4458 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO.

5, SEPTEMBER 2018

A Microgrid Multilayer Control Concept for

Optimal Power Scheduling and Voltage Control
Christof Deckmyn, Tine L. Vandoorn, Jan Van de Vyver, Jan Desmet, and Lieven Vandevelde

Abstract—In this paper, a novel multilayer control structure aggregation of (controllable) loads and DER units including
for microgrids is proposed. A scheduling layer comprehends the dispatchable sources, renewable energy sources (RES) and dis-
minimization of the microgrid operating costs together with the tributed storage devices. They can operate either connected
CO2 emissions produced and provides a sequence of power ref-
erences for the next 24 h. Subsequently, within the executive or isolated from the utility grid. Since the architecture of
layer, an off line ac power flow calculation will be performed to microgrids can change on a regular basis, and in order to
obtain the initial values of the voltage magnitudes of the differ- increase their potential of scalability, control strategies should
ent microgrid buses. The adjustment layer, which is the scope be implemented in a plug-and-play fashion.
of this paper, includes a control strategy to maintain the voltage Controllability is crucial to the concept of microgrids. The
in the network. The purpose of this third layer is to keep the
voltage within a pre-specified tolerance band by adjusting the distributed structure of a microgrid comprises a network of
power provided by the microgrid distributed energy resources local controllers connected to each microgrid element. It is
(DER). Depending on the voltage deviation, the location of the with these power electronic inverters, cooperative actions can
DER units in the network and their distance from the voltage be taken to control the microgrid [2]. The control concept
deviation, an appropriate dynamic gain will be provided to the within microgrids is conceived as a hierarchical control, which
relevant DER units. The renewed settings are then fed back to
the first layer, which performs a new optimization and redis- is adapted from the transmission level [3]–[8]. In literature,
tributes the adjusted reference set points among the DER units. prior work has been done on hierarchical energy management
The performance and effectiveness of the proposed hierarchical systems for microgrids. In [9], an energy management strategy
multilayer control structure were evaluated and demonstrated by was developed for microgrids in grid-connected mode without
several case studies. considering cases of operation in islanded mode. In [10], a
Index Terms—Microgrid, environomic, genetic algorithm, double-layer coordinated control approach for microgrids is
multilayer, voltage control. developed and tested for both operating modes. This strategy
obtains an economic optimization of the microgrid without
considering the environmental aspects. In [11] and [12], the
I. I NTRODUCTION voltage of weak microgrids with inverter-connected sources
was discussed. Both do not consider the economical nor the
ER PLAY a major role in the path towards the future
D energy landscape. These technologies, which are cur-
rently in different stages of development, drive us to a more
environmental optimization of the microgrid in grid-connected
as well as in islanded operating mode.
In this paper, the future operational states of the controllable
decentralized energy world. Fuel cells, wind turbines, pho-
units within the microgrid are determined ahead of time. The
tovoltaics, or more traditional energy sources such as diesel
proposed concept follows the idea of a microgrid day-ahead
generators, are being deployed at a rapid pace. Deploying
planning including the unit commitment problem (scheduling
DER in a widespread, efficient and cost-effective way requires
layer), an off line power flow calculation (executive layer)
advanced control strategies and can help the distribution grid
and a security check with feedback control (adjustment layer).
in the progress towards a smarter grid. The trend towards the
Since the complete multilayer control concept works on a
enhancement of the environomical effectiveness and resilience
day-ahead time scale, the multilayer model can be consid-
of the main grid can be supported by smart microgrids [1].
ered as an off line optimization approach. With this approach,
Microgrids are subsystems of the main grid consisting of an
the problem is solved on a day-ahead basis, allowing an
Manuscript received November 5, 2015; revised April 15, 2016, September online implementation to be achieved via real-time system
12, 2016, and December 26, 2016; accepted January 23, 2017. Date of pub- state updates. The proposed multilayer control is presented
lication January 25, 2017; date of current version August 21, 2018. This in Fig. 1. The scheduling layer finds out the power set points
work of J. Van de Vyver was supported by the SBO Project OptiWind.
Paper no. TSG-01426-2015. (at quarter-hourly intervals) of the DER units and is based the
C. Deckmyn, T. L. Vandoorn, J. Van de Vyver, and L. Vandevelde are day-ahead energy market, the demand bids, and the availabil-
with the Electrical Energy Laboratory, Department of Electrical Energy, ity of renewable energy sources. The executive layer performs
Systems and Automation, Ghent University, 9052 Ghent, Belgium (e-mail:
christof.deckmyn@ugent.be). an off line system load flow calculation in order to determine
J. Desmet is with Lemcko, Department of Electrical Energy, Systems and the voltages at the different buses in the network. The adjust-
Automation, Ghent University, 8500 Kortrijk, Belgium. ment layer provides a dynamic gain to adjust the boundaries
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. of the microgrid generation units according to the data com-
Digital Object Identifier 10.1109/TSG.2017.2658865 ing from the executive layer. This layer is utilized to maintain
1949-3053 c 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
DECKMYN et al.: MICROGRID MULTILAYER CONTROL CONCEPT FOR OPTIMAL POWER SCHEDULING AND VOLTAGE CONTROL 4459

15 minute power generation schedule will be calculated for

the next 24 hours. In order to solve this multi-objective UC
problem, a hybrid goal attainment based genetic algorithm is
used. The algorithm is based on the evolutionary ideas of
natural selection, mutation and recombination. In literature,
genetic algorithms are commonly used for power dispatch
optimization, UC and voltage profile improvement [15]–[20].
The operation of the algorithm will not be further discussed
in detail, since the scope of this paper is voltage control.
Furthermore, in the paper, we did not consider the forecast pro-
cess or techniques nor the accuracy on these forecasts, since
these parameters depends on several factors including local
climate, forecast horizon, whether forecasts apply to a single
point or cover a wide geographic area, etc.
The objective is to minimize the cost and CO2 emissions
of the microgrid local energy production:

