Version of Record: https://www.sciencedirect.

com/science/article/pii/S1366554515302520
Manuscript_53163bfb3b756561b8e1c3e14ab33c0c

Optimization Models for Placement of an

Energy-Aware Electric Vehicle Charging
Infrastructure

Zonggen Yi (Corresponding author)

E-mail: [email protected]
Address: 229 Cushing Hall, Department of Electrical Engineering, University of
Notre Dame, Notre Dame, IN, 46556 USA

Peter H. Bauer
E-mail: [email protected]
Address: 269 Fitzpatrick Hall, Department of Electrical Engineering, University of
Notre Dame, Notre Dame, IN, 46556 USA

© 2016 published by Elsevier. This manuscript is made available under the Elsevier user license
https://www.elsevier.com/open-access/userlicense/1.0/
Optimization Models for Placement of an Energy-Aware
Electric Vehicle Charging Infrastructure
Zonggen Yi, Peter H. Bauer
Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN,
46556 USA

Abstract
This paper addresses the problem of optimally placing charging stations in
urban areas. Two optimization criteria are used: maximizing the number
of reachable households and minimizing overall e-transportation energy cost.
The decision making models used for both cases are mixed integer program-
ming with linear and nonlinear energy-aware constraints. A multi-objective
optimization model that handles both criteria (number of reachable house-
holds and transportation energy) simultaneously is also presented. A number
of simulation results are provided for two different cities in order to illustrate
the proposed methods. Among other insights, these results show that the
multi-objective optimization provides improved placement results.
Keywords:
Charging infrastructure placement, electric vehicles, energy-aware
optimization models, multi-objective optimization.

1. Introduction
In order to achieve emission reduction targets and reduce dependency on
(foreign) oil and fossil fuels in general, electric vehicles have drawn more and
more attentions from governments and the general public. Developing electric
vehicles and creating an electrified transportation system is an effective way
to promote urban sustainable development as pointed out by (Eberle and
von Helmolt (2010), Bouscayrol et al. (2011), Khaligh and Krishnamurthy
(2012) and Bilgin et al. (2015)). Therefore policies to facilitate the growth

Email address: [email protected], [email protected] (Zonggen Yi, Peter H. Bauer)

Preprint submitted to Transportation Research Part E April 28, 2016

and market penetration of electric vehicles (EVs) have been developed in
almost every nation of the industrialized world. While over the last decade
several car producers made EV models commercially available in the US, the
market share trajectory of EVs in the US has been below predicted levels.
The acceptance of EVs by the public depends on a large variety of fac-
tors. An efficient, convenient and economic charging infrastructure system
can enhance the willingness of consumers to purchase and promote industry
development (Hatton et al. (2009), Guo and Zhao (2015)). The availability
and convenient locations of charging stations in metropolitan environments
is a key factor that globally affects not only the adoption process of EVs but
also the sustainability of transportation. The energy-efficiency criterion is
one of the principle considerations for enhancing sustainability (Kates et al.
(2005)). Therefore, it is necessary to employ proper methods to determine
the optimal energy-aware charging station locations. Aiming to construct an
energy-efficient charging infrastructure system for sustainable urban trans-
portation systems, multiple objective energy-aware decision-making models
are introduced in this paper. Two different criteria are proposed in the op-
timization models, which consider several energy-aware constraints:
First, we address the problem of optimal charging station placement from
the viewpoint of reaching the most customers or households by providing an
energy cost constraint, i.e. something a private charging station owner would
typically consider. On the other hand, this problem is also of interest for mu-
nicipalities, power companies, and federal agencies such as the Environmen-
tal Protection Agency (EPA) and the Department of Transportation (DOT).
Given an energy bound, the corresponding reachable contoures in Google
Maps for different possible charging station centers are determined by using
the energy model in (Yi and Bauer (2014a)). Maximizing the number of
households, i.e. EV users, in this range is discussed subsequently.
Second, another criterion, namely minimizing overall transportation en-
ergy consumption to perform charging actions, is addressed. This is an im-
portant aspect to construct future sustainable transportation systems, which
is of utmost interest for agencies such as EPA and DOT. Suppose all EVs
charge once a day at the charging station away from home. Each EV has a
corresponding energy consumption when it travels to a public charging sta-
tion from home. The objective is to find a subset of locations from potential
positions to achieve minimum overall transportation energy cost considering
all charging actions. This objective is crucial for establishing future sus-
tainable cities. Some other constraints, e.g. service capability of charging

2
station, etc. also are included in the optimization model.
Third, we combine both criteria to get a more realistic decision making
framework. The corresponding multiple objective optimization model will
be proposed to obtain more balanced planning strategies under energy cost
constraints. This multi-objective model will balance both introduced energy
related requirements.
This paper is organized as follows. Section 2 provides a literature re-
view, the main contributions of this paper, and a brief description of how
this paper addresses open research problem in relative to other studies. In
Section 3, the energy-aware charging station placement framework will be
formulated. The optimization model for the maximum number of reachable
households will be introduced in Section 4. In Section 5, the charging station
placement problem considering minimum overall transportation energy cost
will be proposed and discussed. In Section 6, the multiple objective opti-
mization model will be constructed by considering both of the introduced
requirements simultaneously. Conclusions are provided in Section 7.

2. Literature Review
The problem of charging infrastructure placement has been investigated
by many researchers. The most powerful and popular techniques are decision-
making methods such as linear/nonlinear programming, multilayer program-
ming and mixed-integer programming. Many very different criteria and con-
straints have been applied in these models.
A maximal coverage model to optimize the demand covered within an
acceptable level of service has been investigated in Frade et al. (2011). The
work in Worley et al. (2012) formulates the problem of locating charging
stations as a discrete integer programming optimization problem based on
the classic vehicle routing problem. In Liu (2012), an assignment model for
different charging infrastructure assignment strategies was proposed by esti-
mating the charging demand of the early electric vehicle market in Beijing.
In Xi et al. (2013) an optimization model was developed to maximize total
fleet-wide charging levels for the location of a public EV charging infras-
tructure. In Lam et al. (2013, 2014), the electric vehicle charging station
placement problem was formulated to minimize the total construction cost
subject to the constraints for the charging station coverage and the driver
convenience for EV charging. Environmental factors and service radius are
considered in Liu et al. (2013) to determine the optimal charging station

