Planning of Electric Vehicle Charging Infrastructure

Dharmakeerthi C.H., Mithulananthan N. and Saha T.K.

School of Information Technology & Electrical Engineering
University of Queensland
Brisbane, Australia
p.dharmakeerthi@uq.edu.au; mithulan@itee.uq.edu.au; saha@itee.uq.edu.au

come into the picture. A charging station is required to be

Abstract— The Electric Vehicle (EV) has become the sustainable located considering both consumer convenience and power
alternative to fossil fuel driven automobiles. As a result, a new system concerns like voltage stability, losses and other
type of load is being seen on power systems. Hence, provision of system constraints. Hence, planning of the EV charging
an EV charging infrastructure has become a new challenge for infrastructure is a multi-objective problem and needs to be
power system engineers worldwide. It has been identified that
solved to obtain the best compromise solution. Particle
the voltage dependent nature of EV load may lead to voltage
instabilities in the system. Furthermore, significant load Swarm Optimization (PSO); one of the most promising
integration into the distribution system may overload the system optimization techniques is considered here for the
components, increase power system losses and may violate identification of the best placement of EV charging stations.
system constraints. Despite these factors EV consumers should The rest of the paper is organized as follows: some
be provided with convenient and reliable charging facilities. important facts to be considered during the planning of EV
Hence, the identification of a charging infrastructure which charging infrastructure are discussed in Section II. The
satisfies requirements of both the EV customer and the power problem solving approach of PSO is given in Section III,
system is of primary importance. A particle swarm optimization while Section IV covers the approach of evaluating optimal
(PSO) based approach is considered here for planning of EV
siting and sizing of the charging station. The network study
charging infrastructure.
and the conclusion of the study are given in Sections V and
Index Terms-- charging infrastructure planning, electric VI respectively.
vehicle, particle swarm optimization, voltage dependent loads,
voltage stability. II. PLANNING OF EV CHARGING INFRASTRUCTURE
Availability of a charging infrastructure is an important
I. INTRODUCTION concern for the penetration of EVs. Hence, the interested
Many governments around the globe encourage people to governments and EV developers consider the development of
use Electric Vehicles (EVs) to overcome the draw backs of a charging infrastructure as the foremost task. Hence, power
fossil fuel usage. Improved energy efficiencies and reduced systems worldwide should be ready to accommodate this new
maintenance costs of EVs make it interesting to the load. A consumer convenient charging approach which does
consumer. However, the “range anxiety” of EV consumers not degrade power grids is required.
together with an underdeveloped charging infrastructure EV charging will generally occur overnight at homes or in
imposes a significant hindrance to EV popularity. Hence, public charging stations mostly in daytime. Home based EV
development of convenient charging facilities is essential. On charging will mostly be level 1(120V, 12/16A, single phase)
the other hand, a number of grid impacts associated with EV or lower level 2 (240V, 16/32A, single phase),where the
grid integration have been revealed [1]. Power quality levels are defined according to the SAE J1772 terminology
problems, distribution system component over loadings, [5]. The authors suggest power system engineers to promote
violations of regulatory voltage limits, increased power public charging stations as they will convey the following
system losses [2, 3] and voltage stability issues [4] are among benefits to the grid.
them. These need to be considered during charging • Help to divert the charging peak from network demand
infrastructure development. peak, while home based EV charging will probably fall
It would be of great benefit if power system engineers on the network demand peak.
could identify the location of the charging station which • Reduce network upgrade requirements to accommodate
could impose least impact to the power grid and the allowable EV charging at the end distribution feeder level.
size of charging station which complies with the grid • Alleviate the overloading impact on distribution
constraints. However this optimal siting and sizing problem lines/cables and transformers. Furthermore, this does not
becomes more complex when the EV consumer requirements disturb normal cooling cycles of distribution