Min(CG (PG ), εG (PG )) (1)

where CG (PG ) and εG (PG ) are the operating cost and the CO2
emission of the microgrid power generation units respectively.
Both should be minimised over the variable PG of which their
Fig. 1. Microgrid multilayer control structure. values can be considered as the active power reference set
points (kW). A fuel cost function describes the cost to oper-
ate a generator to produce an additional unit of power. The
the voltage within the acceptable limits. Subsequently, when most prevalent fuel cost model, is the quadratic model which
closing the loop, the scheduling layer recalculates and updates describes the fuel cost as a quadratic function of the genera-
the power settings of the DER units by considering the output tors active power output. Fuel-based thermal generators where
from the adjustment layer as a feedback. This iterative process fuel is oil, coal and gas (e.g., small internal combustion engine
will continue until all voltage violations are eliminated and a generators, micro turbines and fuel cells) have a quadratic fuel
converged optimal solution is found. cost function [21]. Accordingly, the operating cost, in e/h, of
This paper is organized as follows: Section II, discusses these generating units within the scheduling time horizon can
the multilayer control structure. In Section III, the proposed be presented as follows:
voltage control approach is described and verified with a
N N
 
clear example. A demonstration and validation is given in gen
 
Section IV. Section V, concludes the paper. CG (PG ) = aj (i) + bj (i) · PG j + (i − 1) · Ngen
i=1 j=1
 
II. M ICROGRID M ULTILAYER C ONTROL S TRUCTURE + cj (i) · P2G j + (i − 1) · Ngen (2)
The proposed multilayer control concept, which is presented
where aj , bj and cj are the fuel cost coefficients of the jth gen-
in Fig. 1, is decoupled into three sub-problems, including a UC
eration unit. The form of the emission function model depends
problem (scheduling layer), a load flow calculation (executive
on the types of pollutants. In general, CO2 emissions of fuel-
layer) and a security check with feedback control (adjustment
burning generators are directly related to the consumed amount
layer). As introduced before, the focus of this multilayer model
of fuel, and can be expressed based on the quadratic fuel con-
is on the optimal off line microgrid planning which can pro-
sumption curves [22]. Hence, the amount of CO2 in kg/h can
vide an input for an online microgrid application. The different
be expressed as a quadratic function of the generators active
levels of the microgrid multilayer control structure will be
power output:
described in the following sections.
N N
 gen
 
A. Scheduling Layer εG (PG ) = dj (i) + ej (i) · PG j + (i − 1) · Ngen
The scheduling layer forms the highest level in this micro- i=1 j=1
 
grid multilayer control concept. Within this layer, a day-ahead + fj (i) · P2G j + (i − 1) · Ngen (3)
UC strategy is implemented which involves the simultaneous
optimization of the fuel costs and the CO2 emissions of the where dj , ej and fj are the emission coefficients.
local generators [13], [14]. This layer starts with the initial- The objective function presented in (1) is constrained by
ization of the characteristics of the DER units, the availability physical relations in load flows and system performance
of the (predicted) renewable energy production, the demand requirements. A first constraint is that the sum of each power
bids, and the day-ahead market conditions. Subsequently, a generating unit (including the slack bus generator) must equal
4460 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 5, SEPTEMBER 2018

the actual consumption within the microgrid, and is expressed When the storage unit is discharging, active power will be
below: injected in the microgrid bus and Pstorage has a positive value.
Ngen For discharging, the set point Pstorage during the time span Ts

PGj = PaD (4) of a time step is constrained by:
j=1 Pstorage · Ts ≤ Estorage − Emin (11)
were PaD , represents the actual demand. In this paper, the line Note that Estorage should not be confused with Pstorage · Ts .
losses were included in PaD and are assumed to be 5.05% of Estorage represents the state of charge (SOC) over a period of
the microgrid energy actual demand [23]. Since RES, such as N · Ts . Whereas Pstorage · Ts indicates how much energy the
PV and wind, produce power at zero running cost and zero storage unit (dis)charges at a given time step Ts . Furthermore,
emissions, their output power can be treated as a negative the rate of the storage unit (for charging and discharging) is
load. Therefore, the power generated by renewables PRES is constrained and already included in (7).
subtracted from the total demand Ptotal
D in order to obtain the In addition, a flexible load with the characteristics of an
actual demand PaD : electric vehicle is implemented and can only store energy (act-
ing as a load). The index to the variable name PEV is given
PaD = Ptotal
D − PRES (5) as an indication to the reference set point of the electric vehi-
The total demand includes a fixed part (Pfixed cle (EV). The flexible load can only be committed when it is
D ) which is non-
controllable and a flexible part (Pflexible ): connected to the microgrid (between Tin and Tout ):
D

Tout
D = PD
Ptotal + Pflexible .
fixed
D (6) EEV = EEVTin + PEV (i) · Ts (12)
i=Tin
In this work, as a part of a demand side participation strategy,
a flexible load is introduced which can be controlled within where EEV represents the energy level of the EV. EEVTin is
its limits when it is connected to the microgrid. This will be the energy stored in the EV at the time step when the electric
explained further in this section. vehicle is plugged in. Similar to the storage device, EEV has
The power reference set point of each microgrid unit is a defined range with a minimum and a maximum:
bounded by the upper bound and by the lower bound, PGmin EEVmin < EEV < EEVmax (13)
and PGmax respectively.
An extra constraint was added where the EV should be charged
PGmin < PG < PGmax (7) with a certain amount of energy on the time step it will be
unplugged (EEVTout ) and can be represented as a percentage
In this paper, a storage element is added which participates
of EEVmax .
in the commitment strategy where the charge and discharge
When the EV is committed to charge and acting as a load,
commands are allocated according to the lowest cost and the
PEV has a negative value. For charging, the set point PEV
lowest CO2 emissions. Besides, the commands of charging and
during the time span Ts of a time step is constrained by:
discharging are associated to the energy prices as well which
takes advantage of purchasing power from the upstream grid PEV · Ts ≤ EEVmax − EEV (14)
(slack bus generator) and selling it back according to the most
The active power rate of the EV unit is non-negative and has
favorable microgrid revenue. Pstorage is given as an indication
a defined range with a minimum and a maximum (7).
to the reference set point of the storage unit. Since Pstorage is
Given the input characteristics, the objectives, the
an active power reference set point (kW) and Estorage (kWh)
(in)equality constraints, and the algorithm options, the power
represents the (dis)charged energy over a period with N time
set points of the different microgrid generation units are cal-
steps, Pstorage is multiplied by the time span (Ts ) of the time
culated (for each time step) in a way that over the complete
step as presented below:
time horizon the objectives are optimized.