3
locations. In Wang et al. (2013b), an optimal location model of charging
stations is established based on electricity consumption along city roads. A
mixed-integer programming model was developed in Chen et al. (2013) to de-
termine optimal location assignments of charging stations in Seattle, which
minimized the station access cost of EV users and took the parking demand,
local job, population density and trip attributes as constrains. In Pasha-
javid and Golkar (2013) the charging stations were allocated by minimizing
energy loss and voltage deviation in the distribution system. In Wang et al.
(2013a), the optimal location and size of charging stations were determined
by maximizing the EV traffic flow under the constraint of battery capacity.
The work in Xu et al. (2013) proposed a mathematical model with minimum
total transportation distance to determine the optimal charging station lo-
cations. The work in Ghamami et al. (2014) formulated this problem as a
fixed charge facility location model with charging capacity constraints, allow-
ing unserved demands and considering driver preference for parking lots. A
mixed-integer nonlinear optimization approach was proposed for determining
the optimal place and size of fast charging stations in Sadeghi-Barzani et al.
(2014), which took the station development cost, EV energy loss, electric loss
and the location of electric substations as well as urban roads as constraints.
In Yao et al. (2014), an equilibrium-based traffic assignment model was pro-
posed to maximize the annual traffic flow captured by fast charging stations.
The work in Cavadas et al. (2014) tried to plan the location of charging sta-
tions for EVs in a city by maximizing the number of vehicles served under
a fixed budget for building charging stations. In Khalkhali et al. (2015),
the optimal location of plug-in hybrid electric vehicle charging stations were
determined by maximizing the benefit of the distribution system manager.
Several data-based methods for analyses of driving activities and travel
behaviors were employed in order to improve the placement strategies. In
the work of Sweda and Klabjan (2011, 2015), a decision support system was
presented for identifying patterns in residential EV ownership and driving
activities to enable strategic deployment of a new charging infrastructure. A
similar decision support system was also developed in (Wagner et al. (2013,
2014) and Cai et al. (2014)). It was achieved by analyzing large-scale tra-
jectory data to obtain travel patterns. The work in Andrews et al. (2013)
proposed an optimization model based on a user charging model to find loca-
tions for charging stations by analyzing the needs of vehicle owners. In Dong
et al. (2014), an activity-based assessment method was proposed to evaluate
electric vehicle feasibility for the heterogeneous traveling population in the

4
context of real world driving, and then determined the optimal location for
public charging station sites of electric vehicles. The work in Cavadas et al.
(2015) considered successive activities of travelers and then proposed an im-
proved mathematical model for locating EV charging stations. In Lee et al.
(2014), a optimal location model for fast charging stations was proposed by
taking battery state of charge and user charging and traveling behavior into
account. The work in You and Hsieh (2014) determined the optimal location
of charging stations by analyzing the round-trip itineraries to maximize the
number of people who can complete their trips.
Several other methods have also be investigated to improve the charg-
ing station placement strategies, such as grid partition method(Ge et al.
(2011)), clustering methods (Ip et al. (2010) and Momtazpour et al. (2012)),
Voronoi diagrams for equilibrium arrangements (Koyanagi and Yokoyama
(2010), Koyanagi et al. (2001) and Feng et al. (2012)), simulation-based
methods(Ma et al. (2014) and Hess et al. (2012)), dynamic spatial and tempo-
ral models(Wirges et al. (2012)), equilibrium modeling frameworks(He et al.
(2013)), continuous facility location models(Sathaye and Kelley (2013)) and
flow-refueling location models(Hodgson (1990), Kuby and Lim (2005), Wang
and Lin (2013), Chung and Kwon (2015)). All these methods have provided
important contributions from different viewpoints on the optimal charging
station placement problem.
From the above literature review, it becomes obvious that the placement
of charging stations has been approached from a number of different angles. A
large variety of optimization models for decision-making with different crite-
ria were used. Unfortunately, the previous research paid almost no attention
to transportation energy cost of electric vehicles. Very few research results
thoroughly addressed the energy consumption requirements in the context
of charging station placement. This is an important aspect for achieving a
sustainable transportation system. The main difference and contributions of
our work relative to other studies are as follows:

• An energy-aware optimization framework is introduced that provides a

solution to the problem of charging infrastructure placement. Two
different energy-based criteria are proposed: (1) maximum number
of reachable household with travel energy cost bound; (2) minimum
overall transportation energy cost for charging actions. Several energy-
aware constraints are also incorporated, e.g. reachable range constraint
represented by travel energy cost, a service capability, etc.. Further-

5
 more, an energy-aware multi-objective decision-making model is pro-
posed to find better equilibrium placement strategies. In the literature,
some work also discussed the maximum number of vehicles, e.g. in
(Cavadas et al. (2014); You and Hsieh (2014)), but the travel energy
cost was omitted in these models.

• A detailed energy consumption estimation model for electric vehicles is

employed to construct optimal decision-making models. The work in
Sebastiani et al. (2014) and Baouche et al. (2014) provided the optimal
placement model by considering the energy cost of electric vehicles, al-
though the energy consumption estimation models used in this work are
not explicit and cannot consider the real geographic and environmental
information to obtain an accurate estimation. The energy consumption
estimation model in our previous work (Yi and Bauer (2014a)) con-
siders realistic information and is applied to provide a more accurate
energy cost for finding energy-aware optimal locations.

• In order to solve the proposed mixed nonlinear integer programming

problem with nonlinear constraints, a highly efficient algorithm called
mesh adaptive direct search (Audet and Dennis Jr (2006)) has been
utilized. It is a powerful tool to aid in solving the large-scale problem.

• A data analysis technique, i.e. a clustering algorithm, is used as a

pre-selection procedure to determine the potential charging station set.
This proposed preprocessing method will aid in finding a suitable sub-
set of potential locations for charging stations and then reducing the
computational complexity in the final decision-making process.

• Real world datasets are investigated, including datasets for a large city
(Chicago, IL) and a medium to small size city (South Bend, IN). By
using real world data, the capability and performance of the developed
framework is demonstrated.

3. Problem Formulation
3.1. Notation
The notation used throughout this paper is listed in Table 1. In addition,
variables and parameters are explained in the text where they are first used.

