Applied Energy 189 (2017) 301–309

Contents lists available at ScienceDirect

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

Optimal siting and sizing of soft open points in active electrical

distribution networks
Chengshan Wang a, Guanyu Song a, Peng Li a,⇑, Haoran Ji a, Jinli Zhao a, Jianzhong Wu b
a
Key Laboratory of Smart Grid of Ministry of Education, Tianjin University, Tianjin 300072, China
b
Institute of Energy, School of Engineering, Cardiff University, Cardiff CF24 3AA, UK

h i g h l i g h t s

A model to optimal siting and sizing of SOPs was proposed considering DG impacts.

The problem was converted to an MISOCP model with improved accuracy and convergence.

Economic benefits and algorithm validation were given for different scale systems.

a r t i c l e i n f o a b s t r a c t

Article history: Soft open points (SOPs) are power electronic devices installed to replace normally open points in active
Received 13 September 2016 electrical distribution systems. SOPs can provide active/reactive power flow control and voltage regula-
Received in revised form 12 December 2016 tion under normal operating conditions as well as fast fault isolation and supply restoration under abnor-
Accepted 14 December 2016
mal conditions. The application of SOPs can improve the controllability of distribution systems, thus
further enhances the economy, flexibility and reliability of the grid. In this paper, considering the
long-term operation characteristics of distributed generation, a mixed integer non-linear optimization
Keywords:
problem is formulated to optimally determine the siting and sizing of SOPs based on the typical operation
Active electrical distribution network
Soft open points (SOPs)
scenarios generated by Wasserstein distance. It is then transformed to and solved as a mixed integer
Optimal siting and sizing second-order programming model. Finally, case studies on the IEEE 33-node system and Taiwan Power
Operation scenario generation Company distribution system are used to verify the effectiveness and efficiency of the proposed method.
Mixed integer second-order cone Ó 2016 Elsevier Ltd. All rights reserved.
programming (MISOCP)

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
2. Optimization model to site and size SOPs in an active electrical distribution network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
2.1. Typical operation scenario generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
2.2. Modelling of soft open points. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
2.3. Optimization model to site and size SOPs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
3. MISOCP model conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
4. Case study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
4.1. IEEE 33-node system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
4.2. Taiwan power Company distribution system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
4.3. Algorithm validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308

⇑ Corresponding author.
E-mail address: lip@tju.edu.cn (P. Li).

http://dx.doi.org/10.1016/j.apenergy.2016.12.075
0306-2619/Ó 2016 Elsevier Ltd. All rights reserved.
302 C. Wang et al. / Applied Energy 189 (2017) 301–309