N
Estorage = Epre−storage + Pstorage(i) · Ts (8) B. Executive Layer
i=1
In order to determine the future operational states of the
where Epre-storage is the energy stored in the storage unit at the microgrid units whereby the bus voltages are maintained
end of the previous day. At any time, Estorage or the energy within acceptable limits, an off line load flow calculation
level of the storage unit is non-negative and has a defined should be performed to obtain the initial values of the volt-
range with a minimum and a maximum. age magnitudes of the different microgrid buses. Therefore,
within the executive layer, an AC power flow calculation will
Emin < Estorage < Emax (9) be performed with the power reference set points calculated
In case the storage unit is charging and acting as a load, Pstorage by the scheduling layer to obtain the values of the voltage
has a negative value. For charging, the set point P storage during magnitudes of the different buses in the microgrid network. In
the time span Ts of a time step is constrained by: this paper, a microgrid model was built with Matlab Simulink
to perform a load flow calculation and extract the numerical
Pstorage · Ts ≤ Emax − Estorage (10) values of the bus voltages.
DECKMYN et al.: MICROGRID MULTILAYER CONTROL CONCEPT FOR OPTIMAL POWER SCHEDULING AND VOLTAGE CONTROL 4461

C. Adjustment Layer
After the load flow calculation, the numerical values of the
bus voltages are transmitted to the third layer. The adjust-
ment layer, which is the scope of this paper, is utilized to
maintain the bus voltages and contains a security check with
feedback control to the first layer. The purpose of this third
layer, is to keep the voltage within a pre-specified tolerance
band by producing more or less power provided by the micro-
grid generation units. In case of a severe voltage deviation at
any bus within the microgrid, a dynamic gain will be intro-
duced. Depending on the deviation (positive or negative), and
on the location of the power generation units and their respec-
tive distance from the voltage problem, an appropriate gain
will be provided to adjust the boundaries (7) of the microgrid
generation units. In this way, the range (between the lower and
upper bound) in which the set point can be committed will be
modified. Subsequently, when closing the loop, the renewed
boundaries are then fed back to the scheduling layer, which
recalculates the UC solution. The updated power generation
schedule will again go through the second and third layer of
the control hierarchy. This iterative process will continue until
the voltage violations are eliminated and a converged optimal
solution is found. Due to the introduction of this third layer,
active power reference set points can be provided not only
according to the economical and the environmental objectives,
Fig. 2. Flowchart security check with feedback voltage control approach.
but also taking to account the voltage level at every bus in the
microgrid. The solutions can be used for an online microgrid
system state update. In the next section, the voltage control will be explained in the next section. In case of an undervolt-
method will be presented in detail. age, the lower boundaries will be increased by the dynamic
gain in order to increase the reference set points and to reduce
III. VOLTAGE C ONTROL A PPROACH the voltage deviation. In case of an overvoltage, the upper limit
Microgrids are receiving a growing interest as they can will be decreased and as a result the reference set points will
provide a coordinated approach for the integration of DER be reduced. The output power of RES can only be adjustment
units [24]. In comparison with traditional networks [25], by curtailment (by adjusting their upper boundaries).
microgrids are often based on mainly resistive distribution With this third layer (and feedback loop), active power ref-
network lines [26]. Typical R/X values vary between 2 and 8. erence set points can be provided (for real-time scheduling)
As a result, the active power/voltage (P/V) control strategies according to the economical and the environmental objec-
are becoming more relevant in these microgrids because of tives, and taking to account the microgrid bus voltages. In
the linkage between P and V [27], [28]. Therefore, the voltage the following subsection, the voltage control method will be
control method presented in this work is done by influencing explained in more detail.
the active power set points of the microgrid generation units.
B. Mathematical Model
A. Control Strategy Distribution networks are typically operated loopy or
In this subsection, the dynamic voltage control approach, meshed. However, for practical engineering concerns, the
which is embedded in the adjustment layer, will be explained. majority of distribution grids are operated as radial networks.
The process of the control strategy via a dynamic gain which This radial topology distinguishes distribution networks from
influences the boundaries of the microgrid generation units transmission networks which have multiple loops energized
can be explained using the flowchart depicted in Fig. 2. In a all the time to guarantee continuous delivery of power to
first step, the numerical values of the bus voltages transmitted every node. The microgrid considered in the paper has a
by the executive layer will be initialized and evaluated. If one radial network topology. The structure of a radial distri-
of the bus voltages exceeds the permissible limits, the voltage bution network has important features which motivates the
control approach will be activated in order to adjust the bound- development of this control strategy.
aries of the microgrid generation units. Depending on the The dynamic gain, which influences the boundaries and as
deviation (under- or overvoltage), a ‘factor’ and a ‘network’ a result the reference set points of the power generation units,
matrix will be introduced in order to construct the dynamic will be explained by introducing a network matrix (N ) and
gain which modifies the boundaries of the microgrid gener- a factor matrix (F). The network matrix is represented as
ation units. The definition and construction of both matrices an Nunit -by-Nbus matrix and contains the contribution of the
4462 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 5, SEPTEMBER 2018

