417627345 0.91
Table 4.23: Cost of medium voltage network evolution for random strategy.

+25% 635014 0.03
+50% 1944353 0.08
Sample #1
+75% 3536060 0.14
+100% 5539077 0.20
+25% 1832132 0.04
+50% 5362070 0.12
Sample #2
+75% 9454081 0.19
+100% 15488444 0.28
+25% 357093 0.04
+50% 1143678 0.11
Sample #3
+75% 2075073 0.19
+100% 3206160 0.26
+25% 831466 0.03
+50% 2936596 0.11
Sample #4
+75% 5026907 0.17
+100% 7688212 0.24
+25% 801764 0.03
+50% 2128363 0.09
Sample #5
+75% 3412290 0.13
+100% 5187261 0.19
+25% 543105 0.03
+50% 1433992 0.09
Sample #6
+75% 2588723 0.15
+100% 3822329 0.20
+25% 1104605 0.02
+50% 3521507 0.07
Sample #7
+75% 6534753 0.11
+100% 10042275 0.17
+25% 584039 0.03
+50% 2127943 0.12
Sample #8
+75% 4016394 0.20
+100% 5840352 0.27
+25% 661698 0.05
+50% 1861427 0.12
Sample #9
+75% 3476464 0.20
+100% 5524643 0.29
+25% 314451 0.03
+50% 987993 0.09
Sample #10
+75% 1971402 0.17
+100% 2912718 0.23
+25% 478534 0.03
+50% 1230921 0.09
Sample #11
+75% 2224697 0.14
+100% 3262358 0.20
+25% 1069466 0.03
+50% 3143898 0.07
Sample #12
+75% 5723602 0.12
+100% 8686081 0.18
Table 4.24: Cost of medium voltage network evolution for least distance strategy.
(a) Losses-related metrics. (b) Reliability-related metrics.

Figure 4.11: Comparison to the original networks (fraction of the original) of proper-
ties influencing electricity cost at 1st stage of evolution (i.e., +25% edges).

ties influencing electricity cost at 2nd stage of evolution (i.e., +50% edges).

ties influencing electricity cost at 3rd stage of evolution (i.e., +75% edges).

ties influencing electricity cost at 4th stage of evolution (i.e., +100% edges).
Figures 4.11, 4.12, 4.13, and 4.14 show the improvement as a fraction of the orig-
inal values in the metrics related to the cost of electricity distribution respectively
in the four step of evolution we consider (+25%, +50%, +75%, and +100% of edges)
compared to the original samples. To give a general idea of the improvement, we
use the average results over the 12 medium voltage networks samples. For each
strategy, we look at the improvement, in percentage, for the same metrics compared
to the initial samples. Already from a small increase in connectivity (addition of
25% of edges) the resistance of a path decreases with every strategy with the best re-
sult obtained by the assortative high degree strategy. Adding more lines in general
promotes a reduction of the average resistance of edges compared to the initial situa-
tion. This is even more true in the assortative and least distance strategy where short
distance connections (and therefore less resistive) tend to dominate or more efficient
cables are chosen in the KNN procedure selection. In addition, a considerable reduc-
tion in the losses that are experienced in traversing substations can be avoided with
more dense networks: almost a 30% less station traversed in the graph enhanced
with assortative connections. Of course less benefits take place with the least dis-
tance strategy that reduces the average number of traversed stations in a path by 7%.
This higher connectivity provides benefits to the robustness aspects: resilient paths
(i.e., 10th shortest weighted path) that are less lossy (about 30% reduction for the
least distance strategy). An improvement is also experienced in the resilience of the
network to node disruptions: about 80% more resilience with the random strategy.
Also the amount of current that is supported in a path (the weighted characteristic
path length with maximal supported current as weight of the edges) is definitely
higher (more than 2.5 times for the least distance strategy) compared to the initial
samples.
Similar considerations can be done for the second evolution step (i.e., +50% ca-
bles). Considering the loss aspect the benefits are for the three strategies on average
about 30% in the reduction of the weighted path. For the average weight (i.e., re-
sistance) of a cable the random addition creates networks with much more cables
with an high resistance. This is an indication that the random strategy creates long
distance connections that act as shortcuts in the network from a topological point
of view. However, in a physical system there are no significant benefits since the
resistance grows with the distance. These limits of the benefits are shown by the
weighted path analysis. Considering robustness oriented metrics, the most of the
benefits take place in resilience where the networks on average stay twice as much
as connected than the initial evolution step, except for the assortative strategy that
lags behind. We see a small decrease for all the strategies in the characteristic path
when the weight is the maximal current that can flow in the cables; this is not dif-
ficult to understand since the characteristic path uses the most efficient (i.e., with
smallest weight paths).
When even more connectivity (+75% and +100%) is added one sees even better
results and improvements but smaller in their magnitude. In these last two stages,
the benefits for the losses reach values of reduction above 30%, the same for the
nodes to be traversed on average in a path. Only the least distance strategy has a
small reduction, but this is due to the very nature of such evolution strategy that
avoids long distance (topological shortcuts) cables. Considering network robust-
ness, the main improvements are for resilience to node failures that for the least
distance strategy allow the network to be twice as much as robust than the initial
samples. Concerning the capacity of the network, we see a slight decrease for the
same reasons as above. However, the networks with the least distance evolution are
able, on average, to transport twice the amount of current than the initial samples.
We can now provide a recommendation on how to evolve the analyzed samples
given the cost analysis for cable addition and the benefits that such an higher con-
nectivity brings in the topological aspects related to the price of electricity. Using
our topological-based method, we consider that the least distance strategy with the
addition of 75% or 100% of edges is a very good way of making the network more
connected. Such an evolution strategy provides benefits from a topological point of
view, and it keeps the costs low. For the samples examined, it requires an investment
in cabling costs about 25% of the value of the cables already on the ground. Such
investment can provide benefits in the loss reduction of the network as well as in its
robustness. The energy economic studies show [84, 141] that losses and robustness
are directly related to the cost of electricity. These factors are tight to topological
parameters as shown above, therefore such evolution would provide less cost in
electricity distribution. It is difficult to translate in exact monetary terms for the end
user concerning the savings in the energy bill that an improvement in the topology
parameters that influence the cost of energy distribution might bring. For the real-
4.4. New Topologies for the Smart Grid 123
ization of a smart grid less costs in the distribution of electricity and a local reliable
and robust network are the essential ingredients to enable a paradigm where energy
is produced and distributed locally such as at neighborhood or city level.
In this analysis of the influence of the topology on the price of electricity, we
have not used the α and β metrics themselves (cf. Section 3.4), but the components
of these metrics. In fact, especially for the α metric we see a decrease of the average
weight of the edges compared to the Dutch grid samples for when the evolution is
performed (exception is the random strategy where long, therefore high-weight con-
nections are present). The decrease in the average weight of edges is faster than the
decrease in the average path length, therefore the LlineN part of the α metric grows,
which is counterintuitive given the decrease that all the components characterizing
the losses show. Therefore, we have found more appropriate and meaningful for the
analysis of the evolution to consider the single components influencing losses and
reliability and not their aggregation.
4.4 New Topologies for the Smart Grid

In an evolving electricity sector with end users able to produce their own energy
and sell it on a local-scale market, the grid plays the essential enabling role of sup-
porting infrastructure. Local-scale energy exchange is in fact beneficial for several
aspects such as the increase in renewable-based energy production, the possibility
for the end user to have an economic contentment by selling surplus energy and,
not less important, a step forward to the unbundling of the electricity sector. We
studied how different topologies inspired from technological and social network
studies have different properties and can be (or not) suitable for the future smart
grid networks. We showed that between the various models analyzed, the small-
world model appears to have many supporting characteristics, according to a set of
topological metrics defined for power grids. We also showed how these topological
benefits can be related to economical aspects of electricity distribution.
We have also shown that different strategies to evolve the current physical grid
provide different cost/benefits ratios and have identified which strategies are mostly
promising. The statistical approach that we take, combined with traditional power
engineering approaches, can form the basis of a decision support system for evolv-
ing the current power grid into a “smart grid friendly” network.
We have seen that the random and assortative high degree evolution strategies
are the best in a topological weighted analysis, but they have costs that are extremely
high and unrealistic to realize in practice. In addition, the weighted analysis gave
us the possibility to investigate the elements of the power grid that influence the
cost of electricity distribution (i.e., grid losses and grid reliability) and the results of
this analysis suggest that evolving the network by adding connections between the
nodes with smallest distance is beneficial and provides in many cases better results
compared to the other evolution strategies. Therefore, with an investment about
25% of the actual costs in cables already on the ground the distribution grid can
improve consistently in reducing transportation costs.
From the industrial perspective, where a unique and clear definition of the term
smart grid [139] is missing and where the standardization process is at the early
stages of development, we consider that the present proposal is usful to make gen-
eral decisions on how to evolve the grid and what costs are entailed at least from
a coarse grain point of view. Existing planning techniques will have to be revised
in the future, especially for the distribution grid due to the presence of Advanced
Metering Infrastructure (AMI) (i.e., bidirectional intelligent digital meters at cus-
tomer location) and Distribution Automation (DA) (i.e., feeders can be monitored,
controlled in automated way through two-way communication). In addition, the
medium and low voltage grid will no longer be a layer where only energy is con-
sumed, but distributed energy generation facilities (small-scale photovoltaic sys-
tems and small-wind turbines) will be attached to this segment of the grid; alto-
gether these elements are likely to reshape the way planning for medium and low
voltage is realized [31] and will also call for new instruments such as the one we
propose here.
Chapter 5
ICT Services and Applications for the Smart
Grid
One of the main innovation of the smart grid will be to bring more Information
Information Technology (ICT) capabilities and computation in the power domain,
especially in the low layers of the grid and at the consumer facility.
We consider four different aspects to show how the ICT is important for the suc-
cess of the smart grid and in which scenarios it is a key component. We consider the
benefits and necessity of a software approach based on Software-Oriented Archi-
tectures (SOAs) in Section 5.1. We describe in Section 5.2 an application of services
for the realization of demand-response functionalities and an implementation with
today’s technologies and actual data services. We further describe how these smart
grid services can be used in a real office environment to realize a demand-response
solution with the goal of minimizing the energy expenditure while keeping users’
satisfaction (Section 5.3). We conclude the chapter with another application of the
smart grid enabled energy market where users provision energy through agent tech-
nologies and automated ICT platforms.
5.1 The Services: the Future Energy Landscape

The vision of the future smart grid, however defined, will bring unbundling of en-
ergy markets, appearance of renewable generating facilities at all scale levels, the
standardization of the control elements of the power grid, the diffusion of digi-
tal/data prone smart meters, and especially two way communication among end
users and the grid. This will be a reality only if there are infrastructures that can
support it especially from a communication and information point of view.
5.1.1 Smart grid challenges

A number of challenges needs to be addressed for building such a grid. We identify
the following ones as the key issues to be solved at the software level.
126 5. ICT Services and Applications for the Smart Grid
• Interoperation. The number of actors populating the energy market is con-

stantly increasing and their capabilities as well. There is a strong need to have
standards for interoperation at all levels (not only the control layer of the grid).
Furthermore, standards tend to cover the syntactic part of the interoperation,
while the semantics of the message exchange is scarcely addressed.
• Scalability. The increase of actors also involves scalability issues. If millions of

micro energy producers start trading micro-quantities of energy, there must be
an appropriate infrastructure to manage this, possibly real-time, information
exchange.
• Discovery. If the actors increase and more entities can take on the same role,
one may think of discovering services on the fly. The idea of signing an yearly
contract for a home, may be too limiting and one may want to switch energy
supplier on a much shorter time frame. Furthermore, if anybody can be a
supplier, then one may want to find a provider in the moment the energy is
needed.
• Mobility. In the future, grid energy consumers, and also producers, may be
mobile on the grid. Cars will be electric, but may also have energy producing
and storing facilities (e.g., a solar cell roof, fuel cells powered engine). The
mobile elements need to interact with the power grid in a transparent way.
• Resilience to failure and trust. The electrical power grid is a critical infras-
tructure. A key performance indicator of the current energy distributors is the
down time that should not exceed the few hours per year. When moving to
an open smart grid the delivery of energy must not decrease in quality. This
requires having a trust mechanism among the various players. It may also
require having reliable forecasting of generation and use.
• Service integration and composition. The physical layer, the data layer and
the business layer will have to interact more closely. In fact, any node that pro-
duces energy needs to interact with control/actuation part of the grid and get
paid for the energy produced; the generation might be a part of a larger busi-
ness process relying on the energy (e.g., one could drive an electric car while
lodging at a motel, plug it into the grid and use the car generated electricity to
pay part of the bill [207]).
• Topology. The current infrastructure is highly hierarchical, not only in the

physical infrastructure, but also in the information systems that manage the
electricity system. Very few large energy producers and backbone owners ex-
ist, with few systems that control the plants and the transmission network is
5.1. The Services: the Future Energy Landscape 127
highly centralized. But the new vision of the open grid demands for a flat
peer-to-peer (P2P) network in which all actors are producers and consumers
of energy, data and services.
• Smart Meters. The smart meter is likely to become the “energy gateway”
of the house with more and more functionalities and embedded intelligence.
Smart Meters might work as automatic bidder on the energy market knowing
family energy usage and production patterns together with estimation meth-
ods based on past usage and environmental forecasts (e.g., future weather con-
ditions).
• Real-time. Energy related operation such as control, actuation, distribution

and trading have very strict time-dependent constraints to satisfy. All the
players in the next generation grid must interact following real-time commit-
ments to provide and receive an energy service with the proper quality.
5.1.2 Service-oriented architecture supporting the smart grid

Interestingly, the just listed challenges have a natural counterpart in the Service-
Oriented Architectures. These have been traditionally built to address interoper-
ability and scalability issues for the integration problem of enterprise information
systems (e.g., [172, 41, 171, 173]) or to support business process, especially across
companies borders. Here we take a different look and consider how SOAs are ap-
propriate for the smart grid and look at the challenges just introduced through the
glasses of service-orientation.
• Interoperation. Web services are a technology to build Service-Oriented Ar-

chitectures and address the problem of interoperation being standardized eX-
tensible Markup Language (XML) protocols to describe messages, remote op-
erations and coordination among loosely coupled entities, e.g., [120]. These
are already entering the energy sector as described in Chapter 2.
• Scalability. The basic SOA pattern: publish–find–bind allows to decouple

service consumers from producers and to substitute, even at run-time, one
component for another one. The communication, most often asynchronous,
provides all the ingredients for a highly scalable infrastructure. Examples of
which have already appeared in the area of eBusiness.
• Discovery. Discovery is one of the basic ingredients of a SOA. It needs to

support the publish and find operations and is usually based on registries, but
can also be realized with flooding models.
• Mobility. A SOA supports actor loosely coupling and behavior based binding,
therefore the mobility of the elements is easily supported, e.g., [1].
• Resilience to failure and trust. Protocols exist to enable a Web service based
SOA with trust, privacy and security support. This can provide the basic for a
secure infrastructure. Reliability will also have to be pursued with appropriate
energy technology which is beyond the SOA.
• Service integration and composition. Service integration and composition is
the key added value of a SOA and many examples exist on methodologies to
support this, e.g., [63, 2, 114, 20, 37].
• Topology. SOAs support any kind of topology. The hierarchical client-server
one is less common, but can be realized. The P2P topology is most often the
one realized.
• Smart Meters. In the SOA paradigm the smart meter is basically a service
provider and a service consumer at the same time. It invokes other services to
interact on the market and also provides services to other market participants
interested in energy purchase. It also interacts in a service-oriented fashion
with intelligent home appliances that require energy at a certain time.
• Real-time. Solutions are available to introduce enhancements to SOA paradigm
in order to provide an appropriate quality of service and satisfy real-time con-
straints, e.g., [204, 170].
5.1.3 Traditional service-oriented architecture vs. energy service-

oriented architecture
The smart grid is thus amenable to be supported by SOAs, though there are some
differences with traditional SOA approaches. Table 5.1 shows the main points of
contact and dissimilarity between the traditional SOAs and those for the energy sec-
tor. A natural common point is the use of SOA technologies for integrating hetero-
geneous systems thus enabling their interoperability. Beyond this common feature,
several differences lay that must be taken into account when dealing with energy
systems. Traditional SOA is mainly used in the business process domain managing
complex supply chain and interactions between a multiplicity of actors whose ap-
plications are usually triggered by specific events. Usually, the paradigm of these
interactions is asynchronous. On the other hand, the SOA for energy applications
must tackle some peculiarities of this type of business and systems. First of all the
requirements for real-time interactions between the various subsystems and compo-
nents of the energy-related ICT involve Supervisory Control and Data Acquisition
5.1. The Services: the Future Energy Landscape 129
(SCADA) and Energy Management System (EMS) systems and low level electric
applications embedded systems. Being these systems highly important and mis-
sion critical, real-time constraints together with fault tolerance, security and trust
mechanisms in the service provisioning are essential requirements that a SOA for
the electricity sector needs to satisfy. An interaction with non-strictly related en-
ergy systems such as Customer Relationship Management (CRM) and Enterprise
Resource Planning (ERP) is also required to have a complete interoperability pic-
ture.
Another issue that is central in enabling the SOA solutions for the smart grid
is of course an appropriate communication infrastructure. It is not the focus of the
present work, but it is worth to mentioning the adaptation required by the telecom-
munication/telecontrol infrastructure to support the enhanced amount of informa-
tion data and control signaling that the smart grid requires [226]. The telecommuni-
cation aspects and its infrastructure must not be taken for granted, since they form
the basis to build more complex service-oriented software layers on top.
Another aspect that is likely to appear in the new smart grid landscape is the
increase in the interactions that follow a P2P paradigm. In fact in a grid with many
more energy prosumers and the related information exchange, end users will be
directly involved in a fashion similar to the P2P paradigm in data exchange. Al-
though the high level picture might look the same, some important differences re-
main beyond the main commonality that is the scalability requirement. Traditional
P2P architectures are used in the framework of data exchange with requirements in
the infrastructure that lay between the static and dynamic solutions based on the
specific application. This aspect usually reflects on the network structure that the
P2P system acquires (i.e., structured, unstructured). The energy P2P infrastructure
has to deal with energy flows which imply, at least with actual technologies, the
extreme difficulty in energy storage and the seek of an energy balance equilibrium
all the time; on the other hand when dealing with P2P for information exchange,
data buffering and replication are easy and cheap options to increase system per-
formance. The dynamism in the energy paradigm is high when considering the
demand side, but it is almost static considering the locations of the user. The result
is a static network structure (an exception lies in the extensive usage of electric vehi-
cles that add dynamism also at the location aspects). Performance and security are
two key aspects that differentiate the data and energy P2P worlds: these properties
are tight to the specific application considered for the former, while for the latter
real-time requirements and high security are essential to deliver a valuable energy
service. The level of centralization in a P2P data environment is closely coupled
with the application it has to satisfy; in the energy environment the system tends to
evolve to an hybrid solution since generating companies that own power plants still
Traditional SOA Vs. Energy SOA