1. Introduction AC optimal power flow was used to define a mixed integer

In recent years, an increasing number of distributed generators
(DG), such as wind turbines (WT) and photovoltaic (PV) arrays,
have been integrated into electrical distribution systems. The inte-
gration of DG is able to contribute to the power loss reduction [1],
power supply reliability improvement [2], economic promotion
[3], reduction of CO2 emission, etc. [4,5], while posing new chal-
lenges to the planning and operation of distribution systems as
well, such as a higher risk of network congestion [6] and voltage
violation [7], and bi-directional power flow [8]. Traditional distri-
bution systems cannot effectively address the problems caused
by the high penetration of intermittent DG because of their limited
regulation means. Soft open points (SOPs) are power electronic
devices which replace normally open points (NOPs), effectively
covering the lack of power adjustment ability in a distribution sys-
tem. Instead of simply opening/closing NOPs, SOPs can further bal-
ance load flow and optimize the network voltage profile by
providing fast, dynamic and continuous active/reactive power flow
control among the feeders. Meanwhile, due to the isolation of DC
link and instantaneous control of currents, the short-circuit current
brought by SOP is limited and different from the traditional fault
characteristics [9]. Coordinated with the protection of distribution
network, the duration of fault isolation and supply restoration can
be shortened. The application of SOPs enhances the controllability
of distribution power flows to improve the system operation with a
better economy, flexibility and reliability [10].
Previous studies have investigated both steady-state and tran-
sient performances of SOPs to facilitate distribution network oper-
ation [11–16]. In [11,12], the concept and topology of SOPs were
proposed. Three types of topologies were defined, including
back-to-back voltage source converter (B2B VSC), unified power
flow controller (UPFC) and static series synchronous compensator
(SSSC). In [13], based on the traditional B2B VSCs, the DC capacitor
was replaced by battery energy storage, which further enhanced
the regulating range of the studied SOP. In [14], the applications
of SOPs were analysed and verified in terms of DG accommodation,
feeder load balance and voltage profile improvement. In [15,16],
considering the system fault transient process, different control
modes of SOPs during a fault and post-fault period were investi-
gated. Compared with NOPs, SOPs have more reliable power con-
trol, and can even realize real-time optimal control of the
distribution network. When a fault occurs, the fault recovery mode
of SOPs is used for power supply of the load out of service. How-
ever, SOPs are mainly based on fully controlled power electronic
devices, leading to higher capital and maintenance costs. There-
fore, it is necessary to optimize the siting and sizing of SOPs to
maximize the benefits of capital and operation.
The problem of optimal installation sites and capacities of SOPs
consists of not only continuous variables such as voltages and cur-
rents of a distribution network, but also discrete variables such as
the locations and capacities of SOPs. It is a mixed integer non-
linear programming problem (MINLP). Meanwhile, the connection
of DG will change the power flow of a distribution system, bringing
more challenges to the siting and sizing of the SOPs. Therefore, the
long-term operation characteristics of DG should be considered.
second-order cone programming (MISOCP) model to optimally site
and size the energy storage systems in the distribution network.
In this paper, an MISOCP model is proposed to optimally deter-
mine the siting and sizing of SOPs. First, the yearly DG operation
characteristics are considered in this model using scenario genera-
tion method based on the Wasserstein distance [19]. For the typi-
cal scenarios generated, to minimize the annual expense of the
overall distribution system, a model was developed to optimally
determine the installation sites and capacities of SOPs while con-
sidering the constraints of the radial distribution system, system
power flow, system security and operation of the SOPs, which is
essentially an MINLP problem. Furthermore, a conic conversion is
used to transform the original MINLP model into an MISOCP model.
Finally, the effectiveness and efficiency of the proposed method are
verified on the IEEE 33-node system and Taiwan Power Company
(TPC) distribution system.
The main contributions are summarized as follows: (1) Taking
the long-term operational characteristics of DG and network topol-
ogy changes into account, a model was developed to optimally
determine the siting and sizing of SOPs to minimize the annual
expense of the overall distribution system including capital cost
and annual operational cost of the SOP, and annual energy loss cost
of the distribution system. (2) The siting and sizing of SOPs is a
large-scale MINLP problem, which cannot be efficiently solved.
Firstly, we reduce the size of the original problem using Wasser-
stein distance based scenario generation method. Then by applying
the linearization and conic relaxation, the original nonconvex
MINLP model is converted into an MISOCP model, which guarantee
a global optimality with moderate computational burden.
The reminder of the paper is organized as follows. Section 2
introduces the optimization model to site and size SOPs in an
active electrical distribution network. The original problem is con-
verted into an MISOCP model by the linearization and conic relax-
ation in Section 3. Case studies are given in Section 4 to verify the
effectiveness and efficiency of the proposed method. Section 5 con-
cludes this paper with a discussion.
2. Optimization model to site and size SOPs in an active
electrical distribution network
2.1. Typical operation scenario generation
In order to consider DG influence on the installation sites and
capacities of SOPs, the probability density functions (PDFs) generated
based on the historical data is used [20]. And the discrete PDF distri-
bution is used instead of a continuous distribution, which is called
scenario generation. In this paper, a method based on the Wasser-
stein distance is adopted to generate typical operation scenarios so
as to reduce the computational burden in the optimization process.
Assuming that a continuous PDF of variable x is f ðxÞ, S discrete
scenarios is used to approximate f ðxÞ. The typical scenario zs
(s = 1, 2, . . . , S) is obtained as follows:
Z Z þ1
zs
1 s1 1
f ðxÞ2 dx ¼ f ðxÞ2 dx ð1Þ
1 2S 1

Furthermore, the number of decision variables and constraints is The probability ps of a scenario is calculated as follows:
increasing rapidly with the system scale, making the problem solv- 8 Z zs þzsþ1
ing difficult and time-consuming. Heuristic methods such as >
> 2
>
> p ¼ f ðxÞdx; s ¼ 1; 2; . . . ; S
genetic algorithm, simulated annealing and particle swarm algo- >
> s zs1 þzs
>
>
rithm are usually used to solve such problem. Ref. [17] used a >
>
2
Z z1 þz2
< 2
genetic algorithm to determine the active distribution network p1 ¼ z þz f ðxÞdx ð2Þ
expansion planning, in which the rewiring, network reconfigura- >
>
>
>
0 1
2
>
> Z zS þzSþ1
tion, new protection devices and DG installation were considered. >
>
>
>
2
Recently, some mathematical programming methods are also : pS ¼ z þz f ðxÞdx
S1 S
adopted to solve the problem. In [18], a convex formulation of 2
 C. Wang et al. / Applied Energy 189 (2017) 301–309 303

Fig. 1. SOP in a medium-voltage distribution system.