microgrid units (connected to the appropriate bus) to mitigate The input entries FUVm,i and FOVm,i are presented below:
the voltage deviation at a certain bus. Nunit can be considered
FUVm,i = fUV · vm (%) (21)
as the number of units within the microgrid, such as loads,
controllable loads, RES and DER units including distributed FOVm,i = fOV · vm (%) (22)
generation (DG) units and distributed storage (DS) devices. where the factors fUV and fOV are represented as dimensionless
⎡ ⎤ binaries which indicate an under- or overvoltage. In case of
Cunit1 to bus1 ··· Cunit1 to busNbus
⎢ Cunit2 to bus1 ··· Cunit2 to busNbus ⎥ an undervoltage deviation (Vbus < Vmin ), factor fUV will be
⎢ ⎥
NNunit ,Nbus = ⎢ .. .. .. ⎥ (15) set to ‘1’ and fOV will be set to ‘0’. In case of an overvoltage
⎣ . . . ⎦ (Vbus > Vmax ), fUV will be set to ‘0’ and fUV will be set to
CunitNunit to bus1 · · · CunitNunit to busNbus ‘1’. In case no voltage deviation occurs, both values are set
to ‘0’. Subsequently, as presented in (21) and (22), the factor
where, Cunitk to busm represents the contribution of unit k to the values are multiplied by the percentage deviation with respect
voltage deviation at bus m. This contribution is represented to VN of the corresponding bus (vm (%) = VbusV−V nom
· 100).
by a positive value which depends on the distance (Dk to m ) of nom
This will make the dynamic gain dependent on the voltage
the microgrid unit to the location of the bus with the voltage deviation of the corresponding bus.
deviation, and is dimensionless. In order to build up the dynamic gain which enables the
1 microgrid units to contribute proportionally to the voltage
Cunitk to busm = 1 + (16)
D deviation and with the distance to the location (of the volt-
where D is a dimensionless quantity: age deviation), the network matrix (N ) and the factor matrix
(F) are multiplied. As a result, the dynamic gain for the lower
Lk to m and upper boundaries are constructed as Nbus -by-N matrices
D= (17)
Lmax and are presented in (23) and (24).
⎡ ⎤
with, Alb1,1 ··· Alb1,N
⎢ Alb2,1 ··· Alb2,N ⎥
Lmax = maxk,m (Lk to m ) (18) ⎢ ⎥
N · FUV = ⎢ .. .. .. ⎥ (23)
⎣ . . . ⎦
Lk to m is the line length which reflects the line impedance from AlbNbus ,1 ··· AlbNbus ,N
unit k to bus m. The closer the unit is located to the voltage ⎡ ⎤
deviation, the greater it’s contribution to the voltage deviation. Aub1,1 ··· Aub1,N
⎢ Aub2,1 ··· Aub2,N ⎥
In case of a voltage deviation at the location of the microgrid ⎢ ⎥
N · FOV = ⎢ .. .. .. ⎥ (24)
unit, the contribution of the unit itself will be set to ‘1’. This ⎣ . . . ⎦
is because the voltage deviation caused in such a case, is due AubNbus ,1 ··· AubNbus ,N
to, and can be reduced by the unit itself. At the same time the
contribution of the other units will be set to ‘0’. where, Albk,i and Aubk,i representing the gain of the lower and
Note that, the network matrix (N ) was built up including upper boundaries of the kth microgrid unit on the ith time step.
every unit within the microgrid (as well the non-controllable Note that a dynamic gain is provided for every unit within the
loads who are not participating in the UC). Therefore, within microgrid since this is a result of the multiplication with the
N , the number of microgrid units (Nunit with index k) is used network matrix (N ). However, non controllable loads are not
and not Ngen , which includes only the microgrid units who participating in the UC strategy, and therefore, they cannot be
are participating the UC. Since in this work, only controllable controlled by a certain gain. This was included by providing
loads, RES and DER units (including DG and DS units) are a contribution value of ‘0’ to the units who are not participat-
assumed to contribute to voltage deviations, the contribution ing in the UC strategy as explained in (15). As a result, the
of the non-controllable loads was set to ‘0’ within N . dynamic gain of these units will be zero.
Secondly, 2 factor matrices are introduced which indicate Depending on the deviation (positive or negative), and on
on which bus a voltage deviation occurs. Both, a factor matrix the location of the power generation units and their respec-
for undervoltages (FUV ) and a factor matrix for overvoltages tive distance from the voltage problem, an appropriate gain
(FOV ) will be used. They can be presented as an Nbus -by-N will be provided to adjust the boundaries of the microgrid
matrix (N time steps). units who are participating in the voltage control strategy.
⎡ ⎤ When an undervoltage occurs, the lower bound of the micro-
FUV1,1 ··· FUV1,N grid units should be increased. In case of an overvoltage, the
⎢ FUV2,1 ··· FUV2,N ⎥
⎢ ⎥ upper bounds of the microgrid units should be decreased. The
FUV = ⎢ .. .. .. ⎥ (19) relationship between the dynamic gain and the boundaries is
⎣ . . . ⎦
FUVNbus ,1 ··· FUVNbus ,N presented in the equations below:
⎡ ⎤ PGmink,i = PGmink,i − Prange · Albk,i
FOV1,1 ··· FOV1,N (25)
⎢ FOV2,1 ··· FOV2,N ⎥ PGmaxk,i = PGmaxk,i − Prange · Aubk,i (26)
⎢ ⎥
FOV =⎢ .. .. .. ⎥ (20)
⎣ . . . ⎦ where Prange represents the range between the lower and upper
FOVNbus ,1 ··· FOVNbus ,N bound. Note that the dynamic gain is proportional to the
DECKMYN et al.: MICROGRID MULTILAYER CONTROL CONCEPT FOR OPTIMAL POWER SCHEDULING AND VOLTAGE CONTROL 4463

range, and besides, depends on the type of microgrid unit.