Similarities
System integration
Interoperation
Differences
Supply Chain management Interaction with non-Energy systems (e.g.,
ERP, CRM)
Event-based applications
Real-time requirement
Business process management
High security (authentication, encryption for
market trading)
Asynchronous long-running business process
Trust mechanism for services provided
and transactions

Interaction with low level electrical interface
standards (IEC standards for SCADA/EMS)

Fault tolerance
Table 5.1: Similarities and differences between Traditional SOA and Energy-Oriented
SOA
act as sort of super-peers in the system.
5.2 Today’s Services for Implementing the Smart Grid

The smart grid and the SOA-based vision described in Section 5.1 are far from be-
ing implemented. At most there are local and small scale testbeds (e.g., Hoogk-
erk’s PowerMatching city smart grid project [99]) where some features of the smart
grid are implemented. In addition, business analysis and business cases around
the smart grid are still under investigation [29, 147]; so is the standardization pro-
cess. The smart grid is still rich of uncertainties, however some functionalities are
assumed to be its pillar: demand-response through dynamic pricing, local energy
generation through renewables, and energy forecast to help schedule smart appli-
ances when more favorable. In such scenario, our aim is to design and realize a
smart grid Simulation Engine based on available Web resources. We have devel-
oped ad hoc interfaces to existing services on the Web (e.g., energy tariffs, meteo-
rological forecast for renewable energy) in order to realize the smart grid features
essential for a smart meter/smart home interaction. In particular, information re-
garding energy prices, energy generated, and environmental conditions is accessed
through Web sites or Web services. For each smart grid functionality such as energy
price, energy production metering and meteorological conditions, we describe the
solution we envision to be available in the future, but we also describe what can
be realized today with the currently available information. A recurrent aspect for
5.2. Today’s Services for Implementing the Smart Grid 131
the envisioned ideal solution is the use of SOAs. A typical implementation of SOA
is through Web services. An incarnation of Web services is through protocols like
the Hypertext Transfer Protocol (HTTP), the Simple Object Access Protocol (SOAP),
and through standardized service descriptions interfaces such as the Web Services
Description Language (WSDL). A summary of the vision for the smart grid services,
the standardization process involved, and the nowadays possible implementation
is shown in Table 5.2.
Category Vision Ongoing standardiza- Nowadays realization

tion effort
Energy price Third party Web ser- OpenADR, ZigBee Ad hoc data retrieval in real
vice providing prices Smart Energy Profile wholesale energy market
from several energy 2.0, OASIS EMIX, IEC-
providers 62325, direct interaction
with smart meter
Energy pro- Web service in the OpenADE, ZigBee Ad hoc interaction with produc-
duction/ control unit of the Smart Energy Profile ing/consuming device control
consumption small-scale produc- 2.0, IEEE 1547.3, direct and monitoring equipment, in-
tion plant interaction with smart teraction with smart meter of a
meter specific standard/technology
Environmental Dedicated Web Outcome of NIST B2G Interaction with existing
conditions service from meteo- DEWG, addition to weather Web services (e.g.,
rological companies existing weather stan- Yahoo!, NOAA, The Weather
specialized in energy dards e.g., Weather Channel)
with very localized Information Exchange
forecasting Model (WXXM)
Table 5.2: smart grid services: vision, ongoing standardization processes, and nowa-
days possible realization.
5.2.1 Interface towards energy price service

The availability of updated prices and price forecast within few hours or within few
days are essential to implement the smart grid demand-response functionality.
Vision
The envisioned solution to implement the price mechanism would be a Web service
provided by a third party for the smart grid. This party administers the energy tar-
iffs for the providers that the smart home is allowed (or wants) to supply from. The
smart home subscribes to this service indicating for which providers the tariffs are
required. Either in a pull or push method the smart home obtains the forecast tariffs
for different time granularities. The granularity can be real-time for balancing pur-
poses or hourly day-ahead forecast for mid/long term appliances schedule. The ser-
vice, in addition, signals through a push message a changing tariff compared to the
previous forecast so that energy intensive operations already scheduled in the house
can be adapted on-the-fly to the changing price conditions. In this ideal situation all
the interactions and information exchange take place automatically thorough the
World Wide Web and the smart meter. The OASIS Energy Market Information Ex-
change (eMIX) might be the protocol used to exchange information about energy
prices.
Realization today
The solution proposed above is ideal and not implemented at the moment. How-
ever, it is possible to have a similar behavior by interacting with wholesale energy
markets that already implement price differentiation in the energy trading process.
In particular, the prices generated on the wholesale market vary in accordance with
the congestion and energy availability on the power grid infrastructure. This is real-
ized to keep the balance between demand and supply, in other words the demand-
response mechanism. To have variable energy tariffs it is possible, for instance, to
use prices coming from the PJM Interconnection1 which is a Regional Transmission
Organization (RTO) that coordinates the movement of wholesale electricity in more
than 13 states of Eastern U.S.A., or from another organization such as Independent
System Operator New England (ISO-NE)2 which provides the energy market facil-
ities for the New England region of the U.S.A. For both markets the data that can
be extracted are the real-time prices and the day-ahead energy market locational
marginal pricing (LMP). These prices are respectively the live minute-by-minute
prices and the prices of energy negotiated in the wholesale market for the following
day by energy companies at specific locations where energy is delivered or received.
For the day-ahead market, data contain the energy price for each unit (dollars per
Megawatt-hour) for each hour of the day (for the next day) at the locations of de-
livery. Each location can have in principle a different price from any other one.
These prices can be considered for the smart home as the prices of different energy
providers. Real-time prices can be obtained by repetitively interrogating the PJM or
ISO-NE Web site where prices for the various location in the network are continu-
ously updated; whereas the day-ahead data can be automatically obtained through
PJM or ISO-NE Web site and they are available each day for the following day (day-
ahead) as a comma separated value (CSV) files. Little ad hoc adaptations are re-
1 http://www.pjm.com/
2 http://www.iso-ne.com/
quired to automatize the process of file download and cleaning of the unnecessary
fields, before they can be used as the input of a systems that provides dynamic pric-
ing for demand-response purposes. A feasible solution today is to use day-ahead
prices to forecast the schedule for smart appliances for the following day, and then
correct the scheduling in real-time if considerable difference in prices arise in the
real-time market.
5.2.2 Interface towards energy producing equipment service

Another essential feature for a smart home is the possibility of generating energy
through small-scale renewable plants (e.g., photovoltaic (PV) panels, small wind-
turbine).
Vision
The envisioned solution to implement information retrieval of the amount of energy

and power produced by a small-scale energy plant would be a Web service provided
by the control unit of the plant. The smart home, directly or through the smart meter,
could then in an automated way ask the information about the available instanta-
neous power and the amount of energy generated. Information about the generated
power could be easily and automatically reported via Web service interaction to the
utility company with whom the sell contract is signed if no Advanced Metering
Infrastructure (AMI) system is already implemented.
Realization today
Usually, every modern small-scale plant has an electronic controller that is able to
show information about power and energy produced together with essential techni-
cal and environmental parameters of the plant (e.g., for a photovoltaic plant: air tem-
perature, panel temperature, and solar irradiance). This information is usually pub-
lished on the Internet (or LAN) through the Web server embedded in the controller
of the plant so that the owner of the plant can monitor the performance. This infor-
mation can usually be directly accessed or downloaded in CSV format so with little
ad hoc manipulations these data can be used as input for an application. An exam-
ple is available at the U.S. School Power Naturally data portal.3 Through this portal
it is possible to access solar plant information and data. Another example of a Web
server is available at http://pagani.dyndns.org/html/en/onlineOverWr.
html, it provides real-time and historical information from a solar plant about the
3 http://sunviewer.net/portals/NYSERDA/
power and energy produced. While a comprehensive small-scale power generating

unit composed by a solar, wind turbine and small-scale combined heat and power
turbine plant is available at Grand Valley State University.4
5.2.3 Interface towards environmental service

External environmental and weather parameters are essential to know how external
conditions may influence both energy consumption inside the smart home and the
renewable-based generating equipment. The smart home has then to be able to
access current and forecast environmental conditions.
Vision
The envisioned solution to obtain environmental parameters would be a dedicated

Web service provided by a third-party specialized in environmental conditions fore-
cast for the smart grid with precise geolocalization. In addition to traditional weather
information (e.g., weather conditions, temperature, pressure), other information
such as solar irradiance, wind speed, humidity, cloud coverage, air density, exter-
nal light conditions and others are significant to estimate energy consumption of
buildings and to forecast energy production by small-scale energy equipment. This
information should be extremely localized and dedicated to the specific location of a
smart home. The envisioned service requires that the smart home subscribes to the
service by providing exact location (longitude and latitude coordinates) and orien-
tation towards the sun. Once subscribed, the service should be able to provide the
environmental parameters mentioned above. The service could be implemented in
a push and pull manner so that updates on the conditions and forecasts can be sent
to the smart home regularly, or it can request for an update when needed.
Realization today
On the World Wide Web there are several weather services, with advanced pieces
of functionality that can be accessed in the form of Web services. A primary exam-
ple, considering the levels of detail and abundance of weather-related parameters
(current weather, 7 day forecast and many parameters such as max and min temper-
atures, wind speed, wind direction, wind gust, percentage of sky cover and more),
is provided by the National Weather Service realized by the National Oceanic and
Atmosphere Administration (NOAA)5 in the U.S.A. From a technical point of view
the service provides an interaction through SOAP requests. Other services that use
4 http://datamonitoring.marec.gvsu.edu/
5 http://www.noaa.gov/
a Web service approach are provided by commercial services such as Yahoo!.6 This
weather-related information seems the closest to the ideal solution for this kind of
service required by the smart home in his interaction with the smart grid even if the
precise localization is not realized yet.
5.2.4 Implementation
Proof-of-concepts implementations of the smart grid, simulations of customer be-
havior or small scale pilot projects have been reported being underway [124, 202,
110], though no actual large implementation exists.
To be able to perform research and development related to smart grid, we have
developed smart grid services considering the case of a smart home located in New
York, U.S.A. with the ability of generating energy with small-scale production units
through a photovoltaic installation of 2.4 kW of power realized using an AstroPower
AP-100 PV module,7 and a small-scale wind turbine namely the Proven 2.5 wind
turbine.8
In our simple implementation scenario for smart grid interaction we decided
to use data and services coming from real markets, real energy installations, and
real weather information. Such choice enables us to realize a realistic simulation
of dynamic pricing functionality, energy production from small-scale renewable
sources, and energy forecast. A representation through a block diagram of the com-
ponents involved and implemented in software using JAVA programming language
is shown in Figure 5.1. The central component is the Smart Grid Simulation Engine
which is responsible for contacting the information providers. Its role is to establish
external connections, gather the data and convert them to the format required by
the internal sub-component that has specific tasks to further process those data.
Figure 5.1: smart grid simulation engine and external services.
6 http://developer.yahoo.com/weather/
7 http://atlantasolar.com/pdf/Astropower/ap-100.pdf
8 http://www.windandsun.co.uk/Wind/wind_proven.htm
To simulate the variable energy tariffs, we use the energy prices coming from the
PJM Interconnection and the data extracted is the Day-Ahead Energy Market LMP.
The Smart Grid Simulation Engine makes an HTTP request to the PJM Web site
where day-ahead LMP prices are stored as CSV files. The data contain the energy
price for each unit (dollars per Megawatt-hour) for each hour of the day (agreed for
the following day) at 20 locations of delivery. We consider that the energy supply
of a home could be fulfilled not just by one provider at a time, but also by the com-
position of several providers that might contribute to satisfy the power required by
a house (e.g., provider A is able to give 1 kW of power at tariff x and provider B is
able to give 2 kW of power at tariff x ∆ for a certain time period).
To account for the energy produced in the smart home, we access a real PV in-
stallation in New York at Dalton School in Manhattan9 that has the same PV array
hypothesized in our smart home and whose real-time data can be accessed through
the U.S. School Power Naturally data portal. The Smart Grid Simulation Engine
connects to the Web site10 providing the performance of the solar installation on top
of the school, and extracts the CSV file with real-time information about the power
generated by the PV array. Data is sampled every 15 minutes. Information about rel-
evant environmental parameters such as solar irradiance, ambient temperature and
wind speed is also available. For our purpose, we assume a simplified situation: the
data about the power is available every hour and constant.
To simulate the power supplied by the wind turbine in the smart home, we use
the data gathered by the anemometer at Dalton School. The CSV file with the data
is obtained through an HTTP request issued by the Smart Grid Simulation Engine
towards the school Web site. The computation of the power produced by a wind
turbine, given the wind speed sensed, is performed by the Energy Calculator Engine
component. The component uses the well-know relationship between wind speed
and power extracted by a wind turbine:
P 21 ρAU 3 Cp (5.1)
where ρ is the air density, A is the rotor swept area, U is the wind speed and Cp is the
power coefficient representing the efficiency of the turbine rotor [72]. Once we have
chosen the turbine, the parameters are known: A π p 3.5 2 q (the turbine blades have
2
a 3.5 meters diameter), ρ 1.225 (typical air density value), Cp 0.35 (a typical
value of rotor efficiency for wind turbines) and the wind speeds (i.e., cut-in and cut-
out speed) between which the turbine works, then the wind to power relationship
can be applied. Also in this case we assume to have the data about the wind speed
available every hour and constant.
9 http://www.dalton.org/
10 http://sunviewer.net/portals/NYSERDA/siteHome.php?sid=17
5.2.5 Pricing energy from renewables
Usually, renewable sources are considered as base load in energy dispatching mech-
anisms: when energy is generated by these types of plants it is fed into the system
and these plants are never shut down [199]. In our smart home, we consider both
solar and wind power to have a cost of production due to the investment that is
required to purchase, install, and maintain the equipment.
The idea, and the related computations performed by the Price Extractor Engine
consider a simplified investment analysis for calculating the cost for each kWh of
produced energy. We simply consider the investment cost, the maintenance and op-
erations costs, governmental incentives and the energy produced over the expected
lifetime of the plant. Therefore, we have the energy cost coming from the PV as
ECP V pCInv COpr CM ai q{EnL (5.2)
where CInv is the total investment costs for the PV array, COpr and CM ai are the
total cost of operation and the total maintenance costs respectively over the invest-
ment lifetime for the PV system. For the PV array we consider 20 years of service
for the AP-100 model. EnL is the estimated overall energy to be produced during
the lifetime of the PV array. First, we estimate a production of energy during the
20 years of panel lifetime that is on average the same as the one produced in the
previous years since the installation at Dalton school. Second, the investment cost is
based on the results by Wise et al. [224] that investigated the cost of PV panels in the
U.S.A. The value that emerges from their analysis considering the cost for PV pan-
els, inverters and installation once the incentives applied by the U.S. government
are subtracted, is 5.1 dollar for each installed watt of power. Third, the maintenance
and operation costs can be considered as an annual expense about 0.84% of the in-
vestment, as found by Rehman et al. [180] for a PV installation. Another component
that might be associated to operation and maintenance costs is the replacement of
inverters which, as suggested in Croxford et al. [55], should be considered every 10
years. According to the findings of Wiser et al. [224] inverters replacement amounts
about 7% of the initial total investment in a PV project for residential and small com-
mercial purposes. Once we feed these parameters in the Energy Calculator Engine
component it provides a constant value that represents the cost of each kWh gen-
erated by the PV plant. Therefore, the PV plant can be considered as an additional
“virtual” energy provider since the produced energy in a future deregulated smart
grid could be sold on the market at the price of returning from the investment (i.e.,
obtained from the computation shown above) or higher to make a profit.
We apply a similar relationship to evaluate the cost of electricity generated by
the small-wind turbine:
ECSW pCInv COpr CM ai IGov q{EnL (5.3)
the only parameter that we have in addition to those described above is IGov that
represents the incentive subsidy from the government in wind energy. We consider
an investment cost for the wind turbine of 28000$ comprising the installation and
required equipment (e.g., installation pole, inverters, workforce) and an annual cost
of maintenance and operations of 1% of the total investment. This is in line with
the findings of [212]. EnL for the turbine is the estimated overall energy to be pro-
duced during its lifetime (according to [186] the lifetime of a wind turbine can be
considered around 20 years). To estimate the production of energy in New York we
consider the average wind speed recorded by NOAA in more than 50 years.11 Using
this data the turbine produces around 1300 kWh of energy per year. Considering the
subsidy, the state of New York has an incentive program for small-scale wind instal-
lation to promote such energy source with a contribution for one year of 3.50$ per
kWh produced.12 Therefore, with all these parameters fed in the Energy Calculator
Engine component one has once again a constant price for each kWh generated, in
this case by the small-wind turbine.
5.2.6 Renewable energy forecast

