Energy 143 (2018) 323e336

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

CVaR risk-based optimization framework for renewable energy

management in distribution systems with DGs and EVs
Jiekang Wu a, *, Zhijiang Wu b, a, Fan Wu c, Huiling Tang a, Xiaoming Mao a
a
School of Automation, Guangdong University of Technology, China
b
South China National Centre of Metrology, China
c
Guangxi Bo Yang Electric Power Survey and Design Co, Ltd, Guangxi Power Grid Co, Ltd., China

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

Article history: A method based on chance constrained second-order cone programming (CCSOCP) is presented for the
Received 28 February 2017 optimal risk value control of power loss in distribution systems with the distributed generation (DG) of
Received in revised form renewable energy systems and electric vehicles (EVs). The charging power of the EV is seen as a random
15 October 2017
variable, and the risk value of the power loss e due to the uncertainties in the power output of
Accepted 18 October 2017
Available online 1 November 2017
distributed generation of renewable energy systems and charging power of electric vehicles e is studied.
Moreover, a second-order cone programming based method is also presented to constrain the potential
risk of power loss to an acceptable range by optimally coordinating the power output of DG and the EV
Keywords:
Distribution systems
charging power in a distribution system. A conditional value at risk (CVaR) model for the power loss of
Risk value control of power loss distribution systems is presented and CVaR is taken as a constraint to control the risk value of power loss
Distributed generation (DG) due to uncertainties in DG and EV charging. The results of a test on a 69-node system are used to verify
Electric vehicles (EV) the validity of the risk control method proposed in this paper.
Chance constrained second-order cone © 2017 Elsevier Ltd. All rights reserved.
programming (CCSOCP)

1. Introduction is a kind of disturbance, which causes the network to operate in the

abnormal state. Moreover, it is a potential risk, resulting in
The large application of renewable energy, energy storage and cascading failures and outage accidents. Power flow transfer in a
electric vehicle (EV) is a development trend of distribution net- power grid is a kind of disturbance, which often leads to the
works. At the user sides, renewable energy, energy storage and cascading tripping of certain components in the power grid, and is
electric vehicles are typically used in a distributed way, and this will also a potential risk of power outages. Changes in load, regulation of
have a great influence on the power supply capability, voltage operation mode of power grids, reactive power and voltage coor-
quality and power loss of the distribution networks. Generally, the dinative control, will cause a change or power flow shift, which may
output or input power of a great number of distributed generation result in the overload of other components of the grid.
(DG) systems, distributed storage (DS) systems and electric vehicles It is a significant trend for DG systems and EVs to penetrate in
at different nodes in different time periods have great uncertainties distribution systems with even accelerated growth. This trend will
and randomness, the problem the power supply capability, voltage add great difficulty to the coordination of DG and EV charging,
quality and power loss becomes more prominent, more serious and which may result in frequent overloads and higher power losses in
more difficult to control with great risk for the operation of the distribution systems. It is therefore highly essential to investigate a
distribution networks. technology for the effective coordination of DG and EV charging.
The normal operation of a large power grid is the key to Enhancing the security, reliability and efficiency of power sys-
ensuring the continuous and reliable supply of electric power. tems with uncertainties, is an important working task [1e3].
However, the existence of various disturbances greatly increases Moreover, it requires the carrying out of necessary risk and security
the potential risk of the normal operation of the power grid. The assessments [4e7], which includes the deterministic evaluation [8],
failure of the components, such as overhead lines and transformers, probabilistic evaluation [9], and risk assessment [10e13]. The
deterministic assessment generally gives a set of the most credible
contingencies, which may result in highly conservative decisions
* Corresponding author. with high operation costs [14]. The probabilistic evaluation uses a
E-mail address: wujiekang@163.com (J. Wu).

https://doi.org/10.1016/j.energy.2017.10.083
0360-5442/© 2017 Elsevier Ltd. All rights reserved.
324 J. Wu et al. / Energy 143 (2018) 323e336

probability method to evaluate uncertain events in forms such as [40,41]. Moreover, the cone programming model for the optimal
loss of load probability and expected unserved energy. Risk reconfiguration of a distribution network is solved by taking the
assessment uses the risk value calculation method to evaluate the switch in the lines as variables [42], and the optimal power flow
impact of uncertain events and consequences [8,10e22], some of model based on second-order cone programming is established
which relate to the DG of renewable energy [23e26]. However, using an objective function for minimizing active power loss,
minimal work has been carried out on the impact on the risk of reactive power loss, and generation cost [49]. In reality, the oper-
power grids in events of uncertainties related to EV charging. ating conditions of the power networks are always changing, and
In a distribution system with a large number of DG systems and the above literature fails to consider the impact of various un-
EVs injected at different nodes, the optimal coordination problem certainties on the power grid.
of DG and EV charging has drawn much attention. Moreover, It is not possible for a distribution system with DG and EVs to
certain experts and scholars have carried out a thorough study on operate with a strict constraint condition of power flow. Therefore,
the optimal penetration in distribution systems [27] and the it can be relaxed for the operation of the power network, and it only
charging influence on distribution networks [28]. The use of smart needs to meet the operating conditions under a certain degree of
metering and demand side management in distribution systems confidence. For example, considering the uncertainties of wind
was studied for charging coordination [29e32], which may alter power, the maximal grid-connected active power is calculated
the charging influence on distribution systems. A method for based on a conditional constraint [50]. Using an objective function
minimizing the power loss of distribution systems by coordinating of minimizing the total cost and power loss of distribution systems
EV charging [33,34] is studied. Some scenarios for controlling the with DG and EVs, and considering the uncertainties of the DG, load
charging of EVs, based on minimizing charging costs [35], were and EVs; a model based on chance-constrained programming is
presented. These scenarios aimed at putting forward a one-day constructed [51]. In this paper, we modify the second-order cone
charging pattern. However, peak load and uncertainties in the programming model, which is introduced in the original cone
charging demand of EVs are not taken into account in these studies. programming model, to solve the uncertain problem by intro-
Moreover, the uncertainties of EV charging are significant, and their ducing the chance-constrained condition to second-order cone
influence on the power loss and nodal voltage is also significant. At programming.
the same time, the power output of DG in distribution systems is In this paper, the charging power of the EV is seen as a random
always uncertain, and its influence on the power loss and nodal variable. Moreover, the risk value of power loss due to the un-
voltage is also significant. The uncertainties of EV charging depend certainties in the power output of the DG of renewable energy
on factors such as the battery characteristics, SOC levels, driving systems and charging power of EVs, is studied. Furthermore, a
distance, arrival times, departure times, and charger ratings. second-order cone programming based method is also presented to
The risk due to the power loss of a distribution network is constrain the potential risk of power loss to an acceptable range, by
described by the potential loss caused by the uncertainties, such as optimally coordinating the power output of DG and EV charging
the power flow change and shift in certain components, which may power in a distribution system. The proposed method can greatly
result in the increase in power loss in many cases. If the disturbance decrease the potential risk value caused by the large disturbance
is large, it will lead to a significant increase in power loss. In order to generated by EVs.
evaluate the potential consequences caused by the disturbance,
some methods for risk measurement are used to characterize the 2. Power loss of distributed generation with DGs and EVs
loss based on value at risk (VaR).
Value at risk is a popular risk measurement technique [36]- [38]. In this paper, it is assumed that a distribution system is installed
The VaR based method is not only widely used in the financial field, with the DG of renewable energy systems and charging systems for
but also in the field of power market, power generation cost, and so EVs at different nodes, as shown in Fig. 1. It is also assumed that the
on. Furthermore, the conditional value at risk (CVaR) based method DG of renewable energy systems includes a DG system and an
is used to set up a decision making model for purchasing electricity energy storage system. A DG system is based on wind power, solar
in the real-time market, long-term contract market and day-ahead energy, etc. The output power of the DG of a renewable energy
market [36]. Considering the randomness of wind power genera- system is dependent on the output power of the DG system, the
tion, the optimal limit capacity of the wind power grid-connected storage power of the energy storage system, and the load power at
to distribution systems is calculated using the CVaR based the local node. The renewable energy is fully utilized either by the
method [37e42]. CVaR risk based approach is used for dynamical local load or by the local storage, and the surplus power will be
scheduling optimization model for virtual power plant connected stored when the power of the local load is smaller than that
with wind-photovoltaic-energy storage system with uncertainties generated from renewable energy. With this assumption, control-
and demand response [43], robust coordinated reserve scheduling ling the renewable energy means controlling the power output of
[44], stochastic unit commitment in isolated systems with renew- renewable energy systems, and controlling the power output of
able penetration [45], agricultural water management [46], multi- renewable energy systems means the controlled power of the
product selective newsvendor problem [47], hedging problem in renewable energy is used for local storage. Controlling the renew-
incomplete markets [48]. able energy aims at minimizing system loss and the risk value of
The second-order cone programming is a special convex pro- power loss. In Fig. 1, it is assumed that Zij is the impedance between
gramming model [49e51], which is one of the important optimi- node i and j, SLossij is the power loss between node i and j due to
zation methods in mathematics, and has been applied to power impedance Zij .
systems [52]. Many scholars have applied the cone programming to
In the distribution system, there are three cases of flow-in po-
the power flow calculation of power systems. A great of nonlinear 0 0
wer, and four cases of flow-out power. Furthermore,SNi or SNj is a
constraint condition is transformed into linear in power flow
calculation, and the relationship between the cone sets of variables flow-in power injected from the main power grid; SDGi or SDGj is a
is presented so that calculation speed can be improved. For flow-in power injected by DG systems at node i or j; SBi or SBj is a
example, a power flow calculation model is proposed based on flow-in power injected by the DG systems in the sub-branch con-
conic programming [39]. A cone programming model for DG is nected to node i or j when the power of the distributed generation
obtained by minimizing the power loss as the objective function systems is greater than the power of the local load in the sub-
 J. Wu et al. / Energy 143 (2018) 323e336 325