6
 Table 1: Notations
EV set and uj = (ujh , ujw , ujt )
j j j
U = {u1 , ..., uj , ..., uN } uh , uw , ut are the EV home address, the work place address
and the type of EV for vehicle j, respectively.
Potential charging station set and ci = (cip , cicap , xi )
C = {c1 , ..., ci , ..., cM } cip is the location and cicap is the capacity for the potential charging station i
xi is an indicator variable
j j
Y = {y1 , . . . , yN } ZIP code zone and yj = (ycp , ypop )
j j
ycp is the center position of yj and ypop is the population of yj
Eij Transportation energy consumption for EV uj to charging station ci
Ut EV type set and Ut = {u1t , ..., ult , ..., uLt }
αe Average market penetration rate for EVs
αel Market share for EV type ult
zj The variable to describe whether yj is covered
Il Coverage indicator matrix
X0 Needed number of charging station
pij Probability of EV j to choose charging station i
dij Distance between work place of EV j and charging station i
Si Service ability of charging station i

3.2. Problem Statement

As shown in Fig. 1, a possible location for a charging station may initially
be determined by the power infrastructure limit. A potential location set for
charging stations should also consider the concerns of EV users, e.g. driving
pattern, most popular parking locations, etc. A “driving pattern” denotes
often used travel routes, i.e. popular sequences of distance segments. The
obtained potential charging station locations can then improve the selection
of locations. As an example, EV owners can charge their vehicles at the
parking lots of their work places. In city planing, DOT and EPA also have
an interest in the transportation energy consumption caused by charging
actions. Global energy savings will be significant if the charging stations
are placed optimally. This is an important aspect for developing sustainable
transportation systems and green cities. An energy-aware set of optimal
charging station locations needs to be determined by both, maximizing the
number of reachable EV home locations under energy cost constraints and
minimizing the overall transportation energy cost for all charging actions. In
order to achieve the goal of this energy-aware optimality, the optimization
framework will involve EVs and potential charging stations and depend on
the real characteristic information of both entities.

7
 Initial Location Set for
Charging Stations
(Power infrastructure Limit) Charging Station
EVs
Driving Pattern
Placement Procedure
Popular visited Locations
Social-economic factors
Etc.

Potential Charging
Station Set
Energy-Aware City Planning(DOT, EPA, etc.)
Maximum Reachable Household Green City
Minimum Overall Energy Cost Sustainable Transportation

Energy Consumption
Model for EVs
Optimal Energy-Aware
Charging Station Locations

Figure 1: Charging Station Placement Procedure

3.2.1. EVs
Define the EV set as U = {u1 , ..., uj , ..., uN }. The cardinality of U is
N , i.e. there are N EVs. Each uj in U is a tuple with three elements
and uj = (ujh , ujw , ujt ), where ujh , ujw , ujt are the EV home address, the work
place address and the type of EV for vehicle j, respectively. In the following
analyses, simulations and case studies, the initial state of battery for electric
vehicles is assumed to be 100% state of charge(SOC) at home, which is known
to usually be the case (Smart and Schey (2012)).

3.2.2. Charging Stations

Define the potential charging station set as C = {c1 , ..., ci , ..., cM }. The
cardinality of C is M , i.e. there are M potential charging station positions.
Each ci in C is a tuple with three elements, i.e. ci = (cip , cicap , xi ), where cip is
the location and cicap is the capacity for the potential charging station i. The
variable xi is an indicator with binary value to show whether the charging
station ci will be placed(xi = 1) or not be placed(xi = 0). Assume that
public transportation is available in all potential charging station locations
and expected to be used to reach final destinations. Also assume that this
additional transport cost is neglected or approximately equal.
Suppose C0 is the set of all possible charging stations in the investigated
area, and the potential charging station location subset is C, C ⊂ C0 . Assume
a selection function g : C = g(C0 ). This function serves as the pre-selection
for the charging station locations according to some additional criteria, e.g.
parking time and driving patterns of EVs, available infrastructure and power
limits. Therefore our model can combine many other constraints into the

8
 (1) Routing Information
Origin(Home)
Route Planning (2) Elevation data
(Google Maps) (3) Driving Cycles
Destination (4) Environmental Information
(Charging Station)

Single EV Energy Energy Consumption

Consumption Model

Figure 2: Single EV Energy Calculation Procedure

optimization framework. In the following study, the average parking time

will be used to determine the potential charging station set C.

3.2.3. EV Energy Consumption

In order to obtain an accurate value of the energy consumption, we need
to know the route followed by the vehicle, the traffic condition along the
route, the driving pattern and environmental information as shown in Fig. 2.
Transportation energy consumption Eij for EV uj at charging station ci can
be calculated by the following function

Eij = f (cip , ujh , ujt , environmental information) (1)

The energy consumption when EVs travel to a charging station depends

on the location of the charging station cip , the home location of EV ujh ,
the EV type ujt and the corresponding environmental information along the
route. The environmental information can include geographic information,
the traffic conditions, weather conditions such as wind speed, temperature
and road surface conditions. The energy consumption can be computed by
integrating out a differential equation which provides instantaneous power
of an EV. This has been discussed in detail in our previous work in (Yi and
Bauer (2014a),Bauer et al. (2012)), where a detailed energy consumption
model was proposed for a single EV:

(Pbat (t) − Ppara (t))ηbat ηconv ηcontr ηm ηmp

(2)
= Pair (t) + Proll (t) + Phill (t) + Pac (t)
Where Pbat (t) is the power provided by the EV battery, Ppara are parasitic
power losses at the battery (lights, heater, stereo, etc.), Pair (t) is the air drag
power component, Proll (t) is the rolling resistance power component, Phill (t)
is the hill climbing power component, and Pac (t) is the acceleration power

9
component at the wheel. The efficiencies of battery, power converter, e-
motor controller, e-motor and mechanical powertrain are denoted by ηbat ,
ηconv , ηcontr , ηm , ηmp respectively, i.e. we are using a lumped model for
efficiencies. By obtaining the route information from Google Maps, typically
the shortest path and using the software EVRE that was developed in Yi and
Bauer (2014a), we can estimate the energy cost between starting location and
destination.

4. The Maximum Number of Reachable Households Problem

In this framework, we focus on maximizing the number of reachable house-
holds with a given battery energy bound. This is a very crucial consideration
for charging station owners. Given the energy constraint, the reachable range
can be derived from Google Maps (Yi and Bauer (2014a)). In order to get
the amount of covered EV users in the reachable range, the population dis-
tribution for the area considered must be known. A reasonable way to divide
the city into subareas is by ZIP code zones and subsequently by using the
population in each ZIP code area.

4.1. Optimization Model

Suppose the possible charging station set is C = {c1 , . . . , cM } and there
are N ZIP code zones: Y = {y1 , . . . , yN }. ZIP codes are used to simplify the
location information for EV home locations. The EVs in the same ZIP code
zone have the same home address information. The model for a ZIP code
j j j j
zone is given by yj = (ycp , ypop ), where ycp is the center position of yj and ypop
is the population of yj . Define the EV type set as Ut = {u1t , ..., ult , ..., uLt },
which means L types of EVs are considered in total. Suppose the average
market penetration rate for EVs is PLαe and the market share for EV type
l l l
ut in overall EV sales is αe , then l=1 αe = 1. For each possible charging
station ci , the function Range(cip , ult , Ebound ) is used to obtain its reachable
range TRil for EV type ult where Ebound is the energy constraint. Furthermore
Ri = Li=1 Ril . This range can be obtained by the software tool we designed
based on Google Maps Javascript API in Yi and Bauer (2014a). Assume
X0 charging stations will be placed in the city area. The corresponding
optimization model can be constructed as:

10
 L X
X
Max zj αe αel ypop
j

l=1 j
M
X
s. t. xi = X0 , xi ∈ {0, 1}
(3)
i=1
X
xi ≥ zj , zj ∈ {0, 1}
j
ycp ∈Ri

i = 1, . . . , M ; j = 1, . . . , N.
where the binary value xi is an indicator for filling a possible charging station
ci (i.e. xi = 0: not placed; xi = 1: placed). zj describes whether yj is covered:
If zj = 1, yj is covered. The objective is to find the indicator variables xi for
determining the set of charging station that covers the maximum number of
reachable households.
The optimization problem (3) is a discrete problem and essentially solves
a coverage problem on the real map. The union of coverage ranges Rl is
always difficult to obtain when X0 is large. In order to make this problem
solvable and decrease the computational complexity, the following procedures
are proposed to relax this problem:
(1) A coverage indicator vector for each charging station can be pre-
calculated. Suppose the coverage indicator vector for ZIP code zone yj under
EV type ult is I~jl and I~jl ∈ RM , the element values in I~jl should be 0 or 1. For
example, for the kth element in this vector, if its value is one, it is covered
by charging station ck . If its value is zero, it is not covered by ck .
(2) There are N ZIP code zones, then N I~jl s will be constructed. Then the
coverage indicator matrix is given by I = [I~l , ..., I~l , ..., I~l ]T and I ∈ RN ×M .
l 1 j N l
(3) Let ~x = [x1 , x2 , ..., xi , ..., xM ]T , then Il ~x ∈ RN represents the covered
number of charging stations for each ZIP code zone. So if ZIP code zone k
is covered at least by one charging station, the corresponding value of (Il ~x)k
will be equal or larger than one. By considering a step function U (.), we
have:
XL
Max αe αel p~T U (Il ~x − 0.5)
l=1
(4)
s. t. |x|1 = X0 , xi ∈ {0, 1}
p~ ∈ RN , Il ∈ RN ×M , ~x ∈ RM

11
Where p~ is the population vector for N ZIP code zones. If we want to
maximize the number of reachable households, each ZIP code zone can only
be calculated at one time. The step function U (Il ~x −0.5) is used to guarantee
this.

4.2. Case Study: South Bend

South Bend, IN, a midsize city in the US, is selected as an example. As
stated in the above optimization models, in order to model the population
distribution, the area can be divided according to ZIP codes. There are 13
ZIP codes in the South Bend area, which means N = 13. Table 2 includes
the ZIP codes and their corresponding population information. For possible
charging station positions, five possible places are selected as examples, e.g.
M = 5. They are listed in Table 3. We need to choose one or more from these
five possible positions to maximize the reachable population. According to
the data from DOT, we have the approximate average penetration rate in
the US of αe = 0.001. The most popular EVs in the US are the Nissan Leaf
and Tesla Model S. All others are neglected in our simulation. So we have
L = 2 and αe1 = 0.66 for Nissan Leaf, αe2 = 0.34 for Tesla Model S according
to the data in Wikipedia (2015).

4.2.1. Single Charging Station Placement Case

Table 2: Population Distribution in South Bend

Y ZIP Code Population
y1 46544 30695
y2 46545 28445
y3 46556 7424
y4 46601 8460
y5 46613 11526
y6 46614 27041
y7 46615 16905
y8 46616 6431
y9 46617 11644
y10 46619 22489
y11 46628 25319
y12 46635 4172
y13 46637 13829

12
 Table 3: Charging Station Placement in South Bend
C Possible Position Nissan Tesla Total
Leaf Model S Coverage
c1 Century Center 73 64 137
c2 University of Notre Dame 73 33 106
c3 McKinley Town Center 65 43 108
c4 University Park Mall 56 21 77
c5 South Bend Airport 44 25 69

With these ZIP codes and their population information, EVRE in Yi

and Bauer (2014a) is used to calculate the covered households with EVs for
each possible position. The energy bound Ebound for the reachable range is
assumed to be 2 kWh. The energy bound is set to such a low value so that
the reachable range can be obtained accurately in an acceptable computation
time. It also provides a good option for a mid size city. Fig. 3 shows the
coverage of two possible charging station positions with EV type Tesla Model
S, one is the Century Center and the other is the University Park Mall. The
markers with “A” are the centers of the covered ZIP code zones, and the
marker with “C” is the possible charging station position. The proposed
five locations have different reachable ranges and will cover different ZIP
code zones. After we know the covered ZIP code zones, we can calculate
the corresponding number of covered EVs. This number is listed in Table 3.
We can see that the Century Center in South Bend has the largest number
of covered EVs, which means that under this gain function and the 2 kWh
battery energy bound constraint, it is the optimal location.

4.2.2. The Multiple Charging Stations Case

As discussed in the single charging station case, there are 5 potential
charging station positions. Two of them will be selected in order to maximize
the number of reachable households, e.g. X0 = 2. According to the relaxed
optimization model in Equation (4), the population vector p~ has 13 elements
and the values can be obtained from Table 2. The energy bound for coverage
is 2kWh. By calculating the energy cost for each ZIP code zone center to
the five potential positions, we can construct the corresponding coverage
indicator matrix I. Only the Nissan Leaf and Tesla Model S are considered

13
Figure 3: Coverage Range in South Bend, IN. Left: University Park Mall, Right: Century
Center

and indicator matrices are I1 and I2 , respectively. I2 is provided as follows:

 
0 0 1 1 1 0 1 1 1 0 0 1 1
0 0 1 1 0 0 1 1 1 0 0 1 1
 
IT
2 = 1 1
 1 1 0 0 1 0 1 0 0 1 0 
0 0 1 0 0 0 1 0 1 0 0 1 1
0 0 0 1 0 0 0 1 0 1 0 0 0

The element Iij in I indicates whether yj is covered by ci . Iij = 1 means it

is covered and Iij = 0 means it is not. For example, I32 = 1 ZIP code zone
No. 3 is covered by c2 .
By solving (4), we can obtain the optimal solution x∗ = (0, 1, 0, 1, 0)T ,
i.e. the optimal pair of charging station locations are the University of Notre
Dame and the University Park Mall.

5. The Minimum Overall Transportation Energy Problem

The objective of minimizing the overall transportation energy required
when EVs owners commute between home and charging station can reduce
the global transportation energy cost. This is a crucial criterion for deciding
the location of charging stations in future sustainable and green cities.