Using (25) and (26), the range (between the lower and upper
bound) in which the set point can be committed will be modi-
fied. In case of an under voltage, Alb is negative which result in
an increase of PGmin . In case of an over voltage Aub is positive
and results in a decrease of PGmax . In the following section, a
demonstration is given in order to validate the performance and
effectiveness of the proposed hierarchical multilayer control
structure.

IV. D EMONSTRATION AND VALIDATION

In power systems, the bus voltages are considered as a
critical performance factor which should lie within a specific
range [29]. The EN50160 is a European standard that defines
voltage characteristics of the electricity supplied by public
distribution systems. According to this standard, the distribu-
tion system operator (DSO) is considered to keep the voltage Fig. 3. Microgrid test network.
between -10% (207.0 V) and +10% (253.0 V) of the nominal
distribution voltage (VN = 230 V) [30]. In purely voltage opti-
mization, the aim is to maintain the system voltage magnitudes
as close as possible to a nominal voltage. The multilayer con-
trol concept proposed in this paper, provides an economically
and environmentally viable UC that is physically feasible in
terms of voltage violations. Due to the introduction of this third
layer (in which the voltage control is embedded), active power
reference set points can be provided (for real-time scheduling)
not only according to the economical and the environmen-
tal objectives, but also taking to account the voltage level at
every bus in the microgrid. The voltage control approach aims
to reduce this deviation by providing a dynamic gain to the
corresponding power generation units in order to adjust their
boundaries in which they can be committed. In fact, there will Fig. 4. Microgrid load profile and RES production.
be deviated from the economical and environmental optimal
set points given by the first layer in order to maintain the TABLE I
voltage at every bus. This can be strengthened by the fact F UEL C OST AND E MISSION C OEFFICIENTS
that security of supply has an overriding priority. On the other
hand, a small tolerance band can increase the deviation from
the economical and environmental optimal set points. In this
paper, the voltage control approach is utilized to maintain the
voltage within ±5% tolerance band.
In the following case study, a day-ahead microgrid
was simulated with and without voltage control approach.
Additionally, events of under and overvoltages will be dis-
cussed and how they are tackled by the voltage control The cost and emission coefficients of the thermal generation
approach. The microgrid test network is presented in Fig. 3 units are listed in Table I. The microgrid generator capac-
and includes 11 low voltage (LV) buses connected downstream ities are listed in Table II. A CO2 emission intensity of
from a distribution transformer. Furthermore, the studied 381 · 10−3 kg/kWh [31] was considered for the power supply
system includes renewable energy resources, a storage unit, of the slack bus generator. The cost coefficient of the slack
two micro turbines, a fuel cell and a controllable load includ- bus generator is related to a variable pricing scheme which
ing 4 electric vehicles. Serving the microgrid load (Load 1 + is presented in Fig. 5. For the storage unit, Emin was set to
Load 2 + Load 3 + Controllable Load), active power can be 0 kWh, Emax was set to 40 kWh, and EEV max of each EV
produced either by the thermal generation units, the slack was set to 80 kWh. The charging rate of each EV is lim-
bus generator, the storage device, or the renewable energy ited between -19.2kW and -7.2kW, which results in a total
resources. (4 EVs) minimum and maximum charging rate of -28.8kW
The profiles of the different microgrid loads, and the renew- and -76.8kW respectively. The length (L) between the buses
able energy production during the complete time horizon (24 are assumed to be 200m with an R/X value of 7.7 (R = 0.642
hour, 96 time steps of 15 minutes) are depicted in Fig. 4. and X = 0.083) [32].
4464 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 5, SEPTEMBER 2018

TABLE II
M ICROGRID G ENERATOR C APACITIES

Fig. 5. Market conditions.

Fig. 6. Set points UC without voltage control.

A. Day-Ahead Unit Commitment With Voltage Control

In this section, a demonstration is given where the volt-
age control approach is embedded in the multilayer control
structure and maintains the microgrid bus voltages within
pre-specified limits.
1) Scheduling Layer: Based on the day-ahead energy mar-
ket, the demand bids, and the availability of renewable energy
sources, a 15 minute power generation schedule for the next 24
hours was calculated by the scheduling layer. The active power
is scheduled among the different microgrid units while min-
imising the operating costs together with the produced CO2
emissions. The reference set points are presented in Fig. 6. Fig. 7. Busvoltages without voltage control.
The total microgrid load (Ptotal
D ) is presented by the thick black
line. The actual load (PaD ) (excluding the power produced by
the renewable sources) is presented by the thin black line. set points calculated by the scheduling layer. The bus voltages
Between 11:00 a.m. and 14:30 p.m. (time step 44-58), and are presented in Fig. 7. During the complete time horizon,
between 18:00 p.m. and 19:30 p.m. (time step 72-78), a con- the voltage at bus 5 and bus 6, on which load 2 and RES1
gestion signal is introduced by the DSO. These time steps are are connected, are the lowest. Especially, during periods of
chosen arbitrary in order to present the influence on the com- high energy demand, the voltage drops below 5%. It is clear
mitment strategy. During this signal, the slack bus generator that during time steps 70-84 an undervoltage occur at bus
is committed to zero and no energy is exchanged between 5 and bus 6, on which load 2 is connected. This can be
the microgrid and the main grid. As a result, a transition to explained by the fact that during this period (time step 79),
islanded mode is scheduled. Because of the lack of the slack the energy demand becomes excessively high and causes
bus generator, the buying and selling price were neglected dur- and undervoltage. The same happens between time steps 22
ing these periods. The maximum power which can be delivered and 42.
depends on the upper bounds of the controllable thermal gen- During periods of low microgrid demand (time steps 1-6
erators and the stored energy. During the islanded operation and 90-96) and while the power produced by the wind turbine
mode, the fuel cell has the greatest contribution in the unit becomes very high, an overvoltage occurs at bus 11 on which
commitment. This can be explained by the fact that the fuel RES2 is connected.
cell has the lowest fuel cost and lowest emission rate. We can conclude that power scheduling without consider-
When the power produced by the wind turbine becomes ing the bus voltages can result in voltage violations. In the
high (time steps 1-6 and 90-96), the storage device will be following section, the adjustment layer is presented, which is
charged. During periods of islanded operation, the storage utilised to maintain the bus voltages and contains a security
unit will be discharged which reduces additional output power check with feedback control to the first layer.
from the thermal generation units, and as a result reduces 3) Adjustment Layer: The adjustment layer, which is the
the total cost and emission of the microgrid operation. The scope of this paper, includes a control strategy to maintain
EVs were plugged in between 09:00 a.m. and 17:00 p.m. the voltage in the network. The purpose of this third layer
(time step 36-68). At 17:00 p.m, the EVs are fully charged. is to keep the voltage within a pre-specified tolerance band
During islanded mode, the EVs are charged at their maximum by adjusting the boundaries of the microgrid units. Due to
rate in order reduce additional output power from the thermal the introduction of this third layer, the active power reference
generation units. set points will be provided (for real-time scheduling) not only
2) Executive Layer: Within the executive layer, an AC according to the economical and the environmental objectives,
power flow calculation is performed with the power reference but also taking into account the voltage level at every bus in
DECKMYN et al.: MICROGRID MULTILAYER CONTROL CONCEPT FOR OPTIMAL POWER SCHEDULING AND VOLTAGE CONTROL 4465

