ITEC Asia-Pacific 2014 1569951429

Planning of Electric Vehicle Charging Station

Based on Hierarchic Genetic Algorithm
Xiangwu Yan, Member, IEEE, Cong Duan, Xiao Chen, Zhengyang Duan
State Key Laboratory of Alternate Electrical Power System with Renewable Energy Sources (North
China Electric Power University), Changping District, Beijing 102206, China
E-mail: xiangwuy@163.com, duancongysu@126.com

Abstract- With an increasing number of Electric Vehicles (EV), basis of branch-chain [7] is adopted in this paper, which can
the optimal planning of electric vehicle charging station (EVCS) effectively make up for the deficiency of both above method.
into distribution system has become more and more important.
However, the problem of optimal planning is complex, nonlinear In addition check operator is introduced, changing infeasible
and combinatorial optimization problem. Thus a multi-objective individuals into feasible ones and avoiding ring network or
and multivariate planning model based on hierarchic genetic island net.According to the general principles of distribution
algorithm (HGA) is proposed, considering the investment costs system planning, combined with the characteristics of the
of EVCS and feeder and the energy losses and constraint EVCS, this paper puts forward a novel method of HGA based
conditions. The innovation points of this paper are encoding
method and checking operator, changing the infeasible solutions on the mathematical model of optimal planning.
in filial populations into feasible solutions. Finally the proposed II. THE MATHEMATICAL MODEL OF EVCS PLANNING
approach is tested on 33 nodes distribution power system of
IEEE with EVCS, based on the MATLAB programming
language. Study results indicate the validity of this method. According to [5,6], the objective function is obtained by
considering the investment costs of EVCS and feeder and the
Keywords- electric vehicle charging station (EVCS); optimal
energy losses, in addition the capacity limits and network
planning; distribution system; hierarchic genetic algorithm
(HGA); encoding method ; checking operator operating conditions should be satisfied. The mathematical
model required in HGA is as follows:
I. INTRODUCTION A. Functions:
Defined Finner as the objective function in the inner layer
Faced with the increasing oil price and environmental which is obtained by equation 1 as follows:
problems , energy conservation and emissions reduction have
become a significant issue. The widely use of EV instead of Finner =CL +Closs (1)
fossil-fueled ones is the most promising way to tackle the Where CL means the investment of feeder, which is
problem. As the number of EV increases, the deployment of obtained by equation 2 as follows:
EVCS in the distribution system has become an issue yet to nL
be solved by the transportation and power supply department.
A lot of study efforts have been devoted to the charging
=CL ∑ y (α
i =1
i Lf C Lfi + C Lmi ) (2)
modes of EV[1,2], principle of site selection of EVCS[3], and
the development situation of EV charging facilities[4]. It And Closs means the energy loss, which is obtained
proposd an optimal model of regional sizing and locating of equation 3 as follows:
nL
EVCS, based on particle swarm algorithm[5] and genetic
algorithm[6].However, only a few research efforts focus on Closs = ∑ peτ max i Plossi (3)
i =1
quantitative analysis combined with distribution system
planning.There are several kinds of planning methods such as
Pi 2 + Qi2
enumeration method[7], artificial intelligence algorithm (i.e., where Plossi =yi ri (4)
genetic and immune algorithm[6,8-13] and particle swarm U i2
optimization algorithm[5,14], etc.) and the hybrid algorithm[15].
As intelligent optimization method, hierarchic genetic Defined Finnerbest as the minimum objective function in the
algorithm (HGA) [8] introduces natural selection and genetic inner layer,which is obtained by genetic ordinal optimization
mechanism into the mathematical theory, with many suitable algorithm.
characteristics for solving the problem of optimal planning of Defined CC as the investment costs of EVCS,which is
EVCS and distribution system. While how to encode is the obtained by equation 4 as follows:
nC
key to intelligence algorithm. In general there are three kinds
of coding method. The first method chooses the combination =CC ∑ x (α
i =1
i Cf CCf + CCm + α s s pu plandi ) (5)
of switch states of all feeders as variable, with low efficiency
and large solution space. The second method is based on loop
[13]
, with poor level of automation. The last method on the

