Energy 164 (2018) 560e574

Contents lists available at ScienceDirect

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

Allocation optimization of electric vehicle charging station (EVCS)

considering with charging satisfaction and distributed renewables
integration
Jin-peng Liu a, b, *, Teng-xi Zhang a, b, Jiang Zhu c, Tian-nan Ma d
a
School of Economics and Management, North China Electric Power University, Beijing 102206, China
b
Beijing Key Laboratory of New Energy and Low-Carbon Development (North China Electric Power University), Beijing, 102206, China
c
Strategic and Economic Research Department, Central Research Institute, State Power Investment Corporation, Beijing 102209, China
d
The Economic and Technological Research Institute of Sichuan Electric Power Company of State Grid Corporation, Chengdu 610041, China

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

Article history: Under the background of large-scale electric vehicle (EV) development, it is necessary to design and
Received 20 July 2018 deploy the EVCS more scientific. Among various factors influential to the EVCS allocation, charging
Received in revised form satisfaction and distributed renewables integration were mainly considered in this paper. First, with
1 September 2018
System Dynamics (SD) model, the key factors affecting the EVCS allocation were identified from the
Accepted 5 September 2018
Available online 8 September 2018
conduction mechanism. Then, focusing on the site selection of EVCS from the aspect of user satisfaction,
k-means clustering method was used to illustrate the relationship between charging distance and
satisfaction degree. On this basis, considering with renewables integration and stable operation of power
Keywords:
Multi-objective optimization
system, the paper constructed a multi-objective function including voltage fluctuation, load fluctuation
EVCS and connected capacity of energy storage in EVCS. Third, under the feeder framework of an IEEE 33-node,
Distributed renewables GA-PSO was employed to determine the best solution of EVCS allocation., i.e. the optimal allocation
Charging satisfaction number of EVCS, the site and capacity of EVCS, and the access nodes of renewables and EVCS. Combing
k-means with the analysis results, suggestions from the aspects of technology standard, finance subsidy, land use
GA-PSO support and energy management were proposed for accelerating the generalization of EVs and
strengthening the supporting infrastructure construction.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction the level of 13.7% and 6% in 2017 [2]. The phenomenon shows that
lots of clean energies were not being used effectively in China.
To limit climate change and achieve sustainable growth, it's Under the dual constraints of power supply and demand, some
imperative to produce and use more renewables around the world. areas have become the new poles of energy development and
International Renewable Energy Agency (IRENA) shows that the utilization, such as the electric vehicle (EV) in transport sector. Due
world needs to increase the share of renewables in total final en- to its advantages of reducing the emissions of greenhouse gas and
ergy consumption (TFEC) from 19% in 2017 to two-thirds by 2050 increasing the consumptions of abandoned renewables, EV in China
[1]. In the era of energy transition, progresses in transport, building, has entered into a rapid development phase. Based on the Electric
industry and power sectors need to accelerate further. Especially, Vehicle World Sales Database, more than one million EVs were sold
the curtailment of renewables emerging in recent years challenges in 2017 which reached one-third of global electric vehicle owner-
the realization of energy transition more seriously. For example in ship [3]. In the sales increment of 2017, about half was contributed
China, by the end of 2017, 164 GW wind power and 130 GW solar by China. Its positive effects on the usage of clean energy and the
power were connected to the power grid in total. However, the promotion of electrification are emerging gradually.
curtailment rate of wind power and solar power still remained at In order to ensure the orderly development of electric vehicles,
governments in many countries have enacted policies encouraging
EV sales, including China, France, Germany, India, Italy, Japan,
* Corresponding authorSchool of Economics and Management, North China Norway, South Korea, Spain, Sweden, the United Kingdom, and the
Electric Power University, Beijing 102206, China. United States. These policies range from direct monetary incentives
E-mail addresses: [email protected], [email protected] (J.-p. Liu).

https://doi.org/10.1016/j.energy.2018.09.028
0360-5442/© 2018 Elsevier Ltd. All rights reserved.
 J.-p. Liu et al. / Energy 164 (2018) 560e574 561