The amount of energy that can be produced by renewable systems is strongly corre-
lated to the meteorological conditions. It is beyond the scope of this work to build
an exact model of how meteorological and environmental conditions influence the
energy production of wind and solar plants; here we aim at a realistic estimation of
possible energy production in the broader context of simulation of realistic services
for the smart grid. The Smart Grid Simulation Engine interacts with the NOAA na-
tional weather service13 Web service requesting information about the wind speed,
cloud coverage and temperatures. The service provides a week forecast given the
location of interest (through longitude and latitude coordinates); for the first three
days the forecast data has a 3 hours interval, while for the following days the inter-
val is 6 hours. The Energy Forecast Engine has the task to adjust the meteorological
data to an hour-by-hour information (in our simple solution we use interpolation)
and to compute an hour-by-hour energy forecast for the week. The available power
for each hour is computed in the following way:
11 http://lwf.ncdc.noaa.gov/oa/climate/online/ccd/avgwind.html
12 http://www.dsireusa.org/incentives/incentive.cfm?Incentive_Code=NY35F&re=
0&ee=0
13 http://graphical.weather.gov/xml/
• for the wind turbine, we apply P 12 ρAU 3 Cp , given the wind speed from the
forecast and the other wind turbine parameters;
• for solar panels, we apply P Hpw21 pCc Tf q where Hpw21 is the hour-
by-hour average of the historical power produced the previous year during
the 10 previous and 10 following days of the very same day of the previous
year. Therefore, we have an average value of produced power in comparable
days, i.e., similar sun horizon condition. The terms in brackets Cc and Tf are
used to correct the power considering cloud coverage and temperature esti-
mations. Cc represents the percentage of cloud coverage and Tf pThis T qδ
is a temperature factor that considers, based on the difference in the average
hourly temperature in the same 21-day period in the previous year, the in-
crease/decrease in efficiency of δ percentage. We consider a linear relation
which establishes a decrease/increase of δ 0.5%{ C in PV efficiency for
a polycrystalline silicon panel as the temperature increases/decreases com-
pared to factory test conditions [106].
Following these models the Energy Forecast Engine provides for one week avail-
able power hour-by-hour for the two sources of renewable energy installed in our
hypothetical smart home.
5.2.7 Examples
Figure 5.2 shows an example of the output of the Price Extractor component of the
Smart Grid Emulation Engine. It represents the price of electricity on the PJM mar-
ket for a sub set of the zones in the U.S. electrical system. One notes that there are
significant variations in price during the day: from the minimum in the middle of
the night, the price almost doubles in the evening. Therefore, with such information
available beforehand (we recall that these prices are day-ahead) one can create au-
tomation to provide consistent savings over time. Think for instance of the schedul-
ing of a washing machine, a dryer, and, a dishwasher, or the charging of an electric
car.
Figure 5.2: Energy prices for several locations of the U.S. electrical system on 5th
October 2012.
Having information about the amount of renewable energy to be produced (fore-

cast) is essential to better plan on a week basis the activities that can wait the opti-
mal environmental conditions for self production. Figure 5.3 shows the amount of
energy that is forecast at the installation site we consider (Dalton school) and by
using the weather forecast of NOAA and the model proposed in Section 5.2.6. In
the figure, one week of power forecast is shown. The continuous line represents
the power provided by the PV installation with a typical bell-shaped power output
during the daytime and flat night profile. The dashed line represents the forecast of
the wind turbine that has an irregular pattern. One can see how 6th , 7th and 10th
October are more favorable to the self-production of energy by using solar panels,
while the night between 6th and 7th and the days 10th and 11th are very favorable
to power production through the wind turbine. By exploiting this information the
smart home can have an idea about the right time to schedule the usage of devices.
Figure 5.3: Energy forecast for solar power and wind power in the week 5th -11th
October 2012.
5.3. Smart Grid Aware Buildings 141
Naturally, in an implementation of the smart grid services on the field, a recon-

ciliation system to evaluate the gaps between forecast and real prices will be needed
a posteriori for billing purposes and to inform the user. The same reconciliation will
be required in the actual amount of energy extracted from the grid in case of non
exact forecasts of generated renewable energy. This aspect is out of the scope of this
work, however we envision that the smart meter will be the center of this reconcili-
ation by storing the in- and out-flow of energy and the tariffs applied at each point
in time.
5.3 Smart Grid Aware Buildings

The smart grid simulation tool described in Section 5.2 has been the smart grid en-
gine to realize demand-response functionalities in a real living lab environment at
the University of Groningen. A set of software communicating modules have been
developed to realize a real demand-response implementation involving real appli-
ances and end users. The components realized are shown in Figure 5.4. In the
context of this work we just want to give the essential details and focus on the ap-
plication of the smart grid functionalities rather than the details of the architecture
and the optimization algorithms to schedule the devices. For a more thorough cov-
erage of the architecture and the details concerning the optimization details we refer
to [76].
Figure 5.4: smart grid enabled architecture to implement demand-response fucntion-

alities.
The smart meter component shown in Figure 5.4 has exactly the functionalities
described in Section 5.2. This component is the engine that provides the prices of
energy and the amount of power each provider can supply; furthermore, it provides
the quantity of power available thanks to the small-scale renewable sources that
are supposed to power the office in addition to traditional energy supply. These
information, in addition to a set of rules characterizing the operation of the devices
(i.e., device policies), are essential to enable the schedule of the appliances with the
goal of minimizing energy costs to realize demand-response experiments.
5.3.1 Realization of a smart office

In the facility hosting the Faculty of Mathematics and Computer Science, we have
realized a living lab office environment where we have tested the scheduling of a
set of appliances (see Figure 5.5). Each of the appliances is connected to the power
outlet through a plugwise socket. Plugwise14 devices are plugs that are attached to
a common power outlet on one side and provide a socket to connect an appliance on
the other side. A plugwise device is shown in Figure 5.6. The plugwise outlet is able
to perform two operations: 1) measure in real-time the amount of power used by an
attached device, 2) enable/disable the power flow to the connected appliance acting
de facto as an on/off switch. These operations of sense and actuation can be issued
to the plugwise sockets via a base station with ZigBee communication capabilities
running an appropriate software.
Figure 5.5: Appliances and communication infrastructure in living lab.
For each appliance/device, there is an associated policy. A policy is a set of con-

sistent rules that hold for device operations. We have identified these rules by mea-
suring the power consumption of the devices in normal usage conditions and from
our usage experience. For example, “a fridge must work at least 15 minutes per
hour” to be able to maintain its internal temperature below a certain threshold tem-
perature level. Policies can have different parameters, a few of which are common
14 http://http://www.plugwise.com/
Policy Associated Description

type device
REPEAT Fridge, Boiler Device should be put to a specified state repeat-
edly with a certain periodicity.
TOTAL Laptop Device should operate for at least a certain
amount of time.
MULTIPLE Printer Device should operate for the time that allows
for all scheduled jobs to be performed.
STRICT Beamer A strict schedule is given in advance.
PATTERN Microwave An expected pattern of device operations.
SLEEP Any device No demand for device during the scheduling pe-
riod.
Table 5.3: Device policies
to all: ptBegin, tEndq – time period, when the policy is active; and sid – state ID
that the policy is applied to. State IDs are unique per device. In general, we assume
several possible states per device, together with associated actions to move a device
to these states. In the presented setting, each device has two states: “on” and “off,”
and two associated actions: “turn on” and “turn off.”
Figure 5.6: A plugwise Circle plug device.
5.3.2 Benefits of the smart grid-enabled system

In order to have prices for energy that are similar to those available for the end
user market, the prices extracted by the price extractor engine of the Smart Grid
Simulation Engine have been rescaled (from $/MWh to $/kWh) for this living lab
experiment. We have used the system over three weeks in the months of October
2011 and November 2011 performing measurements from Monday to Friday (as in
the weekend there is irregular presence). In particular, in the first 2 weeks (W1-
W2) we measured energy use in order to define a baseline. The third week (W3),
we let the scheduling component control the environment in order to measure the
actual savings. Next we present the results in terms of economic savings (due to the
varying prices of the smart grid) and of energy savings (due to the introduction of
device policies).
Figure 5.7: Average price ($ per kWh) comparison between non-scheduled (upper
chart) and scheduled (lower chart) appliances for each work day.
Figure 5.8: Average price ($ per kWh) comparison between scheduled (continuous
line) and non-scheduled (dashed line) situation.
(a) Price of energy ($ per kWh) during (b) Price of energy ($ per kWh) during
non-scheduled day October 27t h 2011. scheduled day November 3r d 2011.
Figure 5.9: Energy price in non-scheduled and scheduled situations.
Economic savings
As we have mentioned before, the goal of the system is to save money for the of-
fice taking advantage of the smart grid. Therefore the first evaluation we make is
based on taking the energy bill for a week using the developed demand-response
system versus a week without it. To make the comparison fair, we use the energy
prices of the third week (W3) and apply those same retrieved prices for the energy
consumed in the other two weeks. The situation between each working day of the
two weeks (average) without scheduling policies and the week where the policy has
been applied is shown in Figure 5.7, where the cost in term of $ per kWh is shown
versus the time of the day (from Monday to Friday). It is interesting to notice the
difference in the average price paid for each kWh of energy in the situation with-
out device scheduling and, on the other hand, considering scheduling, the chart is
shown in Figure 5.8. On average the price in $ per kWh drops by more than 27%
in the two situations. An interesting day where the savings on energy expenses
are particularly significant is between the three consecutive Thursdays monitored
(October 20t h, 27t h and November 3r d). Comparing these three days the money
savings are on average more than 50%. A comparison between the price paid for
energy in each hour between the situation in October 27t h and November 3r d is
shown in Figures 5.9a and 5.9b. In particular, one can see the cut of unnecessary en-
ergy expenses related to those consumptions that happen during non-working time
(late evening or during the night) by non strictly necessary devices (most notably
the hot water boiler). Another optimization the system achieves is the most effi-
cient schedule of devices when the energy generated by photovoltaic panel is more
intense and whose cost is generally smaller than energy provisioning on the market.
Energy savings
Although energy saving is not the primary aim of the system, but rather economic
savings based on dynamic pricing, the use of policies for devices alone provides for
energy saving in absolute terms. Figure 5.10 shows the average energy consumption
(kWh) considering the use and the absence of the scheduling system. One can see
that the scheduling reduces the consumption of devices that are not used during
non working hours and that do not impact the habits of the user (e.g., keeping hot
water boiler working at night); in addition the scheduler tries to use at best the cheap
electricity coming from the solar panels during day-light hours. Figure 5.11 visually
reinforces the idea of reducing loads when unnecessary among the normal (upper
chart) and the scheduled solution (lower chart): one notices a more compact chart
in which energy is mostly used during daytime (8 a.m. 6.30 p.m.) in each day of the
week. On average the savings in energy consumed between the situation without
the scheduling policy and the situation considering it, is more than 15%.
Figure 5.10: Average energy usage (kWh) comparison between scheduled (continu-
ous line) and non-scheduled (dashed line) situations.
Figure 5.11: Energy (per kWh) comparison between non-scheduled (upper chart) and
scheduled (lower chart) appliances for each work day.
5.4 Agent-Based Energy Market

Our vision for the future is a completely free energy market. In this market con-
sumers, prosumers and tradional energy companies can participate offering energy
directly on a short term time horizon (e.g., 15 minutes to 1 hour). In addition, we
envision that the buildings of the future will be aware of their energy consump-
tion and of the energy tariffs through their energy management systems. Buildings
will regulate their energy use and appliance operations based on the price signals
for energy from the smart grid as we showed in Section 5.3. However, in a totally
free market the building (via its energy management system) has to provision for
its energy consumption and access directly to the energy market. Other buildings
and actors of the smart grid with generating capabilities will do the same creating
a market where energy is traded. In order to test this concept and interactions we
have realized a software solution based on agent technology to test the feasibility
of this kind of energy market on a real software platform able to interact with real
smart meter devices too. Smart meters are likely to be an essential component of
the future energy management system of buildings. Our second aim is to test the
existence of procurement strategies that minimize the average price paid for energy
by consumers.
5.4. Agent-Based Energy Market 147
5.4.1 Roles in the smart grid energy market

In the market, we distinguish a number of roles. The fundamental actors are: few
traditional big generating companies (i.e., gencos), a limited number of prosumers
that are able to produce small quantities of energy (compared to gencos) and a high
number (compared to gencos and prosumers) of buyers that are interested in pur-
chasing energy at the cheapest price possible. The market also contains a third party
authority, known as balancer. This authority has a role similar to the role today
played by ISOs or RTOs. In fact, the balancer has information about the quantities
of energy produced and demanded by the various parties. Its role is to act as a
intermediary with the aim of maintaining as much as possible the energy balance
equilibrium on the power grid.
In such a model, the described actors have, possibly conflicting, goals. The pro-
sumers have the goal to sell any surplus power on the energy market at the highest
possible price. We consider that the energy produced by prosumers has a price usu-
ally significantly lower than the price offered by the gencos. In addition, the main
issue is that the power provided by prosumers is not sufficient to supply the whole
demand, but only a fraction of it. We consider this condition to be realistic given the
adoption rate of small-scale distributed energy resources. The gencos have the goal
to sell energy optimizing price per unit, that is, since production costs do not grow
linearly, they want to sell energy at the price yielding the highest possible revenue
for them [142]. The balancer has the goal to keep (a geographically identified por-
tion of) the market in balance to avoid failures on the grid. Therefore, the amount
of demand not met by the offer by the prosumers should be backed up by the gen-
cos. The external balancer entity calculates the sum between every capacity of every
prosumer and forecasts the amount of extra energy that needs to be produced by the
gencos to satisfy the expected demand. Weather conditions are essential informa-
tion for the balancer. To compute the power equilibrium condition the balancing
authority uses an equation showing the relation between the total energy demand
of an area and the supply capacity of the sellers for that area. There are several
mathematical models to describe balancing equations (e.g., [109, 112, 201]), though
all have the same basic underlying idea which is to set to zero the algebraical sum
between energy demand and supply.
Let D∆t be the sum of all the demands of all consumers in a given time interval,
we define its balancing equation as
¸
C ¸
Ng
¸
Mp

∆t
Gci Pr
D∆t Dk S∆t S∆t j (5.4)

k 1
i 1
j 1
where:
• C∆t is the number of consumers at time interval ∆t;
• Dk is the energy demand by the k th energy buyer;

z
• S∆t is the energy supply provided by z th source of energy at time interval ∆t;
• Ng is the total number of number of gencos;
• Gci is the ith genco;
• Mp total number of prosumers; and
• P rj is the j th prosumer.
The equation simply states that the sum between the two different production sources
(gencos and prosumers) should be equal to the total consumer demand. This is a
simple equation that does not take into account other aspects such as energy losses
in transportation and production, forecasting errors in demand, nor any quality of
services of the energy producers. However, for the purpose of this work more fo-
cused on realizing the platform and the market, the relaxation of all the details of
the real situation are adequate.
5.4.2 Agent modeling

Agents are software programs that are programmaed to act with a specific purpose
on behalf of an authority or user or role [91]. Therefore, software agent technology
is the perfect match to implement the smart grid energy market given the several
actors and roles given the description above. We distinguish between main agents
and auxiliary agents. The former represent energy consumers and energy genera-
tors, whereas the latter do not directly deal with the energy purchase and sell, but
provide information and mediation to support the behavior of the main agents.
In synthesis the main agents are:
• Consumers: their goal is to buy energy.
• Prosumers: their goal is to produce and consume electricity being able to both
buy and sell it. They can produce a limited quantity of energy (compared to
a genco) thanks to small-scale energy production devices such as small wind
turbines, solar panels, and micro-combined heat and power (micro-CHP).
• Gencos: their goal is to sell energy at various scale. They are the traditional
big energy generating companies which can be (or not) also in charge of the
distribution of the energy.
As for the auxiliary agents, we initially identify three:
• Balancer: it has a mediating role among main agents to keep the balance of
supply and demand.
• Time Agent: it defines the starting and closing of a time interval for negotiation
(i.e., ∆t interval in Equation 5.4).
• Weather Agent: it provides localized information about weather forecast, espe-

cially parameters useful to predict small-scale generation.
In addition to these agents, for the simulation purposes two more agent roles
need to be defined since we resort to the Java Agent Development Environment
(JADE) [44]. The first agent created is the Creator Agent that generates all the other
agents. Another essential agent that is present in the architecture is the Directory
Facilitator Agent (DF) that provides a yellow pages-like service for the platform.
5.4.3 Agent interactions

All the interactions between agents take place by message exchanges. The message
exchanges need to be standardized in order to make sure that any new agent can
join an existing infrastructure. We chose to go for the most accepted standard for
agent communication, the Foundation for Intelligent Physical Agents (FIPA) stan-
dard [154]. We consider two kind of messages: synchronization messages and mar-
ket related messages.
From the implementation point of view, the chosen platform (i.e., JADE) follows
an asynchronous message passing paradigm and it is based on the Agent Communi-
cation Language (ACL), while being fully FIPA compliant. Every agent has a queue
that stores the incoming messages, the extraction of the message from the queue is
then up to the programmer.
The essential parts composing a message according to the FIPA/ACL language
are:
• Content is the payload of the message itself.
• Performative represents the purpose of the message, or better the communica-

tive intention. Independently from the content, an agent could decide to try
to better understand what a sender is saying by just fetching this part of the
message. In particular, for the smart grid energy market, we consider the fol-
lowing constructs:
– INFORM and INFORM REF performatives represent generic informative

messages used for synchronization purposes. The latter is used as syn-
chronization signal for final operations (e.g., a buyer informs a balancer
that he has stipulated a contract successfully).
– PROPOSE, REFUSE and ACCEPT PROPOSAL performatives are used in
market related messages. The first one is used for a bid proposal sent by
a buyer agent or it provides information about the starting price of a ne-
gotiation when sent by a energy seller agent. The second performative is
used to refuse a single offer, while the last performative is used to accept
an offer proposal.
– CANCEL performative is used by a prosumer in order to abort the cur-
rent auction with the specified buyer.
• Language represents the syntax used to express the content, however this as-
pect is not used in our project.
• Ontology represents the knowledge representation of the domain. This aspect

is not used in the current version of our project, though might be considered
in future extensions.
5.4.4 Agent behavior in the market