Fig. 1. A simplified 2-node distribution system with distributed generation and electric vehicles.

branch or a flow-out power outpouring of node i or j when the

power of the distributed generation systems is smaller than the X  
power of the local load in the sub-branch. Moreover,SDi or SDj is a Gjj Vj2  Vj Vk Gjk cos qjk þ Bjk sinqjk
k2UMj
flow-out power outpoured by the local general load at node i or j, X   X 
which is uncontrollable at any time;SEVi or SEVj is a flow-out power  Vj Vk Gjk cos qjk þ Bjk sinqjk þ Vj Vk Gjk cos qjk
outpoured by the changing load of the EVs, which is controllable at k2UBj k2UDj
00 00   
any time; SNi or SNj is a load power flowing in the main branch,
þ Bjk sinqjk  Vi Vj Gij cos qij þ Bij sinqij
which depends on the natural physical power flow or network
impedance (including generator and load), and its flowing direction ¼ PDGj  PDj  PEVj
is generally from the side of the main power grid to the side of the (3)
local load in a radial distribution network.
The power flow equation for node i and node j can be respec-
X  
tively expressed as follows: Bjj Vj2  Vj Vk Gjk sinqjk  Bjk cos qjk
k2UMj
X   X 
 Vj Vk Gjk sinqjk  Bjk cos qjk þ Vj Vk Gjk sinqjk
X
Gii Vi2  Vi Vk ðGik cos qik þ Bik sinqik Þ k2UBj k2UDj
  
k2UMi
X X  Bjk cos qjk  Vi Vj Gij sinqij  Bij cos qij
 Vi Vk ðGik cos qik þ Bik sinqik Þ  Vi Vk ðGik cos qik
k2UBi k2UBi
¼ QDGj  QDj  QEVj
  (4)
þ Bik sinqik Þ þ Vi Vj Gij cos qij þ Bij sinqij
¼ PDGi  PDi  PEVi 0 0 0 0
where PNi and QNi , and PNj and QNj are the active and reactive
(1) powers of the main branch between node i at the power supply side
and node j at the load side, respectively; PDGi and QDGi , andPDGj and
QDGj are the active and reactive output powers of the DG systems
connected to node i and node j; PBi and QBi , andPBj and QBj are the
X
Bii Vi2  Vi Vk ðGik sinqik  Bik cos qik Þ active and reactive powers of the sub-branch connected to node i
k2UMi and node j, respectively;PDi and QDi , andPDj and QDj are the active
X X
 Vi Vk ðGik sinqik  Bik cos qik Þ þ Vi Vk ðGik sinqik and reactive powers of the local load at node i and node j, respec-
k2UBi k2UDi tively; PEVi and QEVi , andPEVj and QEVj are the active and reactive
  powers of the changing load of EVs at node i and node j, respec-
 Bik cos qik Þ þ Vi Vj Gij sinqij  Bij cos qij
tively; Vi ,Vj and Vk are the voltage of node i, node jand nodek,
¼ QDGi  QDi  QEVi respectively; Gii and Bii , Gjj and Bjj are the respectively self
(2) conductance and self susceptance at node i and PL0 , respectively;
P DG
CVaRaloss ðPDG Þ and PL0 are the mutual conductance and mutual
326 J. Wu et al. / Energy 143 (2018) 323e336

P DG P DG Furthermore,VaRa is defined as the risk value that is less than Qa

susceptance between node CVaRaloss ðPDG Þ and CVaRaloss ðPDG Þ  PL0 ,
in these cases, in which the probability of the uncertain event is a.
respectively; Gik and Bik are the mutual conductance and mutual
The consensus risk measure rðXÞmay be expressed as:
susceptance between node i and k, respectively; Gjk and Bjk are the
mutual conductance and mutual susceptance between node jand k, Z1
respectively; qij is the voltage phase between voltage Vi at node i 1  
rðXÞ ¼ CVaRa ðXÞ ¼ VaRg X dg (7)
and voltage Vj at node j,qik is the voltage phase between voltage Vi 1a
a
at node i and voltage Vk at node k, qjk is the voltage phase between
Generally, CVaR is defined as the average value of the excess of
voltage Vj at node j and voltage Vk at node k; UMi , UMj make up a
VaR, and CVaR can be mathematically expressed as:
nodal set of power flowing in node i and node j from the side of an
infinite supply system beyond the radial distribution system, Z1
respectively; UBi , UBj is a nodal set of power flowing in node i and 1  
rðXÞ ¼ CVaRa ðXÞ ¼ VaRg X dg
node j, respectively, from the DG systems installed in the sub- 1a
a
branch; UDi , UDj make up a nodal set of power flowing out of Z
1
node i and node j for supplying the load at node i and node j, ¼ f ðx; yÞpðyÞdy (8)
1a
respectively; and Lij is the main branch. f ðx;yÞb 计算VaR 还有 CVaR本来就很难
Using power flow computation, the amplitude and phase of the
voltage of a node in the distribution system may be determined, As for the general distribution function, it is always very difficult
and then the active and reactive power loss of any branch between to calculate the value of VaR and CVaR. Many experts and scholars
nodes iand j is respectively computed using [52]: have done a lot of research in this field, and have made much
progress. An effective method for calculating VaR and CVaR is
X
n X
n   proposed by constructing a special function [53e58]. In the pro-
Ploss ¼ Gij Vi2 þ Vj2  2Vi Vj cos qij (5) posed method, a special function Fa ðx; bÞ[55e58] is first
i¼1 j¼1
formulated:
n o
Fa ðx; bÞ ¼ b þ ð1  aÞ1 E ½f ðx; yÞpðyÞdyþ (9)
3. Risk value due to power loss
The VaR and CVaR can be effectively calculated by minimizing
The risk value of power loss is due to uncertainties in the EV the value of the proposed special function Fa ðx; bÞ:
charging and DG of renewable energy systems. A conditional VaR
VaRa ðXÞ ¼ ba ðf ðx; yÞÞ ¼ argminðbÞ ðFa ðx; bÞÞ (10)
model for the power loss of distribution systems is presented in the
following section, using the probability model of the charging po-
wer of EVs and output power of the DGs of renewable energy CVaRa ðXÞ ¼ minðFa ðx; bÞÞ (11)
systems.
In equation (10), the arg minðbÞ ðFa ðx; bÞÞ means that the value of