Here, z0 and zSþ1 are the lower and upper limits of x, which are min f ¼ C CAP þ C OPE þ C LOSS ð6Þ
1 and þ1 if there is no additional information. The Wasserstein
distance is the integral of two PDF gaps [21]. The annual expense of the overall distribution system is com-
Weibull distribution [22] and Beta distribution [23] are usually posed of the following three parts.
used to describe PDFs of wind speed and light intensity for a long
time scale, which are used frequently in the study of electric power
(1) C CAP : capital cost of SOP
system planning problems. With the PDFs illustrated in the refer-
ences, the typical operation scenarios considered in siting and siz-
dð1 þ dÞ X N Xy
ing SOPs can be generated based on the Wasserstein distance. C CAP ¼ y cSOP SSOP
ij ð7Þ
ð1 þ dÞ  1 i¼1 j2XðiÞ
2.2. Modelling of soft open points
where d is the discount rate, and y is the SOP lifetime. N is the num-
An SOP is able to control the active power flow between the ber of nodes, XðiÞ is the set of all adjacent nodes of node i, and cSOP is
connected feeders and supply or absorb the reactive power at its the capital cost per unit capacity.
interface terminals under normal operation conditions. Fig. 1
shows a typical integration of an SOP. (2) C OPE : annual operational cost of SOP
This paper uses B2B VSCs in the optimization model for siting X
N X

and sizing an SOP under steady-state conditions, and the PQ-

C OPE ¼ g cSOP SSOP
ij ð8Þ
i¼1 j2XðiÞ
VdcQ control is selected as the SOP control mode [24]. Because
SOP can accurately control the active/reactive power of the feeder where g is the coefficient of the annual operational costs.
to which they are connected [11], the active/reactive power of the
two converters are taken as the decision variables. Although the (3) C LOSS : annual energy loss cost of distribution system
operation efficiency of SOP is sufficiently high, they inevitably pro-
duce losses when there is a large-scale transfer of active power. C LOSS ¼ 24  365  c  ELOSS ð9Þ
!
Therefore, a loss coefficient is considered in the model. Moreover, X
S X
N X X
N
SOP SOP
the constraint of transmitted active power between the two con- ELOSS ¼ 2
r ij Is;ij þ Ai jPs;i j  pðsÞ ð10Þ
verters is described using Eq. (3). The reactive power outputs of s¼1 i¼1 j2XðiÞ i¼1

the two converters are independent of each other because of the

where S is the set of all scenarios. c is the electricity price. ELOSS is
DC isolation and should satisfy their own capacity constraints.
the loss expectation of the network and all SOPs for an hour. pðsÞ
Based on the typical operation scenarios, an SOP model is obtained
is the probability corresponding to the sth scenario. Moreover,
that includes the following constraints:
zij ¼ rij þ ixij is denoted as the complex impedance of line ij, and
(1) SOP active power constraints Is;ij is the current of line ij in the sth scenario. In (9), the multiplier
24  365 denotes 24 h and 365 days.
SOP SOP The operation constraints of the distribution system are as
PSOP SOP
s;i þ P s;j þ Ai jPs;i j þ ASOP
j jPSOP
s;j j ¼ 0 ð3Þ
follows:
(2) SOP capacity constraints
(1) Radial distribution system constraints
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 SOP 2 SOP
ðPSOPs;i Þ þ ðQ s;i Þ 6 Sij ð4Þ
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi aij ¼ bij þ bji ; j 2 XðiÞ ð11Þ
X
ðPSOP
2
ðQ SOP
2
SSOP bij ¼ 1; 8i 2 N n NS ð12Þ
s;j Þ þ s;j Þ 6 ij ð5Þ
j2XðiÞ
X
where s is the index of the scenario. PSOP SOP SOP SOP
s;i , P s;j , Q s;i , and Q s;j are the
bij ¼ 0; 8i 2 N S ð13Þ
j2XðiÞ
transmitted active/reactive power of the SOP at node i and j in the
sth scenario. ASOP
i and ASOP
j are the loss coefficients of the two con- where N S denotes the subset of nodes that are substations. Let a sin-
verters at nodes i and j. SSOP
ij denotes the capacity of the installed gle binary variable aij designate the switch status of the line. For
SOP between nodes i and j. example, aij ¼ 1 implies that the switch of line ij is closed, whereas
aij ¼ 0 indicates that it opens. Since the binary variable aij is not
2.3. Optimization model to site and size SOPs directional, the summation of aij over all lines is exactly the number
of closed switches. Each switch status of the line is also associated
The minimum annual expense of the overall distribution system with two continuous line orientation variables bij and bji , which rep-
is used as the objective function, which is formulated as follows: resent the power flow direction.
304 C. Wang et al. / Applied Energy 189 (2017) 301–309