In the domestic energy market, all the agents have a specific goal that is, trade a
good and maximize profit. Therefore, the agents are competing against each other
for obtaining the energy at the best price and selling it at the highest one. The actual
trade is based on an auction system in which all main agents participate at every
round of trading. Thus the situation is as follows.
• A buyer can stipulate a contract with a prosumer after winning an auction
round, realized with sealed bids; every buyer can send several offers to suit-
able prosumers in any given round.
• Prosumers’ energy starting price is considerably lower than gencos’ initial

contract prices. The underlying assumption is that any positive amount is
helpful in the return from a sunk investment on renewable energy generation.
• Weather conditions during an observed interval can prevent a prosumer to

generate enough electricity to be sold.
• Prosumers communicate to buyers an initial starting price that is influenced by

contracts with energy transmission and distribution operators and a random
cost due to the devices used to produce electricity (e.g., maintenance costs).
• The energy produced by a prosumer has to be sold quickly and cannot be

stored or buffered for selling at a later auction round.
Every prosumer is in direct competition with other energy sellers therefore they
propose an appealing starting price and make an intelligent use of refusing bids in
order to rise the price without letting buyers contact other sellers. On the other hand,
gencos are big energy generating companies and they have a theoretical infinite
amount of energy supply, but they also have some peculiarities to be taken into
account: i) gencos sell energy with contracts lasting for one time interval or more at
a given fixed price; ii) gencos contracts can be stipulated much faster since there is
no auction process; iii) gencos prices are in general higher than prosumers’ starting
prices; iv) gencos prices depend on exceeding production threshold known a priori.
This last aspect implies that energy exceeding the threshold will be more expensive
for the genco and thus for the buyer. This additional cost models the operation
of extra expensive power plants to compensate an excessive demand. Therefore,
the unit price of energy the gencos sell on the market can be represented by the
following threshold function:
#
Euc Costenergy if below supply threshold
(5.5)
Costenergy pEC Aq otherwise
where EC ¡ 1 is an external costs constant and A P N1 is the number of energy

units above the threshold. In other words, asking gencos energy contracts when the
threshold is already exceeded leads to more expensive contract prices. Those prices
rises as we get further from the specified threshold.
In these conditions, the challenge is strictly related with the ability of agents to
quickly obtain an energy contract either with a prosumer or with a genco since as
time passes it is more likely that gencos will exceed their production threshold thus
increasing the price of a contract. In fact, if every buyer follows a natural strategy
that suggest him to contact the prosumer first hoping in cheap contracts. The buyer
is then involved in spending time in the auction process with offers and bids rising.
If no contract with a prosumer is found it is likely that the gencos in the meanwhile
have exceeded their threshold and only high priced energy is available.
A modeling of the best strategies for the agents, is beyond the scope of the
present treatment which aims at providing a modeling and a software platform for
the domestic market. We remark, however, that the family of minority games is
a good way of representing the interaction and finding suitable agent strategies in
such a market. In minority games, two different behaviors for a single player are ad-
mitted and s/he wins if s/he chooses the path taken by the minority of the players.
Applying this game to the domestic energy market means that each buyer agent can
learn and adjust its behavior (i.e., contacting a prosumer or a genco) reducing the
average cost of energy paid by the buyers [35]. In particular, the most similar game
theoretic approach that presents a situation similar to the one just explained for the
energy market model developed is the El-Farol-Bar problem [69]. It is possible to use
the solution identified for the El-Farol-Bar game by Whitehead [223] and extend it to
take into account the even higher level of complexity of the energy market model
presented here [35].
5.5 Bits and Electrons Hand in Hand

In this chapter, we have seen how ICT is essential for the realization and success
of the smart grid. To add value for the end user and make energy provisioning
and distribution more efficient, effective, and sustainable ICT must be at the core
of the smart grid to enable new scenarios of energy use. ICT opens new services
and opportunities in the energy world. First of all, by using flexible information
oriented paradigms such as SOA, it is easier and more efficient to share information
between the many actors in the smart grid panorama. SOA is also an approach able
to easily and fast adapt to the changes that a still uncertain framework such as the
smart grid requires in the current. Since standards are under development and test
of solutions are underway. SOA has also the capability to adapt solutions easily
especially in heterogeneous environments such as the power industry of today and
tomorrow where more technologies have to be integrated in a final solution. We
have shown through this chapter how today parts of the ingredients to realize the
demand-response service for the smart grid are available. By just providing little
adaptations, the ingredients can be used in real smart grid service contexts. In our
tests of an office connected to the smart grid, we have shown that real benefits for
the end user are possible in monetary terms, and at the same time more renewable
sources can be used as energy production sources. ICT opens also new scenarios
and possibilities such as prosumer based energy markets. Users can interact by
using software applications that work on their interest (i.e., agents based) and auto-
matically trade energy to achieve profits (in the prosumer case) or lower the energy
bill (in the consumer case). Proof of concepts such as the one realized in our faculty,
or small living labs (e.g., PowerMatching City Hoogkerk [22]) show the feasibility
of smart grid ideas and scenarios. The efforts are now devoted to scale these proofs
of concept to country-wide solutions and investigate additional services. These new
ICT services have the aim of improving energy efficiency and increment the pene-
tration of renewable sources.
Chapter 6
Conclusion
We envision a more sustainable, reliable, efficient, decentralized, and affordable en-

ergy supply where users can provide for their own supply and exchange energy in
a free energy market at local scale, such as neighborhood or city level. This vision
can be reconciled with the various meanings [139] and visions [207] of the smart
grid. Fostered by this inspiration, we have studied a new approach to the electrical
grid, both in its infrastructural characteristics proposing evolutions, and in realizing
applications for the future energy services and market.
6.1 The Emergence of a New Grid

The smart grid is not yet implemented nor standardized and still misses a unique
definition. In essence, it deals with enhancing the electrical system with the func-
tionalities of Information and Communication Technologies (ICT) to achieve more
flexibility, more resilience, and more efficiency. The aim is to implement a high share
of renewables in the energy generation sources and provide the end user with even
a better service (e.g., more reliable, better tariffs) than the traditional grid.
Our work focuses on two aspects: (i) the modeling and evolution of the distribu-
tion grid to better accommodate the smart grid vision using the complex network
analysis approach and (ii) software and applications in the smart grid context.
Considering the first topic of research, we have focused on the medium and low
voltage layers of the power grid infrastructure, that is, the part of the grid that is
closer to the end user. The novelty of the work has been severalfold. A first novelty
lies in looking at the medium and low voltage grid that is a layer of the electrical
systems that has not received much attention. A second novelty consists of analyz-
ing the power grid using complex network analysis taking into account the physical
parameters of topology that influence the price of electricity through a new set of
topological metrics. A third one consists of proposing novel smart grid related met-
rics to evaluate the topology of the grids. A fourth point deals with using complex
network as a design tool for the design and expansion of the grid by defining a set
of metrics for the smart grid. Fifth, this approach sets the base for a decision support
154 6. Conclusion
system to guide the planning and evolution of the grid towards a smart grid where
local energy exchange is prominent. Topological studies of the power grid have so
far focused only on the high voltage grid and almost always in a traditional con-
ception (i.e., not related to the smart grid) of the power grid in order to understand
failures related to topological properties. Our work has focused on the medium
and low voltage network and we have investigated the topological properties of
physical samples belonging to the Dutch grid. In particular, we have been able to
associate topological measures with the price of distributing electricity and compare
the different Dutch topologies according to those measures. In the literature, com-
plex network approaches have been mainly used to analyze networks in order to
grasp the fundamental properties of the system represented in the network. These
fundamental properties are extracted through measures and indicators of the whole
network. Another effort in the complex network community has been to create mod-
els to mimic the dynamics of network evolution systems (e.g., the Barabási-Albert
preferential attachment model). We have used complex network tools also to design
medium and low voltage grids for the future energy system by considering a set of
properties the networks have to satisfy. Our investigation focused on both models
coming from complex network analysis literature which might be used for planning
the infrastructures in settlements to be realized (e.g., developing countries), and on
the evolution of current distribution grid topologies as we do for the Dutch sam-
ples. Considering the whole work on the grid infrastructure side, we have set the
basics of realizing a decision support system for analysis and planning of the smart
grid. A skeleton diagram of the envisioned decision support system is shown in
Figure 6.1. In the figure, several phases and input are considered to plan the evolu-
tion of the infrastructure where a local energy exchange is the overarching goal. It
starts with a pre-processing phase where the input data of the grid is converted into
a graph; the output of this initial phase is a Power Grid graph. The following phase
consists of the analysis of the topological properties characterizing the graph. The
output of this phase consists of a set of values representing the metrics related to the
power grid that influence the price of electricity (α and β metrics in the figure). The
process continues with the generation of a network model. The number of nodes
and edges of this reference model are provided according to the targets for the cost-
related parameters (α and β) and the will to invest by the stakeholders. Based on the
theoretical model identified, the physical network under assessment is then fitted to
a topological structure similar to the one of the model. Several solutions are pro-
vided that differ in the topology and the α and β metrics. All of these solutions are
then input into a computer-assisted decision support system that presents the ben-
efits/costs of the evolution of the network. An expert is involved in the selection
of the most promising candidates for the evolution process that is built by the com-
6.2. Open Issues and Future Directions 155
puter because there are other political, geographical, and legal considerations that
may rule in favor or against any possible evaluation. Once the decision is made,
the adaptation of the physical grid can begin. We have realized several steps of this
system, but it is still not a fully automated piece of software.
Considering the second topic, we have looked into the software approach for an
heterogeneous environment such as the smart grid and built scenarios and applica-
tions for the future. Our finding, in line with other recent proposals, e.g. [220, 53],
is that software-oriented architectures (SOAs) are an ideal candidate for integrating
different systems coming from traditional power engineering telecontrol (e.g., su-
pervisory control and data acquisition) and new systems featuring smart grid func-
tionalities (e.g., dynamic energy pricing, demand-response). The applications that
we have realized go into the direction of exploring scenarios and services for the
smart grid. On one side, we have tested appliance scheduling and automation in
a dynamic pricing environment with realistic energy prices and renewable energy
production in an office environment with real users. On the other side, another ap-
plication considered an essential ingredient of the smart grid in our vision: a free
market where consumers, prosumers and power companies can participate. We
have tested agent technologies to enable such a scenario coupled with real smart
meter devices and identified a promising strategy for the users to supply their elec-
tricity based on minority games.
6.2 Open Issues and Future Directions

The smart grid is an evolving subject with test beds and pilot projects appearing all
over the world. The subject investigated in this thesis is open to further research
and improvements both on the grid side and on the software/application side.
Considering the power grid, several aspects can be considered to follow up. First
of all, in order to move from a theoretical high level planning and decision support
system to a more fine grained planning tool close to traditional power engineer-
ing, power flow analysis should be interpreted. With more digital reading, precise
information on the power flows in the medium and low voltage grid will be avail-
able. The goal is to evaluate the performance of the synthetic networks proposed
and the Dutch grid topologies evolved using complex network analysis principles.
Such analysis could give a better insight into the real application on the field of our
findings. Along the same lines, we propose the evaluation of the findings of this
thesis with engineers from the power industries to understand how to improve the
decision support systems skeleton that we have developed so far into a fully usable
tool to test and plan distribution grid evolution for the smart grid.
6. Conclusion
Figure 6.1: Decision support system for smart grid planning.

156
6.2. Open Issues and Future Directions 157
Other research topics involve the identification or design of topological mod-

els tailored to the smart grid. The models used for the synthetic generation (in
Chapter 4) were based on existing complex network analysis. For the smart grid,
a specific model could be developed that enables topological and spatial efficiency
and robustness on the one hand, and that is electrically valid on the other hand;
small-world networks seem as a good starting point for this further investigation.
The decision support system depicted in Figure 6.1 needs refinement. Especially the
translation from economic targets to topological targets has to be fine-tuned to make
a translation in quantitative objectives. An open issue is the influence on the elec-
tricity price paid by the end user versus the economic advantage of end user topolo-
gies. A detailed economic analysis should be performed to understand the impact
of topologies that should reduce the price of electricity distribution and the effects
on the energy bills of the end users. Last, but not less important, is the comparison
of the analysis and evolution proposed for the Dutch grid to other distribution grid
topologies coming from other geographies and countries.
Considering the application and ICT side of the smart grid, two lines of im-
provement of the software realized are foreseeable. On the engineering side, the
developed scheduling system should be reduced to its essential components and
embedded in a home energy management system as well as the agent technology
application for market interaction. This is more of an industrial research and devel-
opmental issue. The software system should also be enhanced with artificial intel-
ligence and machine learning techniques to learn user preferences and be aware of
external conditions in order to act or react. A series of tests on home energy man-
agement system and intelligent appliances in the home environment should be per-
formed in collaboration with sociologists and psychologists to obtain the maximum
results and feedback on the approach to energy savings and products when dealing
with end users. The future of services involving the smart grid have to be broad
(i.e., large number of services available) and personalized for the user (e.g., dedi-
cated tariffs based on user profile and his building). We envision cloud computing
services that provide the user with personalized information where the back-end of
these services is based on the SOA approach. Tests and scaling of these applications
applied in the power domain are another field of research and development.
There are other open points that go beyond the technical and ICT aspects of the
smart grid that, anyway, must be considered especially on the influences that they
might pose on the evolution of the smart grid. From a sociological side, open points
are the acceptance of the smart grid by the end users accustomed to the traditional
way of dealing with the electrical system. Another aspect from the psychological
side that deserve attention concerns the incentives to provide to the users to imple-
ment energy efficiency policies or demand-response initiatives. Price alone could
158 6. Conclusion
not be the best to approach all the segments of the population [23]. Open points
exist in other fields, such as law. A full unbundling such as the one we envision,
have to pass through the legal institution that have to modify the current legal sys-
tem and maybe create new subjects with legal responsibilities over the operations
of the local grid (today this is in place for the high voltage grid). Economists and
policy makers are also investigating the pricing mechanisms to be applied in a sce-
nario with more distributed generation and potentially more small markets that are
economically viable and at the same time that do not have to impact on the stability
of the electrical system.
We have found that complex networks methodology prove to be a good tool to
analyze and design networks. However, given the peculiarity of the physical prop-
erties that lay in the electrical system, a more detailed model of the lines need to
be considered. In particular, the analysis of power flows is important to translate
the theoretical models to networks that can be later engineered. Other interesting
ways of evolving networks are, for instance, through multi-objective optimization
on complex networks or complex network evolution using a genetic algorithm ap-
proach. We also consider the scientific discipline of complex systems as an interest-
ing framework to use in the study of smart grid, given the studies performed and
the results obtained in the study of city and infrastructure scaling [21]. The complex-
ity of the smart grid lies not only in the technical aspects of a system, but also in the
social dynamics overlying the system: user interactions and their attitude to buy or
sell energy on the grid, the mutual influence through social media modifying their
energy behavior might have effects on the electrical system (macro) not predictable
by the analysis of the single individual (micro) [27]. On the software side, we have
seen that the drivers are available to realize demand-response and appliance au-
tomation functionalities related to the smart grid. The aspect that deserves more
attention is the involvement of the users, their feedbacks and their reactions to the
new electrical system. Implementation and tests on large-scale are the necessary
steps to proceed towards smart grid realization.
Obviously, all of the speculations for future direction of research and develop-
ment associated with the topics described in this chapter depend heavily on the
transition towards the smart grid. The path is, in the broad sense, set. Different
countries are adopting different approaches at different speeds; the adoption is in-
fluenced by the natural resources available, the energy policies, and the economy
involved in the energy sector. Until new solutions for harvesting clean energy be-
come available, the smart grid is the solution on the horizon that can enable the
transition towards a sustainable future providing high integration of distributed re-
newable energy sources into the system with higher flexibility and enabling new
opportunities for energy utilities.
Appendix A
Graph Theory and Complex Network
Fundamentals
A.1 Graph Theory and Complex Network Definitions

and Properties
The approach used in this thesis to model the power grid and its evolution is based
on Graph Theory and Complex Networks Theory. Here we recall the basic defini-
tions that we use throughout the thesis and refer to standard Graph Theory text-
books such as [24, 25] or those focusing on Complex Networks such as [153, 49] for
a deeper introduction.
First, for the sake of completeness we define once again a graph for the power grid
and its weighted representation as we described in Chapter 3.
A.1. D EFINITION (G RAPH ). A graph G is a pair of sets G(V,E) where V is the set of
vertexes and E is the set edges. An edge ei,j is a pair of vertexes pvi , vj q. If pvi , vj q P E then
vi and vj are said to be adjacent or neighboring and are called end-vertexes of the edge.
A.2. D EFINITION (P OWER G RID GRAPH ). A Power Grid graph is a graph GpV, E q
such that each element vi P V is either a substation, transformer, or consuming unit of
a physical power grid. There is an edge ei,j pvi , vj q P E between two nodes if there is
physical cable connecting directly the elements represented by vi and vj .
One can also associate weights to the edges representing physical cable properties
(e.g., resistance, voltage, supported current flow).
A.3. D EFINITION (W EIGHTED GRAPH ). A weighted graph is a pair GpV, E q where V

is the set of vertexes and E is the set of edges. An edge ei,j,w pvi , vj , wq is a triple where
vi , vj P V and w P R. w is called weight of the edge.
A.4. D EFINITION (W EIGHTED P OWER G RID GRAPH ). A Weighted power grid graph
is a Power Grid graph Gw pV, E q with an additional function f : E Ñ R associating a real
160 A. Graph Theory and Complex Network Fundamentals
number to an edge representing the physical property of the corresponding cable (e.g., the
resistance, expressed in Ohm, of the physical cable).
A first classification of graphs is expressed in terms of their size.
A.5. D EFINITION (O RDER AND SIZE OF A GRAPH ). Given the graph G the order is given
by N |V |, while the size is given by M |E |.
From order and size, it is possible to have a global value for the connectivity of the
vertexes of the graph, known as average node degree . That is xk y 2M
N . To character-
ize the relationship between a node and the others it is connected to, the following
properties provide an indication of the bond between them.
A.6. D EFINITION (A DJACENCY, NEIGHBORHOOD AND DEGREE ). If ex,y P E is an edge

in graph G, then x and y are adjacent, or neighboring, vertexes, and the vertexes x and y
are incident with the edge ex,y . The set of vertexes adjacent to a vertex x P V , called the
neighborhood of x, is denoted by Γx . The number dpxq |Γx | is the degree of x.
When considering a weighted graph it is possible to extend the definition of degree

to have an idea of the importance of a node based on the incident edges.
A.7. D EFINITION (W EIGHTED DEGREE ). Let x PV be a vertex in a weighted graph G,

the weighted degree of x, dw pxq is:
¸
dw pxq wx,y
P pq
y Γ x
where wx,y is the weight of the edge joining vertexes x and y, and Γpxq is the neighborhood
of x.
A measure of the average ‘density’ of the graph is given by the clustering co-
efficient, characterizing the extent to which vertexes adjacent to any vertex v are
adjacent to each other.
A.8. D EFINITION (C LUSTERING COEFFICIENT (CC)). The clustering coefficient γv of