5.1. Probability Model for Charging Station Selection

In the multiple charging stations placement case, X0 (X0 ≥ 2) charging
stations will be placed. Each EV owner can only choose one of them for

14
 Energy consumption Convenience to work place
to charging station e.g. distances to work place

Potential charging Decision Service quality of

station set Model charging stations

Selected Charging Station

Figure 4: Charging station selection scheme for multiple charging station case

charging service each time. The decision is affected by many factors, for
example, the distance dij between the charging station ci and work place
ujw of uj , the waiting time for charging service, the transportation energy
consumption Eij , etc.. This is shown in Fig. 4.
Suppose each EV owner has a certain probability to visit anyone in the
potential charging station set. Assume the probability of EV j to choose
charging station i is pij , where pij ≥ 0. In total, M charging stations can be
selected by each EV j. So the probability set for EV j is {p1j , ..., pij , ..., pM j }.
If M
P
pij = 1, each EV definitely will select one of M charging stations. Of
i=1 P
course, M i=1 pij could be less than 1 if not everyone always charges his/her
EV. For example, M
P
i=1 pij = 0.5 means that EV j only have the probability
of 50% to charge in the commute. Our proposed decision making models can
handle this case, too.
The transportation energy consumption Ej of EV j at the charging sta-
tion is
Ej = Eij w. r. t. pij (5)
The expectated value of the transportation energy consumption for EV j is
M
X
E(Ej ) = Eij pij (6)
i=1

The total transportation energy consumption for all EVs in an urban area
is therefore:
XN N X
X M
Etotal = E(Ej ) = Eij pij (7)
j=1 j=1 i=1

For establishing the probability model, the key point is the construction
of a method to calculate the probability values {pij }. Two methods are
introduced:

15
 (a) Suppose the distance from parking place to work place is dij and
the probability pij has an inverse relationship with dij . This is because EV
owners may not(or have low probability to) go to charge at a location which
is far away from the work place. An exponential dependency is highly useful
for modeling such a relationship and for generating the probability values
respectively. The exponential model can be expressed as follows:

xi e−dij
pij = PM (8)
−dij
i=1 xi e

Where xi is the indicatorP variable to show whether the charging station is

placed. Each pij ≥ 0 and M i=1 pij = 1.
(b) Similarly, when an EV goes to a charging station, it has a smaller
probability to charge at this station if this increases the energy cost. The
model in terms of energy can be expressed:

xi e−Eij
pij = PM (9)
−Eij
i=1 xi e

5.2. Optimization Models

Assume X0 charging stations are placed. Equation (10) formulates the
constructed optimization problem that finds the optimal charging station
positions by minimizing the total expectated transportation energy cost.
M X
X N
min pij Eij
i=1 j=1

xi e−dij
s. t. Eij = f (cip , ujh , ujt ); pij = PM ∀i, j
−dij
i=1 xi e
M
X (10)
xi = X0 where xi = 0 or 1
i=1
XN
pij ≤ Si ∀i
j=1

ci ∈ C, i = 1, . . . , M ; uj ∈ U, j = 1, . . . , N

The optimization variable is xi . The objective function is the expectation

of the overall transportation energy consumption, which depends on pij . The

16
probability pij is a nonlinear function with binary variables. A constraint
PN
j=1 pij ≤ Si is used to make sure the expected number of EVs at a specific
charging station i is smaller than its service capacity Si . Such service capacity
limits exist for many reasons, e.g. if there are infrastructure and power
limitations in specific locations. The optimal solution of xi provided by this
model can ensure the selected places have enough capacity to serve all EVs.
This constraint is also nonlinear. Hence Equation (10) is a nonlinear integer
programming problem.
The total distance between charging stations and work place is also a
crucial factor. The charging station should be placed at the most convenient
positions for EV owners to reach the work place. Equation (11) is the model
that takes this factor into account. The coefficient λ is used to adjust the
weight of the total distance cost in the optimal decision making process.
M X
X N M X
X N
min pij Eij + λ pij dij
i=1 j=1 i=1 j=1

xi e−dij
s. t. Eij = f (cip , ujh , ujt ); pij = PM ∀i, j
x e−dij
i=1 i
M
X (11)
xi = X0 where xi = 0 or 1
i=1
N
X
pij ≤ Si ∀i
j=1

ci ∈ C, i = 1, . . . , M ; uj ∈ U, j = 1, . . . , N

Model (10) and (11) are nonlinear integer programming problems with
nonlinear constraints. To solve such a problem, NOMAD in (Audet and Den-
nis Jr (2006); Le Digabel (2011)) is a good choice. It uses a Mesh Adaptive
Direct Search algorithm to solve non-differentiable and global nonlinear pro-
grams. Algorithm 1 includes the detailed procedures to solve the problem of
multiple charging station placement.

5.3. Simulations
The Simulated City (SimCity) in Fig. 5 will be used to demonstrate the
proposed optimization models. It provides 50 × 50 avenues and streets. In
total 2500 locations are considered as potential charging station positions.

17
 Algorithm 1: Solver for Multiple Charging Station Problem via Min-
imum Overall Energy Consumption
(1) Obtain the profile set U for all EVs.
(2) Determine the potential charging station positions and their
corresponding profiles using set C.
(3) Calculate the energy consumption set {Eij } and the distance set
{dij }.
(4) Construct the charging station selection probability model {pij }.
(5) Establish the optimization model based on the obtained data and
formulate it in matrix form with vector variable x = (x1 , ..., xM )T
(6) Apply the obtained optimization model using the NOMAD solver.
(7) Transfer the optimal solution of x∗ into real charging station
locations.

Eight suburban areas are connected to the city center by highways. Each
suburban area has its own specific number of EVs, EV types, altitudes, etc, as
shown in Fig. 5. By applying the SimCity data to the proposed optimization
model, the model parameters can be obtained, e.g. M=2500, N=940. This
is the complete set of possible charging station locations in this simulation,
while in a real situation, only a subset of these locations would be used due
to other practical considerations. The information such as home and work
address of EVs and the type of EV will be used for the energy cost calculation.

(1) The Single Charging Station Case

In this case, only one charging station will be placed according to the
optimization model in Equation (10). Since all 2500 intersection points can
be considered as a potential charging station location, the overall transporta-
tion energy consumption for all EVs to each possible position needs to be
calculated. Then the position with minimum overall transportation energy
consumption for all EVs can be solved according to the optimization model.
With the current suburban area information in Fig. 5, position (1,15)(at
the lower left of the SimCity) is the optimal location for this single charging
station placement problem to achieve minimum overall energy cost. Chang-
ing the number of EVs in some suburban areas, e.g. from 50 to 150 in
Suburban Area 3, from 60 to 160 in Suburban Area 4, from 50 to 150 in
Suburban Area 5, from 100 to 200 in Suburban Area 6, the same calculation
and optimization procedures are performed. In this case the optimal location

18
 Distance: 30km Distance: 40km
Speed limit:100km/h Speed limit:80km/h
EV:100 EV:50
Altitude:-80m Altitude:-50m
Suburban Area 5 Distance: 60km
Suburban Area 6 Speed limit:80km/h
Distance: 50km 10 km 50 streets
Speed limit:100km/h EV:60
Altitude:200m
EV:180
Altitude:-150m Suburban Area 4

