Applied Energy 230 (2018) 175–188

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Impacts of a medium voltage direct current link on the performance of T

electrical distribution networks☆
⁎
Qi Qia, Chao Longa, Jianzhong Wua, , James Yub
a
Institute of Energy, School of Engineering, Cardiff University, Cardiff CF24 3AA, UK
b
SP Energy Networks, Blantyre, Scotland G72 0HT, UK

H I GH L IG H T S

• Assessing impacts of an MVDC link on network performance with high DG penetrations.

• Investigating the capability of an MVDC link in increasing the DG hosting capacity.
• AControl
real-time control method for MVDC link was proposed.
• strategies considering multiple operational objectives were developed.

A R T I C LE I N FO A B S T R A C T

Keywords: With an increasing number of distributed generators (DGs) integrated into distribution networks, operational
Electrical distribution network problems such as excessive power losses, voltage violations and thermal overloads have occurred. Medium
Medium Voltage Direct Current (MVDC) link Voltage Direct Current (MVDC) technology represents a candidate solution to address these problems as well as
Distributed generation to unlock the capacity of existing electrical network assets. In this paper, the capability of using an MVDC link to
Control strategy
improve the performance of a distribution network, i.e. reducing power losses and increasing the hosting ca-
Multi-objective optimization
pacity for DG connections was investigated. A grid transformer (GT)-based control method was developed, in
which the real-time data of the active power flow at GTs was used to specify the set-points of an MVDC link. The
control strategies considered multiple objectives, i.e. power loss reduction, feeder load balancing, voltage profile
improvement, and trade-off options among them. The response curves of these control strategies were developed
through offline studies, where a multi-objective Particle Swarm Optimization (MOPSO) method was used. Case
studies on a real distribution network were conducted to analyze the impacts of the MVDC link. The perfor-
mances of the network were evaluated and compared between the proposed control strategies, using real de-
mand and generation profiles. Results revealed that, for an MV distribution network, it might be beneficial to
switch between different control strategies with the variations in demand and generation conditions. Results also
showed that, regardless of the control strategy used, the MVDC link can significantly increase the network
hosting capacity (up to 15%) for DGs, and reduce about 50% of power losses compared to a conventional
alternative current (AC) line for the test network.

1. Introduction capacity for distribution networks to host DGs are required.

With the rapid development of power electronic technologies, and
In recent years, an increasing number of distributed generators their applications to High Voltage (HV) transmission networks and Low
(DGs) have been integrated into electrical distribution networks [1,2], Voltage (LV) distribution networks, analogies are being made with the
which pose challenges for Distribution Network Operators (DNOs), such use of direct current (DC) in Medium Voltage (MV) networks [4]. DC
as excessive power losses, voltage violations, and thermal overloads and power electronic technologies provide controllability and flexibility
[3]. Solutions to address these issues whilst providing enhanced to distribution networks, and can be used to increase the hosting

☆
The short version of the paper was presented at ICAE2017, Aug 21-24, Cardiff, UK. This paper is a substantial extension of the short version of the conference
paper.
⁎
Corresponding author.
E-mail address: wuj5@cardiff.ac.uk (J. Wu).

https://doi.org/10.1016/j.apenergy.2018.08.077
Received 5 February 2018; Received in revised form 14 August 2018; Accepted 15 August 2018
Available online 24 August 2018
0306-2619/ © 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/BY/4.0/).
Q. Qi et al. Applied Energy 230 (2018) 175–188

capacity for DGs of existing networks. infrastructures required for centralized control accounts for a large
Research on distribution-level power electronic devices has been portion of the total costs [20], which makes centralized control less
conducted. The use of shunt voltage source converters (VSCs) in rural attractive. A real-time control strategy of an MVDC link which uses only
networks for voltage regulation was carried out in [5]. In [6], an in- the measurements at grid transformers was investigated in [21].
telligent node, which is implemented by several VSCs connected to a However, the MVDC link set-points were obtained by using sensitivity
common DC-bus, was proposed to extend the current limits of feeders, analysis, in which a range of active power values were examined with
as well as to improve the voltage profiles. In [7] distribution-level the one inducing lowest network power loss selected as the MVDC link
power electronic devices with different topologies were assessed and set-point. The drawback of using sensitivity analysis is that only limited
compared, considering their capabilities in relieving network con- number of candidate values, rather than the entire solution space are
straints and accommodating DGs. More recently, the capability of back- considered. This might result in underutilization of MVDC links.
to-back VSCs based Soft Open Point (SOP) to regulate voltage and Therefore, effective control strategies for MVDC link need to be in-
therefore to increase DG penetration was quantified in [8]. The benefits vestigated. Also, as the performances of an MVDC link depend on the
of using SOPs in a distribution network were analyzed in [9]. A sensi- control strategies applied, different control strategies for an MVDC link
tivity method to define the optimal operating region of SOP was pro- need to be assessed and compared.
posed in [10]. A few initiative pilot projects have been trialed using In this paper, a real-time control method for an MVDC link was
SOPs in MV and LV distribution networks in the UK, such as ‘Network proposed, in which the active power flowing through the grid trans-
Equilibrium’ [11] and ‘Flexible Urban Networks Low Voltage’ [12] formers (GTs) is used to determine the set-points of an MVDC link. GTs
projects. are the transformers that supply power to the network. This method is
The above studies assessed the impacts of power electronic devices called GT-based control. The GT-based control requires only measure-
on the performances of distribution networks, and the topologies were ments at the GTs rather than the load and generation data at each load
mainly back-to-back VSCs. The benefit of deploying a point-to-point point (i.e. substations) of the network. This work considered control
application of VSCs, e.g. an MVDC link, rather than a back-to-back strategies with multiple objectives, i.e. power loss reduction (PLR),
application is that it enables flexible power and voltage control over a feeder load balancing (FLB), voltage profile improvement (VPI), and
wider area. Recent studies on MVDC links have been conducted from compromise strategies providing trade-offs among them. The response
different perspectives. In [13], the utilization of DC links to enhance the curves of these control strategies were developed through offline stu-
integration of DGs, and to reduce power losses in distribution systems dies, where a multi-objective Particle Swarm Optimization (MOPSO)
was investigated, where the DC link control set-points were determined method was used. Assessments and comparisons between different
by the optimal power flow. The benefits of incorporating DC links to control strategies were carried out. To evaluate the effectiveness of
radial distribution networks were assessed in [14], where the maximum implementing MVDC link with the GT-based control method, case
load or DG that can be served by the network were determined. The studies on a real distribution network were conducted.
benefits of using multi-terminal DC links to reduce power losses and This work has the following contributions: (1) proposing a novel
improve voltage profiles, as well as to mitigate transient power quality real-time control method for MVDC link, namely the GT-based control
perturbations were analyzed in [15]. A cost-benefit evaluation of using method, which only requires communication links between the grid
DC links as interconnectors in dense-load urban networks was carried transformers and the controller of the MVDC link; (2) control strategies
out in [16]. The capabilities of using MVDC links to increase network considering multiple objectives were developed and the network per-
hosting capacity for DG connections were investigated in [17,18], and formance of these control strategies were compared; and (3) impacts of
different levels of communication functionalities were considered in an MVDC link on the performances of distribution networks in terms of
[18]. These studies have mainly focused on the evaluation of benefits of the capability of reducing losses and increasing the network’s DG
DC links in reducing power losses, increasing system loadability or in- hosting capacity were investigated.
tegrating more DGs, whereby centralized control schemes are required.
However, the centralized control schemes rely on measurement and 2. MVDC link in distribution networks
communication infrastructures which might be subjected to malfunc-
tions or failure. To account for this, operational strategies to keep DC Fig. 1 shows a schematic diagram of an MVDC link connecting two
links connected in a safe manner in the case of a communication failure distribution networks. The MVDC link is constructed via fully con-
were proposed in [19]. trollable power electronic converters. A voltage source converter (VSC)
Due to the limited number of available real-time measurements in station is used for the conversion between AC and DC at each terminal
distribution networks and the intermittence of demand and distributed of the MVDC link. The MVDC link allows for real power exchange be-
generation, distributed control strategies requiring fewer measurements tween the two terminals as well as reactive power supports at both
are more viable than the centralized ones. In addition, the cost of as- sides. The VSC can perform as the control of: (a) DC voltage Vdc ; (b)
sociated ICT (information and communication technology) active power P ; (c) AC voltage Vac ; (d) reactive power Q and (e) AC