to time-saving measures [4]. Meanwhile, in some emerging coun- effectiveness examination. At last, some issues were discussed from
tries of EV, the governments also realize that the planning and the different perspectives in order to accelerate the EVCS construction
construction of EV charging infrastructure does not match up to the orderly.
quantity increase of EV well. In a long run, lack of electric vehicle
charging stations (EVCS) or disordered planning of EVCS may cause
2. Literature reviews
a collapse of electric transport system. To avoid this risk, some
countries focus on the top-level design and the technical standard
In the previous studies, most of the experts and scholars focused
making. Take China as an example, the proposed policies for EV
on how to make the planning of EVCS more rational considering
charging infrastructure development in recent years are shown in
with different affecting factors. These factors were proposed from
Fig. 1 [5].
different standpoints which can be classified into three fields, i.e.
Driven by the increment of EV ownership and positive policies,
optimal EVCS planning, EV scheduling and charging demand
the EV charging piles in China has increased rapidly. In 2017, the
analysis, EV load forecasting.
ratio of EV vehicles to charging piles has reduced to 1.8. The sup-
From the perspective of planning, EVCS should be planned and
porting construction situation of EV charging infrastructure has
constructed harmoniously with the urban planning or power grid
been improved widely. However, the planning method of EVCS still
planning, especially with the access of renewables. Wang Hengsong
needs to be optimized, that is, more factors and complicated
et al. presented a multi-objective planning model which includes
influencing mechanism should be taken into account. Thus, the
the power grid and factors of municipal planning [6]. Liu Zifa et al.
paper focused on how to plan the EVCSs reasonably by considering
used geographic information, construction cost and running cost to
with multiple factors, such as geographic information of EVCS,
building an objective function to comprehensively reflect the sub-
generation fluctuations of renewables, the power balance in dis-
stance problem of locating and sizing of electric vehicle charging
tribution area, charging satisfaction degree of EV users and so on.
station [7]. Sadeghi Barzani. P et al. constructed a Mixed-Integer
First, a wide review of relevant literature was given to summarize
Non-Linear (MINLP) model for optimal placing and sizing of the
the research results of EV and EVCS. Then, based on the hierarchical
fast charging stations which includes the station development cost,
structure of energy management system, an integrated research
EV energy loss, electric gird loss as well as the location of electric
methodology was built for exploring the optimal allocation plan of
substations and urban roads. Especially, geographic information
EVCS. The conduction mechanism of factors influential to EVCS
were also used to determine EV energy loss and station electrifi-
allocation was explored by System Dynamics (SD) model. K-means
cation cost [8]. Habiballah Kheradmand. K and Mohsen Gitizadeh
method was employed to cluster the charging demand points and
evaluated the reliability changes of distribution network in the
the relationship between the numbers of EVCS and the charging
presence of EVs and managed the plug-in hybrid EVs charging and
satisfaction was described under the constraints of distribution
power injection to the grid using system well-being analysis [9].
network. Meanwhile, take EVCS as an energy storage system, a
Take EVCS as an energy storage unit, it can provide service for
multi-objective function was constructed to select the EVCS sites
power grid, such as shifting load from peak to valley, frequency
and deploy the capacity by considering with the distributed re-
regulation, energy saving and so on [10]. Thus, fulfilling the flexible
newables integration, the satisfaction degree of EV users and the
scheduling of EVs with charging demand is also a critical factor
technical indicators in distribution network. Third, the proposed
influential to the EVCS planning. For example, Sulabh Sachana and
model was simulated and analyzed by using Genetic Algorithm-
Nadia Adnan analyzed the temporal characteristic of EV loads in
Particle Swarm Optimization (GA-PSO) under the framework of
order to ensure the stability of distribution network. Meanwhile,
IEEE 33-nodes test feeder. The best solution was calculated with the
wind power flow and resulting variation in electricity price was

Fig. 1. Published policies of EV charging infrastructure construction in China.

562 J.-p. Liu et al. / Energy 164 (2018) 560e574

considered with stochastic availability of EVs, i.e. arrival and de- Swarm Optimization (GA-PSO) calculation method when consid-
parture times [11]. To mitigate the undesirable fluctuations of wind ering with the renewables. Finally, a case study was used to
power, M. Raoofat et al. proposed a power smoothing service using examine the validity of the proposed methods. The research pro-
the demand response of EVs connected to the adjacent networks. A cess is shown in Fig. 2.
multi-hierarchical controller was used to measure how ready each Based on the research process in Fig. 2, the main methods
vehicle is to participate in mitigating large positive and negative adopted in this paper will be briefly introduced as follows.
fluctuations [12]. For improving the resilience of distribution sys-
tems, Kaveh Rahimi and Masoud Davoudi generated different load 3.1. Hierarchical structure of EVCS energy management system
profiles for various consumption scenarios to model the severity of
conditions and estimate the required energy of EVs. They also In order to understand the allocation environment clearly, an
considered seasonal conditions and developed the load profiles of energy management system with EVCS was constructed in Fig. 3. In
EVs based on characteristic curves of major household appliances, the EVCS energy management system, the electric equipment and
instead of their rated powers [13]. With different amounts of power control system are located on the top level of distribution
charging infrastructure, Till Gnann et al. modeled the market network which contains different power sources, energy storage
diffusion of EVs and their load shifting potential. They forecasted system (ESS) and other load in power grid. EVCSs and EVs are
that 25e30% more excess renewable electricity can be integrated respectively at the middle and bottom level of energy management
with EV load shifting, but the effects of public and private EVs to system which stand for the power transformation and power
load shifting are different. Nonetheless, the demand response of consumption agents.
EVs can still reduce 2% of the system load in the scenario analysis Especially, at the top level of the system, the working status of
[14]. Zhang Weige et al. proposed the charging strategy of EVs to ESS should be set. On current phase, there are mainly two working
ensure high-efficiency charging, reduce load variations to the po- status of ESS, i.e. sequence charging (SC) and vehicle to grid (V2G)
wer grid during charging periods, achieve charging completion [19]. In SC mode, EVCS only provides the charging service for EVs.
with high probability, and accomplish approximate valley-filling However, in V2G mode, EVs are taken as energy storage agents that
[15]. they can discharge their remaining power back to power grid. The
In addition, for understanding the charging demand deeply, mutual interaction between charging and discharging is critical to
advanced methods were employed to forecast EV load. Using the realization of V2G. Meanwhile, the sites and the numbers of
Monte Carlo simulation algorithm, Ye Jiancheng et al. established EVCS are also determined by physical structure of energy man-
the load calculation model with different charging ways for buses, agement system.
taxis and sedans, thereby drawing their appropriate charging time
and characteristics of the interaction with grid [16]. Chiş A. et al. 3.2. System dynamics model
[17] took electricity tariff into account in the charging behaviors
and decisions. Based on the day-ahead data of electricity tariff, the System dynamics was first proposed by Prof. Forrester of the
charging behaviors were estimated by Markov chain. It proved that Massachusetts Institute of Technology in 1956 (Forrester, 1969). It
charging cost was the most important factor influential to the was introduced to China in the late 1970s and has made great
charging decisions of EV users. Li Yunyan et al. proposed an inno- achievements in regional and urban planning, energy management,
vative prediction model based on a convolutional neural network and industrial research. Systems dynamics theory strongly believes
and lion algorithm, improved by niche immunity. The results in this idea: the system must have a structure, and the structure of
indicate that the new hybrid proposed model offers better accuracy, the system determines its function. It argues that the root of the
robustness, and generality in short-term load forecasting for elec- problem should be found from the internal structure of the system
tric vehicle charging stations [18]. based on the feedback characteristics of the elements within the
In summary, there are valuable results in the issue of EVCS site system, rather than using external disturbances or random events
selection. As an intelligent part of the smart energy interconnec- to illustrate the nature of the system. It also firmly believes that
tion, EVCS should meet the requirements and constraints of power many of the variables within the system have causal links in their
grid planning. Especially when more renewables are connected to interaction. The systematic interconnection of feedback poses the
the power system, EVCS should play greater roles in the fields of structure of the system, which is the fundamental determinant of
smoothing the power fluctuations and satisfying the EV charging system behavior [20].
demands. Thus, under the structure of distribution network and The relationship between the causes and outcomes of the ele-
considering with the access of renewables, the storage capacity and ments in the system is named as causal relationship. It can be
the rational sites of EVCSs will be focused on in the research. described by the connection of the causal elements and the direc-
tion of the arrow. When both the cause and the outcome move in
3. Methods the same direction, this relationship is called the positive causal
relationship, and is represented by “þ”. On the contrary, it is called
In order to obtain the best allocation plan of EVCS, several negative causal relationship, using “-” to represent it. When mul-
advanced methods were employed and integrated in this paper. tiple causal chains are connected end to end, they form a causal
Based on the literature reviews, the main factors influential to the feedback loop. As shown in Fig. 4, the causal feedback loop can be
EVCS allocation were explored by System Dynamics (SD) model. divided into positive feedback and negative feedback. The causal
Then, on the basis of planning and supporting environment of circuit diagram (CLD) composed of multiple feedback loops can
EVCS, a hierarchical structure of EVCS energy management system clearly show the complex causal relationship between multiple
was taken into account to analyze the connection constraints of variables, and extract the key elements that affect the structure of
EVCS to the distribution network. Meanwhile, considering with the the system. It simplifies the problem of high order, nonlinear,
distribution of EV charging demand, the allocation site and number multiple feedback complex time-varying systems [21]. EVCS allo-
of EVCS was measured by k-means clustering method and the cation are affected by multiple factors, involving energy system, EV
average charging satisfaction degree. With the site optimization user system, scheduling system and society system. The use of the
results, an objective function was constructed to simulate the best system dynamics model can help us to better comb the main loop
allocation capacity of EVCS by using Genetic Algorithm-Particle between the factors influential to the allocation and grasp the
 J.-p. Liu et al. / Energy 164 (2018) 560e574 563