Γv is
γv
|E pΓv q|
kv
2

where |E pΓv q| is the number of edges in the neighborhood of v and kv
2 is the total number
of possible edges in Γv .
This local property of a node can be extended to an entire graph by averaging over
all nodes.
A.1. Graph Theory and Complex Network Definitions and Properties 161
Another important property is how much any two nodes are far apart from each
other, in particular the minimal distance between them or shortest path. The con-
cepts of path and path length are crucial to understand the way two vertexes are
connected.
A.9. D EFINITION (PATH AND PATH LENGTH ). A path of G is a subgraph P of the form:
V pP q tx0 , x1 , . . . , xl u, E pP q tpx0 , x1 q, px1 , x2 q, . . . , pxl1 , xl qu.
such that V pP q V and E pP q E. The vertexes x0 and xl are end-vertexes of P and

l |E pP q| is the length of P . A graph is connected if for any two distinct vertexes
vi , vj P V there is a finite path from vi to vj .
A.10. D EFINITION (D ISTANCE ). Given a graph G and vertexes vi and vj , their distance
dpvi , vj q is the minimal length of any vi vj path in the graph. If there is no vi vj path
then it is conventionally set to dpvi , vj q 8.
A.11. D EFINITION (S HORTEST PATH ). Given a graph G and vertexes vi and vj the short-
est path is the the path corresponding to the minimum of to the set t|P1 |, |P2 |, . . . , |Pk |u
containing the lengths of all paths for which vi and vj are the end-vertexes.
A global measure for a graph is given by its average distance among any two nodes.
A.12. D EFINITION (AVERAGE PATH LENGTH (APL)). Let vi PV be a vertex in graph

G. The average path length for G Lav is:
¸
Lav N pN1 1q dpvi , vj q

i j
where dpvi , vj q is the finite distance between vi and vj and N is the order of G.
A.13. D EFINITION (C HARACTERISTIC PATH LENGTH (CPL)). Let vi P V be a vertex

in graph G, the characteristic path length for G, Lcp is defined as the median of dvi where:
¸
dvi pN 1 1q dpvi , vj q

i j
is the mean of the distances connecting vi to any other vertex vj in G and N is the order of G.
When considering a weighted graph the definition of CPL can be easily extended to
account for weights characterizing the edges.
A.14. D EFINITION (W EIGHTED CHARACTERISTIC PATH LENGTH (WCPL)). The weighted

characteristic path length for graph G, Lwcp is the median for all pvi , vj q P V of dwi where:
¸
dwi pN 1 1q ewi,j

i j
is the mean of the weighted distances connecting vi to any other vertex vj and ewi,j is an
edge in the minimal weighted path between vi and vj in G and N is the order of G.
To describe the importance of a node with respect to minimal paths in the graph, the
concept of betweenness helps. Betweenness (sometimes also referred as load) for a
given vertex is the number of shortest paths between any other nodes that traverse
it.
A.15. D EFINITION (B ETWEENNESS ). The betweenness bpv q of vertex v PV is
bpv q
¸ σst v pq

v s t
σst
where σst pv q is 1 if the shortest path between vertex s and vertex t goes through vertex v, 0
otherwise and σst is the number of shortest paths between vertex s and vertex t.
Looking at large graphs, one is usually interested in global statistical measures

rather than the properties of a specific node. A typical example is the node degree,
where one measures the node degree probability distribution.
A.16. D EFINITION (N ODE DEGREE DISTRIBUTION ). Consider the degree k of a node in

a graph as a random variable. The function
Nk tv P G : dpvq ku
is called probability node degree distribution.
The shape of the distribution is a salient characteristic of the network. For the power
grid, the shape is typically either exponential or a power-law [15, 6, 158, 184]. More
precisely, an exponential node degree (k) distribution has a fast decay in the proba-
bility of having nodes with relative high node degree. The relation:
P pk q αeβk
A.2. Graph properties example 163
follows, where α and β are parameters of the specific network considered. On the
contrary, a power-law distribution has a slower decay with higher probability of
having nodes with high node degree. It is expressed by the relation:
P pk q αk γ
where α and γ are parameters of the specific network considered. We remark that
the graphs considered in the power grid domain are usually large, although finite,
in terms of order and size thus providing limited and finite probability distributions.
A Graph can also be represented as a matrix, typically an adjacency matrix.
A.17. D EFINITION (A DJACENCY MATRIX ). The adjacency matrix A ApGq pai,j q of

a graph G of order N is the N N matrix given by
#
if pvi , vj q P E,
aij 1
0 otherwise.
A graph can be also represented by its Laplacian matrix:
A.18. D EFINITION (L APLACIAN MATRIX ). Let D pDij q be a diagonal matrix with

Dii dpvi q the degree of vertex vi in graph G and A the adjacency matrix of G.
The matrix L D A is the Laplacian matrix of graph G.
A.2 Graph properties example

For the sake of clarity, we apply the definitions introduced in Section A.1 to a simple
graph example and compute the main properties that characterize it. Figure A.1
shows the simple graph that we consider.
Figure A.1: A simple graph.

Order and Size The graph G shown in Figure A.1 is characterized by the set of
vertexes V :
V t1, 2, 3, 4, 5u
and by the set of edges E:
tp1, 2q, p1, 5q, p2, 3q, p2, 5q, p3, 4qu
E
the order and size of G are |V | 5 and |E | 5 respectively.
Neighborhood The neighborhood of vertex 1 in Figure A.1 is:
Γp1q t2, 5u
its degree is:

dp1q |Γp1q| 2
The average node degree for G is:
xky 2 5 5 2
Clustering The clustering coefficient for vertex 1 is:
γ1 1
while for the entire graph it is the average of tγ1 1, γ2 31 , γ3 0, γ4 0, γ5 1u
that is:
γG 0.467
Path, Distance and Path Length Example of paths between vertex 1 and vertex 4
are:
P1,4 t1 2 3 4u
whose length lP 3 but also:
1
P1,4 t1 5 2 3 4u
is a valid path whose length lP 1 4, therefore the shortest path between vertex 1
and vertex 4 is P1,4 .
The distance between vertex 1 and vertex 4 is:
dp1, 4q 3
while distance between vertex 1 and vertex 5 is:
dp1, 5q 1
A.2. Graph properties example 165
Average and Characteristic Path Length The average path length is given by:
LAV 5 1 4 p7 5 6 9 7q
34
20
1.7
The characteristic path length is the median of:
tdv 74 , dv 45 , dv 64 , dv 94 , dv 74 u
1 2 3 4 5
that is:
LCP 74 1.75
Betweenness Vertex 2 is involved in the following shortest paths:
P1,3 , P1,4 , P3,1 , P3,5 , P4,1 , P4,5 , P5,3 , P5,4
therefore betweenness of vertex 2 is:
bp2q 8
Matrix Representation of a Graph The Adjacency matrix for graph G is:

0 1 0 0 1
1 1
0 1 0
ApGq 0

1 0 1 0

0 0 1 0 0
1 1 0 0 0
The Laplacian matrix for graph G is:

2 1 0 0 1
1 1 1
3 0
LpGq 0 1 1

2 0

0 0 1 1 0
1 1 0 0 2
Appendix B
Complex Network Models
Network models and the algorithms to build them are essential in knowing how
to generate new networks or how to re-arrange connections in already available
networks to obtain networks that satisfy a specific set of properties. Here we give a
brief description of the models used in this work that is based on the more complete
and thorough description in [39] and [151] to which we refer for more details.
B.1 Building Synthetic Networks
Random Graph
A Random Graph is built by connecting each pair of nodes with an edge with prob-
ability p. It is due to the pioneering studies of Erdős and Rényi [67]. More precisely,
there are two ways to built a Random Graph, (a) the GN,p model proposed by Erdős
and Rényi considers a set of N nodes and for each pair of nodes an edge is added
with a certain probability p; (b) the GN,M model considers with equal probability
all the graphs having N vertexes and exactly M edges randomly selected among all
the possible pairs of edges. The models have the same asymptotic properties. In this
work we use the GN,M model since we are interested in setting both the number of
nodes and edges for the networks to generate. A Random Graph with order 199 and
size 400 is shown in Figure B.1.
168 B. Complex Network Models
Figure B.1: A Random Graph.
Small-world Graph
The small-world phenomenon became famous after the works of Milgram in the
sociological context [203] who found short chains of acquaintances connecting ran-
dom people in the USA. More recently, the small-world characterization of graphs
has been investigated by Watts and Strogatz [221, 222], who showed the presence
of the small-world property in many types of networks such as actor acquaintances,
the power grid and neural networks in worms. It is obtained from a regular lattice
that connects the nodes followed by a process of rewiring the edges with a certain
probability p P r0, 1s. The resulting graph has intermediate properties between the
extreme situations of a regular lattice (p 0) and a random graph (p 1). In
particular, small-world networks hold interesting properties: the characteristic path
length is comparable to the one of a corresponding random graph (Lsw Á Lrandom ),
while the clustering coefficient has a value bigger than a random graph and closer
to the one of a regular lattice (CCsw " CCrandom ). A small-world graph with order
200 and size 399 is shown in Figure B.2.
B.1. Building Synthetic Networks 169
Figure B.2: A small-world graph.
Preferential Attachment
The preferential attachment model represents the phenomenon happening in real

networks where a fraction of nodes has a high connectivity while the majority of
nodes has small node degree. This model is built upon the observation by Barabási
and Albert [15] of a typical pattern characterizing several type of natural and artifi-
cial networks. The basic idea is that whenever a node is added to the network and
connects (through edges) to m other nodes, those with higher degree are preferred
for connection. In other words, the probability to establish an edge with an existing
node i is given by Πpki q °kikj where ki is the node degree of node i. One can
j
see then that the more connected nodes have higher chances to acquire more and
more edges over time in a sort of “rich gets richer” fashion; a phenomenon studied
by Pareto [174] in relation to land ownership. The preferential attachment model
reaches a stationary solution for the node degree probability that follows a power-
law with P pk q 2m
2
k3 . A graph based on preferential attachment with order 200 and

size 397 is shown in Figure B.3.
Figure B.3: A preferential attachment graph.
R-MAT
R-MAT (Recursive MATrix) is a model that exploits the representation of a graph

through its adjacency matrix [40]. In particular, it applies a recursive method to
create the adjacency matrix of a graph, thus obtaining a self-similar graph struc-
ture. This model captures the community-based pattern appearing in some real
networks. Moreover, the generated graph is characterized by a power-law node de-
gree distribution while showing a small diameter. The idea is to start with an empty
N N matrix and then divide the square matrix into four partitions in which the
nodes are present with a certain probability for each partition, specifically probabil-
ities a, b, c, d that sum to one. The procedure is then repeated dividing each partition
again in four sub-partitions and associating the probabilities. The procedure stops
when a 1 1 cell is reached in the iterative procedure. The a, b, c, d partitions of
the adjacency matrix have particular meaning: a and d represent the portions con-
taining nodes belonging to different communities, while b and c represent the nodes
that act as a link for the different communities (e.g., in a social network people with
interests both in topics mostly popular in either a or d community). The recursive
nature of this algorithm creates a sort of sub communities at each round. A graph
based on R-MAT model with order 222 and size 499 is shown in Figure B.4.
Figure B.4: A R-MAT graph.
Random Graph with Power-law
A Random Graph with power-law model generates networks characterized by a

power-law in the node degree probability distribution (P pk q k γ ) having the
majority of nodes with a low degree and a small amount of nodes with a very high
degree. Power-law distributions are very common in many real life networks both
created by natural processes (e.g., food-webs, protein interactions) and by artificial
ones (e.g., airline travel routes, Internet routing, telephone call graphs) [13]. These
are also referred to as Scale-free networks [16]. From the dynamic point of view, these
networks are modeled by a preferential attachment model. In addition, reliability is
a property of these graphs, that is, high degree of tolerance to random failures and
high sensitivity to targeted attacks towards nodes with high degree or hubs [5, 138,
56].
This model is characterized by the exponent of the power-law (i.e., γ) which

governs the degree of each node. The edges between the nodes are then wired in a
random fashion. As we have shown earlier, the other way of constructing a graph
that is compliant with a power-law based node degree distribution is through the
growth of the network and preferential attachment based on node degree. A Ran-
dom Graph with power-law with order 200 and size 399 is shown in Figure B.5.
Figure B.5: A Random Graph with power-law graph.
Copying Model
Replicating the structure underlying the links of WWW pages brought to the Copy-
ing Model [107] which captures the tendency of members of communities with same
interests to create pages on the web with a similar structure of links. The basic in-
tuition is to select a node and a number (k) of edges to add to the node. Then with
a certain probability β, the edges are linked independently and uniformly at ran-
dom to k other nodes, while with probability p1 β q the k edges are copied from a
randomly selected node u. If u has more than k edges, a subset is chosen, while if
it has less than k edges they are anyway copied and the remaining are copied from
another randomly chosen node. It leads to a distribution for the incoming degree
that follows a power-law with a characteristic parameter γin 11 β . A graph based
on Copying Model with order 200 and size 200 is shown in Figure B.6.
Figure B.6: A Copying Model graph.
Forest Fire
In order to capture dynamic aspects of the evolution of networks, Leskoveck et

al. [118] proposed the Forest Fire model. The intuition is that networks tend to
densify in connectivity and shrink in diameter (i.e., the greatest shortest path in
the network) during the growth process. Technological, social and information net-
works exhibit this property. The model requires two parameters known as forward
burning probability (p) and backward burning ratio (r). The graph grows over time
and at each discrete time step a node v is added, then a node w, known as ambas-
sador, is chosen at random between the other nodes of the graph and a link between
v and w is added. A random number x (obtained from a binormal distribution with
mean p1 pq1 ) is chosen and this is the number of out-links of node w that are se-
lected. Then a fraction r times less than the out-links is chosen between the in-links
and an edge is created with these as well. The process continues iterating choosing
a new x number for each of the nodes v is now connected to. The idea, as the name
of the model suggests, resembles the spreading of a fire in a forest that starts from
the ambassador node to a fraction (based on the probability parameters) of nodes it
is connected to and goes on in a sort of chain reaction. This model leads to heavy
tails both in the distribution of in-degree and out-degree node degree. In addition,
a power-law is shown in the densification process: a new coming node tends to
have most of his links in the community of his ambassador and just few with other
nodes. A graph based on Forest Fire model with order 200 and size 505 is shown in
Figure B.7.
Figure B.7: A Forest Fire graph.
Kronecker Graph
A generating model with a recursive flavor similar to R-MAT uses the Kronecker
product applied to the adjacency matrix of a graph [117]. The Kroneker product is a
non conventional way of multiplying two matrices [80].
B.1. D EFINITION (K RONECKER PRODUCT ). Given two matrices A and B with dimen-
sion (n m) and (n1 m1 ) the Kronecker product between A and B is a matrix C with
dimension (n n1 m m1 ) with the following structure:

a1,1 B a1,2 B a1,m B
a2,1 B a2,2 B a2,m B
C AbB .