MVDC Link

AC
C DC
C
D
DC A
AC
VSC Station 1 VSC Station 2

Fig. 1. An MVDC link between distribution networks.

176
Q. Qi et al. Applied Energy 230 (2018) 175–188

Table 1 centralized control schemes, which require observability of the entire

Control modes of an MVDC link under normal operating conditions. network and heavily rely on the corresponding communication infra-
Control Mode VSC Station 1 VSC Station 2 structure [24], the GT-based control is relatively simpler with lower
cost, making the control scheme an attractive interim option [25,26]. In
1 PQ/ PVac Vdc Q/ Vdc Vac Fig. 2, Network 1 has a large amount of DG connections, and therefore
2 Vdc Q/ Vdc Vac PQ/ PVac Network 1 is becoming problematic and even approaching its voltage
and thermal limits. In this context, Network 1 is focused on.
For the real-time operation of the MVDC link, a response curve is
frequency f . Typical control modes of an MVDC link under normal
needed. A response curve defines the linear relation between the active
network operating conditions are listed in Table 1, where one VSC
power set-point of the MVDC link (PMVDC ), and the active power flowing
station controls the active power flowing through the link, and the
through the GTs (PGT ). The idea of using the response curves is that, the
other VSC station is used to maintain the DC voltage. In addition, for
most adequate active power provided by the MVDC link is related to the
each VSC station either the reactive power or the AC side voltage can be
active power flowing through the GTs. When the active power mea-
controlled.
sured at the GTs is positive, i.e. the demand in the network is greater
than the DG output, active power is required to be imported to the
3. Modelling of MVDC link
network. In this case the active power provided to the network via the
MVDC link would support the network’s electricity demand. When the
In this paper, PVac−Vdc Vac was selected as the control mode of the
active power measured at the GTs is negative, i.e. the DG output is
MVDC link, and the reactive power outputs were adjusted in real time
greater than the demand, active power is required to be exported from
to maintain the voltage at specified value, e.g. 1 p.u. It is considered
the network. In this case, exporting active power via the MVDC link
that fewer converters with higher power ratings are preferred than
would support to consume the surplus generation in the network. The
more converters with lower ratings of each. Since higher power rating
response curve is obtained by offline studies, where the historical net-
VSC has the merits of higher efficiency, lower power losses and fewer
work data is used.
peripheral devices associated, such as filters and line reactors. In ad-
For the operation of an MVDC link, the network operators may have
dition, the footprint of the VSC station can be reduced with fewer VSCs
different focuses, e.g. losses reduction, or voltage improvement. In this
[22]. For simplicity, the term VSC is used to denote VSC station in this
study, different objectives were considered. These objectives result in
paper.
different response curves, which provide the DNOs with more operation
To fully evaluate the effects of an MVDC link on network operations,
options. In the offline studies, a multi-objective Particle Swarm
a mathematic power injection model of MVDC link was developed by
Optimization (MOPSO) method was used. The active power flowing
considering the following modelling constraints.
through the Grid Transformers (PGT ), when the active power set-point of
the MVDC link was zero, was also recorded. Then a linear approxima-
• Active power constraints: tion of the optimal set-points PMVDC and the corresponding PGT was
PVSC1 + PVSC 2 + PDC − loss = 0 (1) carried out to find their linear relations, which were the response
curves. These curves were used in real-time to specify the set-points of
where PVSC1, PVSC2 are the active power flow through each VSC. PDC − loss
the MVDC link.
is the loss within an MVDC link, which is relatively low (approximate
1–2% of the active power flowing through the MVDC link [23]) com-
pared to the total losses within the network and thus can be neglected. 4.1. Problem formulation
Therefore, Eq. (1) can be simplified as:
The determination of response curves with different objectives was
PVSC1 = −PVSC 2 (2) formulated as a multi-objective optimization problem. Three opera-
tional goals were considered and formulated as:
• Reactive power constraints: Minimize [f1 , f2 , f3 ] (5)
min max
QVSC , n ⩽ QVSC , n ⩽ QVSC , n (n = 1, 2) (3) where f1 is the objective function for Power Loss Reduction (PLR),
where QVSC , n is the reactive power at the nth terminal of the MVDC link. where active power losses in feeder lines and transformers of a network
min
QVSC max
, n and QVSC , n are the lower and upper limits of reactive power
were considered:
max
provided by the VSC at terminal n . QVSC , n is positive indicates that re- Nbranch
min
active power is injected to the network, and QVSC , n is negative indicates f1 = PLR = ∑ Ik2 × rk
that reactive power is absorbed from the network. k=1 (6)

In Eq. (6), Ik is the current flowing through branch k . rk is the resistance

• Capacity constraints: of that branch, and Nbranch is the total number of branches including
2
PVSC 2 lines and transformers.
, n + QVSC , n ⩽ SVSC , n (4)
f2 is the objective function for Feeder Load Balancing (FLB). The
where SVSC , n is the rated capacity of the VSC at the nth terminal of the balance of loads within a network can be achieved by minimizing the
MVDC link. line utilization index:
The MVDC link is modelled using the above equations. Then, the
N Sj
active and reactive power outputs of the MVDC link can be integrated ∑ j =line
1 S
( )2
j, rated
into load flow algorithm without considering detailed design of the f2 = FLB =
Nline (7)
converter controllers.
In Eq. (7), Sj is the apparent power flow in line j , and Sj, rated is the rated
4. GT-based control method capacity of the line. Nline is the total number of feeder lines. The line
utilization index reflects the average degree of utilization of all feeder
Fig. 2 presents an overview of the proposed GT-based control lines in a network.
method, where data from the GTs are provided to the controller of the f3 is the objective function for Voltage Profile Improvement (VPI).
MVDC link by a direct communication link [21]. Compared to the The improvement of voltage profiles can be achieved by minimizing the

177
Q. Qi et al. Applied Energy 230 (2018) 175–188

Network 1

Response Curves
with different control objectives
PMVDC

PGT

AC
DC

MVDC Link
Network 2

DC Meter
AC
Communication Link

Fig. 2. Overview of the proposed GT-based control.