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
(2) System power flow constraints Xn
K ¼ fx 2 Rn : x1 P j¼2 j
x2 ; x1 P 0g ð24Þ
X X
aji ðPs;ji  r ji I2s;ji Þ þ Ps;i ¼ aik Ps;ik ð14Þ X
n
j2UðiÞ k2WðiÞ K ¼ fx 2 Rn : 2x1 x2 P x2j ; x1 ; x2 P 0g ð25Þ
X X j¼3
aji ðQ s;ji  xji I2s;ji Þ þ Q s;i ¼ aik Q s;ik ð15Þ
j2UðiÞ k2WðiÞ As shown above, the second-order cone programming has nota-
U 2s;i  U 2s;j  2ðrij Ps;ij þ xij Q s;ij Þ þ ðr 2ij þ x2ij ÞI2s;ij ¼ 0 ð16Þ bly strict demands on the mathematical formulation. The objective
function needs to be a linear function of the decision variable x,
I2s;ij U 2s;i ¼ P2s;ij þ Q 2s;ij ð17Þ and its feasible region is composed of linear equality constraints
Ps;i ¼ P DG SOP LOAD
s;i þ P s;i  P s;i ð18Þ and convex cone constraints, where Ax ¼ b represents the linear
constraints and x 2 K represents the convex cone constraints.
Q s;i ¼ Q DG
s;i þ Q SOP
s;i  Q LOAD
s;i ð19Þ
Therefore, the optimization model to site and size SOPs has to be
Eqs. (14) and (15) represent the law of power conservation at changed in advance before applying the second-order cone
each node i in the sth scenario, where P s;i and Q s;i denote the real programming.
and reactive power injection at node i. UðiÞ # N denotes the set of The aforementioned optimization model is described with
all parents of node i, and WðiÞ # N denotes the set of all children numerous nonlinear functions, such as the square representation
of node i. The Ohm’s law over branch ij at time t is expressed as of the voltage amplitude and current amplitude. First, additional
Eq. (16). The current magnitude of each line can be determined variables is introduced for each node and line to linearize nonlinear
by Eq. (17). Eqs. (18) and (19) indicate the power injection at each functions using variable substitution.
node i, which is the generation minus the load on node i in the sth e s;i denote the square of the magnitude of
For nodes i; j 2 N, let U
scenario. P DG DG
s;i and Q s;i are the real and reactive power, generated by its complex voltage, and let eI s;ij denote the square of the magnitude
the DG that is connected to node i via inverters. P LOAD and Q LOAD are of the complex current. Then, substituting the new optimization
s;i s;i
e s;i and eI s;ij into the proposed model, the objective func-
variables U
the real and reactive power consumption in the sth scenario. P SOP
s;i
tion and nonlinear constraints become linear except for the equal-
and Q SOP
s;i are the real and reactive power delivered by the SOP. ity constraints described in (17). To cast them as second-order
cone constraints, these nonlinear equality constraints are relaxed
(3) System security constraints to inequality constraints [26]:

ðU min
2
Þ 6 U 2s;i 6 ðU max Þ
2
ð20Þ I2s;ij U 2s;i P P 2s;ij þ Q 2s;ij ð26Þ
i i
2
I2s;ij 6 ðImax Þ ð21Þ Clearly, the original feasible set is enlarged, although previous
ij
studies [16] have shown that the relaxations are exact if moderate
where U max
i and U min
i are the upper and lower limits of the voltage conditions are satisfied, which implies that inequalities (26) can be
amplitude at node i. Imax
ij is the upper limit of the current amplitude tight in the optimal solution. More mathematical proofs can be
of line ij. found in [27,28].
The relationship between the location and capacity of an SOP is The corresponding second-order cone programming formula-
represented as follows: tion is
 
SSOP ¼ mij smodule ð1  aij Þ ð22Þ  2Ps;ij 
ij  
 2Q 
 s;ij  6 eI s;ij þ U
e s;i ð27Þ
where smodule is the minimum optimum capacity, which represents  
 eI  U e 
the capacity of the basic power electronic module that comprises s;ij s;i 2