.. .. ..
.. . . .
an,1 B an,2 B an,m B
B.2. D EFINITION (K RONECKER G RAPH ). Given two graphs G and H with adjacency
matrices A(G) and A(H), a Kronecker graph is a graph whose adjacency matrix is obtained
by the Kronecker product between the adjacency matrices of G and H.
If the Kronecker product is applied to the same matrix, therefore multiplying the
matrix with it elements recursively, a self similar structure arises in the graph. This
situation can be seen as the increase of a community in a network and the further
differentiation in sub-communities while the network grows.
This model creates networks that show a densification in the connectivity with
a shrinking diameter over time. The idea is to apply the Kronecker product to the
B.2. Model Parameters for Synthetic Networks 175
same matrix recursively. The procedure to create a graph based on the Kronecker
product starts with a N N matrix where each xij element of the matrix represents
a probability of having an edge between node i and j. Thereafter, at each time step
the network grows so that at step k the network has N k nodes. By applying the
Kronecker product to the same matrix leads to the emergence of self-similar fractal-
like structures at different scales. This structure mimics a quite natural process that
is the recursive growth of communities inside communities which are a miniature
copy of a big community (i.e., the whole graph structure) [117]. A Kronecker Graph
with order 167 and size 267 is shown in Figure B.8.
Figure B.8: A Kronecker graph.
B.2 Model Parameters for Synthetic Networks

In order to build graphs according to the models introduced in Section B.1 we used
a set of specific parameters that are described below.
• Random Graph. For the GN,M model, the only parameters needed are the
order and size of the graph to be generated. We use the values shown in Ta-
ble 4.2 for the order, and the size is chosen accordingly to obtain an average
node degree of two, four and six, respectively.
• Small-world Graph. In addition to order, the small-world model requires the

specification of the average degree and the edge rewiring probability. For the
first parameter, we simply provide a value to obtain the desired average node
degree (i.e., xk y 2, xk y 4 and xk y 6). The latter parameter represents
the probability of rewiring an edge, connecting a source node to a different

destination node chosen at random. We choose an intermediate approach be-
tween the regular lattice (i.e., rewiring probability p 0) and random graph
extremes (i.e., rewiring probability p 1). In fact, we choose a rewiring proba-
bility p 0.4. This is to give slightly more emphasis to the regular structure of
lattice than to the rewiring, since we expect the future grid to have more em-
phasis on a regular structure than random cabling. This last aspect also helps
to satisfy the qualitative requirement of modularity.
• Preferential Attachment. For the creation of a graph based on growth and

preferential attachment model of Barabási-Albert [15], the only parameters
needed are the order and size of the graph to be generated. We use the val-
ues shown in Table 4.2 for the order parameter, while the size parameter is
chosen accordingly to obtain an average node degree of two, four and six,
respectively.
• R-MAT. The R-MAT model requires several parameters. First of all, order and
size of the network, then the a, b, c, d parameters which represent the probabil-
ities of the presence of an edge in a certain partition of the adjacency matrix.
The order of the graph is chosen so that the nodes are a power of two, in par-
ticular 2n where usually n rlog2 N s. Therefore, we consider for this model
the following values for the order: {32,128,256} for comparison with the low
voltage, and {256,512,1024} for comparison with the medium voltage grids.
For the probability parameters, since we have an undirected graph, we have
b c, in addition the ratio found between a and b, as in many real scenarios ac-
cording to [40], is about 3:1. We assume a more highly connected community
(a 0.46) and a less connected community (d 0.22) and a relative smaller
connectivity between the two communities (b c 0.16).
• Copying Model. The Copying Model requires, in addition to the order of

the graph, a value for the probability of copying (or not) edges from existing
nodes. p1 β q is the probability of copying nodes from another node. In the
present study, we fix β 0.2 so as to have a high probability of having a direct
(just one-hop since with probability 0.8 each new node copies the connections
of another node and attaches directly to them) connection to what might be
considered the most reliable energy sources present in the city or villages (at
medium voltage level), while it represents single users or small aggregation of
users with high energy capacity at low voltage level.
• Forest Fire. The Forest Fire model requires, in addition to the order of the
graph, two values representing the probability of forward and backward spread
of the “burning fire”. We choose the same value for both probabilities since
our graph is not directed. To avoid a flooding of edges, we choose few small
values to assign to forward and backward probability (pf wd pbwd 0.2;
pf wd pbwd 0.3; pf wd pbwd 0.35) that give realistic amounts of average
edges incident to a node that can be compared with the models for which one
is able to directly set order and size.
• Random Graph with Power-law. For the model representing Random Graph
with power-law in node degree distribution, the parameters required are es-
sentially the order of the network and the characteristic parameter of the power-
law (known as the γ coefficient). For the first parameter, we use the usual di-
mensions (see Table 4.2), while for the latter some additional considerations
are necessary. We test different types of power-law coefficients characteriz-
ing real technological networks. For the non-electrical technological networks
(i.e., technological networks which are not power grids) we average the values
of the power-law characteristic parameter described in [47]; the details of the
parameters are shown in Table B.1. For the power grid networks the γ values
represent:
– the findings for the Western and Eastern high voltage U.S. power grid in [42];
the values are averaged to have a single γ, the details are shown in Ta-
ble B.2;
– the findings for the high voltage U.S. Western power grid in [15] which
reports a value γ 4;
– the findings for the medium and low voltage Dutch grid that follow a
power-law in [158]; the values are averaged to have a single γ, the details
are shown in Table B.3.
Type of network γ
Internet degree 2.12
Telephone calls received 2.09
Blackouts 2.3
Email address book size 3.5
Hits to web-sites 1.81
Links to web sites 2.336
Average 2.359
Table B.1: Power-law γ parameters for technological networks [47].
• Kronecker Graph. For the Kronecker model, the required parameter is the
initial dimension of the square matrix to apply the Kronecker product: a 2
Type of network γ
Eastern Interconnection 3.04
Western System 3.09
Average 3.065
Table B.2: Power-law γ parameters for high voltage U.S. power grid [42].
Type of network γ
LV#5 2.402
LV#10 1.494
MV#2 1.977
MV#3 2.282
Average 2.039
Table B.3: Power-law γ parameters for Dutch medium and low voltage power grid
[158].
2 initiation matrix is a good starting model [117]. Once the structure of the
matrix is defined, the initial parameters for the generation matrix need to be
evaluated. With a 2 2 adjacency matrix for the initial graph G:

ApGq
a b
c d
the parameters can be interpreted in a similar fashion as in R-MAT:
– a models the “core” part of the network and the tightness of its connec-
tivity.
– d models the “perifery” part of the network and the connectivity inside
it.
– b, c model the relationships and interconnections between the core and
the periphery.
The findings of Leskovec et al. [117] applying the Kronecker modeling to many
different networks report a common recurrent structure for the parameters of
the 22 Kronecker matrix initiator. In particular, the parameters tend to follow
the empirical rule a " b ¥ c " d and are usually a 1, b c 0.6 and d 0.2.
In this work, we consider two sets of parameters characterizing the Kronecker
initiator matrix. The first set is extracted and averaged from the technolog-
ical and social networks parameters extracted from real sample data [117].
The second set of parameters is obtained applying the fitting procedure to
Type of network 2 2 Kronecker generator parameters

a c b d
Social-technological 0.9578 0.4617 0.4623 0.3162
power grid 0.4547 0.8276 0.8504 0.0186
Table B.4: Probability parameters for the 2 2 Kronecker matrix.
a Kronecker graph to the UCTE high voltage power grid data set used in [184,
191], the high voltage U.S. Western power grid data set used in [222], and the
medium and low voltage samples data set used in [158]. All these values have
been averaged to obtain just one 2 2 Kronecker generation matrix. A sum-
mary of the values for the parameters of the Kronecker matrix used is given
in Table B.4. One notices a very different structure in the matrix parameters
between the social and other diverse technological networks and the power
grid.
Bibliography
[1] M. Aiello and S. Dustdar. A domotic infrastructure based on the web service stack. Per-
vasive and Mobile Computing, 4(4):506–525, 2008.
[2] M. Aiello, M. Papazoglou, J. Yang, M. Carman, M.Pistore, L. Serafini, and P. Traverso.
A request language for web-services based on planning and constraint satisfaction. In
VLDB Ws on Tech. for E-Services (TES02), pages 76–85. Springer, 2002.
[3] W. Aiello, F. Chung, and L. Lu. Random evolution in massive graphs. In J. Abello,
P. M. Pardalos, and M. G. C. Resende, editors, Handbook of massive data sets, pages 97–122.
Kluwer Academic Publishers, Norwell, MA, USA, 2002.
[4] R. Albert, I. Albert, and G. Nakarado. Structural vulnerability of the north american
power grid. Physical Review E, 69, 2004.
[5] R. Albert, H. Jeong, and A. L. Barabási. Error and attack tolerance of complex networks.
Nature, 406(6794):378–382, 2000.
[6] L. A. N. Amaral, A. Scala, M. Barthélémy, and H. E. Stanley. Classes of small-world
networks. Proceedings of the National Academy of Sciences of the United States of America,
97(21):11149–11152, 2000.
[7] C. W. Anderson, J. R. Santos, and Y. Y. Haimes. A Risk-based Input-Output Methodol-
ogy for Measuring the Effects of the August 2003 Northeast Blackout. Economic Systems
Research, 19(2):183–204, 2007.
[8] S. Arianos, E. Bompard, A. Carbone, and F. Xue. Power grid vulnerability: A complex
network approach. Chaos: An Interdisciplinary Journal of Nonlinear Science, 19(1):013119,
2009.
[9] O. Asad, M. Erol-Kantarci, and H. Mouftah. Sensor network web services for demand-
side energy management applications in the smart grid. In Consumer Communications and
Networking Conference (CCNC), 2011 IEEE, pages 1176 –1180, 2011.
[10] K. Atkins, C. Barrett, and A. Marathe. A web services based artificial market. 2009
Winter Simulation Conf. (WSC), pages 3047–3054, 2009.
182 BIBLIOGRAPHY
[11] P. Balducci, J. Roop, L. Schienbein, J. DeSteese, and M. Weimar. Electrical power in-
terruption cost estimates for individual industries, sectors, and US economy. Technical
report, U.S. Department of Energy, 2002.
[12] A. Barabási, H. Jeong, Z. Néda, E. Ravasz, A. Schubert, and T. Vicsek. Evolution of the
social network of scientific collaborations. Physica A: Statistical Mechanics and its Applica-
tions, 311(3–4):590 – 614, 2002.
[13] A.-L. Barabási. Linked: How Everything Is Connected to Everything Else and What It Means.
Plume, reissue edition, 2003.
[14] A. L. Barabási. Scale-free networks: a decade and beyond. Science, 325(5939):412–3, 2009.
[15] A. L. Barabási and R. Albert. Emergence of scaling in random networks. Science,
286(5439):509, 1999.
[16] A. L. Barabási, R. Albert, and H. Jeong. Scale-free characteristics of random networks:
the topology of the World Wide Web. Physica A: Statistical Mechanics and its Applications,
281(1-4):69–77, 2000.
[17] M. Baran and F. Wu. Network reconfiguration in distribution systems for loss reduction
and load balancing. Power Delivery, IEEE Transactions on, 4(2):1401 –1407, 1989.
[18] D. Becker, H. Falk, J. Gillerman, S. Mauser, R. Podmore, and L. Schneberger. Standards-
based approach integrates utility applications. IEEE Computer Applications in Power,
13(4):13–20, 2000.
[19] H. Beitollahi and G. Deconinck. Peer-to-peer networks applied to power grid. In Inter-
national conference on Risks and Security of Internet and Systems (CRiSIS), 2007.
[20] D. Berardi, D. Calvanese, G. D. Giacomo, and M. Mecella. Reasoning about Actions for
e-Service Composition. In ICAPS’03 Ws on Planning for Web Services, 2003.
[21] L. M. A. Bettencourt. The origins of scaling in cities. Science, 340(6139):1438–1441, 2013.
[22] F. Bliek, A. van den Noort, B. Roossien, R. Kamphuis, J. de Wit, J. van der Velde, and
M. Eijgelaar. Powermatching city, a living lab smart grid demonstration. In Innovative
Smart Grid Technologies Conference Europe (ISGT Europe), 2010 IEEE PES, pages 1–8, 2010.
[23] J. Bolderdijk, L. Steg, E. Geller, P. Lehman, and T. Postmes. Comparing the effectiveness
of monetary versus moral motives in environmental campaigning. Nature Climate Change,
2012.
[24] B. Bollobas. Graph theory : an introductory course. Springer Verlag,, New York, 1979.
[25] B. Bollobas. Modern Graph Theory. Springer, 1998.
[26] E. Bompard, R. Napoli, and F. Xue. Analysis of structural vulnerabilities in power trans-
mission grids. International Journal of Critical Infrastructure Protection, 2(1-2):5–12, 2009.
[27] E. Bompard, R. Napoli, and F. Xue. Social and cyber factors interacting over the infras-
tructures: A mas framework for security analysis. In R. R. Negenborn, Z. Lukszo, and
H. Hellendoorn, editors, Intelligent Infrastructures, volume 42 of Intelligent Systems, Control
and Automation: Science and Engineering, pages 211–234. Springer Netherlands, 2010.
BIBLIOGRAPHY 183
[28] E. Bompard, D. Wu, and F. Xue. The concept of betweenness in the analysis of power
grid vulnerability. In Complexity in Engineering, 2010., pages 52 –54, 2010.
[29] F. Bouffard. The challenge with building a business case for smart grids. In Power and
Energy Society General Meeting, 2010 IEEE, pages 1 –3, 2010.
[30] P. Bresesti, M. Gallanti, and D. Lucarella. Market-based generation and transmission
expansions in the competitive market. In Power Engineering Society General Meeting, 2003,
IEEE, volume 1, page 4 vol. 2666, 2003.
[31] R. Brown. Impact of smart grid on distribution system design. In Power and Energy
Society General Meeting - Conversion and Delivery of Electrical Energy in the 21st Century,
2008 IEEE, pages 1 –4, 2008.
[32] C. D. Brummitt, R. M. DSouza, and E. A. Leicht. Suppressing cascades of load in inter-
dependent networks. Proceedings of the National Academy of Sciences, 109(12):E680–E689,
2012.
[33] N. Capodieci, E. Alsina, and G. Cabri. A context-aware agent-based approach for dereg-
ulated energy market. In Enabling Technologies: Infrastructure for Collaborative Enterprises
(WETICE), 2012 IEEE 21st International Workshop on, pages 16 –21, 2012.
[34] N. Capodieci, G. Cabri, G. A. Pagani, and M. Aiello. Adaptive game-based agent negoti-
ation in deregulated energy markets. In Collaboration Technologies and Systems (CTS), 2012
International Conference on, pages 300–307, 2012.
[35] N. Capodieci, G. Cabri, G. A. Pagani, and M. Aiello. An agent-based application to
enable deregulated energy markets. In Computer Software and Applications Conference
(COMPSAC), 2012 IEEE 36th Annual, pages 638–647, 2012.
[36] N. Capodieci, G. A. Pagani, G. Cabri, and M. Aiello. Smart meter aware domestic energy
trading agents. In Proceedings of the 2011 workshop on E-energy market challenge, IEEMC ’11,
pages 1–10, New York, NY, USA, 2011. ACM.
[37] F. Casati, M. Sayal, and M.-C. Shan. Developing e-services for composing e-services. In
Conf. on Advanced Inf. Sys. Engineering (CAiSE), pages 171–186. Springer, 2001.
[38] V. Chaitanya, M. Reddy, and D. Mohanta. Topological analysis of eastern region of in-
dian power grid. In Environment and Electrical Engineering (EEEIC), 2011 10th International
Conference on, pages 1 –4, 2011.
[39] D. Chakrabarti and C. Faloutsos. Graph mining: Laws, generators, and algorithms.
ACM Computing Surveys (CSUR), 38(1):2, 2006.
[40] D. Chakrabarti, Y. Zhan, and C. Faloutsos. R-MAT: A Recursive Model for Graph Min-
ing. In Fourth SIAM International Conference on Data Mining, 2004.
[41] D. Chappell. Enterprise Service Bus. O’Reilly, 2004.
[42] D. P. Chassin and C. Posse. Evaluating North American electric grid reliability using
the Barabási Albert network model. Physica A: Statistical Mechanics and its Applications,
355:667–677, 2005.
184 BIBLIOGRAPHY
[43] P. Chiradeja. Benefit of Distributed Generation: A Line Loss Reduction Analysis. 2005
IEEE/PES Transmission & Distribution Conference & Exposition: Asia and Pacific, pages 1–5,
2005.
[44] K. Chmiel, M. Gawinecki, P. Kaczmarek, M. Szymczak, and M. Paprzycki. Efficiency of
JADE agent platform. Scientific Programming, 13(2):159–172, 2005.
[45] J. Choi, T. Tran, A. El-Keib, R. Thomas, H. Oh, and R. Billinton. A method for trans-
mission system expansion planning considering probabilistic reliability criteria. Power
Systems, IEEE Transactions on, 20(3):1606 – 1615, 2005.
[46] S. Civanlar, J. Grainger, H. Yin, and S. Lee. Distribution feeder reconfiguration for loss
reduction. Power Delivery, IEEE Transactions on, 3(3):1217 –1223, 1988.
[47] A. Clauset, C. Shalizi, and M. E. J. Newman. Power-law distributions in empirical data.
SIAM Review, 51(4):661–703, 2009.
[48] R. Cohen, K. Erez, D. Ben-Avraham, and S. Havlin. Resilience of the internet to random
breakdowns. Physical review letters, 85(21):4626–8, 2000.
[49] R. Cohen and S. Havlin. Complex Networks: Structure, Robustness and Function. Cam-
bridge University Press, 2010.
[50] T. Considine. Ontological requirements of the service oriented grid. In Grid-Interop - The
road to an interoperable grid, Atlanta, Georgia, USA, 2008.
[51] R. Cossent, T. Gómez, and P. Frı́as. Towards a future with large penetration of dis-
tributed generation: Is the current regulation of electricity distribution ready? regulatory
recommendations under a european perspective. Energy Policy, 37(3):1145–1155, 2009.
[52] T. Cover and P. Hart. Nearest neighbor pattern classification. Information Theory, IEEE
Transactions on, 13(1):21 –27, 1967.
[53] W. T. Cox and T. Considine. Architecturally significant interfaces for the smart grid. In
Grid-Interop - The road to an interoperable grid, Denver, Colorado, USA, 2009.
[54] D. Crawford and J. Holt, S.B. A mathematical optimization technique for locating and
sizing distribution substations, and deriving their optimal service areas. Power Apparatus
and Systems, IEEE Transactions on, 94(2):230 – 235, 1975.
[55] B. Croxford and K. Scott. Can pv or solar thermal systems be cost effective ways of
reducing co 2 emissions for residential buildings? In Solar 2006: Renewable Energy - Key to
Climate Recovery, American Solar Energy Society, 2006.
[56] P. Crucitti, V. Latora, and M. Marchiori. Model for cascading failures in complex net-
works. Physical Review E, 69(4):3–6, 2004.
[57] P. Crucitti, V. Latora, and M. Marchiori. A topological analysis of the italian electric
power grid. Physica A: Statistical Mechanics and its Applications, 338(1-2):92 – 97, 2004.
[58] P. Crucitti, V. Latora, and M. Marchiori. Locating critical lines in high-voltage electrical
power grids. Fluctuation and Noise Letters, 5(2):L201–L208, 2005.
BIBLIOGRAPHY 185
[59] F. Dahlfors and J. Pilling. Integrated information systems in a privatized and deregu-
lated electri city market. In 1995 Int. Conf. on Energy Management and Power Delivery EMPD
’95, pages 249–254. IEEE, 1995.
[60] J. Ding, X. Bai, W. Zhao, Z. Fang, Z. Li, and M. Liu. The improvement of the small-
world network model and its application research in bulk power system. In Power System
Technology, 2006. International Conference on, pages 1–5, 2006.
[61] S. N. Dorogovtsev and J. F. F. Mendes. Evolution of networks. Advances in Physics,
51(4):1079–1187, 2002.
[62] S. Dunn. Micropower: The next electrical era. Worldwatch Paper, 2000.
[63] S. Dustdar and W. Schreiner. A survey on web services composition. Int. J. Web Grid
Serv., 1(1):1–30, 2005.
[64] A. Dwivedi, X. Yu, and P. Sokolowski. Identifying vulnerable lines in a power network
using complex network theory. In Industrial Electronics, 2009. IEEE International Sympo-
sium on, pages 18–23, 2009.
[65] A. Dwivedi, X. Yu, and P. Sokolowski. Analyzing power network vulnerability with
maximum flow based centrality approach. In Industrial Informatics (INDIN), 2010 8th IEEE
International Conference on, pages 336 –341, 2010.
[66] J. Ekanayake, N. Jenkins, K. Liyanage, J. Wu, and A. Yokoyama. Smart Grid: Technology
and Applications. Wiley, 2012.
[67] P. Erdős and A. Rényi. On random graphs. I. Publ. Math. Debrecen, 6:290–297, 1959.
[68] S. K. Ethiraj and D. Levinthal. Modularity and innovation in complex systems. Manage-
ment Science, 50(2):pp. 159–173, 2004.
[69] J. Farago, A. Greenwald, and K. Hall. Fair and efficient solutions to the santa fe bar
problem. In In Proceedings of the Grace Hopper Celebration of Women in Computing 2002,
2002.
[70] K. Fekete, S. Nikolovski, D. Puzak, G. Slipac, and H. Keko. Agent-based modelling
application possibilities for croatian electricity market simulation. In Electricity Market,
2008. EEM 2008. 5th International Conference on European, pages 1–6, 2008.
[71] A. Fratkin and L. Ostrov. Ehv transmission grid design in israel electric corporation’s
power system: approaches and experience. In Electrical and Electronics Engineers in Israel,
1996., Nineteenth Convention of, pages 475 –478, 1996.
[72] L. Freris and D. Infield. Renewable Energy in Power Systems. Wiley, 2008.
[73] T. L. Friedman. Hot, Flat, and Crowded: Why We Need a Green Revolution - and How It Can
Renew America. Picador, 2008.
[74] L. Garver. Transmission network estimation using linear programming. Power Apparatus
and Systems, IEEE Transactions on, PAS-89(7):1688 –1697, 1970.
[75] C. W. Gellings. A consumer portal at the junction of electricity, communications, and
consumer energy services. The Electricity Journal, 17(9):78 – 84, 2004.
186 BIBLIOGRAPHY
[76] I. Georgievski, V. Degeler, G. A. Pagani, T. A. Nguyen, A. Lazovik, and M. Aiello. Opti-