voltage profile index: ⎡ ∂P

cal ∂P cal
⎤
∂V ∂θ
N
J = ⎢ cal ⎥
∑i =bus (Vi −Vi, rated )2 ⎢ ∂Q ∂Q cal ⎥
1
f3 = VPI = ⎣ ∂V ∂θ ⎦ (13)
Nbus (8)
where J is the Jacobian matrix whose elements are listed in [29]. The
In Eq. (8), Vi and Vi, rated are the real and nominal voltage magnitudes at Jacobian matrix is evaluated at V = V it and θ = θit . V and θ are the
bus i . Nbus is the total number of buses. The voltage profile index reflects voltage magnitude and phase angle. it is the iteration count.
the average degree of dispersion of all bus voltages from the nominal
ΔV it = V it + 1−V it (14)
value.
In addition to the constraints of an MVDC link as shown in Eqs. Δθit = θit + 1−θit (15)
(2)–(4), the network active and reactive power flows, and the voltage
and thermal limits were also taken into consideration: ΔP it = P sp−P cal . it (16)
The power balance equations for a network are [27,28]:
ΔQit = Q sp−Q cal . it (17)
Vi = ei + jfi (9) where P sp
and Q sp
are the specified net active and reactive power in-
Nbus Nbus jections. P cal, it and Q cal, it are the calculated net active and reactive
Pical = ei ∑ (Gij ej−Bij f j ) + fi ∑ (Gij f j + Bij ej ) power injections at iteration it using V it and θit .
j=1 j=1 (10) The voltage and thermal limits are expressed in Eqs. (18) and (19):

Nbus Nbus Vimin ⩽ |Vi | ⩽ Vimax (18)

Qical = fi ∑ (Gij ej−Bij f j )−ei ∑ (Gij f j + Bij ej )
where Vimin and Vimax are the lower and upper voltage limits at bus i .
j=1 j=1 (11)
|Sk | ⩽ Skmax k ∈ {1, 2, …, Nbranch} (19)
where Vi is the voltage at bus i , and ei and fi are its real and imaginary
components. Pical and Qical are the calculated active and reactive power where Sk and Skmax
are the actual and upper limit of the apparent power
injections at bus i . Gij and Bij are the conductance and susceptance of the flow of branch k .
branch connecting buses i and j .
In the Newton Raphson method, Eqs. (10) and (11) can be expanded 4.2. Solution methodology
into Taylor series and the following first order approximation can be
obtained [28]: 4.2.1. Pareto-dominance principle
It is common that multiple objectives are incommensurable in
⎡ΔP it ⎤ = J ⎡ ΔV ⎤
it
nature and can be conflicting with each other, aggregating multiple
⎢ ΔQit ⎦
⎣ ⎥ it
⎣ Δθ ⎦ (12) objectives into one equation may result in losing significance. Pareto-

178
Q. Qi et al. Applied Energy 230 (2018) 175–188

dominance principle [30], on the contrary, is based on a simultaneous by their resulted objective function value. However, in a multi-objec-
optimization along different objective functions. It provides a range of tive problem, since the solutions now have multiple fitness criteria, the
alternative solutions rather than a single one, leading to more flexibility concept of Pareto dominance is introduced to evaluate their fitness.
in the process of decision making and operation of the network. In this Recalling the Pareto-dominance principle (Eq. (20)) and the objective
study, Pareto-dominance principle was adopted to solve the multi-ob- functions considered in this study (Eqs. (6)–(8)), the fitness value of a
jective optimization problem. particle’s position u is better than position w , if u dominates w (denoted
Assuming a number of Nobj functions are to be minimized, solution as u ≺ w ):
‘A’ is said to dominate solution ‘B’ if:
u ≺ w, if :
∀ n ∈ [1, 2, …, Nobj]: fn (A) ⩽ fn (B ) ∩ ∃ n ∈ [1, 2, …, Nobj]: fn (A) < fn
N
(20) ⎧ f1 = PLR = ∑k =branch
1 Ik2 × r k
⎪
⎪ 2
where fn (A) and fn (B ) are the values obtained by solution A and B N ⎛ Sj ⎞
⎪ ∑ j =line
⎪ 1 ⎜S
j , rated
⎟
respectively along the nth objective function. fn (u) ⩽ fn (w ) for all n, and fn (u) < fn (w ) for some n fn f = FLB = ⎝ ⎠
⎨ 2 Nline
Under this concept, a set of non-dominated solutions that are of ⎪
⎪ N
∑i =bus 2
⎪ f3 = VPI = 1 (Vi − Vi, rated)
equal interests amongst different objectives can be obtained. This set is ⎪ Nbus
⎩
called Pareto optimal set, which provides alternative solutions bringing
out flexibility on network operation. For instance, some solutions may (23)
lead to lower power losses while others cause the network loading more The selection of personal best position is straightforward: if the
balanced. The representation of Pareto optimal set in the solution space current position of the ith particle x iiter dominates its previous personal
iter − 1 iter iter
is called Pareto frontier. best position pbest , i , the new personal best position pbest , i is set to x i . If
iter − 1 iter
x iiter and pbest ,i non-dominate each other, pbest , i is set to either of them
4.2.2. Multi-objective optimization method by applying random sampling (equal probability). Otherwise, pbest iter
, i is
An improved multi-objective Particle Swarm Optimization iter − 1
the same as pbest ,i .
(MOPSO) method based on Pareto-dominance principle developed in Regarding the selection of gbest , research work has been carried out
the authors’ previous work, was used in the offline studies to derive the using approaches that aggregate all the objectives into a single function
response curves of an MVDC link. The main contents and key for- [34,35], or approaches that assign the objectives in order of importance
mulations of the method are presented in this section, and more de- [36,37]. Then during each iteration, the particle’s position that has
tailed explanations can be found in reference [31]. A number of de- optimal value along the aggregated function or the most important
mand and generation scenarios were first generated from the historical function is selected to be gbest . There are also approaches that select the
data of a network. For each scenario (with defined demand and gen- guide based on the Pareto dominance [38,39]. In the proposed MOPSO,
eration levels), the proposed MOPSO method was applied to find op- an archive is used to store non-dominated solutions, and is updated
timal set-points in the offline studies. The active power set-point of an iteratively based on the Pareto-dominance principle. The archive is
MVDC link (PMVDC ) was taken as the decision variable. Independent empty at the beginning of the search. Then the non-dominated solutions
variables included the demand and generation of the network, and the of each iteration are added to it, and any solutions in it which are
reactive power outputs at both terminals of the MVDC link, which were dominated by the new solutions are deleted from the archive. Eq. (24)
adjusted accordingly to maintain the terminal voltage at specified illustrates the rules of selecting the global best position ( gbest ) to update
value, e.g. 1 p.u. The novelty of this method is integrating both global particles in each iteration. A is the archive that stores non-dominated
and local search techniques to search for optimality. The MOPSO al- solutions, and a is a solution in A . When updating the position of the ith
gorithm is adopted to explore the solution space globally. A local search particle ( x i ), x i is first checked whether it is dominated by any solution
technique, namely the Taxi-cab method, is used for solution space ex- in the archive. If so, assuming the set of solutions in the archive that
ploitation, which refines the quality of solutions searched by MOPSO in dominate x i is A xi , and the leader of the ith particle is selected from this
each iteration. The improved MOPSO method is an effective tool for set with equal probability, i.e. |A xi |−1. If x i belongs to the archive,
solving multi-objective optimization problems by providing qualified clearly no solution in the archive dominates x i . Therefore, a solution
and diversified solutions with enhanced search capability. from the archive is selected randomly with equal probability, i.e. |A|−1
Particle Swarm Optimization (PSO) [32] is used for global search to lead the ith particle.
due to its easy implementation, effective memory use, and an efficient
maintenance of the solution diversity. As a multi-point search algo- a ∈ A with probability |A|−1 if x i ∈ A
gbest , i = ⎧
rithm, PSO can provide a set solutions in a single run [33]. The search ⎨ a ∈ A x with probability |A x |−1 otherwise
(24)
⎩ i i
starts with a population of random search points named particles. Each
particle is encoded by a position vector ( x ) containing the decision In the proposed MOPSO method, ε -dominance method [40] was
variable information. x is updated with the particle’s velocity vector (v ) used to restrict the archive size while maintaining the diversity of non-
successively. In each iteration, v is updated with two best values. The dominated solutions in the archive. A mutation operation using Cauchy
first one is the individual/personal best position ( pbest ) achieved by each method [41] was also included to avoid premature convergences.
particle itself. The other one is the global best position ( gbest ) obtained Despite the global exploring capability, evolutionary algorithms are
by any particle among the population, which is used as a guide leading comparatively inefficient in solution space exploitation [42]. To over-
the population towards optimum. Updating equations of the ith particle come this deficiency and to enhance the search capability of evolu-
are formulated as: tionary algorithms, the appropriate integration of global and local
search techniques, which maintains the balance between exploration
viiter + 1 = ωviiter + c1 r1 (pbest
iter iter iter iter
, i −x i ) + c2 r2 (gbest , i−x i ) (21) and exploitation, appears to be an ideal solution. In this method, a local
search technique, namely the Taxi-cab method, is integrated to fine
x iiter + 1 = x iiter + viiter + 1 (22)
tune the non-dominated solutions, overcoming the drawback of original
where ω is the inertia weight. c1 and c2 are the cognitive learning factor MOPSO in local optima trapping. The Taxi-cab method does not require
and the social learning factor respectively. r1 and r2 are two random information of the derivatives of objective functions, in which the
numbers ∈ [0, 1] used to keep away from entrapment in local optimum. search is performed by moving the decision variable along standard
MOPSO is a multi-objective version of PSO. In PSO, the selection of base vectors. The Taxi-cab method is applied to the solutions in the
pbest and gbest relies on the fitness value of particles, which is determined archive at the end of each iteration. New solutions obtained by the Taxi-