Literature Reviews

Hierarchical Structure of EVCS Key Factors influential to the

Charging Demand Distribution
Energy Management System EVCS Allocation

K-means Site Optimization and Charging Constraints of EVCS connected to Conduction Mechanism of EVCS
Satisfaction Power Grid System Dynamics (SD)
Probability Distribution Function Allocation

Multi-objective Optimization Genetic Algorithm-Particle Swarm

Objectives and Constraints
Function Construction Optimization (GA-PSO)

Case Study of Best Allocation Plan IEEE 33-nodes test feeder in a distribution network
of EVCS Parameter setting of simulation

Discussions

Fig. 2. The research process of optimal allocation of EVCS.

Fig. 3. Hierarchical structure of EVCS energy management system.

Fig. 4. Examples of positive and negative feedback loop.

dynamic conduction mechanism briefly. minimum difference in threshold of the two consecutive iterations
[23]. The calculation flow of k-means algorithm is shown in Fig. 5.
Due to the advantage of high efficiency, k-means algorithm will
3.3. K-means clustering method be employed to quickly determine the numbers of EVCSs. To solve
the issue of unclear initial clustering numbers, a range of possible
The KM algorithm was first explored by MacQueen J. [22]. Ac- clustering numbers of EVCS according to the charging demand will
cording the Euclidean distance, the data are clustered in several be set in order to find the best division of clusters. Meanwhile, k-
classes. The data in each cluster have common features but means algorithm must have an objective function which should be
different from other clusters. It starts with an initial clustering and a minimization problem. In previous study, distance was usually
defines initial clusters, then iterates the approach until reach a final
564 J.-p. Liu et al. / Energy 164 (2018) 560e574

Fig. 5. The calculation flow of k-means algorithm.

used to describe the minimization function between the centroids will go down following with the Increasing distance.
and the points around them. Considering with the charging de-
mand, a correlation function between distance and demand was 3.5. Particle swarm optimization (PSO) model
constructed, i.e. a distribution function of charging satisfaction
degree, which will be introduced as follows. In the construction process of optimal allocation model, it can be
found that the objective functions are nonlinear. For calculating and
3.4. Satisfaction distribution function measuring the best capacity of energy storage system (ESS) in EVCS,
lots of advanced simulation algorithms were used to explore the
Satisfaction distribution function is built to reflect the charging similar problem. Among these methods, Particle Swarm Optimi-
convenience of EV users. According to the EV charging demand, the zation (PSO) has the advantages of strong adaptability, less calcu-
satisfaction degree is corresponding to the distance from demand lation and high accuracy which can deal with the multi-objective
location to the site of EVCS. Thus, meeting the demand differences problems in complex system [25]. To optimal allocation model, the
of different charging points is also an important part in the allo- calculation flow of PSO is shown in Fig. 7. With different constraint
cation optimization model of EVCS. conditions, PSO will be employed to calculate the optimal energy
Besides the geographic information of EVCS, the acceptance storage capacity of EVCS. The hybrid optimization algorithm of GA-
degree of EV user to the EVCS location should be taken into account PSO will be specifically introduced in section 6.
as well [24]. In a real-life scenario, the charging satisfaction degree
will be promoted influencing by the shortening of the distance 4. Conduction mechanism of factors influential to EVCS
between the demand point to the site of EVCS. Following this rule, allocation
EV users prefer the location of EVCS closer to themselves. So the
charging satisfaction degree can be described by probability dis- Based on the cause and effect feedback features in the system,
tribution function, such as the curve in Fig. 6. Its probability dis- the method of System Dynamics (SD) is good at finding the source
tribution function is set as equation (1). of the problem from the inner structure. In the allocation model of
EVCS, different load characters will lead to the adjustment of site