1
 Finally, the objective function of each individual in outer
layer is obtained by equation 1 to 5, which is shown in
equation 6 as follows:
Fouter = CC + Finnerbest
Moreover, the constraint of the capacity limits and grid
structure should be satisfied. The constraint is given in
equation 7 as follows:
 si ≥ Sneedi ,i ∈ nc

 Pi ≤ Pimax ,i ∈ nL (7)
U
 imin ≤ U i ≤ U imax，i ∈ nN
B. Parameters:
nN The total number of candidated and existing feeders
nL The total number of candidate EVCS
nC The total number of nodes
αCf The average coefficient of construction cost of a set of
charging equipment
αLf The average coefficient of construction cost offeeder
αs The average coefficient of land price of EVCS
CCf Construction cost of EVCS i，the value of whom is
determined by the level of EVCS i
CCm Maintenance cost of EVCS i ，the value of whom is
determined by the level of EVCS i
plandi The land price of EVCS i ，the value of whom is
determined by the level of EVCS i
si The number of actual service capability of EVCS i，the
value of whom is determined by the level of EVCS i
pe The electricity price
τmaxi The maximum load loss hour of feeder i
CLfi Construction cost of feeder i
CLmi Maintenance cost of feeder i
ri Impedance of feeder i
Pi Active of feeder i
Pimax Maximum active of feeder i
Qi Reactive of feeder i
Ui Voltage
Uimin Maximum voltage of feeder i
Uimax Minimum voltage of feeder i
spu The occupied area of a set of charging equipment
Sneedi The number of needed service capability of EVCS
i，the value of whom is determined by necessary
C. Variables:
li The variable in outer layer which means the level of
EVCS i and determines the CCf, CCm, plandi, si
xi The variable in outer layer which means the number of
sets of charging equipment of candidated EVCS i,which is an
integer greater than or equal to zero(i.e. charging station is
not constructed).
yi The variable in inner layer which means the state of
feeder i, which is a 0-1 variable( 0 means the feeder i is open
or not constructed).
III. HIERARCHIC GENETIC ALGORITHM
A. The theory of hierarchic GA:
HGA is mainly composed of three parts of coding strategy,
(6)
genetic operators and tabu table in this section. While how to
encode and keep feasibility are important to intelligence
algorithm. The three parts in detail are as follows:
i Coding strategy
1)Outer layer: Four kinds of levels of charging station is
settled and shown in Table1, according to the standard file of
Technical specifications of electricity supply and assurance
for electric vehicle: electric vehicle charging station [16],
which is published in Beijing.
TABLE 1
INFORMATION OF CANDIDATE STATIONS
Level Service Construction Number Floor Required
Capability Cost [Ten of Space power
[Car/Day] Thousand Charger [m2] [MW]
Yuan ]
1 360 690 45 1085 0.39
2 240 520 30 693 0.26
3 100 310 15 337 0.13
4 60 210 8 165 0.07
In terms of the outer layer, the objective function is the sum
of investment of EVCS and optimal objective function from
the inner layer. While the number of sets and level of
candidate EVCS are considered as variable. So each
chromosome is divided into two parts by decimal coding
method. The first part means the level of every candidate
EVCS, while the second part means the number of sets. For
example, the form of one chromosome is [l1, l2, x1, x2] and
value is [1 3 3 4]. Then it means that there are two candidate
EVCS. And the level of candidate EVCS1 is 1 and the
number of sets of charging equipment is 3, while the level of
candidate EVCS2 is 3 and the number is 4.
2) Inner layer: This layer is on the basis of the level and
number of sets of each candidate EVCS are constant and all
swiches are closed. The object of this layer is to obtain the
optimal combination of switch states under what the total cost
of candidate feeders and energy losses is least. So the
combination of switch state of feeders is considered as
variable, which include each of the existing line and
candidate line.
The decimal coding method is adopted. Each chromosome
represents a combination of branches besides spanning-tree.
So the length of every chromosome is equal to the number of
branch-chains. Each gene in a chromosome represents a
branch-chain, which stores the branch serial number in this
branch-chain. The switch state of every chosen branch is
closed.
Take 33 nodes distribution power system of IEEE as
example which is shown in figure 1. There are 14 branch-
chains in total. Give serial number to every branch-chain. For
example, branch-chain 1 only includes branch T1. Then the
value of first gene can only be 0 or 1, which respectively
means the swich of branch T1 is open and closed.Branch
chain2 contains branch T3, T25 and T26. Then the value of