179
Q. Qi et al. Applied Energy 230 (2018) 175–188

Demand Response Curves

PMVDC
PGT with different control objectives
obj 1 obj 2 obj n C1 C2
Sce. 1 Sce. 1 Linear Approximation PMVDC
Sce. 1 a1 b1,1 b1,2 b1,n b1,c1 b1,c2
Sce. 2 Sce. 2
Operating Distribution Sce. 2 a2 b2,1 b2,2 b2,n b2,c1 b2,c2
Generation
Scenarios Network
Sce. m Sce. m Sce. m am bm,1 bm,2 bm,n bm,c1 bm,c2 PGT

Historical demand &

generation data
Multi-objective
optimization
model

Fig. 3. Process of developing the response curve of an MVDC link.

cab method that are not dominated by any members in the archive are by the utopian point method, and Un is the value of the utopian point
added into the archive, and any members in the archive which are along the nth objective function.
dominated by the new solutions are deleted. When applying the proposed MOPSO method to the network, the
It is to be mentioned that, the proposed MOPSO was used to find the size of archive A was set to 50, i.e. the maximum number of Pareto
optimal set-points of an MVDC link in offline studies. Other optimiza- optimal solutions stored in the archive in each iteration was 50. At the
tion algorithms can also be used for the same purpose, and the algo- end of the search process, the stored Pareto solutions in the archive
rithms with better search capability were considered to obtain more were 50 or about 50, and they were used for the determination of the
appropriate response curves. best compromise solutions. The number, diversity and distance to the
ideal points of Pareto solutions all have impacts on the quality of the
4.2.3. Selecting compromise solutions from Pareto optimal set Pareto frontier. Hence, Pareto frontiers with higher diversity and
When solving real-word problems, a compromise solution may be quality were considered to provide better compromise solutions.
required. Ideally, such solution belongs to the Pareto optimal set and The compromise solutions obtained by using the reference point
takes into account the preferences of decision makers. In this study, two approach (C1) and the utopian point method (C2), together with the
methods, namely the reference point approach [43] and the utopian optimal solutions along each single objective, were used to derive dif-
point method [44] were used for selecting two compromise solutions, ferent response curves of an MVDC link.
which are trade-offs among multiple criteria.
The reference point approach aims to reach a solution on the Pareto 4.3. Process of obtaining responsive curves
frontier that locates near to a specific reference point, which is normally
pre-defined by decision makers. In this study, since the information For the GT-based control, the active power provided by an MVDC
from decision makers is unknown, a hypothetical point that with the link (PMVDC ) is related to the active power flowing through the grid
minimum values of each objective function is considered as the re- transformers (PGT ). In this study, their relationship was simplified to a
ference point. The normalized Euclidian distance between each solution linear relation, i.e. PMVDC = αPGT + β . From a number of offline studies
on the Pareto frontier and the reference point is calculated as: using the historical data of the test network, such relationship was
quantified.
2
fn (a)−Rn ⎞ A schematic diagram showing the process of obtaining the response
∑n=obj1 wn ⎜⎛
N
da = max min ⎟ curve is presented in Fig. 3, and the process is illustrated as follows:
⎝ n −fn ⎠
f (25)

where Nobj refers to the number of objective functions considered, and (1) Historical data of both demand and generation of the network
in this study Nobj = 3. da is the fitness value of the ath solution on the under study are categorized into several levels. For instance, de-
Pareto frontier defined by the reference point approach. fnmax and fnmin mand levels include summer minimum and maximum, winter
are the maximum and minimum values of the nth objective function minimum and maximum. Levels with scaled-up magnitude are used
obtained from the Pareto optimal set. fn (a) is the value of the ath so- to reflect the forecasted demand and generation in the future. The
lution along the nth objective. Rn is the nth component of the reference combination of these demand and generation levels represents
point. wn is the weighting factor of the nth objective reflecting the re- different scenarios of a network with different demand and gen-
lative importance of different objectives. For the identification of a eration conditions.
compromise solution, the one with lower distance to the reference point (2) For each scenario, the active power flowing through GTs is first
and, simultaneously, with higher weighting factor, will be selected obtained. This is carried out by considering the active power set-
preferentially. point of the MVDC link is zero (i.e. with no power flowing through
The utopian point method first searches for the optimal value along the MVDC link). Next, with the MVDC link in operation, the im-
each objective function. Then the intersection of these single-objective proved MOPSO method is applied to obtain the optimal set-points
optimal values is defined as the utopian point, which is normally lo- of the MVDC link in the offline studies. The optimal set-points when
cated outside of the actual solution space. The point on the Pareto considering one objective are divided into one group. This means
frontier with the shortest distance to the utopian point is selected as the there would be 5 groups of optimal set-points (i.e. objectives for
compromise solution: power loss reduction, feeder load balancing and voltage profile
improvement, and two compromise solutions).
N
Da = ∑n=obj1 (fn (a)−Un)2 (3) Repeat the procedure (2) for all scenarios. As a result, for each
(26)
scenario, the active power at GTs and a group of set-points of the
where Da is the fitness value of the ath Pareto optimal solution defined MVDC link are obtained. A linear approximation of the GT active