Suburban Area 7 Distance: 20km

Speed limit:80km/h
10 km
EV:50
50 avenues Altitude:100m
EV:100
Altitude:-120m
Suburban Area 3

Suburban Area 8
Distance: 40km Distance: 50km
Speed limit:80km/h Distance: 40km Speed limit:100km/h
Speed limit:100km/h

EV:200 EV:200
Altitude:-100m Altitude:0

Suburban Area 1 Suburban Area 2

Figure 5: An example of SimCity (Yi and Bauer (2014b))

(4,35) is obtained. The two different results show that the optimal charg-
ing infrastructure position definitely depends on the distribution of EVs in
suburban areas.
(2) Multiple Charging Station Case
Multiple charging stations will be placed based on the model in (11). In
order to make this problem scalable, a pre-selection has been applied. Only
100 positions (shown in Fig. 6) in SimCity have been used as the potential
location set. Each potential charging station position has been randomly
assigned with service capability Si (the maximum servable number of EV
during one day). The other parameters for the optimization model are the
same as those in the single charging station case. Then 10 optimal charging
station positions will be selected from the 100 potential locations.
In model (11), the value of λ serves to change the weight between energy
cost and distance to the work place. Different values of λ (e.g. 0, 0.01, 0.1, 1,
10) have been used to perform this simulation. Different values of λ generate
totally different placement strategies. The results in Fig. 6 show that the
selected optimal charging station locations become more scattered when the
value of λ becomes larger. This is because the total distance cost related to
work place location contributes more to the total optimization cost.

19
 100 Potential Positions X0=10 with λ=0 X0=10 with λ=0.01
50 50 50

45 45 45

40 40 40

35 35 35

30 30 30

25 25 25

20 20 20

15 15 15

10 10 10

5 5 5

5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50

X0=10 with λ=0.1 X0=10 with λ=1 X0=10 with λ=10

50 50 50

45 45 45

40 40 40

35 35 35

30 30 30

25 25 25

20 20 20

15 15 15

10 10 10

5 5 5

5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50

Figure 6: Placement of Multiple charging stations for different values of λ

5.4. Case Study: Chicago

The Chicago metropolitan area is used to demonstrate the optimal charg-
ing station placement which aims to minimize overall transportation energy
cost.
(1) Pre-selection of the Set of Potential Charging Station Position
In order to obtain a potential position set for optimal charging station
placement, data mining techniques are applied to analyze the data set–
Chicago Regional Household Travel Inventory(CRHTI) on CMAP (2006).
This data set is a comprehensive study of the demographic and travel be-
havior characteristics of residents in the greater Chicago area. The criterion
“Positions with larger average parking time have higher probability as poten-
tial charging station locations” is used to implement the pre-selection. The
parking time analysis is perform on the whole data set to obtain the parking
time distribution for all the places visited. Fig. 7 provides histogram statis-
tics of parking times. The cases with abnormally low or high parking time
of more than 12 hours or less than 30 minutes are discarded. Hence, initial
cleaning actions are necessary to obtain more accurate analyses results. In
order to obtain potential charging station positions, the following steps are
performed:
a) Use the K-means clustering algorithm to classify the sample set into
N classes.
b) Calculate the average parking time for each class.

20
 Histogram of Parking Time

1000 2000
Frequency
0
100 200 300 400 500 600 700
Parking Time(min)

Figure 7: Histogram of parking time for samples with less than 12 hours and more than
30 minutes

Figure 8: 50 potential charging station positions in Chicago and the corresponding his-
togram for average parking time

c) Classes with larger parking time have higher priority for placing charg-
ing stations and will be selected more likely.
In order to determine the set of potential charging station locations in
this case study, the K-means clustering algorithm with 50 classes is used to
classify the whole data set. The corresponding centroid of each class is used
to represent a whole class. The histogram of average parking times and the
positions on the map for these 50 clusters are provided in Fig. 8. The result
shows that most of these clusters have an average parking time between 3
and 4 hours.
(2) Charging Station Placement in Chicago
The Chicago metropolitan area is divided by the ZIP code areas. For each
ZIP code area, the corresponding population is obtained from the dataset
on UnitedStatesZipCodes.org (2015). Consider each ZIP code area as an

21
entirety with the location and population information. When we calculate
the transportation energy for traveling to potential charging station positions,
EVs in the same ZIP code area have the same location information. In the
greater Chicago area, the same 10 counties as investigated in dataset CRHTI
(CMAP (2006)) are considered and the detailed information is displayed in
Table 4. In summary, there are totally 344 ZIP code areas with a total
population of 7500299 in the entire city area.

Table 4: ZIP codes and Population Information in the greater Chicago Area
County ZIP Code Number Population
Cook County 163 4093345
DuPage County 36 869041
Grundy County 7 43937
Kane County 18 365911
Kendall County 6 86351
Lake County 51 978188
McHenry County 19 294188
Will County 29 552783
La Porte County 9 85518
Porter County 6 131037

Assume the entire greater Chicago area will be considered for the over-
all energy cost calculation. By considering the obtained potential charging
station position set, we can use M = 50 and then N = 7500299. Each en-
ergy consumption value Eij is obtained by using the developed software tool
EVRE (Yi and Bauer (2014a)). After all the energy components are cal-
culated, the probability models will be constructed. Then the optimization
model can be established to find the optimal position subset among the 50
potential charging station positions.
1) The Single Charging Station Case
Only one particular position will be selected from the potential set. Fig. 9(a)
gives the optimal position(red marker) among the 50 potential positions. The
result appears reasonable due to the optimal location in the center of down-
town which has the highest population density. The other blue markers are
the potential positions in Cook county, and the corresponding energy con-
sumption for travel to these positions are provided in the sensitivity analysis
in Fig. 9(b). The result shows that an increase in energy consumption has
an approximately quadratic relationship with driving distance to the opti-

22
 7
4.5 ×10
4

Energy Difference(kWh)
3.5
3
2.5
2
1.5
1
0.5
0
0 10 20 30 40 50 60
Driving Distance to Optimal Position(km)

(a) (b)

Figure 9: (a) Single optimal charging station placement in Chicago; (b) Sensitivity analysis

mal location. If the charging stations are placed far away from the optimal
positions, energy cost for EVs will increase dramatically. This shows that
the energy cost component has a strong dependency on location.
2) Multiple Charging Stations Case
In this case, multiple positions will be selected for charging station place-
ment. Particularly 10 out of 50 potential positions are selected in this case
study. Fig. 10 provides the placement results. Fig. 10(a) shows the 10 opti-
mal positions for charging station placement. More of the selected positions
are located around the city center, which appears intuitively reasonable. The
positions are scattered enough outside the city center to minimize the energy
cost for EVs in suburban areas. This is due to the lower population density
in the suburban areas.
The other results in Fig. 10(b)–Fig. 10(c) are used to show the sensitiv-
ity of our optimization model. The sensitivity is checked by increasing the
desired location number from 10 to 12 and 15. The number propagation
process shows that, most of the selected positions are the same, only a small
part of positions are added or changed. All these results indicate a very good
robustness of our optimization model to find the optimal subset even when
the required number of charging stations are increasing over time.