2
first gene can only be 0, 1, 2 or 3. While 0 means all the three iii Tabu table
branches are closed. And 1, 2 or 3 respectively means the In order to improve the search ability and avoid repeated
open branch is T3, T25 or T26. calculation, the tabu table is intrduced, which is expressed
19 20 21 22 T18 with matrix TT. TT is used to deposit a number of optimal
T3 T25 T26 T27 T12 individual and their objective function value.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
T1 T2 T4 T6 T7 T8 T10 T11 T13 T15 T16 T17 T19 T20 T21 T22 T23 B. Flow chart of HGA
T5
T9
T31 T32 T33 T34 T35 T36 T37
T14
T24
The flow chart of HGA is as follows:
T28 T29 26 27 28 29 30 31 32 33
T30
23 24 25
Select 1000 feasible solutions of
Fig. 1. 33 nodes distribution power system of IEEE inner layer randomly ,and calculate
begin individual fitness，get set θn
ii Genetic operators
1)Select operator: Select operator plays the role of Initail outer population,genouter=1 Judge the type of ordered performance
choosing excellent chromosomes which have the opportunity curve(OPC) and rank set θn ,get initail
inner population and maximum genetic
to pass to the next breeding. In order to choose excellent algebra geninnermax,geninner=1

chromosomes, a selection will be made among parent Calculate individual fitness，i.e. Fouter

individual and offspring according to rank-based fitness

Keep elite Calculation individual fitness，i.e. Finner
assignment. However, not all chromosomes are chosen, and
the proportion of selected chromosomes is known as select genouter=genouter+1 Genetic operator Keep elite
probability.
N
2)Crossover operator: Crossover plays the role of genouter>genoutermax Genetic operator
geninner=geninner+1
exchanging partial of two chosen chromosomes in such a Y
N
manner that the offspring generated.The proportion of End and output result geninner>geninnermax
Y
exchanging partial is known as crossover probability. End and output Finnerbest
3)Check operator: The infeasible individuals, which can’t Outer layer
satisfy the constraint of grid structure, may be generated after
selection and crossover operation. To solve this problem, the
feasible trees matrix Tnew based on topological analysis is Inner layer
used. According to graph theory, assuming that the number of Fig. 3. Flow chart of HGA
nodes is n and branches is b in a diagram, there are n−1 tree-
branches. Choose n−1 from all b lines as possible spanning- IV. CASE STUDY
tree. The total number of all combinations is Cbn-1, which are
stored in matrix T. Each row of T represents a kind of Take 33 nodes distribution power system of IEEE with
combination. There are n−1 elements (i.e., tree-branch) in EVCS as example which is shown in figure 4.
each row. Take a row from T as spanning-tree one by one. 19 20 21 22 T18
The row will be stored in matrix Tnew, if all the nodes can be T42 CS3
T43
T3 T25 T26 T27 T12
traversed (i.e., no isolated net) according to the tree-base. The 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
matrix Tnew is used to check every individual. T1 T2 T4 T6 T7 T8 T10 T11 T13 T15 T16 T17 T19 T20 T21 T22 T23
T38 T9 T14
Take a row from T T5 T39 T31 T32 T33 T34 T35 T36 T37
CS1
T40 T4133 T24
T28 T29 26 27 28 29 30 31 32
T30 CS2
Get spanning-tree 23 24 25

Fig. 4. 33 nodes distribution power system of IEEE with EVCS

Traverse node
The black nodes represent the nodes of power supply , load
NO and contact switches in figure 4. While the red nodes
if all the nodes
can be traversed represent the candidate electric vehicle charging stations. The
YES black solid lines represent close branches, and the black
stored in Tnew dotted lines represent disconnect branches. While the red
dotted lines represent candidate lines.
NO The data of 33 nodes distribution power system of IEEE is
If last row
given in appendix and other data needed in the planning of
YES electric vehicle charging station based on hierarchic genetic
End
algorithm are given as follows:
Fig. 2. Flow chart of searching for feasible trees Tnew