180
Q. Qi et al. Applied Energy 230 (2018) 175–188

Start

Input historical
Input objectives &
demand &
constraints
generation data

Combine the demand & generation

data into a number of m scenarios Build the proposed
MOPSO model
Set Sce=1

Network with the

scenario data Sce

Record PGT
when PMVDC=0

Apply the proposed MOPSO to

determine the Pareto optimal set of
Fig. 5a. The 33 kV Anglesey network.
PMVDC among multiple objectives

Obtain compromise
literature, this paper followed the commonly used one, which is that DG
solutions using Eqs.
penetration is the percentage of total capacity of DG units over the
25-26
maximum loading capacity of a network. The same definition of DG
penetration was used in [8,46].
Record the optimal &
Fig. 5a presents a schematic of the 33 kV Anglesey network and a
compromise PMVDC
portion of the network on the mainland in North Wales. As shown in
Sce=Sce+1 Fig. 5a, Bangor substation (mainland) and Llanfair PG substation (An-
glesey) were selected to install VSC stations, and in between are the
existing AC circuits that to be trailed to demonstrate the application of
No
Sce>m? an MVDC link. A detailed single-line diagram is given in Fig. 5b, where
locations of the existing installation of DGs are shown. Details of the
Yes network data and the demand and generation data can be found
Apply liner through the link in the Acknowledgements. The meshed 33 kV Anglesey
approximation to PGT network is supplied by three 132/33 kV GTs at Amlwch and Caergei-
and PMVDC to derive liog. Currently the Anglesey network has a high penetration of DG and
response curves is sensitive to the configuration of the transmission network. There will
also be a significant amount of additional demand on Anglesey due to
End planned regional redevelopment, and the peak load of Anglesey net-
work is forecasted to be 81.82 MVA in 2023. Three wind farms with a
Fig. 4. Flowchart for developing the GT-based response curves of an MVDC total capacity of 34.7 MW, and two solar farms with a total capacity of
link. 28.5 MW are already installed on Anglesey as shown in Fig. 5. Three
additional DGs, with a total capacity of 67.8 MW have been contracted
power and the corresponding MVDC set-points is then carried out to and will be installed by 2023. Therefore, the total DG penetration rate
derive the response curve. This would result in 5 response curves of the Anglesey network is forecasted to reach 152% by 2023. The main
for the 5 different objectives considered. challenge on Anglesey is that uncontrolled power flows are predicted to
exceed the thermal limits of cables and overhead lines. Managing the
A flowchart for the overall procedure of developing the GT-based network within voltage and thermal limits is becoming problematic,
response curves is illustrated in Fig. 4. due to the network topology, high levels of demand and DG penetra-
tions. The adoption of MVDC link represents a solution to expand
network operational flexibility, and to provide extra capacity for the
5. Case studies integration of renewable generations.
A model of real Anglesey network was built and analyzed in IPSA 2
5.1. ANGLE-DC project and challenges on Anglesey network software. The improved MOPSO method and the proposed GT-based
control strategies were implemented with python scripting. In the si-
As part of the Ofgem Electricity Network Innovation Competition mulation, the capacity of each VSC is rated at 30.5 MVA, with the active
(ENIC) award, ANGLE-DC [45] project aims to demonstrate a novel power ranging from −25 MW to 25 MW, and the reactive power ran-
network reinforcement technique by converting an existing double ging from −15 MVAr to 15 MVAr. The decision variables are con-
33 kV AC circuit into DC operation, i.e. an MVDC link, between the isle tinuous values within these limits. The voltage limit is set to be ± 6% of
of Anglesey and North Wales. the nominal voltage. The capacity limit of each branch is according to
Although there is no unified definition of DG penetration rate in the its real rating.

181
Q. Qi et al. Applied Energy 230 (2018) 175–188

Fig. 5b. Single-line diagram of the 33 kV Anglesey network.

5.2. GT-based control for the MVDC link Table 2

MW flows through GTs on Anglesey under 25 scenarios of demand and gen-
To derive response curves of the MVDC link, the procedure illu- eration.
strated in Section 4.3 was applied. Firstly, active power flows at the DEMAND (MVA)
Anglesey GTs were calculated under different network operating sce-
PGT (MW)
narios, while assuming no active power transfer was provided by the SMIN SMAX WMAX WMAX_2019 WMAX_2023
MVDC link. Historical and forecasted demand and generation data of 24.56 37.64 74.70 77.94 81.82

Anglesey were used to form different scenarios. For the demand, five 0 23.45 36.06 75.08 78.47 82.58
levels from 24.56 MVA to 81.82 MVA were taken into account, corre- 31.25 −8.04 4.45 43.12 46.48 50.51
sponding to: summer minimum (24.56 MVA); summer maximum DG (MW) 62.50 −39.08 −26.69 11.61 14.94 18.89
(37.64 MVA); existing winter maximum (74.70 MVA); existing winter 93.75 −69.66 −57.39 −19.42 −16.13 −12.20
125 −99.69 −87.55 −49.94 −46.67 −42.80
maximum uniformly increased by 5%, denoting the forecasted winter
maximum in 2019 (77.94 MVA); and existing winter maximum uni-
formly increased by 11%, denoting the forecasted winter maximum in
2023 (81.82 MVA). For the generation, five levels between 0 MW and (Table 2) and the corresponding set-points of the MVDC link were
maximum capacity of 125 MW were considered, which included both firstly plotted by many dots (i.e. marked by circles, crosses, triangles,
connected and contracted DGs on Anglesey. rhombuses, and squares to differentiate the control strategies). The
Table 2 shows the active power flows (in MW) at the GTs. A positive circles correspond to the MVDC link set-points when considering power
sign means the power is imported from the upstream transmission loss reduction (PLR) as the objective. The crosses represent the set-
system to Anglesey, and a negative sign means a reverse power flow points when considering feeder load balancing (FLB) as the objective.
from Anglesey to the transmission system. The triangles are the set-points when improving the voltage profiles
For each scenario, the MVDC link was in operation with its set- (VPI) was considered as the objective. The rhombuses are the com-
points calculated by using the improved MOPSO method. Two com- promise set-points (C1) obtained by using the reference point approach,
promise set-points were also obtained by applying the reference point and the squares are the compromise set-points (C2) obtained by using
approach, and the utopian point method to the Pareto frontier of each the utopian point method. These five sets of dots were then used to
scenario. derive five straight lines by applying the least squares method, and the
Fig. 6 shows the response curves of the MVDC link with different parameters of each line are listed in Table 3. These straight lines are the
control objectives, where the active power flowing through GTs most appropriate operational curves for the active power set-points of

182
Q. Qi et al. Applied Energy 230 (2018) 175–188

25

20

15

MVDC link set-point (MW) 10 Response Curves for:

Power Loss Reduction

5
Feeder Load Balancing

0 Voltage Profile Improvement

Compromise 1
-5
Compromise 2

-10

-15

-20
-120 -100 -80 -60 -40 -20 0 20 40 60 80 100

Net power flow through GTs (MW)

Fig. 6. Response curves of the MVDC link in Anglesey network with different control objectives.

Table 3 strategies achieved medium performances among these objectives.