Fig. 9. Set points and boundaries on time step 5.

Fig. 8. Set points and busvoltages between time steps 1-15.

the microgrid. In fact, a deviation will be made from the eco-

nomical and environmental optimal set points given by the first
layer in order to maintain the voltage at every bus. In this work,
the tolerance band of the voltage control approach is set to 5%.
Based on the voltage deviation (vm (%)), it can take several
iterations until the bus voltages are controlled within the tol-
erance band. Within every iteration, a loop through each layer
of the control hierarchy will be performed. As a result, each
iteration comes with an updated set of reference set points for
each time step N of the complete time horizon. Additional, for
every set of reference set points per iteration, an updated set
of bus voltages will be provided. In order to clarify the oper-
ation of the voltage control strategy and it’s dynamic gain, we Fig. 10. Set points and busvoltages on time step 5.
will focus on the iterations within the hierarchical multilayer
control approach each time for a certain time step a voltage
deviation occurs.
The sequence of power references and the corresponding
bus voltages are presented in presented in Fig. 6 and Fig. 7.
Two events with a voltage deviation (greater that 5%) will be
discussed in this work. One between time step 1-15 and one
between time step 70-84.
The first event of a voltage deviation (greater that 5%)
occurs between time step 1 and 15 and a detailed plot of the
power reference set points and the corresponding bus voltages
during this period is presented in Fig. 8.
On time step 5, Load 1 consumes 100.68 kW, Load 2 con-
sumes 91.38 and Load 3 consumes 94.4 kW. During this time Fig. 11. Set points and busvoltages between time steps 70-84.
step, RES2 (Wind) produces 270.17 kW. As a result, the volt-
age bus 11 rises above 5% and as a result, the voltage control the bus voltages, between iteration 1 and 4, are presented in
approach will be activated which can be verified on Fig. 8. Fig. 10.
The voltage control approach aims to reduce this deviation by Note that the reference set points of the other generation
providing a dynamic gain to the corresponding power gener- units are increasing in order to cover the reduced wind power
ation units in order to adjust their boundaries, and as a result, and to maintain the power balance. After 4 iterations the
their range in which they can be committed. In this particu- overvoltage at bus 11 is reduced within the 5% tolerance band.
lar case, the upper bound of RES2 will be decreased by the A second event of a voltage deviation (greater that 5%)
dynamic gain in order to decrease the reference set points occurs between time step 70 and 84. This can be explained by
and to reduce the voltage deviation. This process continues the fact that during this period, the energy demand becomes
until the microgrid bus voltages are within acceptable limits. excessively high and causes and undervoltage in the network.
The reference set points of RES2 throughout the iterations are A closer look of the power reference set-points and the
presented in Fig. 9. corresponding bus voltages during this period is presented
It is clear that the upper bound of RES2 is decreasing in in Fig. 11.
order to decrease the reference set points and to reduce the On time step 79, Load 1 consumes 168.47 kW, Load 2
voltage deviation at bus 11. The redistribution of the reference consumes 210.59 kW and Load 3 consumes 42.11 kW. During
set points among the different microgrid units together with this time step, an undervoltage occurs at bus 5 and 6 due
4466 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 5, SEPTEMBER 2018

Fig. 14. Set points UC with voltage control.

Fig. 12. Set points and boundaries on time step 79.

Fig. 15. Busvoltages with voltage control.

in the microgrid are reduced within the 5% tolerance band.