mizing energy costs for offices connected to the smart grid. Smart Grid, IEEE Transactions
on, 3(4):2273–2285, 2012.
[77] J. K. Gershenson, G. J. Prasad, and Y. Zhang. Product modularity: definitions and bene-
fits. Journal of Engineering Design, 14(3):295–313, 2003.
[78] H. Ghenniwa. Web-services infrastructure for information integration in power systems.
In IEEE Power Engineering Society, page 8 pp. IEEE, 2006.
[79] G. Gilbert, Y. Chow, D. Bouchard, and M. Salama. Optimization of high voltage sub-
stations using a random walk technique. In Transmission and Distribution Construction,
Operation and Live-Line Maintenance (ESMO), 2011 IEEE PES 12th International Conference
on, pages 1 –7, 2011.
[80] A. Graham. Kronecker products and matrix calculus: with applications. Ellis Horwood series
in mathematics and its applications. Horwood, 1981.
[81] L. L. Grigsby, editor. The Electric Power Engineering Handbook, Second Edition. CRC Press,
2007.
[82] Z. Guohua, W. Ce, Z. Jianhua, Y. Jingyan, Z. Yin, and D. Manyin. Vulnerability assess-
ment of bulk power grid based on complex network theory. In Electric Utility Deregula-
tion and Restructuring and Power Technologies, 2008. Third International Conference on, pages
1554–1558, 2008.
[83] P. Han and M. Ding. Analysis of Cascading Failures in Small-world Power Grid. Inter-
national Journal of Energy Science, 1(2):99–104, 2011.
[84] C. Harris. Electricity Markets: Pricing, Structures and Economics. Wiley, 2006.
[85] J. A. Harris and F. G. Benedict. A biometric study of human basal metabolism. Proceed-
ings of the National Academy of Sciences, 4(12):370–373, 1918.
[86] P. Hines and S. Blumsack. A centrality measure for electrical networks. In Hawaii Inter-
national Conference on System Sciences, Proceedings of the 41st Annual, page 185, 2008.
[87] P. Hines, E. Cotilla-Sanchez, and S. Blumsack. Do topological models provide good in-
formation about electricity infrastructure vulnerability? Chaos: An Interdisciplinary Journal
of Nonlinear Science, 20(3):033122, 2010.
[88] T. E. Hoff. Using distributed resources to manage risks caused by demand uncertainty.
The Energy Journal, 18(Special I):63–84, 1997.
[89] A. J. Holmgren. Using Graph Models to Analyze the Vulnerability of Electric Power
Networks. Risk Analysis, 26(4):955–969, 2006.
[90] IEA. Key world energy statistics 2013. International Energy Agency, 2013.
[91] N. R. Jennings. On agent-based software engineering. Artificial Intelligence, 117(2):277 –
296, 2000.
[92] Y. Jia-hai, Y. Shun-kun, and H. Zhao-guang. A multi-agent trading platform for electric-
ity contract market. 2005 International Power Engineering Conf., 102206:1024–1029 Vol. 2,
2005.
BIBLIOGRAPHY 187
[93] P. L. Joskow. Lessons learned from electricity market liberalization. The Energy Journal,
29(Special I):9–42, 2008.
[94] S. Karnouskos. The cooperative internet of things enabled smart grid. In Proceedings
of the 14th IEEE International Symposium on Consumer Electronics (ISCE2010), June 07-10,
Braunschweig, Germany, 2010.
[95] S. Karnouskos. Future smart grid prosumer services. In Innovative Smart Grid Technologies
(ISGT Europe), 2011 2nd IEEE PES International Conference and Exhibition on, pages 1 –2,
2011.
[96] S. Karnouskos, P. Da Silva, and D. Ilic. Energy services for the smart grid city. In Digital
Ecosystems Technologies (DEST), 2012 6th IEEE International Conference on, pages 1 –6, 2012.
[97] S. Karnouskos and T. N. D. Holanda. Simulation of a Smart Grid City with Software
Agents. 2009 Third UKSim European Symposium on Computer Modeling and Simulation,
pages 424–429, 2009.
[98] S. Karnouskos, O. Terzidis, and P. Karnouskos. An advanced metering infrastructure
for future energy networks. In IFIP/IEEE 1st International Conference on New Technologies,
Mobility and Security (NTMS 2007), Paris, France, pages 597–606, 2007.
[99] S. Karnouskos, A. Weidlich, K. Kok, C. Warmer, J. Ringelstein, P. Selzam, A. Dimeas,
and S. Drenkard. Field trials towards integrating smart houses with the smart grid. In
Energy-Efficient Computing and Networking, volume 54 of LNCS, pages 114–123. Springer,
2011.
[100] R. L. Keeney. Decision analysis: An overview. Operations Research, 30(5):pp. 803–838,
1982.
[101] W. Kehe, S. Qianhui, D. Cheng, and Y. Haiyani. The research of power grid data inte-
gration and sharing platform based on soa. Computer Science and Engineering, 2009. WCSE
’09. Second Int. Ws on, 1:106 –109, 2009.
[102] M. Kendall. Rank correlation methods. Griffin, London, 1948.
[103] A. Khan and H. Mouftah. Web services for indoor energy management in a smart grid
environment. In Personal Indoor and Mobile Radio Communications (PIMRC), 2011 IEEE
22nd International Symposium on, pages 1036 –1040, 2011.
[104] T. Khoa, P. Binh, and H. Tran. Optimizing Location and Sizing of Distributed Gen-
eration in Distribution Systems. 2006 IEEE PES Power Systems Conference and Exposition,
pages 725–732, 2006.
[105] C. J. Kim and O. B. Obah. Vulnerability assessment of power grid using graph topolog-
ical indices. International Journal of Emerging Electric Power Systems, 8(6):1–15, 2007.
[106] D. King, W. Boyson, and J. Kratochvil. Analysis of factors influencing the annual energy
production of photovoltaic systems. In Photovoltaic Specialists Conference, 2002. Conference
Record of the Twenty-Ninth IEEE, pages 1356 – 1361, 2002.
[107] J. Kleinberg, R. Kumar, P. Raghavan, S. Rajagopalan, and A. Tomkins. The web as a
graph: Measurements, models, and methods. In Proceedings of the 5th annual international
conference on Computing and combinatorics, pages 1–17. Springer-Verlag, 1999.
188 BIBLIOGRAPHY
[108] J. M. Kleinberg. The small-world phenomenon: an algorithm perspective. In Proceed-

ings of the thirty-second annual ACM symposium on Theory of computing, STOC ’00, pages
163–170, New York, NY, USA, 2000. ACM.
[109] P. D. Klemperer and M. A. Meyer. Supply function equilibria in oligopoly under un-
certainty. Econometrica, 57(6):pp. 1243–1277, 1989.
[110] K. Kok, Z. Derzsi, M. Hommelberg, C. Warmer, R. Kamphuis, and H. Akkermans.
Agent-based electricity balancing with distributed energy resources, a multiperspective
case study. In Hawaii International Conference on System Sciences, Proceedings of the 41st
Annual, page 173, 2008.
[111] K. Kok and G. Venekamp. Market-based control in decentralized electrical power sys-
tems. In First International Workshop on Agent Technologies for Energy Systems, ATES2010,
Toronto, 2010.
[112] K. Kok, C. J. Warmer, and I. G. Kamphuis. Powermatcher: multiagent control in the
electricity infrastructure. In Proceedings of the fourth international joint conference on Au-
tonomous agents and multiagent systems, AAMAS ’05, pages 75–82, New York, NY, USA,
2005. ACM.
[113] L. L. Lai, T. Motshegwa, H. Suasinghe, N. Rajkumar, and R. Blach. Feasibility study
with agents on energy trading. In Advances in Power System Control, Operation and Manage-
ment, 2000. APSCOM-00. 2000 International Conference on, volume 2, pages 505–510 vol.2,
2000.
[114] A. Lazovik, M. Aiello, and M. Papazoglou. Planning and monitoring the execution of
web service requests. Journal on Digital Libraries, 6(3):235–246, 2006.
[115] A. D. Le, M. Kashem, M. Negnevitsky, and G. Ledwich. Optimal Distributed Gener-
ation Parameters for Reducing Losses with Economic Consideration. 2007 IEEE Power
Engineering Society General Meeting, pages 1–8, 2007.
[116] S. Lee, K. Hicks, and E. Hnyilicza. Transmission expansion by branch-and-bound inte-
ger programming with optimal cost - capacity curves. Power Apparatus and Systems, IEEE
Transactions on, PAS-93(5):1390 –1400, 1974.
[117] J. Leskovec, D. Chakrabarti, J. Kleinberg, C. Faloutsos, and Z. Ghahramani. Kronecker
graphs: An approach to modeling networks. The Journal of Machine Learning Research,
11:985–1042, 2010.
[118] J. Leskovec, J. Kleinberg, and C. Faloutsos. Graphs over time: densification laws,
shrinking diameters and possible explanations. In Proceedings of the eleventh ACM
SIGKDD international conference on Knowledge discovery in data mining, KDD ’05, pages
177–187, New York, NY, USA, 2005. ACM.
[119] F. Leuthold, H. Weigt, and C. von Hirschhausen. Efficient pricing for european electric-
ity networks - the theory of nodal pricing applied to feeding-in wind in germany. Utilities
Policy, 16(4):284 – 291, 2008. European Regulatory Perspectives.
[120] F. Leymann, D. Roller, and M.-T. Schmidt. Web services and business process manage-
ment. IBM Systems Journal, 41(2):198–211, 2002.
BIBLIOGRAPHY 189
[121] D. Liben-Nowell and J. Kleinberg. The link-prediction problem for social networks. J.
Am. Soc. Inf. Sci. Technol., 58(7):1019–1031, 2007.
[122] R. Lincoln, S. Galloway, and G. Burt. Open source, agent-based energy market simu-
lation with python. In Energy Market, 2009. EEM 2009. 6th International Conference on the
European, pages 1–5, 2009.
[123] A. B. Lovins, E. K. Datta, T. Feiler, K. R. Rabago, J. N. Swisher, A. Lehmann, and
K. Wicker. Small is profitable: the hidden economic benefits of making electrical resources the
right size. Rocky Mountain Institute, 2002.
[124] G. Lu, D. De, and W.-Z. Song. Smartgridlab: A laboratory-based smart grid testbed. In
Smart Grid Communications (SmartGridComm), 2010 First IEEE International Conference on,
pages 143 –148, 2010.
[125] L. Lü and T. Zhou. Link prediction in complex networks: A survey. Physica A: Statistical
Mechanics and its Applications, 390(6):1150 – 1170, 2011.
[126] A. Machias, E. Dialynas, and C. Protopapas. An expert system approach to designing
and testing substation grounding grids. Power Delivery, IEEE Transactions on, 4(1):234 –
240, 1989.
[127] R. Mackiewicz. The Benefits of Standardized Web Services Based on the IEC 61970
Generic Interface Definition. In IEEE Power Systems Conf. and Expo., pages 491–494, 2006.
[128] M. Marmiroli and H. Suzuki. Web-based framework for electricity market. In Electric
Utility Deregulation and Restructuring and Power Technologies, 2000. Proceedings. DRPT 2000.
International Conference on, pages 471–475, 2000.
[129] C. Marnay and G. Venkataramanan. Microgrids in the evolving electricity generation
and delivery infrastructure. In In IEEE Power Engineering Society General Meeting. IEEE,
2006.
[130] J. Massey, Frank J. The kolmogorov-smirnov test for goodness of fit. Journal of the
American Statistical Association, 46(253):pp. 68–78, 1951.
[131] F. McDonald. Insull. Beard Books, 1962.
[132] S. Mei, X. Zhang, and M. Cao. Power Grid Complexity. Springer, 2011.
[133] A. Mercurio, A. Di Giorgio, and P. Cioci. Open-source implementation of monitoring
and controlling services for emsscada systems by means of web services - iec 61850 and
iec 61970 standards. Power Delivery, IEEE Transactions on, 24(3):1148–1153, 2009.
[134] A. Merlin and H. Back. Search for a minimal-loss operating spanning tree configuration
in an urban power distribution system. In Proc. of the Fifth Power System Conference (PSCC),
Cambridge, pages 1–18, 1975.
[135] S. Milgram. The small world problem. Psychology Today, 2:60–67, 1967.
[136] M. Mitchell. Complexity - A Guided Tour. Oxford University Press, 2009.
[137] D. Moneta, L. Bisone, G. Mauri, and R. Meda. New interactions between lv customers
and the network: further possibilities for home automation functions. In Robotics and
Automation, 2007 IEEE International Conference on, pages 2886–2891, 2007.
190 BIBLIOGRAPHY
[138] Y. Moreno, R. Pastor-Satorras, A. Vazquez, and A. Vespignani. Critical load and con-
gestion instabilities in scale-free networks. Europhysics Letters, 62, 2003.
[139] M. G. Morgan, J. Apt, L. B. Lave, M. D. Ilic, M. Sirbu, and J. M. Peha. The many
meanings of “smart grid”. Technical report, Carnegie Mellon University, 2009.
[140] K. Moslehi, A. Kumar, E. Dehdashti, P. Hirsch, and W. Wu. Distributed autonomous
real-time system for power system operations - a conceptual overview. In IEEE PES Power
Systems Conf. and Expo, volume 1pp, pages 1705–1712. IEEE, 2004.
[141] M. Munasinghe. Engineering-economic analysis of electric power systems. Proceedings
of the IEEE, 72(4):424 – 461, 1984.
[142] B. Murray. Power Markets and Economics: Energy Costs, Trading, Emissions. Wiley, 2009.
[143] A. Nasiruzzaman and H. Pota. Critical node identification of smart power system us-
ing complex network framework based centrality approach. In North American Power
Symposium, 2011, pages 1 –6, 2011.
[144] S. Nath and R. Somnath. Network Reconfiguration for Electrical Loss Minimization.
Inernational Journal of Instrumentation, Control and Automation, 1, 2011.
[145] National Energy Technology Laboratory. A system view of the modern grid. Technical
report, U.S. Department of Energy - Office of Electricity Delivery and Energy Reliability,
2007.
[146] National Energy Technology Laboratory. A vision for the modern grid. Technical re-
port, U.S. Department of Energy - Office of Electricity Delivery and Energy Reliability,
2007.
[147] National Energy Technology Laboratory. Building a smart grid business case. Technical
report, U.S. Department of Energy - Office of Electricity Delivery and Energy Reliability,
2009.
[148] National Grid. Undergrounding high voltage electricity transmission - the technical
issues. Technical report, National Grid, 2009.
[149] Netbeheer Nederland. Betrouwbaarheid van elektriciteitsnetten in nederland - resul-
taten 2012. Technical Report RM-ME-13L10440006, Netbeheer Nederland, 2013.
[150] M. E. J. Newman. Assortative mixing in networks. Phys. Rev. Lett., 89:208701, 2002.
[151] M. E. J. Newman. The structure and function of complex networks. SIAM REVIEW,
45:167–256, 2003.
[152] M. E. J. Newman. Analysis of weighted networks. Phys. Rev. E, 70(5):056131, 2004.
[153] M. E. J. Newman. Networks: An Introduction. OUP Oxford, 2010.
[154] P. D. O’Brien and R. C. Nicol. Fipa towards a standard for software agents. BT Technol-
ogy Journal, 16(3):51–59, 1998.
[155] Office of the National Coordinator for Smart Grid Interoperability. Nist framework and
roadmap for smart grid interoperability standards, release 1.0. Technical Report NIST
Special Publication 1108, National Institute of Standards and Technology, 2010.
BIBLIOGRAPHY 191
[156] G. A. Pagani and M. Aiello. Energy market trading systems in G6 countries. Technical
Report JBI preprint 2010-6-01, JBI, University of Groningen, 2010. Available at http:
//www.cs.rug.nl/˜andrea/publications/energyMarketG6.pdf.
[157] G. A. Pagani and M. Aiello. Towards a service-oriented energy market: Current state
and trend. In E. Maximilien, G. Rossi, S.-T. Yuan, H. Ludwig, and M. Fantinato, editors,
Service-Oriented Computing, volume 6568 of Lecture Notes in Computer Science, pages 203–
209. Springer Berlin Heidelberg, 2011.
[158] G. A. Pagani and M. Aiello. Towards decentralization: A topological investigation of
the medium and low voltage grids. Smart Grid, IEEE Transactions on, 2(3):538 –547, 2011.
[159] G. A. Pagani and M. Aiello. Power grid network evolutions for local energy trading.
Technical Report Available at arXiv:1201.0962, JBI, University of Groningen, 2012.
[160] G. A. Pagani and M. Aiello. Service orientation and the smart grid state and trends.
Service Oriented Computing and Applications, 6(3):267–282, 2012.
[161] G. A. Pagani and M. Aiello. A complex network approach for identifying vulnerabil-
ities of the medium and low voltage grid. International Journal of Critical Infrastructures,
(To appear), 2013.
[162] G. A. Pagani and M. Aiello. Cost and benefits of denser topologies for the smart grid.
In E. Gelenbe and R. Lent, editors, Computer and Information Sciences III, pages 73–81.
Springer London, 2013.
[163] G. A. Pagani and M. Aiello. From the grid to the smart grid, topologically. Technical
Report Available at arXiv:1305.0458, JBI, Univ. of Groningen, 2013.
[164] G. A. Pagani and M. Aiello. Generating realistic dynamic prices and services for the
smart grid. Technical Report JBI preprint 2013-12-01, JBI, University of Groningen, 2013.
Available at http://www.cs.rug.nl/˜andrea/publications/techRepBB.pdf.
[165] G. A. Pagani and M. Aiello. Modeling the last mile of the smart grid. In Innovative
Smart Grid Technologies (ISGT), 2013 IEEE PES, pages 1–6, 2013.
[166] G. A. Pagani and M. Aiello. The power grid as a complex network: A survey. Physica
A: Statistical Mechanics and its Applications, 392(11):2688 – 2700, 2013.
[167] G. A. Pagani and M. Aiello. Power grid complex network evolutions for the smart grid.
Physica A: Statistical Mechanics and its Applications, 396(0):248 – 266, 2014.
[168] S. Pahwa, A. Hodges, C. Scoglio, and S. Wood. Topological analysis of the power grid
and mitigation strategies against cascading failures. In Systems Conference, 2010 4th Annual
IEEE, pages 272–276, 2010.
[169] P. Paiva, H. Khodr, J. Dominguez-Navarro, J. Yusta, and A. Urdaneta. Integral plan-
ning of primary-secondary distribution systems using mixed integer linear program-
ming. Power Systems, IEEE Transactions on, 20(2):1134 – 1143, 2005.
[170] M. Panahi, W. Nie, and K.-J. Lin. A Framework for Real-Time Service-Oriented Archi-
tecture. IEEE Conf. on Commerce and Enterprise Computing, pages 460–67, 2009.
192 BIBLIOGRAPHY
[171] M. P. Papazoglou. Web Services: Principles and Technology. Pearson, 2007.