3.1. CVaR calculation b is obtained by minimizing the value of the proposed special
function.
VaR is a risk assessment method that is widely used in the In the calculation, the CVaRa value of the uncertain event is
current risk management, and the principle of VaR is that it uses the calculated by minimizing the function value of Fa ðx; bÞ, and then the
quantile of the loss distribution to evaluate the risk. Furthermore, corresponding value of b is the VaRa value of the uncertain event
because of this simple principle, i.e., being easy to use and easy to when minimizing the function value of Fa ðx; bÞ.
implement, VaR is widely used for the risk assessment of power
grids, and may also be used to estimate the maximum possible loss 3.2. Calculation method for risk value due to power loss
of a decision or grid configuration in a particular time period.
However, it ignores the thick tail incident, which makes the risk It is supposed that pDG ¼ fp1 ; p2 ; …pk g is a set of active output
assessment to the ideal. Moreover, because VaR does not satisfy the powers of DG, made up of k renewable generation systems.
second additivity, it is not a perfect tool for risk measuring. CVaR Moreover, pEV ¼ fpEV1 ; pEV2 ; …pEVn gis a set of active charging
was proposed after making some improvements to the traditional powers, made up of n electric vehicles. Taking the active output
VaR method. Conditional value at risk is used to evaluate the part of powers of renewable energy systems in different energy stations as
the loss distribution that is in excess of VaR; therefore, CVaR can be decision variables, and taking the charging powers of EVs as the
more fully considered as the risk loss beyond a certain limit suf- random variable set y ¼ pEV , then a loss function f ðPDG ; yÞ made up
fered by the event, and can be used to effectively evaluate the po- of decision variables and random variables is constructed:
tential risk of the event, which makes up the deficiency of the VaR
method for the tail estimation of risk. Moreover, CVaR is also widely X
n X
n  
used in various risk assessments, because it is the consistent risk. f ðPDG ; yÞ ¼ Gij Vi2 þ Vj2  2Vi Vj cos qij (12)
Given thatxis a strategy for the investment portfolio, f ðx; yÞ is a i¼1 j¼1

loss function that is represented by a decision variable x and a If the confidence level is a, a special function Fa ðpDG ; bÞ can be
random variable y with a probability functionpðyÞ, the probability proposed as follows:
of f ðx; yÞ that does not exceed a given limit of b may then be
Z
expressed in the following form: þ
Fa ðPDG ; bÞ ¼ b þ ð1  aÞ1 ½f ðPDG ; yÞ  b pðyÞdy (13)
Z
y2R
jðx; aÞ ¼ pðyÞdy (6)
f ðx;yÞb Using the proposed special function, CVaRa and VaRa may be
calculated.
 J. Wu et al. / Energy 143 (2018) 323e336 327

Generally, the analytic expression of the probability density- as in equation (19). Given a risk limit ofbfor power loss, the con-
pðyÞof the random variable y is more complex, which makes the ditional risk value due to power loss is defined as a function of the
integral operation very difficult. Therefore, the historical data of yis output power of the DG of the renewable energy system and
used, or a number of sampling data obtained by using the Monte random variables in a confidence level a, and is formulated as
Carlo simulation method are used for solving the estimated values equation (20).
of equation (13).
If random variables are obtained from historical data or by using P DG
CVaRaloss ðPDG Þ  PL0 (19)
the Monte Carlo simulation method to obtain a data set such
that:y1 ; y2 ; :::; yl (lis the number of random variables), then the
proposed special function may become: P DG X
l
CVaRaloss ðPDG Þ ¼ b þ ½lð1  aÞ1 Zi (20)
X
l i¼1
F~a ðPDG ; bÞ ¼ b þ ð1  aÞ1
þ
½f ðPDG ; yÞ  b (14)
i¼1
Zi  ½f ðPDG ; yÞ  bpðyi Þ (21)
Based on equation (14), the risk value of CVaR and VaR due to
the power loss in distribution systems can be obtained, respec- Zi  0 (22)
tively, by the following two equations:
  where PL0 is an acceptable loss value according to an expectation of
CVaRa ðPDG Þ ¼ miny2R F~a ðPDG ; bÞ (15) distribution systems under a certain operation state, which is set as
a value of 7% in this paper.
 
VaRa ðPDG Þ ¼ arg minðbÞ F~a ðPDG ; bÞ (16) 2) Equal constraint condition

These calculations need to satisfy equations (1)e(4) and the

Power flow as an equal constraint condition must satisfy such
following condition:
power flow equations as those in (1)e(4).
0  PDGi  P DGi (17)
3) Nodal voltage constraints

where P DGi is the maximum limit of the output power of the DG of

As for each node in the distribution system, its voltage value
the ith renewable energy system.
must be not greater than its maximal limit, and not less than its
minimal limit:
4. Risk value control model of power loss based on CCSOCP
V i  Vi  V i (23)
In distribution systems with DGs and EVs, the risk value of po-
wer loss is controlled by optimally coordinating the charging power
where Vi , V i and V i are respectively the actual value, the maximal
of EVs and the output power of the DG of renewable energy sys-
limit and the minimal limit of nodal voltage at node i, respectively.
tems. A control method based on chance constrained second-order
cone programming (CCSOCP) is presented in the following section.
4) Branch power constraints
4.1. General model for risk value control of power loss
The branch power transferred from one node to another must be
not greater than its maximal limit:
In order to achieve risk control in the acceptable range, and to
make the power grid running in the best economic benefits, an
optimal model is presented in this paper. In the proposed optimal SLij  SLij (24)
model, the risk of power loss in the distribution system is
controlled to achieve optimal coordinative control of the operating where SLij and SLij are the actual value and maximal limit of the
efficiency and operational risk, by using an objective function to branch power transferred from nodes i to j, respectively.
minimize the power loss, and by a constraint of risk value in CVaR. The transferred power in branch Lij depends on its voltage and
current:
4.1.1. Objective function pffiffiffi
The objective function to minimize the power loss of the dis- SLij ¼ 3  Vi  Iij (25)
tribution system with DGs and EVs:
where Iij is the current value of branch Lij from nodes i to node j.
min PLoss (18)
ðPDG Þ Replacing electric power with electric current, Equation (24)
becomes:
   2
Iij2 G2ij þ B2ij Vi2 þ Vj2  2Vi Vj cos qij  Iij (26)
4.1.2. Constraint conditions

1) Power loss constraints where I ij is the maximal limit current of branch Lij .

First, the conditional risk value due to power loss needs to be 5) Output power constraints for DGs
limited to an expected value, that is to say, the conditional risk
value needs not to be greater than its permitted maximum value, The output power of the DG of renewable energy systems must
and it may be formulated in such an unequal constraint condition satisfy constraint condition (17).
328 J. Wu et al. / Energy 143 (2018) 323e336

4.2. CCSOCP model for risk value control of power loss

2
Vimin 2
Vimax
The second-order cone programming is to minimize or maxi- pffiffiffi  Xi  p ffiffiffi (37)
2 2
mize a linear function on the intersection of the finite number of
two cone Descartes product of an affine subspace, and some  pffiffiffi pffiffiffi  2
problems can be transformed into second-order cone programming Iij2 ¼ G2ij þ B2ij 2Xi þ 2Xj  2Yij  I ij (38)
solutions. By transforming the complex optimization model into
the cone model, the complex relationships between variables may
X
n X
n pffiffiffi pffiffiffi 
be represented by a cone of special structure. Moreover, this
Zi  Gij 2Xi þ 2Xj  2Yij  b (39)
method can greatly simplify the solution of the original model,
i¼1 j¼1
increase the calculation speed and enhance its convergence.
The cone programming based model is generally expressed as 2
follows: where Ui ¼ UMi þ UBi þ UDi ; Uj ¼ UMj þ UBj þ UDj ; . I ij is the
maximal limit current of branch Lij .
 
min cT xjAx ¼ b; x_ k 0 (27) According to the requirement of the standard form of cone
optimization, the relationship between the variables must satisfy
the cone requirement, and after the substitution with the variables
where x is a multidimensional variable,cT and bare constant
of cone optimization in Equations (29)e(31), the new variables
vectors,Ais constant matrix,kis the Descartes product of a finite non-
must satisfy the following condition:
empty convex cone, and can be expressed with a rotating cone:
( )
X
m 2Xi Xj ¼ Yij2 þ Zij2 (40)
m
k¼ x2R : 2x1 x2  x2i ; x1 ; x2 0 (28)
i¼3 Equation (40) is adapted to meet the form of a rotating cone
condition:
where x_ k 0indicates that the variable x belongs to the cone set k.
Furthermore, because the above optimization model is not the 2Xi Xj  Yij2 þ Zij2 (41)
standard form of the cone programming model, it needs to be
transformed into the following form [51]: The output power of the DG of a renewable energy system as a
variable also needs to be transformed into the cone set form:
V2 Vj2
Xi ¼ piffiffiffi or Xj ¼ pffiffiffi (29) 2 2 2
2 2 SDGi  PDGi þ QDGi (42)