23
 (a)

(b)

(c)

Figure 10: Multiple charging station placement in Chicago: (a) 10 charging stations; (b)
12 charging stations; (c) 15 charging stations
24
6. Multi-objective Optimization Model for Charging Station Place-
ment
6.1. Optimization Model
Two energy-aware objective functions have been discussed thoroughly in
previous sections, including maximum number of reachable households and
minimum overall transportation energy consumption for charging actions.
Both are crucial for charging station owners and EV users and definitely
play a role in real world placement strategies and planning. Indeed these two
objective functions are in conflict with each other, because a larger reachable
range causes more energy cost for charging actions of EVs that are far away
from charging station.
The multiple-objective optimization is a promising technique to obtain
better decision making when dealing with different objective functions. It
provides a compromise between the considered objectives. Assume the new
objective function vector is f (x) with two required scalar objective functions:
f (x) = (−f1 (x), f2 (x))T , where f1 (x) =Pp~T u(I~
x − 0.5) is for maximizing
M PN
number of reachable households, f2 (x) = i=1 j=1 pij Eij is for minimizing
overall transportation energy consumption for charging actions. Due to the
maximizing requirement for f1 (x), we use the negative objective function for
reachable households in the overall function vector. This operation aims to
use the two conflicting objective functions in the same vector function. Then
the multiple objective optimal charging station placement decision making
framework is given by:
M X
X N
T
min f (x) = (−~p u(I~x − 0.5), pij Eij )T
i=1 j=1

s. t. |x|1 = X0 where xi = 0 or 1
N
(12)
X
pij ≤ Si ∀i
j=1

ci ∈ C, i = 1, . . . , M ; uj ∈ U, j = 1, . . . , N

where the corresponding constraints are the combination of required con-

straints in two single objective optimization models. Hence, this optimization
model aims to find the optimal placement strategy x∗ for charging stations
to satisfy both objective requirements.

25
6.2. Scalar Method for Optimal Solution
The Pareto optimal solution will be used as the optimal strategy x∗ for
Equation (12). Pareto optimality is a resource allocation method in which it
is impossible to make any one individual objective component better without
making at least one other objective component worse.
Several methods can be used to solve multiple objective optimization
problems for Pareto optimal solutions (Collette and Siarry (2003)). The
scalar method is one of the highly efficient methods. The weighted sum
of objective functions method in the scalar method framework is simple to
implement for multi-objective optimization. This type of method will be
used in this paper in order to solve the multiple objective charging station
placement problem.
The goal is to transform Equation (12) to a single objective optimization
problem. In case the transformed problem is within the single objective
optimization framework, we can use the same solver proposed in Algorithm 1
to find the optimal solution. Suppose there are two weights ω1 and ω2 , ω1 ≥ 0,
ω2 ≥ 0 and ω1 + ω2 = 1. Then the corresponding scalar objective function
can be expressed as

fw (x) = −ω1 f1 (x) + ω2 f2 (x) (13)

Given a pair of weights (ω1 , ω2 ), we can solve a single objective optimiza-

tion problem to obtain the optimal solution. This single objective function
has objective function fw (x) and the same constraints as in Equation (12).
Different pairs of (ω1 , ω2 ) provide different optimal solutions for single ob-
jective optimization problems. From these obtained solutions, the Pareto
optimal solution for the original multiple objective problem will be found.
Algorithm 2 is proposed to summarize the procedure for solving the charging
station placement problem.

6.3. Case Study: Chicago

As in the minimum transportation energy consumption case, the greater
Chicago area is used to demonstrate our multiple objective optimization
model. The same 50 locations in Fig. 8 are used as potential charging station
positions. Fig. 11 provides the results for the single objective optimization
model with each single criterion applied seprately, including the maximum
number of reachable household and minimum overall transportation energy
cost. Both of the results are the 10 optimal locations selected out of the

26
 Algorithm 2: Pareto Optimal Solution for Multiple Objective Charg-
ing Station Placement
(1) Transform the multiple objective optimization problem into the
single objective optimization problem by the weighted sum of
objective function method.
(2) N different values of ω1 in [0, 1] will be used and ω1 + ω2 = 1.
(3) For each pair (ω1i , ω2i ), i = 1, ..., N , the scaled single objective
i
optimization problem is solved to obtain the optimal solution fwp .
i i i
(4) For each fwp , the optimal values of f1p and f2p for two objective
values can be found.
i i
(5) With all N pairs of (f1p , f2p ), the Pareto optimal solution point
can be obtained.

Maximum number Minimum

of Reachable Overall
Households
Transportation
Energy Cost

!
Figure 11: Charging station placement in Chicago for two different criteria

potential 50 locations. We can see the result for maximum reachable house-
holds is much more scattered, while the selected positions will become more
dense by using the minimum overall transportation energy cost. Both are
reasonable results due to the corresponding characteristic of the used crite-
rion.
In order to obtain the Pareto optimal solution for the proposed multiple
objective decision-making model, different ω1 s are used to find different op-
timal solutions for all the obtained single objective optimization problems in
the weighted Algorithm 2. The corresponding values of −f1 (x) and f2 (x) are
shown in Fig. 12. The results provide the possible optimal solutions for all
the checked weights. Then the Pareto optimal solution point is the one with
the red circle in Fig. 12.
With respect to this Pareto optimal solution point, we can obtain the
corresponding optimal value of x∗ , and then mark it on Google Maps, as

27
 Objective Value of Total Energy Consumption(f 2(x))(J)
7
8.2 ×10

7.8

7.6

7.4

7.2

6.8

6.6
-7.5 -7.45 -7.4 -7.35
Negative Objective Value of Total Reachable Households(-f1(x)) ×106

Figure 12: Two objective optimization with different values of ω1 and ω2

Figure 13: Charging station placement in Chicago for multiple objectives

shown in Fig. 13. Comparing this result to Fig. 11, the selected charging
station locations in Fig. 13 have an equilibrium (not too scattered and at same
time not too dense) that is between the results from application of a single
criterion. This result provides a more balanced decision making strategy
for charging station placement by considering two important requirements
simultaneously.