The total cost is slightly decreased with 8.02% and the total
CO2 emission is slightly increased with 2.3%. This can be
Fig. 13. Set points and busvoltages on time step 79.
explained by the fact that the use of the slack bus generator
is reduced in order to reduce the undervoltages over the com-
to the high power demand of Load 2 which can be verified plete time horizon. In order to compensate this reduction, the
on Fig. 11. During the same iterations, the voltage control thermal generators (especially the micro turbines) were com-
approach kicks in at this time step, and aims to reduce this mitted in a higher output range, and with a higher emission
deviation by providing a dynamic gain to the corresponding rate. The difference in total cost and total emission is very less.
power generation units in order to adjust their boundaries, and This can be explained by the fact that the redistribution of the
as a result, their range in which they can be committed. The power reference set points among the different microgrid units
lower boundaries will be increased by the dynamic gain in is dependent on the voltage deviations and their location in the
order to increase the reference set points and to reduce the network. As a result, the total cost and total emission for sim-
voltage deviation. The range and the reference set points of ulations including the voltage control strategy is dependent on
the microgrid units are presented in Fig. 12. the voltage deviation.
The lower bounds of the microgrid units are increasing
in order to increase the reference set points and to increase
the voltage at bus 5. The lower boundary of the fuel cell has V. C ONCLUSION
the steepest increase since this unit has the closest location A novel multilayer control structure for microgrids is
to the voltage deviation (load 2). The increase of the lower demonstrated in this paper. The objective of this multilayer
bounds of micro turbine 1 and 2 are quite similar since their control concept is to provide an economically and environmen-
distance to bus 5 is equal. Note that the boundaries of the EVs tally viable UC that is physically feasible in terms of voltage
are not presented since they do not participate in the UC at violations. A scheduling layer comprehends the minimization
time step 79 (already unplugged). After 4 iterations the under- of the microgrid operating costs together with the CO2 emis-
voltage at bus 5 and 6 (on time step 79) is reduced within sions produced and provides a sequence of power references
the 5% tolerance band. The redistribution of the reference set for the next 24 hours. Subsequently, within the executive layer,
points among the different microgrid units together with the an AC power flow calculation will be performed to obtain the
bus voltages are presented in Fig. 13. initial values of the voltage magnitudes of the different micro-
The converged sequence of power references and the cor- grid buses. The adjustment layer, which is the scope of this
responding bus voltages are presented in Fig. 14 and Fig. 15. paper, includes a control strategy to maintain the voltage in
During the complete time horizon, the voltages at every bus the network. The purpose of this third layer is to keep the
DECKMYN et al.: MICROGRID MULTILAYER CONTROL CONCEPT FOR OPTIMAL POWER SCHEDULING AND VOLTAGE CONTROL 4467