Yij ¼ Vi Vj cos qij or Yik ¼ Vi Vk cos qik (30) where SDGi is the maximal power limits of the DG of renewable
energy system i.
Zij ¼ Vi Vj sin qij or Zik ¼ Vi Vk sin qik (31) The output power of the DG of a renewable energy system needs
to meet the following conditions:
Replacing variables X,Yand Zin the original optimization prob-
lem with the variables of cone optimization, a second-order cone S2DGi  PDGi
2 2
þ QDGi (43)
programming model for the original optimization problem is ob-
tained with an objective function of minimizing the power loss of where SDGi is the minimal power limits of the DG of renewable
the distribution system with DGs and EVs: energy system i.
Considering the uncertainties of load fluctuation and EV
X
n X
n pffiffiffi pffiffiffi 
min DG
PLoss ¼ min Gij 2Xi þ 2Xj  2Yij (32) charging, it is not realistic to keep the grid at the normal level at
ðPDG ;X;Y;ZÞ every moment, which requires a higher control cost and much
i¼1 j¼1
more advanced control technology. Therefore, the constraint con-
pffiffiffi X ditions may be relaxed. In other words, it is accepted as long as the
2Gii Xi þ Gij Yij þ Bij Zij þ Gik Yik þ Bik Zik ¼ PDGi  PDi  PEVi constraint conditions, which related to the equal constraint con-
k2Ui
ditions (1)e(4), nodal voltage constraints in (23), branch power
(33) constraints in (24), output power constraints for DGs in (17), and
power loss constraints in (19) and (20), are satisfied under the
pffiffiffi X
 2Bii Xi þ Gij Zij  Bij Yij þ Gik Zik  Bik Yik confidence level. After the verification in this paper, the distribu-
k2Ui tion system operating under the chance constraint requirement
may be sufficient to supply the user in a normal way and to reduce
¼ QDGi  QDi  QEVi (34)
the power supply costs of the distribution system.
pffiffiffi X A general model for chance constraint optimization may be
2Gjj Xj  Gij Yij  Bij Zij þ Gjk Yjk þ Bjk Zjk ¼ PDGj  PDj  PEVj expressed as:
k2Uj

(35) Prðf ðxÞ  aÞ  a (44)

pffiffiffi X It indicates that the probability of meeting the constraint f ðxÞ 

 2Bjj Xj  Gij Zij þ Bij Yij þ Gjk Zjk  Bjk Yjk a is greater than the a value, and the confidence level a is based on
k2Uj the security of the system, and the economic requirements are
¼ QDGj  QDj  QEVj (36) chosen.
As for conditions (37)e(38), a chance constraint model may be
constructed:
 J. Wu et al. / Energy 143 (2018) 323e336 329

! accommodated simultaneously, the nodes installed with distrib-

2
Vimin 2
Vimin uted generation systems are selected to increase generation, and
Pr p ffiffiffi  Xi  p ffiffiffi  aV (45)
2 2 the nodes with heavy load power are selected to reduce load. With
continuous increase of the output power of distributed generation
h pffiffiffi pffiffiffi  2
i systems and continuous decreases of the load power at the nodes
Pr G2ij þ B2ij 2Xi þ 2Xj  2Yij  I ij  aI (46) with heavy load power, the value of the nodal voltage, branch
P DG
current/power, PL0 and CVaRaloss ðPDG Þ and so on are constantly
where aV and aI is an acceptable probability range respectively for updated, and the chance constraint conditions for nodal voltage
the changes of the nodal voltage and the branch current. and branch power are checked using equation (48), and the value of
P DG
PL0 and CVaRaloss ðPDG Þ are compared using equation (19) in the al-
5. Solving method and procedures
P DG
gorithm. When CVaRaloss ðPDG Þ  PL0 , there is no need to increase
There are many methods that can used to solve cone optimi- the output power of the distributed generation systems or reduce
zation problems. Software, such as MOSEK and MATLAB can be the load power at the nodes with heavy load power. When the
used to solve the cone optimization problem. Some methods, such constraint conditions for nodal voltage and branch power, and
as the interior point method, primal dual interior point method, CVaR constraints are satisfied, the optimization process will
genetic algorithm and particle swarm optimization algorithm, can converge.
also be used to solve the cone optimization problem. Furthermore, It can only increase the output power of the distributed gener-
because of the complexity of the model, the proposed optimization ation systems, or only reduce the load power at the nodes with
problem is transformed into a simple linear or cone set in this heavy load power. When only increasing the output power of
paper, thus making the optimization process more efficient. distributed generation systems, if the chance constraints and CVaR
cannot still be accommodated simultaneously, it can further reduce
5.1. Processing method of chance constraints the load power at the nodes with heavy load power. Or when only
decreasing the load power at the nodes with heavy load power, if
The establishment of the chance-constrained conditions can be the chance constraints and CVaR cannot still be accommodated
verified using the Monte Carlo simulation method. Moreover,N sets simultaneously, it can further increase the output power of
of random data for the charging power of EVs are generated ac- distributed generation systems.
cording to a given charging probability of the EVs. Among Ngroups
of random data, there are Nr data satisfying all the constraint
conditions. In this paper, Nr =Nindicates the probability satisfying all
5.4. Solving steps
the constraint conditions. When the number N of simulations is
very large, Nr =Ntends to be stable and close to the real value. If Nr =N
The steps for solving the risk value control problem of power
is greater than the given confidence level, it is considered that the
loss in distribution systems with DGs and EVs by the optimal co-
chance constraint is established.
ordination of the output power of the DG of renewable energy
systems and charging power of EVs, using the CCSOCP method, are
5.2. Processing method for CVaR constraints
presented as follows:

For an electric vehicle, the CVaR risk value is calculated using

a) EV charging power data initialization: simulate Ngroups of
groups of random data, and is verified using the Monte Carlo
data for the initial active and reactive powers of electric
simulation method. Moreover, the CVaR risk constraint conditions
vehicles.
can be tested using the Monte Carlo simulation method. Consid-
b) Search for optimal power of DGs: search for optimal active
ering the uncertainties of EV charging, the CVaR risk value is
and reactive powers of the DG of renewable energy systems,
calculated according to formula (15), and a group of random data
using an objective function minimizing the power loss of the
generated by the probability density function. If the CVaR risk
distribution system with DGs and EVs.
constraint conditions are satisfied, it is considered that the CVaR
c) Test for optimality: test for the optimal power of the DGs of
risk constraint is established.
renewable energy systems, using CVaR and chance
constraints.
5.3. Processing method for chance constraints d) Monte Carlo simulation for EV charging power: simulate
Ngroup data for active and reactive power of electric
The changes of the nodal voltage and the branch current may be vehicles.
seen as stochastic variables, the chance constraints for these sto- e) Test for chance constraints and CVaR risk constraints: test for
chastic variables are formulated as: chance constraints and CVaR risk constraints, using optimal
output power of the DGs of renewable energy systems and
Prðx  xÞ  ax (47)
optimal EV charging power.
f) Simultaneous test: test whether two test conditions of
where x, x and ax is respectively stochastic variable, limited upper
chance constraints and CVaR risk constraints are simulta-
value of stochastic variable and acceptable probability range for
neously satisfied.
stochastic variable.
g) If the test conditions are simultaneously satisfied, the
Supposed that these stochastic variable follows normal distri-
optimal output power of the DGs of renewable energy sys-
bution with a mean of mand a variance of s, then the chance con-
tems and EV charging power are determined.
straints for these stochastic variables are transformed as:
h) If the test conditions are not simultaneously satisfied, go to
pffiffiffi step b).
m þ sF1 ð1  ax Þ  x  0 (48)
When the chance constraints and CVaR cannot be The solution steps are shown in Fig. 2.
330 J. Wu et al. / Energy 143 (2018) 323e336

1.2
EV charging power data
initialization
1

Search for Optimal power of DGs

0.8

load power/p.u
0.6
Test for optimality

0.4

Monte Carlo simulation for EV

charging power 0.2

0
No 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
Test for chance Test for CVaR risk hour/h
constraint constraint
Fig. 4. Load curve in a day.