Control parameters of different response curves.
Response curves PMVDC = αPGT + β 5.3.2. Case 2: Daily network operation with modified demand and
generation profiles
α β To consider the impact of high DG penetrations, the demand profile
in Fig. 7 was scaled down to the summer minimum, while the gen-
Power loss reduction 0.1225 8.91
Feeder load balancing 0.1216 10.16
eration was scaled up by multiplying the normalized daily profile of
Voltage profile improvement 0.2833 12.69 each DG. This case is recognized as the worst-case scenario [47,48] and
Compromise 1 0.1582 11.45 provides the most vulnerable network operating condition. The mod-
Compromise 2 0.1727 11.58 ified daily profiles are shown in Fig. 9, and the half-hourly power loss
over a day under the worst-case scenario is shown in Fig. 10. The
corresponding average power losses and energy losses when using dif-
the MVDC link. ferent strategies are listed in Table 5.
It can be seen from Fig. 6 and Table 3 that, different control para- Comparing the results in Figs. 8 and 10, it can be found that the
meters result in different slopes and intercepts on axes of the response control strategy leading to minimum power loss over a day was not the
curves, which lead to distinct performances of the MVDC link. How- one for power loss reduction at all times. When the outputs from DGs
ever, the trends of these response curves are similar. were much greater than the demand (e.g. during the hours
13:00–16:00), the control strategy for feeder load balancing resulted in
5.3. Daily operation of different cases lower power losses than others. However, when the demand were
greater than the generation (e.g. from 17:00 to 24:00), the control
5.3.1. Case 1: Daily network operation with the MVDC link using different strategy for power loss reduction became the one with lowest losses
control strategies again. Results indicated that there was no particular control strategy
The GT-based control strategies use the measurements at GTs to bringing minimum power losses to the network at all times. However,
provide the active power set-points of the MVDC link, while the oper- switching between different strategies as the net power demand within
ating condition of the wide area of the network is unknown to the the network varied would be necessary. The results of switching be-
MVDC link. Therefore, it is essential to assess the real-time performance tween strategies that incurred lowest power loss in Fig. 10 are shown in
of the network. the last row of Table 5.
One-day demand and generation profiles of Anglesey obtained from
measurements on the real network are shown in Fig. 7. Taking half- 5.4. Impacts of DG penetration
hourly time steps, power losses of the Anglesey network with the MVDC
link using different control strategies are shown in Fig. 8. The power In the simulations the worst-case scenario was applied, in which
losses of the network when the link is operated in AC are used as a Anglesey network was at its minimum demand. The maximum allow-
reference case. The daily average power losses and energy losses when able DG penetration reflects the maximum capacity of DG units the
using different strategies are listed in Table 4. It can be seen in Fig. 8 network can accommodate before any violation occurs, i.e. the DG
and Table 4 that, with the MVDC link, regardless the control strategy hosting capacity. DG penetration was increased from 0 with an incre-
used, power losses of the network were reduced significantly compared ment of 5%. Three network performance metrics: daily energy loss; the
to those when the link is operated in AC. Comparisons between control maximum line utilization over a day; and the maximum bus voltage
strategies showed that, the one for voltage profile improvement led to over a day were evaluated.
higher power losses than other strategies. This is due to the extra power Firstly, the DG hosing capacity of the network was investigated with
injections from the MVDC link for voltage regulation. Power losses results shown in Fig. 11.
obtained with the strategy for power loss reduction remained the lowest From Fig. 11, it can be seen that MVDC link has the capability to
over the day, and losses obtained with the strategy for feeder load increase the DG hosting capacity over the AC operation. The control
balancing were the second lowest. The two compromise control strategy for feeder load balancing achieved the highest DG hosting

183
Q. Qi et al. Applied Energy 230 (2018) 175–188

40

35

30

25
Demand
P (MW)
20
Generation
15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (half-hourly)
Fig. 7. Case 1: daily profiles of total demand and generation on Anglesey.

capacity, followed by the control strategy for power loss reduction, and Table 4
the two compromise control strategies. With the MVDC link, an in- Power and energy losses by using different control strategies as shown in Fig. 8.
crease of DG hosting capacity up to 15% can be achieved in the An- Daily average power loss Daily energy loss
glesey network. (MW) (MWh)
Impacts of DG penetrations on the network performance metrics
AC operation 1.556 37.35
were also investigated. Results are shown in Figs. 12a–12c. It can be
Control strategy for PLR 0.562 13.49
seen that, the degree of improvements along the performance metrics is Control strategy for FLB 0.578 13.87
different, depending upon the control strategy used. The maximum Control strategy for VPI 0.673 16.15
allowable DG penetration is also different among control strategies. Compromise control strategy 1 0.607 14.58
With the control strategy for voltage profile improvement, a maximum Compromise control strategy 2 0.613 14.72

DG penetration of 50% can be obtained, followed by the AC operation,

of which the maximum DG penetration was 45%. The control strategy
for feeder load balancing achieved the highest DG penetration of 60%. DG outputs. The control strategy for feeder load balancing led to the
These results are consistent with the results shown in Fig. 11. lowest line utilization under all DG penetrations. Although the control
It can be seen in Fig. 12a that, with the MVDC link, daily energy strategy for voltage profile improvement led to higher energy losses and
losses first decreased slightly as the DG penetration increased from 0 to line utilizations under all DG penetrations, it outperformed in miti-
10%. Then as the DG penetration further increased, energy losses gating the voltage rise issue as shown in Fig. 12c. Such control strategy
started to increase due to the large amount of reverse power flows. The can be used when voltage issues are the major concern of a network. It
control strategy for power loss reduction led to minimum daily energy is also shown that, although not outperforming along any single op-
losses under all DG penetrations from 0 to 55%. Fig. 12b shows that, as erational metric, the compromise control strategies can always achieve
DG penetrations increased, the line utilization first decreased due to the medium performances along different criteria and relatively high DG
offset of demand and generation. Then line utilizations became satu- penetrations. Such control strategies can be used as trade-off solutions
rated gradually due to large amounts of reverse power flows brought by for the MVDC link operation.

1.8

1.6
AC operation
1.4
control strategy for PLR
Power Loss (MW)

1.2
control strategy for FLB

1 control strategy for VPI

compromise control strategy 1

0.8

compromise control strategy 2

0.6

0.4

0.2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (half-hourly)
Fig. 8. Power losses over a day of the network by using MVDC link and the original AC operation.

184
Q. Qi et al. Applied Energy 230 (2018) 175–188

70

60

50

40 Demand
P (MW) 30 Generation

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (half-hourly)
Fig. 9. Case 2: modified daily profiles of total demand and generation on Anglesey representing the worst-case scenario.

2.7

2.5

2.3
AC operation
2.1
Power Loss (MW)

1.9 control strategy for PLR

1.7 control strategy for FLB
1.5
control strategy for VPI
1.3
compromise control strategy 1
1.1

0.9 compromise control strategy 2

0.7

0.5

0.3
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Time (half-hourly)

Fig. 10. Power losses over a day of the network by using MVDC link and the original AC operation (worst-case scenario).