3
 TABLE 2 The total cost is 1.3749 ten million yuan. The final result
INFORMATION OF CANDIDATE LINES
after optimal planing based on HGA is shown in figure 6:
number Construction Cost Maintenance Cost
19 20 21 22
[Ten Thousand Yuan ] [Ten Thousand T18
Yuan ] T25 T26 T27
T42 CS3
T12
1 64 6.4 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
2 45 4.5
T1 T2 T4 T6 T7 T8 T11 T15 T16 T17 T19 T20 T21 T22 T23
3 52 5.2
T38 T9
4 45 4.5 T5 T31 T32 T33 T34 T35 T36 T37 T41
5 54 5.4 CS1
6 45 4.5 T28 T29 26 27 28 29 30 31 32 33 CS2
T30
TABLE 3 23 24 25
PARAMETERS OF THE AREA OF EACH CANDIDATE EVCS
Fig. 6. The final result after optimal planing based on HGA
number land price Number of Chargers Needed
1 0.4 350 V. CONCLUSIONS
2 0.3 250
3 0.2 300 The problem of optimal planning of EVCS and distribution
TABLE 4 system is complex, nonlinear and combinatorial optimization
PARAMETERS OF HGA problem, with many constraint conditions, which has the
Parameter value Parameter value characteristics of multi-objective function and multivariate.
αCf 0.177 Population 80 The improved method of HGA not only has many suitable
size
αLf 0.177 max gen 30
characteristics for solving this problem but also effectively
αs 0.177 select 0.9 solve the problem of blind search, low efficiency and so on in
probability traditional GA. The paper gives not only the mathematical
τmaxi 8760h crossover 0.2 model of optimal planning ,but also the flow chart of HGA
probability
pe 0.5 after brief introduction of the theory of HGA.
Yuan/kWh
ACKNOWLEDGMENT
A. The inner layer result
Assume that the level and number of each candidate EVCS This paper is supported by The Natural Science Fundings
are constant. For instance, the level and number of EVCS1 of Hebei (E2014502109) and The Fundamental Research
are 3 and 3, the level and number of EVCS2 are 2 and 2, the Funds for the CentralUniversities (2014ZD30)
level and number of EVCS3 are 2 and 2. While there are 6 REFERENCES
kinds of combinationes of candidate feeders. For instance, the
[1] [1] XIN Jianbo, WEN Yubin, LI Rui. Discussion on demand forecast
candidate feeders are branch 38, 41,42. The active loss of method for electric vehicle charging facilities. Jiangxi Electric Power,
every generation based on GA is shown in figure 5: vol. 34, pp. 1-5, Apr. 2010.
2.5
x 10
8
The active loss of every gen. based on GA [2] KANG Jiguang, WEI Zhenlin, CHENG Danming, XU Fan. Research
minimum active loss on electric vehicle charging mode and charging stations construction.
average active loss Power Demand Side Management, vol. 11, pp. 64-66, May. 2009.
2 [3] XU Fan, YU Guoqin, Gu Linfeng, ZHANG Hua. Tentative analysis of
layout of electrical vehicle charging stations. East China Electric Power,
vol. 37, pp. 1678-1682, Oct. 2009.
1.5
Ploss/W [4] LU Mang, ZHOU Xiaobing, ZHANG Wei. Research on development
of charging facilities for electric vehicles at home and abroad. Central
1 China Electric Power, vol. 23, pp. 16-30, Aug. 2010.
[5] Liu Zifa, Zhang Wei, Wang Lize. Optimal planning of charging station
for electric vehicle based on quantum PSO algorithm[J]. Proceedings of
0.5 the CSEE,2012,32(22):39-45(in Chinese)．
[6] Gao Ciwei, Zhang Liang, Xue Fei, Liu Hongchao. Grid planning
0
considering capacity and site of large-scale centralized charging
0 5 10 15 20 25 30 35 stations [J]. Proceedings of the CSEE,2012,32(7):40-46(in Chinese)．
gen
[7] Ou Dihong,Chen Haoyong,He Ruihui ． An optimal distribution
Fig. 5. Optimal active loss with GA network reconfiguration algorithm based on division of solution
space[J] ． Power System Technology,2012,36(3):140-145(in
B. The final result Chinese)．
The simulation results of EVCS planning into IEEE33 [8] Xiong Hao,Liu Qisheng,Huang Yanhao ． Research of hierarchic
genetic algorithm for power distribution system
power distribution system based on HGA is shown in table 5: reconfiguration[J] ． High Voltage Engineering,2004,30(3):56-59(in
TABLE 5 Chinese)．
THE FINAL RESULT [9] Carreno E M,Romero R,Padilha-Feltrin A．An efficient codification to
solve distribution network reconfiguration for loss reduction
Level and Level and Level and candidate branches with problem[J]．IEEE Trans on Power Systems,2008,23(4):1542-1551．
number of number of number of feeders closed switch [10] de Macêdo Braz H D,de Souza B A ． Distribution network
reconfiguration using genetic algorithms with sequential
EVCS1 EVCS2 EVCS3 encoding:subtractive and additive approaches[J] ． IEEE Trans on
3 and 3 2 and 2 2 and 2 38, 41,42. 10,13,14,24,32 Power Systems,2011,26(2):582-593．