constraint in the form of CVaR is 95%. The risk constraint limits pL0
are set to the power loss values when the power loss rate of the
distribution system with EV charging is 7%. The confidence degree a
Simultaneously of the chance constraint is 95%.
satisfy? A load curve is shown in Fig. 4, and the charging power of EVs
and power output of the DG of renewable energy systems at
6:00e8:00, 14:00e16:00, and 20:00e22:00, are analysed. The
average loads are 3.020 þ j1.870 MVA, 3.455 þ j2.131 MVA, and
Yes 4.3622 þ j2.734 MVA, respectively, for the three time periods. In
this network, 200 EVs are considered, each of which is charged with
Output a charging power of 5 kW, and the total charging power of 200 EVs
in a day is shown in Fig. 5. It is assumed that these 200 EVs are
Fig. 2. Solving steps and solution flow chart. randomly switched to nodes 9, 26, 32, 38, 48 and 64 of the distri-
bution system, and the charging power of these EVs is controlled to
minimize the risk value due to power loss in the distribution sys-
6. Calculation and analysis for study case tems with DG and EVs.
In the following section, the output power coordination of the
A 69-node distribution system is used as a study example, as DG of renewable energy systems installed in different nodes in the
shown in Fig. 3. distribution system, is studied for two cases. The first case is that of
The average load power is 3.802 þ j2.694 MVA, the minimal load a 69-node distribution system only with EVs, and the other is that
power is 2.627 þ j1.57 MVA, and the maximal load power is of a 69-node distribution system with DGs and EVs.
4.684 þ 2.807 MVA. There is a charging station of EVs at nodes 9,
26, 32, 38, 48, and 64. A distributed generation system respectively
6.1. Study case for a 69-nodes distribution system only with EVs
with solar, wind is located at node 27, 50 and 54. The total active
power is 2.4 MW, and the rated capacity of each distributed
Table 1 shows a simulation result of the charging power of EVs
generator is 0.8 MW. The confidence level aof the power loss risk
in the above three time periods. The charging power of the EVs is

          

     

                           

   
            

       

Fig. 3. A distribution system with 69 nodes.

 J. Wu et al. / Energy 143 (2018) 323e336 331

1800 Table 2
Risk value of power loss in the distribution system only with EVs under different
nodal voltage.
1600
Node No. 6:00e8:00 14:00e16:00 20:00e22:00
1400 amplitude/p.u amplitude/p.u amplitude/p.u

28 1.0030 0.9983 0.9599

charging power/P (kW)

1200 29 1.0030 0.9983 0.9599

30 1.0447 1.0447 1.0445
1000 40 1.0401 1.0396 1.0379
41 1.0397 1.0391 1.0372
42 1.0264 1.0216 1.0125
800
53 0.9723 0.9540 0.9258
54 0.9720 0.9536 0.9253
600 55 0.9705 0.9519 0.9231
56 0.9700 0.9514 0.9224
59 1.0165 1.0110 0.9937
400
60 1.0165 1.0110 0.9937
Ploss/MWh 0.1635 0.2192 0.4262
200 CVaR/MWh 0.1882 0.2509 0.4980
0 5 10 15 20 25 VaR/MWh 0.1796 0.2408 0.4747
hour/h PL0/MWh 0.2826 0.3256 0.4642

Fig. 5. Curve of the charging power of an EV in a day.

measures is to store the surplus power in the local energy storage

only 0.5 MW at the time period from 8:00e11:00, which is a peak system in some cases: 1) it is required to decrease the power output
time for an increased amount of travel using EVs. However, much of the renewable energy systems for preventing the voltage from
less EVs are going to charging stations for recharging in this time rising; 2) it requires a fixed power output of renewable energy
period. The number and the charging power of EVs increase with systems; 3) the power of the local load is smaller than the power of
time, and increase a little more from 12:00e14:00, and arrive at a the local DG. Moreover, another one of the controlling measures is
maximum charging power in the time period of 20:00e22:00. The to change the operation mode of the local energy storage system
time period of 20:00e22:00 is a peak charging time, in which the from charging to discharging, when it is required to increase the
number and the charging power of EVs are maximal. power output of the DG of renewable energy systems. The regu-
Table 2 and Fig. 6 show simulation results of the power flow of a lation results for the output power of DG systems are shown in
distribution system with only EVs. Table 3. Using the method based on CCSOCP proposed in this paper,
It is seen from Table 2 that EV charging has a great impact on the the optimal coordination model of the DG in distribution systems
distribution system in each time period, and the voltages of some with DGs and EVs, is built. Using the interior point method to solve
nodes in the distribution system exceed the maximum allowable the problem, the optimal coordinative output power of DG in dis-
value of the voltage offset. Moreover, the power loss may reach a tribution systems with the DGs and EVs is obtained, as shown in
very serious level, especially in the EV peak charging time when the Table 3.
peak power of the general loads is added to the distribution system Table 4 shows the sampling number ratios obtained from a
in this time period. The risk of power loss may further increase distribution system with a satisfied constraint of the nodal voltage
because of the uncertainties of EV charging. In the time period from and branch current, and maximal output power of distributed
20:00e22:00, the power loss of the distribution system is lower generation. In Table 4, NV is sampling number satisfying voltage
than the risk value of power loss, but the forecasting values of VaR requirement,NI is the sampling number satisfying current require-
and CVaR are 0.4747 MWh and 0.4980 MWh, respectively. The ment, andN is the total sampling number. It can be seen that the
potential risk of power loss exceeds the limit value of the risk chance constraints are satisfied with a confidence level of 95%,
constraint, and it needs to take measures to reduce the power loss. under such a condition that the output power of DG is optimal. As
shown in Table 3 and Table 4, the higher the confidence level, the
6.2. Study case for a 69-nodes distribution system with DGs and more stringent the voltage and current constraints, and the lower
EVs the DG output. For example, the optimal powers of the DG of the
renewable systems installed at node 50 are 0.90 MWh, 0.85 MWh,
In order to reduce the risk of EV charging, it is necessary to take and 0.75 MWh in the time period of 20:00e22:00.
some measures to control the power output of renewable energy The simulation results of the power flow of the distribution
systems with DGs and DSs at the load sides. One of the controlling system with DGs and EVs are shown in Table 5 and Fig. 7. There are

Table 1
Charging power of electric vehicles at different time period.

Node No. 8:00e10:00 12:00e14:00 20:00e22:00

active power/MW reactive power/Mvar active power/MW reactive power/Mvar active power/MW reactive power/Mvar

9 0.0850 0.0279 0.1150 0.0379 0.2750 0.0903

26 0.0800 0.0263 0.1050 0.0346 0.2650 0.0871
32 0.0950 0.0312 0.0950 0.0313 0.2800 0.0920
38 0.0750 0.0247 0.1100 0.0363 0.2750 0.0904
48 0.0800 0.0263 0.1000 0.0329 0.2700 0.0887
64 0.0850 0.0279 0.1100 0.0363 0.2800 0.0920
total 0.5000 0.1643 0.6350 0.2094 1.6450 0.5407
332 J. Wu et al. / Energy 143 (2018) 323e336

0.5
Power loss/MWh
0.45 VaR Value of power loss/MWh
CVaR Value of power loss/MWh
0.4 Risk value of power loss/MWh

0.35
Risk value/MWh

0.3

0.25

0.2

0.15

0.1

0.05

0
1 2 3
Time period("1":6:00-8:00,"1":14:00-16:00,"1":20:00-22:00)
Fig. 6. Comparison in risk value due to power loss of distribution systems only with EVs.

four cases that are used to calculate the risk value: 1) the cases only 0.0421 MWh, 0.0380 MWh, 0.2163 MWh; and 0.0424 MWh,
with EVs; 2) the cases with EVs and DGs; 3) the cases with EVs, DGs 0.0508 MWh, 0.0481 MWh, 0.1791 MWh, respectively. Figs. 8 and 9
and DSs (distributed energy storage system) without storage loss, shows the change results of the nodal voltage and power loss when
that is to say the efficiency of the DS is not considered; 4) the cases distributed generators are injected in some nodes of the distribu-
with EVs, DGs and DSs considering DS efficiency. For these four tion system, and Table 6 shows the risk value of the system with
cases, the power loss, the CVaR value of power loss, the VaR value of EVs and DGs. It can be seen from the data in Table 5, Table 6,
power loss, and the risk value of power loss are 0.4299 MWh, Figs. 7e9 that the nodal voltage is lowered, and the risk due to
0.5022 MWh, 0.4775 MWh, 0.4731 MWh; 0.0803 MWh, power loss is decreased by coordinating the output power of DG,
0.0975 MWh, 0.0915 MWh, 0.2661 MWh; 0.0328 MWh, the power loss, the CVaR value of power loss, the VaR value of

Table 3
Optimal output power of DGs.