978-1-4799-1303-9/13/$31.00 ©2013 IEEE

 transformers as home based EV charging does. Hence, In PSO, a population (swarm) of potential solutions called
this imposes less impact on distribution transformer life. particles. Particles fly through the search space to look for
• The unpredictable mobile EV load (when and where it is potential solutions. Consider N number of particles having n
appearing) can be converted into a more predictable number of dimensions search the space for potential
stationary load. solutions; the position of a particle i, at an instant t, is given
• It is much easier to impose regulations on charger power by Xti and its velocity is given by Vti. Both Xti and Vti are
factor, harmonics etc., to a bulk charging load than to bounded by their maximum and minimum limits. The
several distributed chargers. objective function is evaluated for each generation. Each
• More convenient to implement V2G concepts, which particle has a memory to keep track of its best position so far,
does not require integration of sophisticated measuring, PbestXi. The best of all particles are given by Gbest and
control and communication devices up to the end is the position of the particle at Gbest. ,
consumer level. , are updated after each iteration. The best
Furthermore, the considerable time taken by the household position of each particle and whole of the population is used
level 1 and level 2 chargers to charge the EVs, makes fast to calculate the next velocity and the position of the particles
charging the preferred option for the EV consumers too. (V i t+1, Xit+1), as shown in (3) and (4).
Hence, it would be of great benefit, if utilities were to take
prior action to develop a charging infrastructure well before X = X , ,X , ,………..X , ∀ = 1 … … . . N (1)
the wide adoption of EV. Proper planning is of major V = {V , , V , , … … … . . V , } ∀ = 1, … … . . N (2)
importance. Identification of the best siting and the optimum , = × , + × () × − , + ×
size of the charging load, which fits the planning constraints, () × − , (3)
is one primary requirement. However, a more realistic where ∀ = 1, … … . . , ∀ = 1, … … . .
planning approach may search for a best charging station
location considering both consumer and power system , = , + , ∀ = 1, … … . . , ∀ = 1, … … . . (4)
requirements. For that, selecting a location which satisfies the where is the inertia weight. , and are acceleration
following would be ideal. constant, () is a random number.
• It should be located in close proximity to the consumer. Large values of and , cause global exploration, while
• The location should cover most productive EV sites small values demonstrate local exploration. Hence, dynamic
such as road networks, town centres, residential areas, inertia weight ( ) and dynamic maximum velocities ( , )
office complexes etc to maximize the profit for the are defined to facilitate particles to cover the whole solution
charging station owner. space at the beginning and to confirm a better convergence to
• The location of the charging station should cause least the optimum at the end.
system losses, place least impact to power system
voltage stability and it should not violate regulatory ( )
w =w − × Curr_iter (5)
voltage limits and distribution line MVA flow limits. _

PSO is used to address the above planning concerns in this ( , _ , _ )

V, = V, _ − × Curr_iter (6)
study. _

Where and are the initial and final weighting

III. PARTICLE SWARM OPTIMIZATION factors. , _ and , _ are the initial and final
Most linear and nonlinear conventional optimization maximum velocities. _ and _ are the
techniques are deterministic; randomness is not incorporated number of iterations and the current iteration number.
within the algorithm. Hence, they can end up with local The multi objective optimization is to identify the best
optimums despite the number of iterations performed. On the compromise solution considering several objectives which
other hand, randomness is integrated within modern heuristic are conflicting with each other. Fuzzy set theory is applied
approaches. Hence, a global optimum or a near global here which follows the natural way of human reasoning.
optimum is guaranteed depending on the number of iterations Fuzzy membership functions ( ) are defined for each
performed. There exists a number of heuristic techniques; objective. Linear membership functions are defined by
genetic algorithm (GA), simulated annealing (SA), virtual ant assigning 0 to the most desirable value of the objective, while
algorithms (VAA) and PSO to name a few. PSO is used in the most unacceptable solution is granted with 1, since the
this study for the planning of EV infrastructure. problem is formulated as
a minimization function.
PSO originated by Kennedy and Eberhart in 1995 0 ≤ ,
considering swarm behavior. PSO is not only much simpler
compared to most other heuristic approaches like GA, SA, − ,
= ( − ) , ≤ ≤ , (7)
VAA etc, but it also produces better results faster and , ,