the SOP. mij is the quantity of the modules. Eq. (27) ensures that the decision variables are of the quadratic
The optimization variables in the model include the installation cone form and limits the search space within the convex cone
sites and capacities, switch states, and the transmitted active/reac- range.
tive power of the SOPs in each scenario. As a consequence, Eqs. (3)– The capacity constraints of an SOP are transformed as follows
(22) form the optimization model to determine the installation according to the SOCP requirements:
sites and capacities of SOPs in an active distribution system.
2 SOP 2 SSOP
ij SSOP
ij
3. MISOCP model conversion ðPSOP
s;i Þ þ ðQ s;i Þ 6 2  pffiffiffi  pffiffiffi ð28Þ
2 2
Considering typical operation scenarios, the optimization model SOP 2 SOP 2 SSOP
ij SSOP
ij
ðPs;j Þ þ ðQ s;j Þ 6 2  pffiffiffi  pffiffiffi ð29Þ
to determine the siting and sizing of SOPs becomes complicated, 2 2
making it more difficult to be solved. Second-order cone program-
ming (SOCP) is a mathematically convex programming approach Eqs. (28) and (29) are equivalent to the original constraints (4)
that can be considered as the generalisation of both linear and non- and (5) while satisfying the requirement of the rotated quadratic
linear programming. SOCP can solve minimum linear objective cone form.
functions based on a convex cone in a linear space. It has excellent For the absolute terms jP SOP SOP
s;i j and jP s;j j in Eqs. (3) and (10), aux-
performance in terms of global optimality and computation effi-
iliary variables M SOP SOP
s;i and M s;j are introduced to represent and lin-
ciency. The SOCP standard form is written as follows [25]:
earize them as follows:
minfc T xjAx ¼ b; x 2 Kg ð23Þ
MSOP SOP
s;i P 0; M s;j P 0 ð30Þ
where x is the dimensional decision variables. c, b and A are the con-
stant vectors and matrix. K denotes the Cartesian product of a lim- MSOP
s;i P PSOP SOP
s;i ; M s;i P PSOP
s;i ð31Þ
ited nonempty-pointed convex cone, which is generally expressed MSOP SOP SOP SOP
s;j P P s;j ; M s;j P P s;j ð32Þ
using a quadratic cone (24) or a rotated quadratic cone (25):
 C. Wang et al. / Applied Energy 189 (2017) 301–309 305

PV WT WT

23 24 25 26 27 28 29 30 31 32 33

PV WT PV WT WT

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

19 20 21 22

Fig. 2. Structure of the IEEE 33-node system.

Table 1
Installation capabilities of DGs.

Parameters WT PV
Location 10 16 17 30 32 7 13 27
Capacity, kVA 500.0 200.0 150.0 200.0 300.0 500.0 300.0 400.0

25

20
Wind speed, m/s

15

10

0
0 2000 4000 6000 8000 8760
Time, h

1200
Light intensity, W/m2

900

600

300

0
0 2000 4000 6000 8000 8760
Time, h

Fig. 3. Annual wind speed and light intensity curves.

After the conic conversion, the MISOCP model to determine the 4.1. IEEE 33-node system
installation sites and capacities of SOPs is formulated.
The modified IEEE 33-node system is presented in Fig. 2. It has
32 normally closed branches and 5 normally open branches. The
4. Case study IEEE 33-node system has been widely used to verify and demon-
strate various types of methods and algorithms developed for the
In this section, the modified IEEE 33-node system and TPC dis- planning and operation of medium voltage distribution networks.
tribution system are used to demonstrate the effectiveness and More detailed parameters are shown in [30].
efficiency of the proposed method. The proposed method in this In order to consider the impact of high penetration of DGs, five
paper was implemented in the YALMIP optimization toolbox [29] wind turbines and two photovoltaic cells are integrated into the
with MATLAB R2013a, and solved by IBM ILOG CPLEX 12.6. The networks, respectively. And the basic configuration parameters
computation is performed on a PC with an Intel Xeon CPU E5- are shown in Table 1. All the DGs are operated with a power factor
1620 @3.70 GHz processor and 32 GB RAM. 1.0.
306 C. Wang et al. / Applied Energy 189 (2017) 301–309

The annual wind speed and light intensity curves used are According to the wind speed and light intensity from Fig. 3, the
shown in Fig. 3. The parameters of the studied case are shown in Weibull distribution parameters can be obtained: the scale param-
Table 2. eter c is 2, and the shape parameter k is 10. The Beta distribution
parameters k1 and k2 are 0.95. Considering the operation charac-
Table 2 teristics of WT and PV, the typical operation scenarios are obtained
Parameters of the studied case. based on the Wasserstein distance from their PDFs.
According to the Weibull distribution parameters, the WT
Parameters Value
parameters and the rated output power (PWT;N ¼ 1:0), the WT
Discount rate 0.08
PDF is obtained as follows:
SOP economical service life, year 20.00
8
< 0:088  dðPWT  1Þ
> ðPWT ¼ 0Þ
SOP minimum optimum capacity, kVA 100.00
SOP unit capital cost, $/kVA 308.80
f ðPWT Þ ¼ 1:5P0:333
0:667
SOP loss coefficient [31] 0.02  e2:25PWT ð0 < PWT < 1Þ ð33Þ
>
:
WT
Coefficient of annual operational cost 0.01 0:103  dðPWT  1Þ ðPWT ¼ 1Þ
Cut-in wind speed, m/s 3.00
Rated wind speed, m/s 15.00 According to the Beta distribution parameters, the PV parame-
Cut-out wind speed, m/s 25.00
ters and the rated output power (PPV;N ¼ 1:0), the PV PDF is as
Electricity Price, $/kWh 0.08
follows:

Table 3
Probability distribution of wind and photovoltaic power.