Table 5 DC operation, in order to enable improved power flow and voltage

Power and energy losses by using different control strategies as shown in control while enhancing the thermal capability of the circuit in a timely
Fig. 10. manner. A point-to-point MVDC link was considered in this paper and it
Daily average Daily energy loss is consistent with the project. However, the proposed GT-based control
power loss (MW) (MWh) method is applicable to other converter topologies as well, i.e. either a
back-to-back, a multi-terminal, or a series connected MVDC link.
AC operation 1.875 45.01
There are also limitations of the work. Since historical demand and
Control strategy for PLR 0.810 19.44
Control strategy for FLB 0.831 19.95 generation data of a network are required in the offline studies to derive
Control strategy for VPI 0.887 21.28 control curves of an MVDC link, the historical data would affect the
Compromise control strategy 1 0.865 20.77 performances of the control curves. In addition, when the network
Compromise control strategy 2 0.867 20.82 configuration is changed, the control curves may not be applicable to
Switching to the strategy incurring 0.807 19.38
lowest power loss accordingly
the new configuration and new offline studies need to be carried out.
Future works can be carried out with a modified network based on
the real network data of Anglesey (e.g. modified network topology and/
or DGs in modified locations) to expand the learning of this study.
6. Discussions Future works also include analyzing the impact of a multi-terminal
MVDC link on distribution networks, and to compare a multi-terminal
The GT-based control significantly reduces the costs associated with MVDC link with the two-terminal one. Using different optimization
communication and measurement, compared to centralized control algorithms in offline studies to obtain the response curves of an MVDC
schemes, because the GT-based control only requires a few measure- link, and comparing their performances are also considered as one of
ment points in a network (i.e. at the grid transformers). The exact costs the future works to be undertaken.
and savings were not considered, since the scope of this work is to
propose an effective control method of MVDC link and assess its ben-
efits in increasing the network’s hosting capacity for DG connections, 7. Conclusions
and in reducing the network losses. A cost/benefit analysis of con-
structing an MVDC link in Anglesey network was carried out by SP In this paper, impacts of an MVDC link on the performances of an
Energy Networks in [49], but it is out of the scope of this work. electrical distribution network with high DG penetrations were in-
The ANGLE-DC project aims to convert an existing AC circuit into vestigated. A real-time control method for MVDC link was proposed, in

185
Q. Qi et al. Applied Energy 230 (2018) 175–188

65

60

DG Hosting Capacity (%)

60 AC operation
control strategy for PLR
55 55 55
55 control strategy for FLB
control strategy for VPI
50
50
compromise control strategy 1

45 compromise control strategy 2

45

40

Fig. 11. DG hosting capacity of the network by using MVDC link and the original AC operation.

50
Daily Energy Losses (MWh)

45
40
AC operation
35
control strategy for PLR
30
control strategy for FLB
25
control strategy for VPI
20
15 compromise control strategy 1

10 compromise control strategy 2

5
0
0 5 10 15 20 25 30 35 40 45 50 55 60
DG penetration (%)
Fig. 12a. Impacts of DG penetration on daily energy losses.
Daily Maximum Line Utilization (%)

100

90
AC operation
80
control strategy for PLR
70 control strategy for FLB
control strategy for VPI
60 compromise control strategy 1
compromise control strategy 2
50

40
0 5 10 15 20 25 30 35 40 45 50 55 60
DG penetration (%)

Fig. 12b. Impacts of DG penetration on daily maximum line utilizations.

Daily Maximum Bus Voltage (kV)

35
34.8
AC operation
34.6
control strategy for PLR
34.4
control strategy for FLB
34.2 control strategy for VPI
34 compromise control strategy 1
compromise control strategy 2
33.8
33.6
33.4
0 5 10 15 20 25 30 35 40 45 50 55 60

DG penetration (%)

Fig. 12c. Impacts of DG penetration on daily maximum bus voltages.