8
> 1 ; dab  La
>
>

 ><1 1 
p

Ua þ La

p

F dab ¼ þ cos dab  þ ; La  dab  Ua (1)
>
> 2 2 Ua  La 2 2
>
>
:
0 ; dab  Ua

In equation (1), dab is the distance between charging point a to selection and capacity allocation of EVCS. Meanwhile, economy,
EVCS b. F(dab) is the satisfaction degree of a to b. La and Ua are the transportation, land, population will influence the EVCS allocation
critical value of satisfaction degree. When dab  La, the satisfaction model as well. From the energy storage perspective of EVCS, energy
degree is 100%. Ua is the smallest distance when the satisfaction structure, electricity consumption during peak and valley periods,
degree equals 0. Among the range of [La, Ua], the satisfaction degree incentive policy, scheduling strategy and etc. will have effects on
 J.-p. Liu et al. / Energy 164 (2018) 560e574 565

Energy Subsystem User Subsystem

Electricity during Peak

Charging habits
and Valley Periods
Energy Structure
Income Level

Scheduling Subsystem Society Subsystem

Charging Demand Subsidy

Period Division of Power Land Use
Supply
Fig. 6. The curve of charging satisfaction degree.

Fig. 8. The system framework of influencing factors to EVCS.

EVCS planning. Moreover, from the perspective of site selection of

EVCS, charging convenience, charging habits, price mechanism,
income level of EV users and etc. will also impact the EVCS allo-
cation. Thus, the influencing factors of EVCS allocation should be
integrated into a comprehensive system in order to explore the key
components.
According to the division principle of system boundary, the
factors influential to the EVCS can be classified into four subsystem,
i.e. energy subsystem, user subsystem, scheduling subsystem and
society subsystem. The system framework of influencing factors to
EVCS is shown in Fig. 8.
By the framework construction, the interactive relationships
between the factors in the four subsystem can be seen in general.
For example, the transformation of energy structure and the up-
grade of transportation system will increase the ownerships of EV.
Then, the increment of EVs will cause more power consumption
and further influence the optimization of power grid planning and
scheduling. According to such path between factors, a conduction
mechanism of factors influential to EVCS allocation is built based on
the divided framework (see Fig. 9).
In Fig. 8, there are several feedback paths in the conduction
mechanism. Compared with a single influencing path in other
studies, focusing on the factor of “site selection of charging station”,
two cause and effect conduction paths were found in this con-
duction mechanism. They are shown as follows.

(1) Site selection of charging station/(þ) Acceptance degree of

EV users to charging station distribution/(þ) Charging and
discharging behaviors of EV/(þ) Load consumption of
electricity /(þ) Electricity sales/(þ) Generation quantity
of renewable connected to the power grid/(þ) Voltage
fluctuations/() Safety and stability evaluation on power
grid/(þ) Subsystem capacity of energy storage/(þ) Ca-
pacity determination of charging station/(þ) Distribution
of charging station/(þ) Site selection of charging station
(2) Site selection of charging station/(þ) Acceptance degree of
EV users to charging station distribution/(þ) Charging and
discharging behaviors of EV/() Electricity tariff/(þ)
Scheduling ability of power grid/(þ) Safety and stability
evaluation on power grid/(þ) Planning of energy storage
system/(þ) Subsystem distribution of energy storage/(þ)
Site selection of charging station

From these two cause and effect conduction paths, “Distribution

Fig. 7. The calculation flow of PSO.
566 J.-p. Liu et al. / Energy 164 (2018) 560e574

Fig. 9. Conduction mechanism of factors influential to EVCS allocation.

of charging station” and “Planning of energy storage system” play evaluated by probability distribution function. With the parameters
important roles in the conduction mechanism. So they should be of optimal site selection and charging satisfaction, capacity alloca-
taken a serious consideration in EVCS allocation model. In the next tion optimization model of EVCS will be constructed through Par-
section, the site selection model will be built by k-means method ticle Swarm Optimization (PSO) method.
firstly. Then, the acceptance degree of EV users to EVCS will be

Fig. 10. Bubble diagram of charging demand locations and their EV quantities.
 J.-p. Liu et al. / Energy 164 (2018) 560e574 567

5. Charging satisfaction degree analysis satisfaction degree in the whole area can be integrated as shown in
equation (3).
5.1. Distribution of charging demand
X Nb

FTotal ¼ Fb  (3)
Considering with the distribution of charging demand, 50 points NTotal
were generated randomly in an area of 50 km  50 km. Meanwhile,
In equation (3), FTotal is the total satisfaction degree in the whole
the EV quantity in every demand point were also set. The distri-
area. NTotal is the total EV ownership in the area. Based on the
bution of EV charging demand is shown by bubble diagram in
calculation method from equations (1)e(3), and setting the dis-
Fig. 10. In Fig. 10, the charging demand points can be located by x-
tance range as [2,15], i.e. La ¼ 2 and Ua ¼ 15, the total satisfaction
axis and y-axis and their EV quantities are represented by the
degrees with different clusters are shown in Fig. 12.
bubble areas.