P WT P PV
Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5
0.0966 0.2971 0.5000 0.7029 0.9034
Scenario 1 0.0000 0.0180 0.0174 0.0173 0.0174 0.0180
Scenario 2 0.0795 0.0916 0.0885 0.0882 0.0885 0.0915
Scenario 3 0.3163 0.0468 0.0452 0.0451 0.0452 0.0468
Scenario 4 0.7094 0.0268 0.0259 0.0258 0.0259 0.0268
Scenario 5 1.0000 0.0210 0.0203 0.0203 0.0203 0.0210

PV WT WT

23 24 25 26 27 28 29 30 31 32 33

PV WT PV WT WT

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

19 20 21 22

(a)

PV WT WT

23 24 25 26 27 28 29 30 31 32 33

PV WT PV WT WT

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

19 20 21 22

(b)
Fig. 4. (a) Location of SOPs in Scheme I and (b) Location of SOPs in Scheme II.
 C. Wang et al. / Applied Energy 189 (2017) 301–309 307

Table 4
Location and capacity of the SOPs without/with network reconfiguration.

Scheme I Location 12–22 25–29 18–33 8–21 9–15

Capacity, kVA – 300 400 – 100
Scheme II Location 6–7 9–15 11–12 31–32 25–29
Capacity, kVA – – – 400 –

f ðP PV Þ ¼ 0:904P0:05
PV ð1  PPV Þ0:05 ð34Þ The optimal installation sites and capacities of the SOPs on IEEE
33-node system are shown in Fig. 4 and Table 4, and the objective
Based on the Wasserstein distance, five wind power scenarios, function values are shown in Table 5.
five photovoltaic scenarios and their corresponding probabilities Table 5 shows that Scheme II has better economic benefits. The
are simultaneously obtained. Then, the combined 25 scenarios annual total cost of Scheme II is $25,210.7 less than that of without
and corresponding probabilities are also derived and presented in SOP, i.e., it decreased by 29.7%. Among the expenses, the annual
Table 3. energy loss cost of the distribution system is reduced by $40251.0
The following two schemes are used to determine the optimal (reduction of 47.4%). Thus, the operational economy of the power
installation sites and capacities of the SOPs on IEEE 33-node distribution system is significantly improved. Based on the dynamic
system. payback period, it takes 9.20 years to recover the SOP capital cost in
Scheme I. However, only 3.55 years are needed to recover the SOP
Scheme I: Without the consideration of network topology capital cost by considering network reconfiguration.
changing, only normally open points are considered as the can-
didate locations of SOPs.
4.2. Taiwan power Company distribution system
Scheme II: With the consideration of network reconfiguration,
both normally open points and normally closed points are con-
Taiwan Power Company (TPC) distribution system is a typical
sidered as the candidate locations of SOPs.
multi-sectioned electrical distribution network, which has 11 feed-
ers, 83 normally closed branches and 13 normally open branches.
Compared with IEEE 33-node system, this case is more challenging
in system scale and complexity, which is used to further verify the
Table 5 scalability of the proposed method. The detailed parameters of the
Annual cost of different schemes. TPC distribution system are provided in [32].
C CAP , $ C OPE , $ C LOSS , $ C, $ Considering multiple operation scenarios of DG, six wind tur-
bines and twelve photovoltaic cells are integrated into the system.
Without SOP – – 84,926.2 84,926.2
The capacities of wind turbines and photovoltaic arrays are
Scheme I 25,145.3 4,935.3 44,458.1 74,538.7
Scheme II 12,572.7 2,467.6 44,675.2 59,715.5 800 kVA and 600 kVA, respectively. The optimal installation sites
and capacities of the SOPs are shown in Fig. 5 and Table 6. The

PV WT 8 9 PV WT PV

A 10 G
1 2 3 4 5 6 7 55 54 53 52 51 50 49 48 47

WT PV PV WT PV

B H
12 13 64 63 62 61
11 60 59 58 57 56
14

PV WT PV
22 PV

C I
15 16 17 18 19 20 21 23 24 72 71 70 69 68 67 66 65

D J
25 26 27 28 29 76 75 74 73

PV WT PV 39 40 PV

E K
30 31 32 33 34 35 36 37 38 41 42 83 82 81 80 79 78 77

F
43 44 45 46

Fig. 5. Location of the SOPs in TPC distribution system.

308 C. Wang et al. / Applied Energy 189 (2017) 301–309

Table 6
Location and capacity of the SOPs in the TPC distribution system.

Location 5–55 7–60 12–72 20–83 40–42

Capacity, kVA 600 600 600 600 200

Table 7
Comparison of the performance of the MISOCP model and original model.