cheaper[6]. It uses real number randomness and global

1 ≥ ,
communication among particles to optimize its performance.
The normalized overall objective function can be defined as, sites, the lower the weigh dis . Hence, minimization of
= ∑ (8) weigh dis is one of the objectives of a multi-objective
where is the number of objectives and is the weighting optimization problem.
factor of objective i.
∑ = 1, ∀ ∈ (0,1) (9) ii. EV load Model
Hence, the best compromise solution can be obtained by It has been identified that voltage dependent power
assigning suitable weighting factors for each individual consumption of a load provides a major contribution to
objective. power system voltage instability[7]. The 1987 power system
failure in Tokyo was identified to be contributed by the
IV. FORMULATION OF THE OPTIMAL SITING AND SIZING OF EV negative exponential load characteristics of air conditioners
LOAD [8]. Hence, incorporating realistic load models in system
A. Problem Description studies is crucial. The EV load model is selected as (11)
considering the EV load model development concept given in
A primary system investigation can be carried out to evaluate
[4].
the maximum allowable charging station size that complies .
P V
with the system constraints as given in scenario I. A more P = 0.07092 V + 0.9279 (11)
comprehensive planning approach which considers both the
grid and EV consumer requirements is described in scenario iii. Voltage Stability Formulation
II. Voltage stability has become an increasingly concerning
matter all over the world, since most power systems are
Scenario I heavily loaded and operate closer to their secure operation
The highest charging load that can be accommodated while limits due to various environmental and economical
complying with a given set of network constraints is constraints. There are a number of indices to evaluate voltage
considered in scenario I. stability of a power system. However, some of them have
properties which make them not suitable to determine relative
Scenario II voltage stability; the Eigenvalues of system jacobian is one
The best location for an EV charging station is required to example. The most popular L index is simple and having a
be determined to cover several potential EV sites, while clear physical meaning, but it is suitable for the systems
causing the least impact on the power grid. Best siting having only constant power loads. Hence, a different
problem considers the relative importance of each of the evaluation method is considered here by taking into account
potential EV sites considering factors like coverage of road one of the most common causes of system voltage instability
networks, town centres, residential areas etc,. The extent of phenomena.
the impact caused to the grid is determined in terms of real Voltage instability is primarily caused by insufficient
and reactive power losses, the remaining reactive power reactive power support under stressed system conditions.
reserve, bus voltages and line MVA flows. Instability may happen when one or more generators hit their
maximum reactive power limit[9] or when the transmission
B. Problem formulation system is unable to transfer the required reactive power to the
i. Distance Model loads due to high reactive power loss in lines. Hence, total
The best siting of the charging station on the power grid is reactive power losses in all system buses and the total
required to be identified, to cover n number of EV sites. Each remaining reactive power reserves of all the system
EV site has different relative importance. Different weights generators are taken as measures for voltage stability of the
are assigned to each EV site to reflect their relative system as formulated in (12) and (13).
importance (RI), considering factors like the sites’ capability Q = ∑ Q , (12)
of capturing EV consumers for optimizing charging station _
Q =∑ Q −Q (13)
operators profit, coverage of road networks, town centres, Reactive power reserve associated voltage stability is a
office complexes, residential areas, etc,. The distance from local phenomenon. It is assumed here that all the generators
each feasible network bus (i) to each of these EV sites (j) is contribute to the Q since the studied power system is a
obtained (D ). The RI and the distance are then used to local distribution network.
calculate weighted distance (weigh dis ) for each network
bus, as given in (10).
C. Optimization Problem