5.2. Clustering analysis by k-means 5.4. Limits caused by distribution network

Based on the calculation flow of k-means algorithm in Fig. 5, the From Fig. 12, it obviously shows that the charging satisfaction
clustering results of charging demand points were simulated from degree has been significantly enhanced when the number of EVCS
2 clusters to 7 clusters. With the help of SPSS software, the itera- continuously increases. However, it's impossible to increase the
tions were set as 50 times. Convergent standard was set as 0. In the EVCS number with no limits. Besides the construction cost of EVCS,
clustering process, the outliers will be handled and the clusters will the constraint conditions include the voltage fluctuation, load
be optimized by average distance in each circulation until the fluctuation, system power flow, and so on [26]. Thus, the voltage
clusters are not changing. The clustering results are shown in fluctuation and load fluctuation of distribution network will be
Fig. 11. mainly taken account into the objective function after the
connection of EVCSs. IEEE 33-node test feeder will be adopted to
explore the optimal allocation plan of EVCS. The structure of test
5.3. Satisfaction degree of EV charging
feeder is shown in Fig. 13.
Based on the clustering results in Fig. 11, the distances between
each charging demand point and its centroid can be measured. 6. Optimal EVCS allocation model construction
Weighting by the charging demand in each point, the satisfaction
degree in one cluster can be calculated via equation (2). 6.1. Objective function
" #
X
 N When renewables connect to the power grid, their unstable
ab
Fb ¼ F dab  (2) output will cause the voltage and load fluctuations. Take EVCS as an
N b
energy storage system, the power quality will be improved. The
In equation (2), Fb stands for the satisfaction degree in the bth participation of multiple electric power bodies makes the operation
cluster. F(dab) is the satisfaction degree of the ath charging demand of distribution network more complicated. Thus, the voltage fluc-
point to its centroid in the bth cluster. Nab is the EV ownership in tuation, load fluctuation, and the capacity of energy storage system
the ath demand point and Nb is the EV ownership in the bth cluster. will be considered comprehensively in the process of multi-
Taking with the satisfaction degree of each cluster, the total objective function building.

Fig. 11. Clustering results with different clusters.

568 J.-p. Liu et al. / Energy 164 (2018) 560e574

Fig. 12. Total satisfaction degree influenced by different clusters.

In equation (5), PS(j) is the load power of distribution network at

the jth time. ddP represents the average load power during the
period of [1, T].

(3) Capacity of energy storage system

More energy storage system has better regulation performance,

Fig. 13. IEEE 33-node test feeder. but its cost should be properly considered in the EVCS planning.
Small capacity means the relative low cost. Thus, the total capacity
of the energy storage system will be also taken as an objective in
(1) Voltage fluctuation order to measure the rated capacity of each EVCS [27,28]. The total
rated capacity of energy storage system can be calculated via
Node voltage stability in distribution network is one of the equation (6).
important indicators in power quality evaluation. When renew-
ables connect to the distribution network, the power voltage will be X
N X
store t0 þDt
b
fluctuate more extremely. The sum of the voltage fluctuation in f3 ¼ Pstore ðnÞ$Dt (6)
distribution network can be measured via equation (4). b¼1 n¼t0

In equation (6), Nstore is the number of energy storage system or

X
N T 
X  EVCS. t0 is the initial time of charging or discharging. Dt is the in-
 
node

f1 ¼ Vij  Vi  (4) b
terval between the initial time and end time. Pstore ðnÞ is the dis-
i¼1 j¼1 charging or charging power at the nth time.
In equation (4), Nnode stands for the number of node in distri- In view of the voltage fluctuation, load fluctuation, and the ca-
bution network. T is the number of times. Vij is the power voltage of pacity of energy storage system, the multi-objective function can be
the ith node at the jth time. VðddÞi is the average value of the ith described as equation (7).
node during the period of [1, T].
minF ¼ ½f1 ; f2 ; f3  (7)
(2) Load fluctuation
6.2. Constraints
With the integration of renewables, the local load curve will be
Besides the operation indicators in distribution network, the
also influenced by their fluctuation. Meanwhile, when EVCS is
power balance, voltage magnitude, power limit should be taken as
added to the network, its energy storage system can effectively
constraints. Specific constraints are shown as follows.
smooth the load fluctuation. Thus, the load fluctuation should be
taken into account and calculated via equation (5).
(1) System power balance constraint

X
T
2 X
Nnode X
NDG X
N
f2 ¼ PS ðjÞ  P (5) PS ðjÞ ¼ Pload ði; jÞ  PDG ðm; jÞ  Pstore ðb; jÞ (8)
j¼1 i¼1 m¼1 b¼1
 J.-p. Liu et al. / Energy 164 (2018) 560e574 569

In equation (8), Pload(i,j) is the load value of node i at the jth time. to Step 5.
NDG stands for the number of connected distributed generation Step 3: Selecting the random number rid for particle i in each
system. PDG(m,j) is the output power of the mth distributed gen- dimension of position vector. If rid < pm, initialization should be
eration system at the jth time. Pstore(b,j) is the output power of the operated on the position vector in the dth dimension of particle i.
bth energy storage system at the jth time. Step 4: Crossover treatment on the position vector of mutant
particle. If rid < pc, crossover should be taken on the dth dimension
(2) Node voltage constraint of position vector. The crossover object is the global optimization
solution.
In the process of EVCS allocation, the power voltage of distri- Step 5: Finishing the crossover and mutation. At the end of the
bution network should keep at a regulated range. iterations, the final state of position vector is shown in equation
(14).
Vmin  Vij  Vmax (9)
xid ¼ xmin þ ðxmax  xmin Þ$r (14)
In equation (9), Vmin and Vmax stand for the upper limit and
lower limit of the node voltage respectively in the whole system. In equation (14), xid is the final state of the position vector. xmin
and xmax are the minimum and the maximum value of the position
(3) Power constraint of energy storage vector. r is a random number between 0 and 1.