Test case Scale of problem Optimization Time, s Result, $

model/solver
IEEE 33-node system (Scheme I) Integer variables: 5 MISOCP/CPLEX 13.92 74,538.7
Variables: 4,880 MINLP/KNITRO 68.38 74,538.7
Constraints: 8,680
IEEE 33-node system (Scheme II) Integer variables: 37 MISOCP/CPLEX 2124.53 59,715.5
Variables: 4,880 MINLP/KNITRO Divergent –
Constraints: 8,680
TPC distribution system Integer variables: 13 MISOCP/CPLEX 126.70 272,108.1
Variables: 13,288 MINLP/KNITRO Divergent –
Constraints: 22,263

annual total cost, including capital cost and annual operational cost
of the SOP, and annual energy loss cost of the distribution system,
is $272,108.1. It is less than that of without SOPs, which is
$292,013.5. Among the costs, the annual energy loss cost of the dis-
tribution system is reduced by $68,804.3 (reduction of 23.6%).
4.3. Algorithm validation
To verify the effectiveness and efficiency of the model and test
the correctness of the conic conversion, KNITRO [33] in GAMS
package is used as a reference. KNITRO is an optimization package
based on KKT algorithm and interior point method, which can
obtain the global optimum and has been widely applied in solving
the MINLP problem. In Scheme I of IEEE 33-node system, all of the
algorithms converge and obtain the same installation sites and
capacities of the SOPs with 5 integer variables. However, KNITRO
does not converge with more variables and constraints. Table 7
shows the solving time and results for the different algorithms.
Table 7 shows that in the case of fewer integer variables and
constraints such as the Scheme I of the IEEE 33-node system, all
algorithms converge to the identical result, but the MISOCP model
is more efficient. When the quantity of integer variables and con-
straints increases such as the Scheme II of the IEEE 33-node system
and the TPC distribution system, the rapid growth in problem scale
causes the divergence of KNITRO. Because of the linearization and
convex relaxation of the original model, the MISOCP model still
shows good accuracy and convergence by reducing the complexity
of the problem. Therefore, the MISOCP model in this paper is more
effective than the original MINLP model.
5. Conclusion
In this paper, a large-scale MINLP model is proposed to opti-
mally determine the optimal installation sites and capacities of
SOPs, considering the characteristics of DG and network topology
changes. Based on the typical scenarios generation and conic con-
version, an MISOCP model is adopted to solve the problem. The
effectiveness and efficiency of the proposed method are verified
in case studies. Results showed that the application of SOPs can
significantly improve the operational economy of the distribution
systems. With the integration of renewable energy resource, the
flexible control mode of SOPs will also increase the DG hosting
capacity of distribution networks, improve the voltage profile
and enhance the reliability of power supply.
With the development of power electronic technologies and
increasingly complex distribution systems, more flexible
distribution devices such as SOPs will be integrated into the active
electrical distribution networks. The work conducted by this paper
can give a better siting and sizing scheme for the balance of capital
and profit in the engineering practice. Meanwhile, considering sig-
nificant spatial and temporal uncertainties, large-scale mixed inte-
ger nonlinear optimization problems in distribution systems
become more difficult to solve, and the advantages of the proposed
method will become more obvious.
Acknowledgements
This work was supported by the National Natural Science
Foundation of China (51261130473, 51307116) and conducted in
coopetation of APPLIED ENERGY UNiLAB-DEM: Distributed Energy
& Microgrid. UNiLAB is an international virtual lab of collective
intelligence in Applied Energy.
References
[1] Hung DQ, Mithulananthan N, Bansal RC. Integration of PV and BES units in
commercial distribution systems considering energy loss and voltage stability.
Appl Energy 2014;113:1162–70.
[2] Lopes JAP, Hatziargyriou N, Mutale J. Integrating distributed generation into
electric power systems: a review of drivers, challenges and opportunities.
Electric Power Syst Res 2007;77:1189–203.
[3] Conti S, Nicolosi R, Rizzo SA, et al. Optimal dispatching of distributed
generators and storage systems for MV islanded microgrids. IEEE Trans
Power Delivery 2012;27(3):1243–51.
[4] Rueda-Medina A, Padilha-Feltrin A. Distributed generators as providers of
reactive power support-a market approach. IEEE Trans Power Syst 2013;28
(1):490–502.
[5] Kuroda K, Magori H, Ichimura T, et al. A hybrid multi-objective optimization
method considering optimization problems in power distribution systems. J