4
[11] Bahadoorsingh S,Milanovic J V,et al．Minimization of voltage sag
costs by optimal reconfiguration of distribution network using genetic 12 13 1.4680+j1.1550 60+j35
algorithms[J] ． IEEE Trans on Power Delivery,2007,22(4):2271- 13 14 0.5416+j0.7129 120+j80
2278．
[12] Yu Jianming,Zhang Fan．Distribution network reconfiguration based 14 15 0.5910+j0.5260 60+j10
on improved genetic algorithm[J] ． Power
15 16 0.7463+j0.5450 60+j20
SystemTechnology,2009,33(19):100-105(in Chinese)．
[13] Wang Chaoxue,Li Changhua,Cui Duwu ． A novel immune genetic 16 17 1.2890+j1.7210 60+j20
algorithm for distribution network reconfiguration[J]．Power System
Technology,2008,32(13):25-29(in Chinese)． 17 18 0.3720+j0.5740 90+j40
[14] Zhao Jingjing,Li Xin,Peng Yi,et al ．A comprehensive optimization 2 19 0.1640+j0.1565 90+j40
algorithm for injection power of distributed generation and distribution
network reconfiguration based on particle swarm 19 20 1.5042+j1.3554 90+j40
optimization[J] ． Power System Technology,2009,33(17):162-166(in 20 21 0.4095+j0.4784 90+j40
Chinese)．
[15] Ren Jianwen,Guo Yutian．Reconfiguration of distribution network on 21 22 0.7089+j0.9373 90+j40
loop encoding and hybird intelligent algorithm[J]．Modern Electric
3 23 0.4512+j0.3083 90+j50
Power,2013,30(3):28-31(in Chinese)．
[16] Technical specifications of electricity supply and assurance for electric 23 24 0.8980+j0.7091 420+j200
vehicle: electric vehicle charging station, DB11/Z 728-2010, Aug, 2010.
[17] M. Young, The Technical Writer’s Handbook. Mill Valley, CA: 24 25 0.8960+j0.7011 420+j200
University Science, 1989. 6 26 0.2030+j0.1034 60+j25
26 27 0.2842+j0.1447 60+j25
APPENDIX 27 28 0.0590+j0.9337 60+j20

from to impedance Load of end point[kW] 28 29 0.8042+j0.7006 120+j70

29 30 0.5075+j0.2585 200+j600
1 2 0.0922+j0.047 100+j60 30 31 0.9744+j0.9630 150+j70
2 3 0.4930+j0.2511 90+j40 31 32 0.3105+j0.3619 210+j100
3 4 0.3660+j0.1864 120+j80 32 33 0.3410+j0.5362 60+j40
4 5 0.3811+j0.1941 60+j30 8 21 2+j2
5 6 0.8190+j0.7070 60+j20 9 15 2+j2
6 7 0.1872+j0.6188 200+j100 12 22 2+j2
7 8 0.7114+j0.2351 200+j100 18 33 0.5+j0.5
8 9 1.0300+j0.7400 60+j20 25 29 0.5+j0.5
9 10 1.0400+j0.7400 60+j20
10 11 0.1966+j0.0650 60+j30
11 12 0.3744+j0.1238 60+j35