Pstore min  Pstore  Pstore max (10)

7. Case study
In equation (10), Pstore_min and Pstore_max are the minimum and
maximum power of energy storage system. 7.1. Parameter setting

(4) Power balance constraint of energy storage system Before the simulation by using GA-PSO algorithm, the parame-
ters and the assumptions should be set at first. The output curve of
X
T
b
renewables and the load curve are shown in Fig. 14 and Fig. 15.
Pstore ðjÞ$Dt ¼ 0 (11) To simplify the simulation process, 2 solar power sources and 2
j¼1 wind power sources were assumed to connect to the system.
In equation (11), the bth energy storage system should keep Considering with the resource features in a relatively small area,
power balance at the jth time. the outputs of wind power units were assumed to be the same, so
does as the solar power array. 200 kW was set as the rated power to
each power generation unit. Their output power curves are shown
6.3. GA-PSO algorithm in Fig. 14. At the beginning of simulation, the distributed renew-
ables will be randomly connected to the power network in order to
PSO is a kind of random optimization algorithm which has find the best connection nodes. Based on the IEEE 33-nodes test
better performance for continuous and infinite spaces. In the iter- feeder, the system parameters were borrowed from the literature
ation process, each particle update its speed and position through [30]. The total load of the distribution network is 3715 kW þ j2300
tracking two optimal solution. Its update approach is shown in kvar as shown in Fig. 15. The rated voltage of the whole system is
equations (12) and (13). 12.66 kV and the threshold of the node voltage is set from 0.98pu～


 1.05pu. The available connection nodes are from 2 to 33. The
ðkþ1Þ ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ
vid ¼ wvid þ c1 r1 pid  xid þ c2 r2 gd  xid (12) allowed connection power of each energy storage system was set as
400 kW.
ðkþ1Þ ðkÞ ðkþ1Þ With the parameters setting in distribution network, the pa-
xid ¼ xid þ vid (13) rameters in the simulation model will be assigned then as shown in
In equation (12), w stand for the inertia weight. Acceleration Table 1.
factors are represented by c1 and c2. r1 and r2 are random numbers
ðkÞ
among the range of [0,1]. pid is the dth-dimensional component of 7.2. Main results
ðkÞ
the ith particle in the best position vector at the time k. gd is the
dth-dimensional component of population in the best position After the parameters setting, GA-PSO was employed to simulate
vector at the time k. the search process for the global best solution. The main results are
When solving the problem of multi-objectives, PSO usually de- as follows.
termines the historical optimization solution and updates the Par-
eto optimal set through the Pareto sorting mechanism and the (1) Allocation plan of EVCS. The global best solution of the multi-
dominance relationship between particles. However, in conven- objective problem is [1.0227, 0.9629, 1.3933]. When 5 EVCSs
tional PSO algorithm, rapid loss of population diversity may make connect to the distribution network, the results can meet the
the iterations run into the local optimization solution, instead of requirements of constraints. Their site locations in the
finding out the global optimization solution. Thus, the crossover 50  50 area are [18.76, 43.95], [10.43, 9.99], [41.54, 36.84],
and mutation method from genetic algorithm (GA) was employed [13.14, 25.81] and [37.53, 12.57]. The total satisfaction for the
into the PSO algorithm [29]. Specific steps are as follows. charging distance has reached to 82.3%. The connection
Step 1: Determining the difference value X between the particle nodes of EVCS are node 2, node 8, node 11, node 23 and node
position vector and the global optimization solution of the popu- 29. Their storage capacities are 265.46 kWh, 220.98 kWh,
lation. Then, setting the threshold Xmin of X. Meanwhile, the 246.80 kWh, 338.64 kWh and 321.42 kWh respectively. The
crossover rate (pc) and mutation rate (pm) should be also assumed connection nodes of solar power are node 7 and node 8. The
ahead. connection nodes of wind power are node 25 and node 32.
Step 2: Judging the size of the particle i. If Xi < Xmin, crossover The distribution of Pareto solutions and the node voltage of
and mutation treatment should be taken on the particle i. If not, go best solution are shown in Figs. 16 and 17.
570 J.-p. Liu et al. / Energy 164 (2018) 560e574

Fig. 14. Output power curve of renewables.

Fig. 15. Local load curve.

Table 1 (2) Optimized performance by EVCS. With the result of best

Values assigned to the simulation parameters. Pareto solution, the system operation indicators in distribu-
tion network have been improved in varying degrees. In
Type Value
Fig. 18, after the connections of renewables and EVCSs, the
Times of iteration 200
load curve of the system has been improved that the gap
Number of population 100
Initial value of inertia weight 0.9
between peak and valley load has been reduced 20%. The
End value of inertia weight 0.4 new maximum and minimum load are 3598 kW and
Threshold of the difference value 0.1 2357 kW. The charging and discharging performances of
Mutation rate 0.05 each energy storage system in EVCS are displayed in Fig. 19.
Crossover rate 0.1
Scale of Pareto solution set 100
 J.-p. Liu et al. / Energy 164 (2018) 560e574 571

different algorithms are shown in Fig. 20.

As the convergence results shown in Fig. 20, PSO algorithm is
easy to run into the local best solution. Due to a larger number of
population, the searching effectiveness has a better performance
than PSO. Compared to traditional algorithms, FNN has the ad-
vantages of high fault tolerance, strong robustness, good adaptive
ability and etc. However, when lack of data, FNN cannot explain the
best solution well. In this case study, GA-PSO, a hybrid algorithm
has the best convergence performance. After 39 iterations, GA-PSO
found out the global best solution, and its calculation efficiency has
increased extremely than other employed algorithms. With no
doubt, more smart and effective techniques will improve the
optimization algorithm which will be deeply explored in further
research.