Time period 6:00e8:00 14:00e16:00 20:00e22:00

Chance constrained confidence Node No. Active power/ Reactive power/ Active power/ Reactive power/ Active power/ Reactive power/
level MW Mvar MW Mvar MW Mvar

a ¼ 90% 27 0.80 0.2630 0.80 0.2630 0.80 0.2630

50 0.45 0.1479 0.55 0.1808 0.90 0.2958
54 0.65 0.2137 0.95 0.3123 1.00 0.3287
a ¼ 95% 27 0.80 0.2630 0.80 0.2630 0.80 0.2630
50 0.45 0.1479 0.50 0.1644 0.85 0.2794
54 0.55 0.1808 0.85 0.2794 1.00 0.3287
a ¼ 99% 27 0.80 0.2630 0.80 0.2630 0.80 0.2630
50 0.40 0.1315 0.50 0.1644 0.75 0.2465
54 0.50 0.1644 0.80 0.2794 1.00 0.3287

Table 4
Sampling number ratios of bus voltages and branch currents.

Chance constrained confidence level Sampling number ratio 6:00e8:00 14:00e16:00 20:00e22:00

a ¼ 90% NV =N 91453/10000 92341/10000 91562/10000

NI =N 10000/10000 10000/10000 10000/10000
a ¼ 95% NV =N 96874/10000 97253/10000 95213/10000
NI =N 10000/10000 10000/10000 10000/10000
a ¼ 99% NV =N 99206/10000 99473/10000 99324/10000
NI =N 10000/10000 10000/10000 10000/10000
 J. Wu et al. / Energy 143 (2018) 323e336 333

Table 5 No matter how the charging power of electric vehicles, when

Risk value of power loss in the distribution system with DGs and EVs under different the distributed power supply to the power grid, the voltage levels
nodal voltage.
are greatly improved and the voltage qualification rates are
Node No. 6:00e8:00 14:00e16:00 20:00e22:00 increased to 100%. The power loss of some branches is greatly
Amplitude/p.u Amplitude/p.u Amplitude/p.u decreased, and the risk of power loss due to the charging un-
certainties of EVs is also greatly decreased by the optimum coor-
28 1.0449 1.0448 1.0447
29 1.0448 1.0447 1.0444 dination of the output power of distributed generation.
30 1.0444 1.0443 1.0435 Table 5 shows the risk value obtained in a distribution system
40 1.0412 1.0395 1.0331 with EVs and DGs, and the actual output power of DGs is considered
41 1.0412 1.0395 1.0331
in a day. When compared with Table 1, the output of DGs can
42 1.0411 1.0392 1.0317
53 1.0497 1.0434 1.0198 maintain the net loss at a lower level, and reduce the risk of
54 1.0548 1.0488 1.0253 network loss. From Fig. 9, it is seen that the node voltage fluctuation
55 1.0391 1.0371 1.0290 becomes larger when considering the charging efficiency of the
56 1.0391 1.0371 1.0290 energy storage devices.
59 1.0449 1.0448 1.0447
Taking the case for time period from 20:00e22:00 as an
60 1.0447 1.0446 1.0444
example, and using a chance constrained confidence level of
Ploss/MWh 0.0269 0.0196 0.0717
a ¼ 95%, the net loss of the distribution system is reduced to
CVaR/MWh 0.0393 0.0417 0.0846
VaR/MWh 0.0377 0.0401 0.0811 0.0804 MWh from the 0.4262 MWh. The power loss of
PL0/MWh 0.2826 0.3256 0.4642 0.0804 MWh is obtained in the cases with a power loss rate of
6.43%, when the distributed generators operate with an optimal
output power. If the output power of the distributed generators is
0.7 stable, the risk value of CVaR caused by the fluctuation of the
electric vehicle charging power is 0.0393 MWh, and the power loss
rate is 1.04% in these cases. However, in practice, due to the un-
0.6
certainty of the wind speed calculation, the actual output of the
wind power is often less than the optimal output, which leads to an
Power loss or Risk value/MWh

0.5 increase in the loss of risk value. The net loss risk of CVaR and VaR
increases to 0.2273 MWh (corresponding to a power loss rate of
3.48%) and 0.2071 MWh (corresponding to a power loss rate of
0.4
3.19%), respectively, as shown in Table 6.
It is seen from the Equation (19)e(22) that the values of CVaR
0.3 depend on the output power of the distributed generation systems
or the discharge power of the distributed energy storage system as
0.2 well as the random power of the uncontrollable load and the
controllable charging power of electric vehicles with uncertain
characteristics. Under the large branch power flow, the chance
0.1
constraint of nodal voltage and branch power will be possibly
destroyed, and the branch power may possibly exceed the
0 maximum allowable value. In the forward power flow, the large
1 2 3 4
Case No. branch power will reduce the nodal voltage even more; in the
reverse power flow, the large branch power will make the nodal
Fig. 7. Comparison in risk value due to power loss of distribution systems with EVs and voltage rise even greater.
DGs in the time period of 20:00e22:00 in a day (Note: “1”: Risk vale for cases only During heavy load periods, if the power of the electric vehicles is
with EVs, “2”: Risk value for cases with EVs and DGs, “3”: Risk value for cases with EVs,
DGs and DSs without storage loss, “4”: 'Risk value for cases with EVs, DGs and DSs
very large, it will cause very high power loss and lead to a great
considering DS efficiency). drop in the nodal voltage, which may be lower than its lower limit
in severe cases, in which the CVaR value may be greater than the
limit value and the nodal voltage may be lower than the chance
power loss. Moreover, the risk value of power loss becomes larger constraint limit. During light load periods, a reverse power flow on
when considering the charging efficiency of the energy storage the branch possibly forms due to the greater output power of the
devices.

1.06 1.06 1.06

1.04 1.04 1.04

1.02 1.02 1.02

Voltage/p.u

Voltage/p.u

1 1
Voltage/p.u

0.98 0.98 0.98

Nodal voltage for cases only with EVs

Nodal voltage for cases only with EVs
0.96 0.96 0.96 Nodal voltage for cases with DGs and EVs
Nodal voltage for cases with DGs and EVs
Nodal voltage for cases with DGs,EVs and DSs
Nodal voltage for cases with DGs,EVs and DSs
0.94 0.94 0.94
Nodal voltage for cases only with EVs
Nodal voltage for cases with DGs and EVs
0.92 0.92 0.92
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Node No. Node No. Node No.

Fig. 8. Nodal voltage variation within 20:00e22:00 in a day.

334 J. Wu et al. / Energy 143 (2018) 323e336

0.09 0.09 0.09

Power loss for cases only with EVs Power loss for cases only with EVs
0.08 0.08 0.08 Power loss for cases only with EVs
Power loss for cases with DGs and EVs Power loss for cases with DGs and EVs
Power loss for cases with DGs and EVs
Power loss for cases with DGs,EVs and DSs
0.07 0.07 0.07 Power loss for cases with DGs,EVs and DSs

0.06 0.06 0.06

Power loss/kWh

Power loss/kWh
Power loss/kWh

0.05 0.05 0.05

0.04 0.04 0.04

0.03 0.03 0.03

0.02 0.02 0.02

0.01 0.01 0.01

0 0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Branch No. Branch No. Branch No.