Modern Power Syst Clean Energy 2015;3(1):41–50.
[6] Verzijlbergh R, Vries LD, Lukszo Z. Renewable energy sources and responsive
demand. Do we need congestion management in the distribution grid. IEEE
Trans Power Syst 2014;29(5):2170–8.
[7] Morvaj B, Evins R, Carmeliet J. Optimization framework for distributed energy
systems with integrated electrical grid constraints. Appl Energy
2016;171:296–313.
[8] Huang S, Wu Q, Oren S, et al. Distribution locational marginal pricing through
quadratic programming for congestion management in distribution networks.
IEEE Trans Power Syst 2015;30(4):2119–28.
[9] Long C, Wu J, Thomas L, et al. Optimal operation of soft open points in medium
voltage electrical distribution networks with distributed generation. Appl
Energy 2016;184:427–37.
[10] Bloemink JM, Green TC. Increasing distributed generation penetration using
soft normally-open points. 2010 IEEE PES General Meeting, vol. 1. p. 1–8.
[11] Bloemink JM, Green TC. Benefits of distribution-level power electronics for
supporting distributed generation growth. IEEE Trans Power Delivery 2013;28
(2):911–9.
[12] Cao W, Wu J, Jenkins N. Feeder load balancing in MV distribution networks
using soft normally-open points. Proceedings of the 2014 innovative smart
grid technologies conference Europe, vol. 1. p. 1–6.
 C. Wang et al. / Applied Energy 189 (2017) 301–309
[13] Bloemink JM, Green TC. Increasing photovoltaic penetration with local energy
storage and soft normally-open points. 2011 IEEE PES general meeting, vol. 1.
p. 1–8.
[14] Cao W, Wu J, Jenkins N, et al. Operating principle of Soft Open Points for
electrical distribution network operation. Appl Energy 2016;164:245–57.
[15] Cao W, Wu J, Jenkins N, et al. Benefits analysis of Soft Open Points for electrical
distribution network operation. Appl Energy 2016;165:36–7.
[16] Huang X, Yang Y, Taylor GA. Service restoration of distribution systems
under distributed generation scenarios. CSEE J Power Energy Syst 2016;2
(3):43–50.
[17] Martins VF, Borges CL. Active distribution network integrated planning
incorporating distributed generation and load response uncertainties. IEEE
Trans Power Syst 2011;26:2164–72.
[18] Nick M, Cherkaoui R, Paolone M. Optimal allocation of dispersed energy
storage systems in active distribution networks for energy balance and grid
support. IEEE Trans Power Syst 2014;29:2300–10.
[19] Hochreiter R, Pflug GC. Financial scenario generation for stochastic multi-stage
decision processes as facility location problems. Ann Oper Res 2007;152
(1):257–72.
[20] Zou K, Agalgaonkar AP, Muttaqi KM, et al. Distribution system planning with
incorporating dg reactive capability and system uncertainties. IEEE Trans Sust
Energy 2012;3(1):112–23.
[21] Piccoli B, Rossi F. Generalized Wasserstein distance and its application to
transport equations with source. Arch Ration Mech Anal 2014;211:
1335–58.
[22] Shu Z, Li Q, Chan P. Investigation of offshore wind energy potential in Hong
Kong based on Weibull distribution function. Appl Energy 2015;156:362–73.
309
[23] Karaki SH, Chedid RB, Ramadan R. Probabilistic performance assessment of
autonomous solar-wind energy conversion systems. IEEE Trans Energy
Convers 1999;14(3):766–72.
[24] Flourentzou N, Agelidis VG, Demetriades GD. VSC-based HVDC power
transmission systems: an overview. IEEE Trans Power Electron
2009;24:592–602.
[25] Huang C, Li F, Ding T, et al. Second-order cone programming-based optimal
control strategy for wind energy conversion systems over complete operating
regions. IEEE Trans Sust Energy 2015;6(1):263–71.
[26] Low SH. Convex relaxation of optimal power flow-Part I: formulations and
relaxations. IEEE Trans Control Network Syst 2014;1:15–27.
[27] Farivar M, Low SH. Branch flow model: relaxations and convexification–parts I.
IEEE Trans Power Syst 2013;28:2554–64.
[28] Farivar M, Low SH. Branch flow model: relaxations and convexification–parts
II. IEEE Trans Power Syst 2013;28:2565–72.
[29] Lofberg J. Yalmip: a toolbox for modeling and optimization in MATLAB. IEEE
Int Sympos Comput Aided Control Syst Design 2004;1:284–9.
[30] Baran ME, Wu F. Network reconfiguration in distribution systems for loss
reduction and load balancing. IEEE Trans Power Delivery 1989;4(2):1401–7.
[31] Mostafa N, Rachid C. P Mario. optimal allocation of dispersed energy storage
systems in active distribution networks for energy balance and grid support.
IEEE Trans Power Syst 2014;29(5):2300–10.
[32] Su C, Lee C. Network reconfiguration of distribution systems using improved
mixed-integer hybrid differential evolution. IEEE Trans Power Delivery
2003;18:1022–7.
[33] Bonami P, Biegler LT, Conn AR. An algorithmic framework for convex mixed
integer nonlinear programs. Discrete Optim 2008;5:186–204.