186
Q. Qi et al. Applied Energy 230 (2018) 175–188

which active power flows at the grid transformers were used to de- [15] Barragan M, Mauricio JM, Marano A, Nieves M, Churio J, Maza-Ortega JM, et al.
termine the set-points of the MVDC link. Control strategies considered Operational benefits of multiterminal DC-links in active distribution networks.
Presented at the power and energy society general meeting, 2012 IEEE, San Diego,
multiple objectives, i.e. power loss reduction, feeder load balancing, CA, USA; 2012.
voltage profile improvement, and compromise options among these [16] Sciano D, Raza A, Salcedo R, Diaz-Aguilo M, Uosef RE, Czarkowski D, et al.
criteria. These control strategies were developed through offline stu- Evaluation of DC links on dense-load urban distribution networks. IEEE Trans
Power Deliv 2016;31:1317–26.
dies, where an improved MOPSO method was applied. The effectiveness [17] Qi Q, Long C, Wu J, Smith K, Moon A, Yu J. Using an MVDC link to increase DG
of implementing an MVDC link with the proposed control method was hosting capacity of a distribution network. Presented at the 9th international con-
validated on a real network. Results indicated that, for an MV dis- ference on applied energy (ICAE), Cardiff, UK; 2017.
[18] Long C, Wu J, Smith K, Moon A, Bryans R, Yu J. MVDC Link in a 33kV distribution
tribution network, it might be beneficial to switch among different network. Presented at the 24th international conference on electricity distribution
control strategies. Results also showed that, regardless of the control CIRED, Glasgow, UK; 2017.
strategy used, an MVDC link significantly increases the network hosting [19] Marano-Marcolini A, Villarejo MB, Fragkioudaki A, Maza-Ortega JM, Ramos ER,
Jaén AV, et al. DC link operation in smart distribution systems with communication
capacity for DGs compared to the AC operation. For the distribution
interruptions. IEEE Trans Smart Grid 2016;7:2962–70.
network under study, the MVDC link could bring up to 15% increase in [20] Weckx S, Gonzalez C, Rybel TD, Driesen J. LS-SVM-based on-load tap changer
DG hosting capacity over AC operation. The control strategy for feeder control for distribution networks with rooftop PV's. Presented at the innovative
load balancing resulted in the highest DG hosting capacity, due to that smart grid technologies Europe (ISGT EUROPE) 2013 4th IEEE/PES, Lyngby,
Denmark; 2013.
the main challenges brought by the DGs are thermal overloading. [21] Bryans R, Bebbington M, Yu J, Smith K, Knott J, Moon A. Real time control of a
distribution connected MVDC link (ANGLE-DC). Presented at the 13th IET inter-
Acknowledgments national conference on AC and DC power transmission (ACDC 2017), Manchester,
UK; 2017.
[22] Green TC, Silversides RW, Lüth T. Power electronics in distribution system man-
This work has been partially supported by Ofgem Electricity agement. HubNet position paper series. Available: < http://www.hubnet.org.uk/
Network Innovation Competition project ANGLE-DC 2016-2020 and filebyid/633/PE_Distribution.pdf > [accessed Apr 2018].
[23] Nick M, Cherkaoui R, Paolone M. Optimal allocation of dispersed energy storage
FLEXIS (Flexible Integrated Energy Systems). FLEXIS is part-funded by systems in active distribution networks for energy balance and grid support. IEEE
the European Regional Development Fund (ERDF), through the Welsh Trans Power Syst 2014;29:2300–10.
Government. This work has also been partially supported by JUICE [24] Vovos PN, Kiprakis AE, Wallace AR, Harrison GP. Centralized and distributed
voltage control: impact on distributed generation penetration. IEEE Trans Power
project, ‘Joint UK-India Clean Energy Centre’ (EP/P003605/1). Syst 2007;22:476–83.
Information on the data supporting the results presented here, in- [25] Sansawatt T, Ochoa LF, Harrison GP. Smart decentralized control of DG for voltage
cluding how to access them, can be found in the Cardiff University data and thermal constraint management. IEEE Trans Power Syst 2012;27:1637–45.
[26] Antoniadou-Plytaria KE, Kouveliotis-Lysikatos IN, Georgilakis PS, Hatziargyriou
catalogue at http://doi.org/10.17035/d.2018.0050208156.
ND. Distributed and decentralized voltage control of smart distribution networks:
models, methods, and future research. IEEE Trans Smart Grid 2017;8:2999–3008.
References [27] Iwamoto S, Tamura Y. A fast load flow method retaining nonlinearity. IEEE Trans
Power Apparatus Syst 1978;PAS-97:1586–99.
[28] Keyhani A, Abur A, Hao S. Evaluation of power flow techniques for personal
[1] Wang Q, Zhang C, Ding Y, Xydis G, Wang J, Østergaard J. Review of real-time computers. IEEE Trans Power Syst 1989;4:817–26.
electricity markets for integrating distributed energy resources and demand re- [29] Keyhani A. Study of fast decoupled load flow algorithms with substantially reduced
sponse. Appl Energy 2015;138:695–706. memory requirements. Electr Power Syst Res 1985;9:1–9.
[2] Anaya KL, Pollitt MG. Options for allocating and releasing distribution system ca- [30] Deb K. Multi-objective optimization using evolutionary algorithms. New York, USA:
pacity: deciding between interruptible connections and firm DG connections. Appl Wiley; 2004.
Energy 2015;144:96–105. [31] Qi Q, Wu J, Long C. Multi-objective operation optimization of an electrical dis-
[3] Esmaili M, Firozjaee EC, Shayanfar HA. Optimal placement of distributed genera- tribution network with soft open point. Appl Energy 2017;208:734–44.
tions considering voltage stability and power losses with observing voltage-related [32] Kennedy J, Eberhart R. Particle swarm optimization. In: Proc. IEEE international
constraints. Appl Energy 2014;113:1252–60. conference on neural networks; 1995. p. 1942–8.
[4] Mallwitz R, Engel B. Solar power inverters. Presented at the integrated power [33] Ganguly S. Multi-objective planning for reactive power compensation of radial
electronics systems (CIPS), 2010 6th international conference, Nuremberg, distribution networks with unified power quality conditioner allocation using
Germany; 2010. particle swarm optimization. IEEE Trans Power Syst 2014;29:1801–10.
[5] Ramsay SM, Cronin PE, Nelson RJ, Bian J, Menendez FE. Using distribution static [34] Parsopoulos KE, Vrahatis MN. Particle swarm optimization method in multi-
compensators (D-STATCOMs) to extend the capability of voltage-limited distribu- objective problems. Presented at the ACM symposium on applied computing,
tion feeders. Presented at rural electric power conference, Fort Worth, TX, USA; Madrid, Spain; 2002.
1996. [35] Baumgartner U, Magele C, Renhart W. Pareto optimality and particle swarm opti-
[6] Graaff RAAd, Myrzik JMA, Kling WL, Enslin JHR. Intelligent nodes in distribution mization. IEEE Trans Magn 2004;40:1172–5.
systems - optimizing steady state settings. Presented at IEEE power tech, Lausanne, [36] Xiaohui H, Eberhart R. Multiobjective optimization using dynamic neighborhood
Switzerland; 2007. particle swarm optimization. In: Proceedings of the 2002 congress on evolutionary
[7] Bloemink JM, Green TC. Benefits of distribution-level power electronics for sup- computation; 2002. p. 1677–81.
porting distributed generation growth. IEEE Trans Power Deliv 2013;28:911–9. [37] Xiaohui H, Eberhart RC, Yuhui S. Particle swarm with extended memory for mul-
[8] Bloemink JM, Green TC. Increasing distributed generation penetration using soft tiobjective optimization. In: Proceedings of the 2003 IEEE swarm intelligence
normally-open points. Presented at the power and energy society general meeting, symposium; 2003. p. 193–7.
Providence, RI, USA; 2010. [38] Li X. A non-dominated sorting particle swarm optimizer for multiobjective opti-
[9] Cao W, Wu J, Jenkins N, Wang C, Green TC. Benefits analysis of soft open points for mization. In: Proceedings of the 2003 international conference on genetic and
electrical distribution network operation. Appl Energy 2016;165:36–47. evolutionary computation; 2003. p. 37–48.
[10] Long C, Wu J, Thomas L, Jenkins N. Optimal operation of soft open points in [39] Alvarez-Benitez JE, Everson RM, Fieldsend JE. A MOPSO algorithm based ex-
medium voltage electrical distribution networks with distributed generation. Appl clusively on Pareto dominance concepts. In: Proceedings of the international con-
Energy 2016;184:427–37. ference on evolutionary multi-criterion optimization; 2005. p. 459–73.
[11] Western Power Distribution. Network equilibrium. Available: < https://www. [40] Mostaghim S, Teich J. The role of e-dominance in multi objective particle swarm
westernpower.co.uk/Innovation/Projects/Current-Projects/Network-Equilibrium. optimization methods. In: Proc. the 2003 congress on evolutionary computation;
aspx > [accessed Feb 2018]. 2003. p. 1764–71.
[12] UK Power Networks. Flexible urban networks low voltage. Available: < http:// [41] Pošík P. Preventing premature convergence in a simple EDA via global step size
innovation.ukpowernetworks.co.uk/innovation/en/Projects/tier-2-projects/ setting. In: Proceedings of the international conference on parallel problem solving
Flexible-Urban-Networks-Low-Voltage/ > [accessed Feb 2018]. from nature; 2008. p. 549–58.
[13] Marano-Marcolini A, Romero-Ramos E, Gomez-Exposito A, Maza-Ortega JM, [42] Narang N, Dhillon JS, Kothari DP. Multiobjective fixed head hydrothermal sche-
Martinez-Ramos JL. Enhancing the integration of renewable sources in distribution duling using integrated predator-prey optimization and Powell search method.
systems using DC-links. Presented at the sustainable alternative energy, Valencia, Energy 2012;47:237–52.
Spain; 2009. [43] Deb K, Sundar J. Reference point based multi-objective optimization using evolu-
[14] Romero-Ramos E, Gomez-Exposito A, Marano-Marcolini A, Maza-Ortega JM, tionary algorithms. In: Proceedings of the 8th annual conference on genetic and
Martinez-Ramos Jl. Assessing the loadability of active distribution networks in the evolutionary computation; 2006. p. 635–42.
presence of DC controllable links. IET Gener Transm Distrib 2011;vol:5. [44] Dzung NT. Application of multi-objective optimization by the utopian point method

187
Q. Qi et al. Applied Energy 230 (2018) 175–188

to determining the technological mode of gac oil extraction. Int J Chem Eng Appl [47] Keane A, Ochoa LF, Vittal E, Dent CJ, Harrison GP. Enhanced utilization of voltage
2012;3:18–24. control resources with distributed generation. IEEE Trans Power Syst
[45] SP Energy Networks. Electricity NIC submission: SP Energy Networks – ANGLE-DC. 2011;26:252–60.
Available: < https://www.ofgem.gov.uk/publications-and-updates/electricity-nic- [48] Keane A, Malley MO. Optimal utilization of distribution networks for energy har-
submission-sp-energy-networks-angle-dc > [accessed Feb 2018]. vesting. IEEE Trans Power Syst 2007;22:467–75.
[46] Zou K, Agalgaonkar AP, Muttaqi KM, Perera S. Distribution system planning with [49] SP Energy Networks. ANGLE-DC 2015 electricity network innovation competition.
incorporating DG reactive capability and system uncertainties. IEEE Trans Sustain Available: < https://www.ofgem.gov.uk/sites/default/files/docs/angle_
Energy 2012;3:112–23. submission.pdf > [accessed Aug 2018].

188