Fig. 9. Power loss of the branches within 20:00e22:00 in a day.

distributed generation systems or the discharge power of the special kind of load, EV charging will significantly affect distri-
distributed energy storage systems, and this may lead to the rise of bution systems. As a new power supply, the DGs will greatly
the branch power loss and may make the CVaR value greater than benefit the grid. However, there are significant uncertainties
its limit, the branch power may possibly exceed the maximum concerning EV charging and the DG of renewable energy systems,
allowable value and the nodal voltage is greater than the upper which may result in potential risks in the operation of distribution
bound of its chance constraint. In these cases, it is possible that the systems. In this paper, the potential risk value caused by the po-
chance constraints and CVaR constraints are not satisfied simulta- wer loss due to the uncertainties of EV charging and DG is taken
neously. Some measures must taken for handling chance con- into constraints, and is described using the CVaR method. A
straints and CVaR constraints violation, for example, properly method based on CCSOCP is presented for the optimal risk value
decrease or increase the nodal load power sucking from the grid or control of power loss in distribution systems with DGs, DSs and
decrease or increase the charging power of electric vehicles at the EVs.
related nodes, or properly increase or decrease the output power of The risk index of power loss based on CVaR is constructed for
the distributed generation systems or the distributed energy stor- evaluating the impact of the significant uncertainties of the EV
age systems into the grid at the related nodes. charging and DG of renewable energy systems on the economic
operation of distribution systems, and it shows better results that
7. Conclusions the power loss, the CVaR value of power loss, the VaR value of
power loss, and the risk value of power loss considering EV, DG and
Electric vehicle charging and DGs are the factors that must be DS is lesser than that with EV and DG, as shown in Table 7. More-
considered in the development of distribution systems. As a over, the CVaR based method may provide new effective

Table 6
Risk Value of The system with EVs and DGs.

confidence level Time period 6:00e8:00 14:00e16:00 20:00e22:00

Loss value/MW Loss rate/% Loss value/MW Loss rate/% Loss value/MW Loss rate/%

a ¼ 90% Ploss 0.0240 0.65% 0.0316 0.72% 0.0737 1.16%

CVaR 0.0550 1.47% 0.0785 1.79% 0.2205 3.38%
(EVDG)
VaR 0.0508 1.36% 0.0684 1.56% 0.2043 3.14%
(EVDG)
PL0 0.2763 7.00% 0.3237 7.00% 0.4733 7.00%
a ¼ 95% Ploss 0.0298 0.80% 0.0334 0.77% 0.0804 1.26%
CVaR 0.0644 1.72% 0.0829 1.89% 0.2273 3.48%
(EVDG)
VaR 0.0585 1.56% 0.0733 1.67% 0.2071 3.19%
(EVDG)
PL0 0.2763 7.00% 0.3237 7.00%0 0.4733 7.00%
a ¼ 99% Ploss 0.0334 0.90% 0.0348 0.80% 0.0835 1.31%
CVaR 0.0785 2.09% 0.0836 1.91% 0.2312 3.55%
(EVDG)
VaR 0.0684 1.83% 0.0763 1.74% 0.2089 3.22%
(EVDG)
PL0 0.2763 7.00% 0.3237 7.00% 0.4733 7.00%

Ploss in the table is the loss power obtained in the cases where DGs operate with an optimal output power.

Table 7
The risk index of power loss in different cases.

The power loss/MWh The CVaR value of power The VaR value of The risk value of
loss/MWh power loss/MWh power loss/MWh

Cases with EVs and DGs 0.0803 0.0975 0.0915 0.2661

Cases with EVs, DGs and DSs without storage loss 0.0328 0.0421 0.0380 0.2163
 J. Wu et al. / Energy 143 (2018) 323e336 335

technologies for evaluating the potential risk caused by the unor- multiobjective framework. IEEE Trans Power Syst 2009;24(1):78e85.
[8] Alireza M. A critical evaluation of deterministic methods in size optimisation
dered charging of EVs and unordered DG of renewable energy
of reliable and cost effective standalone hybrid renewable energy systems.
systems. Reliab Eng Syst Saf 2014;130:1247e54.
The most direct way for controlling the risk value due to power [9] McCalley J, et al. Probabilistic security assessment for power system opera-
loss is to reduce the power loss of distribution systems. Taking the tions. In: Proc. IEEE Power Eng. Soc. Gen. Meeting (PES), Denver, CO, USA;
2004. p.212e220.
uncertainties in EV charging and the constraints of the nodal [10] Ni M, McCalley JD, Vittal V, Tayyib T. Online risk-based security assessment.
voltage and branch current into consideration, a CCSOCP method IEEE Trans Power Syst 2003;18(1):258e65.
for controlling the risk value of power loss is presented by optimally [11] Wan H, McCalley J, Vittal V. Risk based voltage security assessment. IEEE Trans
Power Syst 2000;15(4):1247e54.
coordinating the power output of DGs and EV charging. It is [12] Kirschen DS, Jayaweera D. Comparison of risk-based and deterministic secu-
reasonable that the potential risk is controlled to be maintained rity assessment. IET Gener Transm Distrib 2007;1(4):527e33.
below an acceptable power loss value, by coordinating the power [13] Jia HJ, Qi, WJ Liu Z, Wang BD, Zeng Y, Xu T. Hierarchical Risk Assessment of
Transmission System Considering the Influence of Active Distribution
output of DGs. Network. IEEE Trans. Power Syst., to be published.
In this paper, the method based on CCSOCP is presented for [14] Zeng Y, Yu Y. A practical direct method for determining dynamic security
transforming the charging probability model of EVs and the output regions of electric power systems. Proc Chin Soc Elect. Eng (CSEE) 2003;23(5):
24e8.
probability of DGs. This is because it takes a long time and a large [15] Dai YJ, McCalley J, Abi-Samra N, Vittal V. Annual risk assessment for overload
amount of computation for the Monte Carlo simulation to solve security. IEEE Trans Power Syst 2001;16(4):616e23.
probability problem, and it requires much more time for processing [16] Xiao F, McCalley JD. Power system risk assessment and control in a multi-
objective framework. IEEE Trans Power Syst 2009;24(1):78e85.
the nonlinear equation with angle variables. The proposed method
[17] Rossmaier J, Chowdhury B. Further development of the overload risk index, an
is used to carry out the cone transformation for the power flow indicator of system vulnerability. In: Proc. North Amer. Power Symp. (NAPS),
equation with nonlinear constraint functions. Using the proposed Starkville, MS, USA; 2009. p. 1e6.
method, the nonlinear equal equation of power flow is converted [18] Li X, Zhang X, Du D, Cao J. Overload risk assessment in gridconnected in-
duction wind power system. In: System simulation and scientific computing
into a linear equation. The simulation results reveal that the pro- (communications in computer and information science), vol. 327. Berlin,
posed method can greatly reduce the computational complexity Germany: Springer-Verlag; 2012. p. 44e51.
and time for risk value control problem of power loss in distribution [19] Wang Q, McCalley JD, Zheng T, Litvinov E. A computational strategy to solve
preventive risk-based security-constrained OPF. IEEE Trans Power Syst
systems with DGs, DSs and EVs. 2013;28(2):1666e75.
Moreover, very good application results are obtained for the [20] Wang Q, et al. Risk-based locational marginal pricing and congestion man-
proposed method, indicating that it can be used to solve the un- agement. IEEE Trans Power Syst 2014;29(5):2518e28.
[21] Feijoo AE, Cidra J, Dornelas JLG. Wind speed simulation in wind farms for
certainty problem of distribution systems with DGs and EVs. It can steady-state security assessment of electrical power systems. IEEE Trans En-
be seen from the simulation results that the optimal solution is ergy Convers 1999;14(4):1582e8.
reasonable, with a confidence of 95%, although the voltages of a [22] Miao L, Fang J, Wen J, Luo W. Transient stability risk assessment of power
systems incorporating wind farms. J Mod Power Syst Clean Energy 2013;1(2):
minimal number of nodes in the distribution system exceed the 134e41.
limited value. [23] Billinton R, Karki B, Karki R, Gokaraju R. Unit commitment risk analysis of
wind integrated power systems. IEEE Trans Power Syst 2009;24(2):930e9.
[24] Wangdee W, Billinton R. Considering load-carrying capability and wind speed
Acknowledgments correlation of WECS in generation adequacy assessment. IEEE Trans Energy
Convers 2006;21(3):734e41.
This work is supported by the National Natural Science Foun- [25] Karki R, Thapa S, Billinton R. Operating risk analysis of wind integrated power
systems. Elect Power Compon Syst 2012;40(4):399e413.
dation of China (51567002,50767001); The National High Tech- [26] Pinson P, Kariniotakis G. On-line assessment of prediction risk for wind power
nology Research and Development of China (863 Program) production forecasts. Wind Energy 2004;7(2):119e32.
(2007AA04Z197); Specialized Research Fund for the Doctoral Pro- [27] Hajimiragha AH, Canizares CA, Fowler MW, Moazeni S, Elkamel A. A robust
optimization approach for planning the transition to plug-in hybrid electric
gram of Higher Education (20094501110002); Natural Science vehicles. IEEE Trans Power Syst 2011;26(4):2264e74.
Foundation of Guangdong (S2013010012431, 2014A030313509); [28] Gomez JC, Morcos MM. Impact of EV battery chargers on the power quality of
Guangxi Natural Science Foundation (2011jjA60017); Guangdong distribution systems. IEEE Trans Power Del 2003;18(3):975e81.
[29] Strbac G, Gan CK, Aunedi M, Stanojevic V, Djapic P, Dejvises J, et al. Benefits of
special fund for public welfare study and ability construction
advanced smart metering for demand response based control of distribution
(2014A010106026); The Guangdong Applied Science and Technol- networks. London, U.K: Imperial College; 2010.
ogy Research Foundation of P.R.China (2016B020244003); The [30] Hajimiragha A, Canizares CA, Fowler MW, Elkamel A. Optimal transition to
plug-in hybrid electric vehicles in Ontario, Canada,considering the electricity-
Talent Introduction Special Foundation Project of Guangdong High
grid limitations. IEEE Trans Ind Electron 2010;57(2):690e701.
School; The Disciplinary Construction Special Foundation Project of [31] Hadley SW, Tsvetkova A. Potential impacts of plug-in hybrid electric vehicles
Guangdong High School (2012KJCX0045). on regional power generation. Electr J 2009;22:56e68.
[32] Letendre S, Watts RA. Effects of plug-in hybrid electric vehicles on the Ver-
mont electric transmission system. In: Proc. Transportation Research board
References 88th Annu. Meeting, Washington, DC, USA; Jan. 2009.
[33] Deilami S, Masoum AS, Moses P, Masoum MAS. Realtime coordination of plug-
[1] Guha Thakurta P, Belmans R, Van Hertem D. Risk-based management of in electric vehicle charging in smart grids to minimize power losses and
overloads caused by power injection uncertainties using power flow con- improve voltage profile. IEEE Trans Smart Grid 2011;2(3):456e67.
trolling devices. IEEE Trans Power Syst 2015;30(4):1838e47. [34] Sortomme E, Hindi MM, MacPherson SDJ, Venkata SS. Coordinated charging of
[2] Moshari A, Ebrahimi A, Fotuhi-Firuzabad M. Short-term impacts of DR pro- plug-in hybrid electric vehicles to minimize distribution system losses. IEEE
grams on reliability of wind integrated power systems considering demand- Trans Smart Grid 2011;2(1):198e205.
side uncertainties. IEEE Trans Power Syst 2015;99:1e10. [35] Qian K, Zhou C, Allan M, Yuan YY. Modeling of load demand due to EV battery
[3] Wang SX, Han L, Wu L. Uncertainty tracing of distributed generations via charging in distribution systems. IEEE Trans Power Syst 2011;26(2):802e10.
complex affine arithmetic based unbalanced three-phase power flow. IEEE [36] Zhou RJ, Yao LH, Tong XJ. Security economic dispatch in wind power inte-
Trans Power Syst 2015;30(6):3050e62. grated systems using a conditional risk method. Proc CSEE 2012;32(1):56e63.
[4] Negnevitsky M, Dinh Hieu Nguyen, Piekutowski M. Risk assessment for power [37] Peng S. Research on wind power penetration limit caleulation based on
system operation planning with high wind power penetration. IEEE Trans conditional value at risk theroy. Changsha University of Science and Tech-
Power Syst 2015;30(3):1359e67. nology; 2011.
[5] Li GF, Zhang P, Luh PB, Li WY, Bie ZH, Serna C, et al. Risk analysis for distri- [38] Zhou RJ. Optimization of the wind farm capacity based on CVaR. Electr Power
bution systems in the northeast U.S. Under wind storms. IEEE Trans Power 2011;6:58e62.
Syst 2014;29(2):889e98. [39] Jabr RA. Radial distribution load flow using conic programming. IEEE Trans
[6] Jayaweera D, Islam S. Steady-state security in distribution networks with large Power Syst 2006;21(3):1458e9.
wind farms. J Mod Power Syst Clean Energy 2014;2(2):134e42. [40] Zhou JL, Zhao JL, Yu YY, Li P, Sun CB, Dong XZ, et al. Application of Conic
[7] Fei X, McCalley JD. Power system risk assessment and control in a Programming for Optimal Distributed Generation Allocation in Distribution
336 J. Wu et al. / Energy 143 (2018) 323e336