weigh dis = ∑ D ∗ RI (10)
Scenario I
where, ∑ RI = 1 ∀RI ∈ (0,1)
i. Objective function
The PSO algorithm is to identify the maximum allowable
The closer the system bus i to the most productive EV charging load and its location. Hence, each particles position
is defined by two dimensions, the location (bus) and the size The line and bus data of the system is given in Table I[10].
of the charging load. Feeder I Feeder II Feeder III
ii. System constraints 1 2 3
The identification of maximum size and location is
bounded by the following constrains, which can be made 4 8 13
10
according to power system engineers requirements. The
5 9
additional real and reactive power losses due to EV 14
integration should be within the defined values. The reactive 6
11 15
power reserve should not be consumed beyond the pre 12
7
defined levels. These constrains can be formulated as given
16
below.
P ≤P , (14)
Q ≤Q , (15) Fig. 1. The IEEE 16 Bus test system
Q ≥Q , (16)
TABLE I
Further, the bus voltages are compared with the regulatory THE IEEE 16 BUS TEST SYSTEM DATA
voltage limits for any violations, as given in (17). The Bus Resistance Reactance P(MW) Q(MVars) Capacitors
distribution lines’ apparent power flows checked with their From- /(p.u) /(p.u) End End Bus Q(MVars)
maximum MVA flow limits, as given by (18). To Bus End Bus
. . 1-4 0.075 0.1 2 1.6
V ≤V ≤V i = 1,2 … . n (17) 4-5 0.08 0.11 3 1.5 1.1
S ≤ S max (18) 4-6 0.09 0.18 2 0.8 1.2
6-7 0.04 0.04 1.5 1.2 -
2-8 0.11 0.11 4 2.7 -
Scenario II 8-9 0.08 0.11 5 3 1.2
i. Objective function 8-10 0.11 0.11 1 0.9 -
The optimization problem consists of four individual 9-11 0.11 0.11 0.6 0.1 0.6
objectives. It is required to 9-12 0.08 0.11 4.5 2 3.7
1. Maximize the reactive power reserve (QResv). 3-13 0.11 0.11 1 0.9 -
13-14 0.09 0.12 1 0.7 1.8
2. Minimize power system losses (PLoss). 13-15 0.08 0.11 1 0.9 -
3. Minimize reactive power losses (QLoss). 15-16 0.04 0.04 2.1 1 1.8
4. Minimize the distance to the most productive EV sites 5-11 0.04 0.04 - - -
(weigh_dis). 10-14 0.04 0.04 - - -
7-16 0.09 0.12 - - -
Each of these objectives except the first objective are
normalized using (7). The first objective is normalized as Scenario I
given in (19). The PSO is initiated to identify the maximum possible EV
(Q _ −Q , ) load integration and the location, which does not violate the
Q , = (Q −Q ) (19)
_ _ following system constraints. The bus voltages should be
within 0.9 to 1.1 p.u., the power losses should not go beyond
The Q _ and Q _ are the maximum and 10% of the base case power losses (0.5114 ± 10%MW,
minimum reactive power reserves when the EV load is 0.5904± 10%MVars).
connected at different network buses. Each objective is The PSO is initiated with a swarm of 20 particles, and
weighted according to their relative importance to obtain the each of them was assigned with a random bus number and a
final objective function given by F. random charging load (real random number). After
F = Min{σ ∗ P +σ ∗Q + σ ∗ weigh_dis + σ ∗ Q } (20) completing 80 iterations the maximum allowable charging
load is identified as 3.2656 MW at bus 13. It was verified
ii. System constraints through Newton Rapson power flow, and that capacity found
The regulatory limits of the system bus voltages and the to be producing a power loss equivalent to the constrained
distribution lines’ apparent power flow limits are checked for limit.
any violations, as given in (17) and (18).
Scenario II
V. NETWORK STUDY This is to determine the best compromise location of a
2MW EV charging station in the power grid, to cover four
The network study is carried out on the IEEE 16 bus
potential EV sites, each having different relative importance
distribution test system, shown in fig. 1. The test system
(RI) as given in Table (II). A random number between 0-3 is
consists of three distribution feeders (23kV), seven static
generated to represent the distance from network buses to
capacitors and 13 loads of total 28.7 MW and 17.3 MVars.
each of the potential EV sites, as given in Table (III). VI. CONCLUSION
TABLE II The extent of popularity of EV is found to be negatively
THE RELATIVE IMPORTANCE OF EACH POTENTIAL EV SITE affected by the under developed charging infrastructure.
EV Site 1 2 3 4 Hence, EV developers and interested governments expend a
Relative
0.4 0.3 0.2 0.1 lot of effort to build required charging facilities. However,
Importance (RI) many system studies confirm that charging of EV in the