[172] M. P. Papazoglou and D. Georgakopoulos. Service-oriented computing. Commun.
ACM, 46(10):24–28, 2003.
[173] M. P. Papazoglou and W.-j. Van Den. Service oriented architectures : approaches ,
technologies and researc h issues. Vldb Journal, pages 389–415, 2007.
[174] V. Pareto. Manual of political economy (manuale di economia politica). Kelley, New York,
1971 (1906). Translated by Ann S. Schwier and Alfred N. Page.
[175] D. Pepyne. Topology and cascading line outages in power grids. Journal of Systems
Science and Systems Engineering, 16:202–221, 2007.
[176] M. Piraveenan, M. Prokopenko, and A. Y. Zomaya. Local assortativity and growth of
Internet. The European Physical Journal B, 70(2):275–285–285, 2009.
[177] M. Postina, S. Rohjans, U. Steffens, and M. Uslar. Views on service oriented architec-
tures in the context of smart grids. In Smart Grid Communications (SmartGridComm), 2010
First IEEE International Conference on, pages 25 –30, 2010.
[178] R. Kinney, P. Crucitti, R. Albert, and V. Latora. Modeling cascading failures in the north
american power grid. European Physical Journal B, 46(1):101–107, 2005.
[179] L. Ramesh, S. P. Chowdhury, S. Chowdhury, A. A. Natarajan, and C. T. Gaunt. Min-
imization of power loss in distribution networks by different techniques. Inernational
Journal of Electrical Power and Energy Systems Engineering, 3, 2009.
[180] S. Rehman, M. A. Bader, and S. A. Al-Moallem. Cost of solar energy generated using
pv panels. Renewable and Sustainable Energy Reviews, 11(8):1843 – 1857, 2007.
[181] C. Roe, S. Meliopoulos, R. Entriken, and S. Chhaya. Simulated demand response of a
residential energy management system. In Energytech, 2011 IEEE, pages 1 –6, 2011.
[182] M. Rosas-Casals. Power grids as complex networks: Topology and fragility. In Com-
plexity in Engineering, 2010., pages 21 –26, 2010.
[183] M. Rosas-Casals and B. Corominas-Murtra. Assessing European power grid reliabil-
ity by means of topological measures. WIT transactions on ecology and the environment,
121:527–537, 2009.
[184] M. Rosas-Casals, S. Valverde, and R. V. Solé. Topological vulnerability of the Euro-
pean power grid under errors and attacks. International Journal of Bifurcation and Chaos,
17(07):2465, 2007.
[185] V. Rosato, S. Bologna, and F. Tiriticco. Topological properties of high-voltage electrical
transmission networks. Electric Power Systems Research, 77(2):99–105, 2007.
[186] L. Schleisner. Life cycle assessment of a wind farm and related externalities. Renewable
Energy, 20(3):279 – 288, 2000.
[187] A. Schuelke and K. Erickson. Serving solar variations with consumption control of
smart appliances and electric vehicles. In Innovative Smart Grid Technologies (ISGT Europe),
2011 2nd IEEE PES International Conference and Exhibition on, pages 1 –8, 2011.
BIBLIOGRAPHY 193
[188] G. Shakhnarovich, T. Darrell, and P. Indyk. Nearest-Neighbor Methods in Learning and

Vision: Theory and Practice (Neural Information Processing). The MIT Press, 2006.
[189] H. Shariati, H. Askarian Abyaneh, M. Javidi, and F. Razavi. Transmission expansion
planning considering security cost under market environment. In Electric Utility Deregu-
lation and Restructuring and Power Technologies, 2008. DRPT 2008. Third International Con-
ference on, pages 1430 –1435, 2008.
[190] M. H. Shariatkhah and M. R. Haghifam. Using feeder reconfiguration for congestion
management of smart distribution network with high dg penetration. In Integration of
Renewables into the Distribution Grid, CIRED 2012 Workshop, pages 1 –4, 2012.
[191] R. V. Solé, M. Rosas-Casals, B. Corominas-Murtra, and S. Valverde. Robustness of the
European power grids under intentional attack. Physical Review E, 77(2):1–7, 2008.
[192] Y. Song, G. Tang, Z. Yang, and J. Hu. The Technical Implementation of the Electric-
ity Market Operation Syste m. In 2005 IEEE/PES Transmission and Distribution Conf. and
Exhibition: Asia and Pacific, pages 1–6. IEEE, 2005.
[193] P. Sotkiewicz and J. Vignolo. Nodal pricing for distribution networks: efficient pricing
for efficiency enhancing dg. Power Systems, IEEE Transactions on, 21(2):1013–1014, 2006.
[194] P. Stephenson. Electricity market trading. Power Engineering J., 15(6):277, 2001.
[195] M. Strobbe, T. Verschueren, K. Mets, S. Melis, C. Develder, F. De Turck, T. Pollet, and
S. Van de Veire. Design and evaluation of an architecture for future smart grid service pro-
visioning. In Network Operations and Management Symposium (NOMS), 2012 IEEE, pages
1203 –1206, 2012.
[196] S. H. Strogatz. Exploring complex networks. Nature, 410(6825):268–76, 2001.
[197] S. Sucic, A. Martinic, and A. Kekelj. Utilizing standards-based semantic services for
modeling novel smart grid supervision and remote control frameworks. In Industrial
Technology (ICIT), 2012 IEEE International Conference on, pages 409 –414, 2012.
[198] K. Sun. Complex networks theory: A new method of research in power grid. In Trans-
mission and Distribution Conference and Exhibition: Asia and Pacific, pages 1–6, 2005.
[199] D. T. Swift-Hook. Grid-connected intermittent renewables are the last to be stored.
Renewable Energy, 35(9):1967 – 1969, 2010.
[200] H. Takamori, K. Nagasaka, and E. Go. Toward Designing Value Supportive Infrastruc-
ture for Electricity Trading. The 9th IEEE Int. Conf. on E-Commerce Technology (CEC-EEE
2007), pages 167–174, 2007.
[201] H. Takamori, K. Nagasaka, and E. Go. Toward designing value supportive infrastruc-
ture for electricity trading. In E-Commerce Technology and the 4th IEEE International Confer-
ence on Enterprise Computing, E-Commerce, and E-Services, 2007. CEC/EEE 2007., pages 167
–174, 2007.
[202] W. Taqqali and N. Abdulaziz. Smart grid and demand response technology. In Energy
Conference and Exhibition (EnergyCon), 2010 IEEE International, pages 710 –715, 2010.
194 BIBLIOGRAPHY
[203] J. Travers and S. Milgram. An experimental study of the small world problem. Sociom-
etry, 32(4):425–443, 1969.
[204] W. Tsai, Y.-h. Lee, Z. Cao, Y. Chen, and B. Xia o. RTSOA: Real-Time Service-Oriented
Architecture. 2006 Second IEEE Int. Sym. Service-Oriented System Engineering (SOSE’06),
pages 49–56, 2006.
[205] L. Tsoukalas and R. Gao. From smart grids to an energy internet: Assumptions, ar-
chitectures and requirements. In Electric Utility Deregulation and Restructuring and Power
Technologies, 2008. DRPT 2008. Third International Conference on, pages 94–98, 2008.
[206] U.S. Department of Energy - Office of Electric Transmission and Distribution. “grid
2030” a national vision for electricity’s second 100 years. Technical report, U.S. Depart-
ment of Energy - Office of Electric Transmission and Distribution, 2003.
[207] V. Vaitheeswaran. Power to the People. Earthscan, 2005.
[208] K. H. van Dam and E. J. L. Chappin. Coupling Agent-Based Models of Natural Gas
and Electricity Markets. In First International Workshop on Agent Technologies for Energy
Systems, ATES2010, Toronto, 2010.
[209] A. Vázquez, R. Pastor-Satorras, and A. Vespignani. Large-scale topological and dy-
namical properties of the Internet. Physical Review E, 65(6):1–12, 2002.
[210] T. Verschueren, W. Haerick, K. Mets, C. Develder, F. De Turck, and T. Pollet. Architec-
tures for smart end-user services in the power grid. In Network Operations and Management
Symposium Workshops (NOMS Wksps), 2010 IEEE/IFIP, pages 316 –322, 2010.
[211] P. Vytelingum, S. Ramchurn, T. Voice, A. Rogers, and N. Jennings. Agent-based mod-
eling of smart-grid market operations. In Power and Energy Society General Meeting, 2011
IEEE, pages 1–8, 2011.
[212] C. A. Walford. Wind turbine reliability: understanding and minimizing wind turbine
operation and maintenance costs. Technical Report SAND2006-1100, Sandia National
Laboratries, 2006.
[213] D. Wall, G. Thompson, and J. Northcote-Green. An optimization model for planning
radial distribution networks. Power Apparatus and Systems, IEEE Transactions on, PAS-
98(3):1061 –1068, 1979.
[214] D. Wang, B. de Wit, S. Parkinson, J. Fuller, D. Chassin, C. Crawford, and N. Djilali. A
test bed for self-regulating distribution systems: Modeling integrated renewable energy
and demand response in the gridlab-d/matlab environment. In Innovative Smart Grid
Technologies (ISGT), 2012 IEEE PES, pages 1 –7, 2012.
[215] J.-W. Wang and L.-L. Rong. Cascade-based attack vulnerability on the us power grid.
Safety Science, 47(10):1332–1336, 2009.
[216] Z. Wang, A. Scaglione, and R. Thomas. Electrical centrality measures for electric power
grid vulnerability analysis. In Decision and Control (CDC), 2010 49th IEEE Conference on,
pages 5792 –5797, 2010.
BIBLIOGRAPHY 195
[217] Z. Wang, A. Scaglione, and R. Thomas. Generating statistically correct random topolo-
gies for testing smart grid communication and control networks. Smart Grid, IEEE Trans-
actions on, 1(1):28–39, 2010.
[218] Z. Wang, A. Scaglione, and R. Thomas. The node degree distribution in power grid and
its topology robustness under random and selective node removals. In Communications
Workshops (ICC), 2010 IEEE International Conference on, pages 1 –5, 2010.
[219] Z. Wang, R. Thomas, and A. Scaglione. Generating random topology power grids. In
Hawaii International Conference on System Sciences, Proceedings of the 41st Annual, page 183,
2008.
[220] C. Warmer, K. Kok, S. Karnouskos, A. Weidlich, and D. Nestle. Web services for inte-
gration of smart houses in the smart grid. In Grid-Interop 2009, pages 1–5, 2009.
[221] D. J. Watts. Small Worlds: The Dynamics of Networks between Order and Randomness.
Princeton Univ. Press, Princeton, NJ, USA, 2003.
[222] D. J. Watts and S. H. Strogatz. Collective dynamics of ‘small-world’ networks. Nature,
393(6684):440–442, 1998.
[223] D. Whitehead. The el farol bar problem revisited: Reinforcement learning in a poten-
tial game. ESE Discussion Papers 186, Edinburgh School of Economics, University of
Edinburgh, 2008.
[224] R. Wiser, G. Barbose, and C. Peterman. Tracking the sun: the installed cost of photo-
voltaics in the u.s. from 1998-2007. Technical report, Lawrence Berkeley National Labora-
tory, 2009.
[225] Y. Xiaodan, J. Hongjie, W. Chengshan, W. Wei, Z. Yuan, and Z. Jinli. Network recon-
figuration for distribution system with micro-grids. In Sustainable Power Generation and
Supply, 2009. SUPERGEN ’09. International Conference on, pages 1 –4, 2009.
[226] Q. Yang, J. Barria, and T. Green. Communication infrastructures for distributed control
of power distribution networks. Industrial Informatics, IEEE Transactions on, 7(2):316 –327,
2011.

Pagani

Uploaded by

Copyright:

Available Formats

Pagani

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Pagani

Uploaded by

Copyright:

Available Formats

From the Grid to the Smart Grid,

Giuliano Andrea Pagani

Printed by NetzoDruk - www.netzodruk.nl - Groningen

From the Grid to the Smart Grid, Topologically

ter verkrijging van het doctoraat in de

Giuliano Andrea Pagani

geboren op 29 maart 1981

Beoordelingscommissie: Prof. dr. P. Pelacchi

3 Network Models for the Smart Grid 41

4 Network Evolutions for the Smart Grid 67

5 ICT Services and Applications for the Smart Grid 125

5.2.7 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

A Graph Theory and Complex Network Fundamentals 159

B Complex Network Models 167

1.1 The New Grid

1.1.1 Motivations for the future grid

1.1.2 What is the smart grid?

• The electricity distribution companies can have a higher level of automation

Figure 1.2: NIST smart grid framework.

some peculiar aspects, such as demand-side management, where utilities compa-

• self-diagnose and take appropriate recovery actions in case of faults,

• become more resistant to willing and malicious attacks,

• be ready to integrate different sources of power generation,

1.2 Thesis Scope and Organization

capabilities compared to the current situation. Each physical piece of equipment

Figure 1.3: Smart grid physical and information infrastructure.

sented and a direction for the future work.

Chapter Venue Citation

2.1 The Visions of New Grids

2.2 Complex Network Analysis and the Power Grid

2.2.1 Power grid characteristics

2.2.2 Statistical global graph properties

approximation to give the idea of the importance of the sample.

Work Number of Number of Sample Network Geography

[215]  4900  6600  2.69 Unweighted X

behavior of the grid to various attacks to its nodes or edges.

2.2.3 The small-world property

2.2.4 Node degree distribution

Work Cumulative Node Fitted

[184] exponential y pxq  e0.56x

[191] exponential y pxq  e0.61x

[198] exponential y pxq  e0.5x

f2 pxq  x x 0.4085p0.5916qx  t0.3545δ p1q, 0.4499δ p2q, 0.1956δ p3qu xi  1, 2, . . . 16

2.2.5 Betweenness distribution

Work Cumulative Betweenness Fitted

Table 2.4: Comparison of the betweenness cumulative distribution probability func-

Power-law seems the dominating rule for betweenness probability distribution

2.3 Power Grid, Network Design and Evolution

2.3.1 Evolution of complex networks

underlying social aspects of the co-author relationship such as neighbors in com-

topology. In particular, the model is able to capture the dissortativeness of hubs

2.3.2 Complex network analysis for power grid design

2.3.3 Power engineering approach for power lines design

Traditionally power system engineers adopt techniques from electrotechnics, which

2.3.4 Power lines adaptation

2.4 Smart Grid Services and Applications

2.4.1 Smart grid software applications and architectures

2.4.2 Service-oriented architectures for energy

Service-oriented architecture and smart grid

3.1 The Dutch Medium and Low Voltage Grid

[215] 4900 6600 2.69 Unweighted X

[184] exponential y pxq e0.56x

[191] exponential y pxq e0.61x

[198] exponential y pxq e0.5x

f2 pxq x x 0.4085p0.5916qx t0.3545δ p1q, 0.4499δ p2q, 0.1956δ p3qu xi 1, 2, . . . 16

α f pLlineN , LsubstationN q; (3.3)

β f pRobN , RedN , CapN q. (3.4)

RobN |M CCRandom20% | 2|M CCN odeDegree20% | (3.7)

α f pLlineN , LsubstationN q LlineN LsubstationN (3.10)

β f pRedN , RobN , CapN q

– A low value for average betweenness in terms of order of the graph υ N σ