voltage within a pre-specified tolerance band by adjusting the [12] Z. Wang, M. Xia, and M. Lemmon, “Voltage stability of weak power
power provided by the microgrid distributed energy resources distribution networks with inverter connected sources,” in Proc. Amer.
Control Conf. (ACC), Washington, DC, USA, Jun. 2013, pp. 6577–6582.
(DER). Depending on the voltage deviation, the location of the [13] C. Deckmyn, T. L. Vandoorn, M. Moradzadeh, and L. Vandevelde,
DER units in the network and their distance from the voltage “Multi-objective optimization for environomic scheduling in micro-
deviation, an appropriate dynamic gain will be provided to the grids,” in Proc. IEEE PES Gen. Meeting Conf. Expo., National Harbor,
MD, USA, Jul. 2014, pp. 1–5.
relevant DER units. The renewed settings are then fed back to [14] C. Deckmyn et al., “Day-ahead unit commitment model for microgrids,”
the first layer, which performs a new optimization and redis- IET Gener. Transm. Distrib., vol. 11, no. 1, pp. 1–9, Jan. 2017.
[15] S. Kahourzade, A. Mahmoudi, and H. B. Mokhlis, “A comparative study
tributes the adjusted reference set points among the DER units. of multi-objective optimal power flow based on particle swarm, evolu-
The performance and effectiveness of the proposed hierarchi- tionary programming, and genetic algorithm,” Elect. Eng., vol. 97, no. 1,
cal multilayer control structure were tested and presented by pp. 1–12, 2015.
[16] H. R. Magham, M. J. Sanjari, B. Zaker, and G. B. Gharehpetian,
several case studies. It is clear that due to the introduction “Voltage profile improvement in a microgrid including PV units using
of this third layer, active power reference set points can be genetic algorithm,” in Proc. 2nd Iran. Conf. Smart Grids (ICSG), Tehran,
provided (for real-time scheduling) not only according to the Iran, May 2012, pp. 1–5.
[17] D. Devaraj and J. P. Roselyn, “Genetic algorithm based reactive power
economical and the environmental objectives, but also taking dispatch for voltage stability improvement,” Int. J. Elect. Power Energy
to account the voltage level at every bus in the microgrid. Syst., vol. 32, no. 10, pp. 1151–1156, 2010.
In the paper, we did not consider the forecast process or [18] Y. L. Abdel-Magid, M. Bettayeb, and M. M. Dawoud, “Simultaneous
stabilisation of power systems using genetic algorithms,” IEE Proc.
techniques nor the accuracy on these forecasts, since these Gener. Transm. Distrib., vol. 144, no. 1, pp. 39–44, Jan. 1997.
parameters depends on several factors including local climate, [19] S. R. Karnik, A. B. Raju, and M. S. Raviprakasha, “Genetic algorithm
based robust power system stabilizer design using Taguchi principle,”
forecast horizon, whether forecasts apply to a single point or in Proc. 1st Int. Conf. Emerg. Trends Eng. Technol., Nagpur, India,
cover a wide geographic area, etc. However, it might be inter- Jul. 2008, pp. 887–892.
esting to implement a forecast technique and investigate the [20] A. P. A. da Silva and P. J. Abrao, “Applications of evolutionary compu-
tation in electric power systems,” in Proc. Congr. Evol. Comput. (CEC),
impact of the accuracy on the power generation schedule in vol. 2. Honolulu, HI, USA, 2002, pp. 1057–1062.
future research. [21] S. A.-H. Soliman and A.-A. H. Mantawy, Modern Optimization
Techniques With Applications in Electric Power Systems. New York,
ACKNOWLEDGMENT NY, USA: Springer, 2012.
[22] B. Marti, “Emissions of power delivery systems,” Ph.D. dissertation,
This research has been carried out in the frame of the EEH—Power Syst. Lab., Swiss Federal Inst. Technol. (ETH) Zurich,
Interuniversity Attraction Poles Programme initiated by the Zürich, Switzerland, 2005.
[23] (2007). Distribution Line Loss. [Online]. Available:
Belgian Science Policy Office (IAP-VII-02). http://www.hydroone.com/RegulatoryAffairs/Documents/EB-2007-0681/
Exhibit%20A/Tab_15_Schedule_3_Distribution_Line_Losses_Study.pdf
R EFERENCES [24] M. A. Jangjoo and A. R. Seifi, “Optimal voltage control and loss reduc-
tion in microgrid by active and reactive power generation,” J. Intell.
[1] T. L. Vandoorn and L. Vandevelde, “Contribution of microgrids Fuzzy Syst., vol. 27, no. 4, pp. 1649–1658, 2014.
to the development of the smart grid,” in Smart Grids: Clouds, [25] F. Saccomanno, Electric Power Systems. Piscataway, NJ, USA: IEEE
Communications, Open Source, and Automation (Devices, Circuits, and Press, 2003.
Systems), D. Bakken and K. Iniewski, Eds. Boca Raton, FL, USA: CRC [26] T. L. Vandoorn, B. Renders, L. Degroote, B. Meersman, and
Press, 2014, pp. 191–211. L. Vandevelde, “Active load control in islanded microgrids based on
[2] A. Bidram, A. Davoudi, F. L. Lewis, and S. S. Ge, “Distributed adap- the grid voltage,” IEEE Trans. Smart Grid, vol. 2, no. 1, pp. 139–151,
tive voltage control of inverter-based microgrids,” IEEE Trans. Energy Mar. 2011.
Convers., vol. 29, no. 4, pp. 862–872, Dec. 2014. [27] T. L. Vandoorn, J. D. M. D. Kooning, B. Meersman, J. M. Guerrero,
[3] D. E. Olivares et al., “Trends in microgrid control,” IEEE Trans. Smart and L. Vandevelde, “Automatic power-sharing modification of P/V droop
Grid, vol. 5, no. 4, pp. 1905–1919, Jul. 2014. controllers in low-voltage resistive microgrids,” IEEE Trans. Power Del.,
[4] F. Dörfler, J. W. Simpson-Porco, and F. Bullo, “Breaking the hierarchy: vol. 27, no. 4, pp. 2318–2325, Oct. 2012.
Distributed control and economic optimality in microgrids,” IEEE Trans. [28] H. Laaksonen, P. Saari, and R. Komulainen, “Voltage and frequency
Control Netw. Syst., vol. 3, no. 3, pp. 241–253, Sep. 2016. control of inverter based weak LV network microgrid,” in Proc. Int. Conf.
[5] J. M. Guerrero, J. C. Vasquez, J. Matas, L. G. de Vicuna, and M. Castilla, Future Power Syst., Amsterdam, The Netherlands, Nov. 2005, p. 6.
“Hierarchical control of droop-controlled ac and dc microgrids—A [29] M. Juamperez, G. Yang, and S. B. Kjær, “Voltage regulation in LV grids
general approach toward standardization,” IEEE Trans. Ind. Electron., by coordinated volt-var control strategies,” J. Modern Power Syst. Clean
vol. 58, no. 1, pp. 158–172, Jan. 2011. Energy, vol. 2, no. 4, pp. 319–328, 2014.
[6] T. L. Vandoorn, B. Meersman, J. D. M. De Kooning, and L. Vandevelde, [30] (Jun. 2012). SynerGrid C10/11. [Online]. Available:
“Transition from islanded to grid-connected mode of microgrids with http://www.synergrid.be
voltage-based droop control,” IEEE Trans. Power Syst., vol. 28, no. 3, [31] S. Youli, Z. Litifu, and K. Nagasaka, “Efficiency of micro grid with
pp. 2545–2553, Aug. 2013. storage battery in reliability, economy and environment assessments,”
[7] A. Bidram, A. Davoudi, F. L. Lewis, and Z. Qu, “Secondary control Elect. Power Energy Syst., vol. 3, no. 3, pp. 154–162, 2009.
of microgrids based on distributed cooperative control of multi-agent [32] A. Engler and N. Soultanis, “Droop control in LV-grids,” in Proc. Int.
systems,” IET Transm. Distrib. Gener., vol. 7, no. 8, pp. 822–831, Conf. Future Power Syst., Amsterdam, The Netherlands, 2005, p. 6.
Aug. 2013.
[8] Q. Shafiee, J. C. Vasquez, and J. M. Guerrero, “Distributed sec- Christof Deckmyn, photograph and biography not available at the time of
ondary control for islanded microgrids—A networked control systems publication.
approach,” in Proc. 38th Annu. Conf. IEEE Ind. Electron. Soc. (IECON),
Montreal, QC, Canada, Oct. 2012, pp. 5637–5642. Tine L. Vandoorn, photograph and biography not available at the time of
[9] X. Li, G. Geng, and Q. Jiang, “A hierarchical energy management strat- publication.
egy for grid-connected microgrid,” in Proc. IEEE PES Gen. Meeting
Conf. Expo., National Harbor, MD, USA, Jul. 2014, pp. 1–5. Jan Van de Vyver, photograph and biography not available at the time of
[10] Q. Jiang, M. Xue, and G. Geng, “Energy management of microgrid publication.
in grid-connected and stand-alone modes,” IEEE Trans. Power Syst.,
vol. 28, no. 3, pp. 3380–3389, Aug. 2013. Jan Desmet, photograph and biography not available at the time of publica-
[11] S. M. A. Shabestary, M. Saeedmanesh, A. Rahimi-Kian, and tion.
E. Jalalabadi, “Real-time frequency and voltage control of an islanded
mode microgrid,” in Proc. 2nd Iran. Conf. Smart Grids (ICSG), Lieven Vandevelde, photograph and biography not available at the time of
May 2012, pp. 1–6. publication.