Network. In: China International Conference on electricity distribution(- [55] Zhang TT, Liu ZF. Fireworks algorithm for mean-VaR/CVaR models. Phys A Stat
CICED); 2014. Mech Appl 2017;483:1e8.
[41] Yafei Y, et al. Maximizing renewable energy penetration through distribution [56] Balbas A, Balb s R. VaR as the CVaR sensitivity: Applications in risk
as B, Balba
network reconfiguration using mixed-integer conic programming. In: IEEE optimization. J Comput Appl Math 2017;309:175e85.
PES general Meeting|Conference & Exposition, National Harbor, MD; 2014. [57] Liu B. A new risk measure and its application in portfolio optimization: The
[42] Jabr RA, Singh R, Pal BC. Minimum Loss Network Reconfiguration Using SPPeCVaR approach. Econ Model 2015;51:383e90.
Mixed-Integer Convex Programming. IEEE Trans Power Syst 2012;27(7): [58] Xie JL. Research on the accuracy of sample quantile of VaR and CVaR. Guangxi
1106e15. Normal University; 2009.
[43] Tan ZF, Wang G, Ju LW, Tan QK, Yang WH. Application of CVaR risk aversion
approach in the dynamical scheduling optimization model for virtual power
plant connected with wind-photovoltaic-energy storage system with ncer-
tainties and demand response. Energy 2017;124:198e213.
Wu Jiekang was born in China, on Dec. 9, 1965. He graduated
[44] Wang Z, Bian QY, Xin HH, Gan DQ. A distributionally robust co-ordinated
from Zhejiang University.His employment experience included
reserve scheduling model considering CVaR-based wind power reserve re-
as an engineer in power supply bureau; engineer in electrical
quirements. IEEE Trans Sustain Energy 2016;7(2):625e36.
engineering designing institute; a professor in Zhejiang Uni-
[45] Miguel A, Javier C. Stochastic Unit Commitment in Isolated Systems With
versity, Guangxi University, Guangdong University of Technol-
Renewable Penetration Under CVaR Assessment. IEEE Trans Smart Grid
ogy; and an engineer in large enterprises. His special fields of
2016;7(3):1356e67.
interest included renewable power systems.WU Jiekang
[46] Zhang CL, Guo P. An inexact CVaR two-stage mixed-integer linear program-
received honorary degrees from institutions of higher learning
ming approach for agricultural water management under uncertainty
including Zhejiang University, South China University of Tech-
considering ecological water requirement. Ecological Indicators, In Press,
nology and Guangxi University.
Corrected Proof, Available online 3 March 2017.
[47] Mohammad AMAA, Shokri ZS. Risk-averse multi-product selective news-
vendor problem with different market entry scenarios under CVaR criterion.
Comput Ind Eng 2017;103:250e61. Wu Zhijiang is with School of Automation, Guangdong University of Technology,
[48] Madalina D, Antoine L, Khaled S. Approximation of CVaR minimization for Guangzhou, Guangdong, China. His research topics lie at power system operation and
hedging under exponential-Le vy models. J Comput Appl Math 2017;326: analysis.
171e82.
[49] Baha A. Decomposition-based interior point methods for stochastic quadratic
second-order cone programming. Appl Math Comput 2014;249(1):1e18. Wu Fan is with Guangxi Bo Yang Electric Power Survey and Design Co, Ltd, Guangxi
[50] Jabr RA. Radial distribution load flow using conic programming. IEEE Trans Power Grid Co, Ltd, Nanning, Guangxi, China. His research topics lie at power system
Power Syst 2006;21(3):1458e9. operation and analysis.
[51] Baha A. Stochastic optimization over symmetric cones. www.math.wsu.edu/
students/balzalg/davis1.pdf.
[52] Jabr RA. Optimal power flow using an extended conic quadratic formulation. Tang Huiling is with School of Automation, Guangdong University of Technology,
IEEE Trans Power Syst 2008;23(3):1000e8. Guangzhou, Guangdong, China. Her research topics lie at power system operation and
[53] Youcef Ettoumi F, Mefti A, Adane A, Bouroubi MY. Statistical analysis of solar analysis.
measurements in Algeria using beta distributions. Renew Energy 2002;26(1):
47e67.
[54] Mary R, Hardy PFF. An Introduction to Risk Measures for Actuarial Applica- Mao Xiaoming is with School of Automation, Guangdong University of Technology,
tions. In: Education and examination committee of the society of actuaries; Guangzhou, Guangdong, China. Her research topics lie at power system operation and
2006. USA. analysis.