8. Conclusions and advices

As an important energy storage resource, EVCS has positive ef-

fects on the operation of electric power system. However, there are
many factors influential to the optimization allocation of EVCS. In
this paper, by using SD model, two cause-effect conduction paths
were identified in the conduction mechanism which are mainly
Fig. 16. The distribution of Pareto solutions. driven by the key factors of “Distribution of charging station” and
“Planning of energy storage system”. For making the EVCS alloca-
tion plan more rational, charging satisfaction degree and distrib-
7.3. Effectiveness examination uted renewables integration were mainly considered into the
hybrid model construction. By charging demand and k-means
In the simulation, GA-PSO was adopted to calculate the best model, it is found that, though the charging satisfaction degree is
solution. In order to exam the high effectiveness of GA-PSO, tradi- significantly enhanced when the numbers and capacity of EVCS
tional genetic algorithm (GA), particle swarm optimization (PSO) continuously increase, it is still imperative to take the operation
and fuzzy neural network (FNN) were employed as the comparison index changes of power system caused by EVCS connections into
methods [31]. Under 200 iterations, take the objective function (f1) the EVCS planning. Meanwhile, the access of renewables will also
of node voltage as an example, the convergence performances of influence the stability of the distribution network. Thus, a multi-

Fig. 17. The node voltages of best solution.

572 J.-p. Liu et al. / Energy 164 (2018) 560e574

Fig. 18. Load optimization after the connections of renewables and EVCSs.

Fig. 19. Charging and discharging performance of each EVCS at different times.
 J.-p. Liu et al. / Energy 164 (2018) 560e574 573

Fig. 20. Convergence performances of different algorithms.

objective function was developed to minimize the voltage fluctu- affect the planning and construction numbers of EVCS which
ations (f1), load fluctuations (f2) and capacity of energy storage affects the charging satisfaction degree further. For EV users,
system in EVCS (f3). Based on IEEE 33-nodes test feeder, GA-PSO proper finance subsidy to infrastructure can reduce the use-
was utilized to solve the multi-objective optimization problem. cost in some extent. The good news is the central govern-
The best solution calculated the proper allocation numbers of EVCS, ment plans to deduct the capacity tariff for EVCS in order to
the access nodes of distributed renewables and EVCSs, the con- enlarge the investment and construction scales of EVCS. The
nected capacity of each EVCS. The allocation plan of EVCS make the policy will make EVCS to provide better services for EV users.
power system have a better performance than before. The feedback of users will make the EVCS allocation more
The analysis result reveals that the optimization allocation of optimal.
EVCS is a complicated problem which relates to many factors and (3) Land use supports for EVCS. In the generalization process of
needs more supports from different aspects and participants. In EVs, land use is an important influencing factor for the site
China, the central government proposed the Development Guid- selection of EVCS. The flexible EVCS site selection supported
ance for Electric Vehicle Charging Infrastructure. By the end of by land use will accelerate the speed of generalization. For
2020, 12 thousand EVCSs and 4.8 million charging piles were example, government can provide differentiated land use
planned to construct. However, a large-scale connection of EVCSs policies for EVCS and allow the holder of the right to coop-
poses huge challenges for the rational planning of supporting fa- erate with other market entities to use the land. Meanwhile,
cilities and stable operation of the power system. Thus, some ad- government should encourage the charging infrastructure
vices will be given in order to expand the utilization scale of electric projects to obtain the easement in accordance with law and
vehicles and form a healthy development mechanism. develop the construction cooperation with social capital in
order to break the land limit of EVCS site selection.
(1) Formulate uniform standards for product and technology. (4) More accurate energy management level. Lots of factors
Uniform standards are the premise of mass EV use. They can affect the load of EVCS, such as the EV ownership, charging
improve the substitutability of EV batteries and the mode and habit, battery performance, power scheduling and
compatibility and expansibility of power grid to EVs. In etc. Thus, energy management technology, such as fore-
China, enterprise cooperation on technical standards has casting techniques and smart algorithm, should be continu-
been gradually carried out. Joint with Shanghai Automotive ously improved. Accuracy promotions of energy
Industry Corporation (SAIC), State Grid Corporation of China management is helpful to make safe and reliable power
(SGCC) plans to research and implement the standardized scheduling, increase the integration rate of renewables and
V2G technology in order to cope with the emerging changes optimize the emission reduction effects of EVs. In the future,
of power demand. Meanwhile, technology standard system energy management of EVCS is also an important research
and special committee have been established in recent years. field which can help the distribution network and EVCS
As the analysis results suggest, uniform standards can help planners to make the optimal decisions.
more EVCS to realize the power interconnection in multiple
regions.
(2) Finance subsidy for EV infrastructure. On the early phase of
Author contributions
EV development, the construction and operation costs of
EVCS are expensive for investors. Compared to subsidy for
Jin-peng Liu proposed the concept of this research and
EVs, the supports for EV infrastructure are insufficient. It may
completed the manuscript. Teng-xi Zhang and Jiang Zhu analyzed
574 J.-p. Liu et al. / Energy 164 (2018) 560e574