power grid may bring additional stress to the distribution
TABLE III
networks. The importance of giving priority for implementing
THE DISTANCE FROM NETWORK BUSSES TO EACH OF THE POTENTIAL EV SITES
Distance to Distance to Distance to Distance to public charging stations to ease grid impacts at the
Bus distribution system level is highlighted. Further, proper
EV site 1 EV site 2 EV site 3 EV site 4
1 1.112 0.243 2.585 1.415 planning should be done to develop a consumer convenient
4 1.591 0.216 1.006 2.072 charging infrastructure while not putting the power grid at
5 0.242 0.070 2.760 0.281 risk. It was found that this can be solved efficiently by
6 2.113 1.718 0.738 0.173 applying a modern heuristic technique; the particle swarm
7 1.041 0.976 0.233 2.476
optimization.
8 0.337 1.575 0.850 0.272
9 2.826 1.206 0.909 0.619
10 0.622 1.888 1.867 2.380 VII. REFERENCES
11 0.029 0.095 1.745 2.045
[1] C. H. Dharmakeerthi, N. Mithulananthan, and T. K. Saha, "Overview
12 2.606 1.028 2.077 1.410
of the impacts of plug-in electric vehicles on the power grid," in
13 2.036 2.870 0.431 2.693 Innovative Smart Grid Technologies Asia (ISGT), 2011 IEEE PES,
14 0.515 1.000 1.044 1.226 2011, pp. 1-8.
15 2.530 2.013 1.870 1.976 [2] P. Papadopoulos, S. Skarvelis-Kazakos, I. Grau, B. Awad, L. M.
16 2.688 0.079 2.577 2.875 Cipcigan, and N. Jenkins, "Impact of residential charging of electric
vehicles on distribution networks, a probabilistic approach," in
Universities Power Engineering Conference (UPEC), 2010 45th
The initial study is carried out to evaluate best and worst International, pp. 1-5.
values of each objective as given in Table (IV), which are [3] J. A. P. Lopes, F. J. Soares, P. M. R. Almeida, P. C. Baptista, C. M.
required to obtain normalized objectives in the multi- Silva, and T. L. Farias, "Quantification of technical impacts and
environmental benefits of electric vehicles integration on electricity
objective minimization function given in (15). The reactive grids," in Advanced Electromechanical Motion Systems & Electric
power reserve could not be calculated as the required data is Drives Joint Symposium, 2009. ELECTROMOTION 2009. 8th
not available for the studied test system. International Symposium on, 2009, pp. 1-6.
[4] C. H. Dharmakeerthi, Mithulananthan, N., Saha, T. K., "Modeling and
planning of EV fast charging station in power grid," in 2012 IEEE
TABLE IV
Power and Energy Society (PES) General Meeting, San Diego, CA,
THE MAXIMUM AND MINIMUM VALUES OF INDIVIDUAL OBJECTIVES
U.S.A, 2012.
Item Value Bus [5] T. SAE, "Charging Configurations and Ratings Terminology,
_ 0.5394 MW 13 Available Online http://www.sae.org/smartgrid/chargingspeeds.pdf,"
_ 0.6555 MW 12 2011.
_ 0.6184 MVars 13 [6] D. M. a. M. D. S. Sakthivel, "Reactive Power Planning for Voltage
0.7562 MVars 12 Stability Limit Improvement with FACTS Devices in Most Critical
_
0.5938 11 Contingency Condition " European Journal of Scientific Research,
ℎ_
vol. Vol.66 No.3 (2011), pp. pp. 408-420 2011.
ℎ_ 2.1875 15
[7] T. J. Overbye, "Effects of load modelling on analysis of power system
voltage stability," International Journal of Electrical Power &
Minimum real and reactive power losses are reported Energy Systems, vol. 16, pp. 329-338, 1994.
[8] A. Kurita and T. Sakurai, "The power system failure on July 23, 1987
when charging load is connected at bus 13, while ℎ_ in Tokyo," in Decision and Control, 1988., Proceedings of the 27th
becomes minimum at bus 11. However, the best compromise IEEE Conference on, 1988, pp. 2093-2097 vol.3.
location is then identified as bus 14 by assigning , , [9] B. Tamimi, C. A. Canizares, and S. Vaez-Zadeh, "Effect of Reactive
Power Limit Modeling on Maximum System Loading and Active and
with 0.3, 0.3, 0.4 respectively in (15). The results are given in Reactive Power Markets," Power Systems, IEEE Transactions on, vol.
Table (V). 25, pp. 1106-1116, 2010.
TABLE V [10] P. Ravibabu, K. Venkatesh, and C. Sudheer Kumar, "Implementation
THE RESULTS FOR SCENARIO II of genetic algorithm for optimal network reconfiguration in
Description Location ℎ_ distribution systems for load balancing," in Computational
Best compromise Technologies in Electrical and Electronics Engineering, 2008.
Bus 14 0.5460 0.6271 0.8374
location SIBIRCON 2008. IEEE Region 8 International Conference on, 2008,
Best location in grid pp. 124-128.
Bus 13 0.5394 0.6184 2.0309
point of view
Best location in
Bus 11 0.6443 0.7382 0.5936
consumer point of view