the empirical data. Tian-nan Ma gave some suggestions. J Power Sources 2018;390:20e9.
[15] Weige Zhang, Di Zhang, Biqiang Mu, Leyi Wang, Yan Bao, Jiuchun Jiang,
Morais Hugo. Decentralized electric vehicle charging strategies for reduced
Conflicts of interest load variation and guaranteed charge completion in regional distribution
grids. Energies 2017;10(2):147.
The authors declare no conflict of interest. [16] Jiancheng Ye, Yuling Li, Dongliang Zhang, Xiangjing Zhu, Jinda Zhu. Load
forecasting of charging and swapping in large-scale electric vehicle. Appl
Mech Mater 2014;568e570:1969e77.
Acknowledgments
n J, Koivunen V. Reinforcement learning-based plug-in electric
[17] Chiş A, Lunde
vehicle charging with forecasted price. IEEE Trans Veh Technol 2017;66(5):
3674e84.
This study is supported by the National Natural Science Foun- [18] Yunyan Li, Yuansheng Huang, Meimei Zhang. Short-term load forecasting for
dation of China (NSFC) (71501071), Beijing Social Science Fund electric vehicle charging station based on niche immunity lion algorithm and
(16YJC064) and the Fundamental Research Funds for the Central convolutional neural network. Energies 2018;11(5):1253.
[19] Natural Resources Defense Council (NRDC), Energy Research Institute of Na-
Universities (2017MS059, 2018ZD14). tional Development and Reform Commission (CRI, NDRC), China Energy
Storage Alliance (CNESA). Technical potentials and economics of energy
References storage from electric vehicles. Available from: http://nrdc.cn/Public/uploads/
2018-02-07/5a7a73d92989d.pdf.
[20] Xingang Zhao, Yuzhuo Zhang, Yanbin Li. The evolution of renewable energy
[1] International Renewable Energy Agency (IRENA). Global Energy Trans-
price policies based on improved bass model: a system dynamics (SD) anal-
formation: A Roadmap to 2050. Available from: https://www.irena.org/
ysis. Sustainability 2018;10(6):1748.
publications/2018/Apr/Global-Energy-Transition-A-Roadmap-to-2050.
[21] Coyle RG. System dynamics modelling: a practical approach. J Oper Res Soc
[2] National Energy Administration (NEA). National power industry statistics and
1996;48(5):544. 544.
electricity consumption in 2017. Available from: http://www.nea.gov.cn/
[22] MacQueen J. Some methods for classification and analysis of multivariate
2018-01/22/c_136914154.htm.
observations. In: Proceedings of the fifth berkeley symposium on mathe-
[3] Electric Vehicle World Sales Database. Global Plug-in Vehicle Sales & Share.
matical statistics and probability. Berkeley, CA, USA: University of California
Available from: http://www.ev-volumes.com/country/total-world-plug-in-
Press; 1967. p. 281e97.
vehicle-volumes/.
[23] Liu Hao, Mahmoudi Nadali, Chen Kui. Microgrids real-time pricing based on
[4] U.S. Energy Information Administration (EIA). Plug-in electric vehicles: future
clustering techniques. Energies 2018;11(6):1388.
market conditions and adoption rates. Available from, https://www.eia.gov/
[24] Zhao Jie, Fang Suping, Jin Peiquan. Modeling and quantifying user acceptance
outlooks/ieo/pev.php.
of personalized business modes based on TAM, trust and attitude. Sustain-
[5] China Electric Vehicle Charging Infrastructure Promotion Alliance (EVCIPA).
ability 2018;10(2):356.
Annual Report of China EV Charging Infrastructure Development. Available
[25] Mozafar Mostafa Rezaei, Mohammad H, Moradi M, Amini Hadi.
from: http://www.evcipa.org.cn/.
A simultaneous approach for optimal allocation of renewable energy sources
[6] Wang Hengsong, Huang Qi, Zhang Changhua, Xia Aihua. A novel approach for
and electric vehicle charging stations in smart grids based on improved GA-
the layout of electric vehicle charging station. In: International conference on
PSO algorithm. Sustain Cities Soc 2017;32:627e37.
apperceiving computing and intelligence analysis (ICACIA); 2010. p. 64e70.
[26] Aljanad Ahmed, Mohamed Azah, Shareef Hussain, Khatib Tamer. A novel
[7] Zifa Liu, Wei Zhang, Xing Ji, Ke Li. Optimal Planning of charging station for
method for optimal placement of vehicle-to-grid charging stations in distri-
electric vehicle based on particle swarm optimization. In: Conference on IEEE
bution power system using a quantum binary lightning search algorithm.
PES innovative smart grid technologies; 2012. p. 1e5.
Sustain Cities Soc 2018;38:174e83.
[8] Sadeghi Barzani P, Rajabi Ghahnavieh A, Kazemi Karegar H. Optimal fast
[27] Wang Yongli, Huang Yujing, Wang Yudong, Yu Haiyang, Li Ruiwen,
charging station placing and sizing. Appl Energy 2014;125(2):289e99.
Song Shanshan. Energy management for smart multi-energy complementary
[9] Habiballah Kheradmand K, Gitizadeh Mohsen. Well-being analysis of distri-
micro-grid in the presence of demand response. Energies 2018;11(4):974.
bution network in the presence of electric vehicles. Energy 2018;155:610e9.
[28] Veras Jaclason M, Silva Igor Rafael S, Pinheiro Pl
^lo Ricardo AL.
acido R, Rabe
[10] Lund H, Kempton W. Integration of renewable energy into the transport and
Towards the handling demand response optimization model for home ap-
electricity sectors through V2G. Energy Policy 2008;36:3578e87.
pliances. Sustainability 2018;10(3):616.
[11] Sachan Sulabh, Adnan Nadia. Stochastic charging of electric vehicles in smart
[29] Sawle Yashwant, Gupta SC, Bohre Aashish Kumar. Optimal sizing of stand-
power distribution grids. Sustain Cities Soc 2018;40:91e100.
alone PV/Wind/Biomass hybrid energy system using GA and PSO optimization
[12] Raoofat M, Saad M, Lefebvre S, Asber D, Mehrjedri H, Lenoir L. Wind power
technique. Energy Procedia 2017;117:690e8.
smoothing using demand response of electric vehicles. Int J Electr Power
[30] Goswami SK, Basu SK. A new algorithm for the reconfiguration of distribution
Energy Syst 2018;99:164e74.
feeders for loss minimization. IEEE Trans Power Deliv 1992;7(3):1481e91.
[13] Rahimi Kaveh, Davoudi Masoud. Electric vehicles for improving resilience of
[31] Lin Whei-Min, Hong Chih-Ming, Cheng Fu-Sheng. Fuzzy neural network
distribution systems. Sustain Cities Soc 2018;36:246e56.
output maximization control for sensorless wind energy conversion system.
[14] Gnann Till, Klingler Anna-Lena, Kühnbach Matthias. The load shift potential of
Energy 2010;35(2):592e601.
plug-in electric vehicles with different amounts of charging infrastructure.