7. Conclusions
In order to accelerate transportation electrification for future sustain-
able transportation systems, an optimal energy-aware charging infrastruc-
ture placement framework has been introduced. Two main objective re-
quirements are proposed: maximizing the number of reachable households
under an energy constraint and minimizing the overall transportation energy
consumption for charging actions. Single objective optimization models for

28
the two requirements have been discussed. The corresponding algorithms
for obtaining optimal decision variables have been proposed. Detailed sim-
ulations and real city case studies have been provided to demonstrate these
models.The results show that the introduced models are promising to pro-
vide good strategies for charging station placement. Additionally, a multiple
objective optimization model is established to handle both proposed criteria
simultaneously and the corresponding case study for Chicago demonstrates
significant improvements by considering two requirements simultaneously.
The here presented method allows to incorporate additional constraints that
originate from criteria such as power infrastructure, popularity of sites, etc.
by apriori reducing the size of the set of all possible locations. Therefore
additional criteria are easily integrated into the presented approach. Finally,
we would like to point out that quick charging stations are not considered in
this treatise.

Acknowledgment
We gratefully acknowledge the support of NSF grant # 1239224

8. References
Andrews, M., Dogru, M. K., Hobby, J. D., Jin, Y., Tucci, G. H., 2013.
Modeling and optimization for electric vehicle charging infrastructure. In:
Innovative Smart Grid Technologies Conference 2013.

Audet, C., Dennis Jr, J. E., 2006. Mesh adaptive direct search algorithms for
constrained optimization. SIAM Journal on optimization 17 (1), 188–217.

Baouche, F., Billot, R., Trigui, R., El Faouzi, N., 2014. Efficient allocation of
electric vehicles charging stations: Optimization model and application to
a dense urban network. IEEE Intelligent Transportation Systems Magazine
6 (3), 33–43.

Bauer, P. H., Mingo, S., Lantero, B., Larkin, J., Oct. 2012. On fuel economy
bounds. In: Vehicle Power and Propulsion Conference (VPPC). IEEE,
Seoul, South Korea, pp. 816–819.

Bilgin, B., Magne, P., Malysz, P., Yang, Y., Pantelic, V., Preindl, M., Ko-
robkine, A., Jiang, W., Lawford, M., Emadi, A., June 2015. Making the

29
 case for electrified transportation. IEEE Transactions on Transportation
Electrification 1 (1), 4–17.

Bouscayrol, A., Hissel, D., Trigui, R., Emadi, A., November 2011. Special
section on advanced transportation systems. IEEE Transactions on Vehic-
ular Technology 60 (9), 4102–4105.

Cai, H., Jia, X., Chiu, A. S., Hu, X., Xu, M., 2014. Siting public electric
vehicle charging stations in beijing using big-data informed travel patterns
of the taxi fleet. Transportation Research Part D: Transport and Environ-
ment 33, 39–46.

Cavadas, J., Correia, G., Gouveia, J., 2014. Electric vehicles charging net-
work planning. In: Computer-based Modelling and Optimization in Trans-
portation. Springer, pp. 85–100.

Cavadas, J., de Almeida Correia, G. H., Gouveia, J., 2015. A mip model
for locating slow-charging stations for electric vehicles in urban areas ac-
counting for driver tours. Transportation Research Part E: Logistics and
Transportation Review 75, 188–201.

Chen, T. D., Kockelman, K. M., Khan, M., 2013. Locating electric vehicle
charging stations: Parking-based assignment method for seattle, wash-
ington. Transportation Research Record: Journal of the Transportation
Research Board (2385), 28–36.

Chung, S. H., Kwon, C., 2015. Multi-period planning for electric car charg-
ing station locations: A case of korean expressways. European Journal of
Operational Research 242 (2), 677–687.

CMAP, 2006. Chicago regional household travel inventory.

URL https://www.cmap.illinois.gov/data/transportation/
travel-tracker-survey

Collette, Y., Siarry, P., 2003. Multiobjective optimization: principles and

case studies. Springer Science & Business Media.

Dong, J., Liu, C., Lin, Z., 2014. Charging infrastructure planning for promot-
ing battery electric vehicles: An activity-based approach using multiday
travel data. Transportation Research Part C: Emerging Technologies 38,
44–55.

30
Eberle, U., von Helmolt, R., 2010. Sustainable transportation based on elec-
tric vehicle concepts: a brief overview. Energy & Environmental Science
3 (6), 689–699.
Feng, L., Ge, S., Liu, H., March 2012. Electric vehicle charging station plan-
ning based on weighted voronoi diagram. In: Asia-Pacific Power and En-
ergy Engineering Conference (APPEEC). IEEE, Shanghai, China, pp. 1–5.
Frade, I., Ribeiro, A., Gonçalves, G., Antunes, A. P., 2011. Optimal location
of charging stations for electric vehicles in a neighborhood in lisbon, portu-
gal. Transportation research record: journal of the transportation research
board 2252 (1), 91–98.
Ge, S., Feng, L., Liu, H., Sept. 2011. The planning of electric vehicle charging
station based on grid partition method. In: International Conference on
Electrical and Control Engineering (ICECE). IEEE, Yichang, China, pp.
2726–2730.
Ghamami, M., Nie, Y., Zockaie, A., 2014. Planing charging infrastructure for
plug-in electric vehicles in city centers. International Journal of Sustainable
Transportation.
Guo, S., Zhao, H., 2015. Optimal site selection of electric vehicle charging
station by using fuzzy topsis based on sustainability perspective. Applied
Energy 158, 390–402.
Hatton, C., Beella, S., Brezet, J., Wijnia, Y., May13-16 2009. Charging
stations for urban settings the design of a product platform for electric
vehicle infrastructure in dutch cities. In: Towards zero emission: EVS 24
International Battery, Hybrid and Fuel Cell Electric Vehicle Symposium
& Exhibition. European Association of Electric Road Vehicles, Stavanger,
Norway.
He, F., Wu, D., Yin, Y., Guan, Y., 2013. Optimal deployment of public charg-
ing stations for plug-in hybrid electric vehicles. Transportation Research
Part B: Methodological 47, 87–101.
Hess, A., Malandrino, F., Reinhardt, M. B., Casetti, C., Hummel, K. A.,
Barceló-Ordinas, J. M., Dec. 2012. Optimal deployment of charging sta-
tions for electric vehicular networks. In: Proceedings of the first workshop
on Urban networking. ACM, Nice, France, pp. 1–6.

31
Hodgson, M. J., 1990. A flow-capturing location-allocation model. Geograph-
ical Analysis 22 (3), 270–279.

Ip, A., Fong, S., Liu, E., Nov. 2010. Optimization for allocating bev recharg-
ing stations in urban areas by using hierarchical clustering. In: 6th Inter-
national Conference on Advanced Information Management and Service
(IMS). IEEE, Seoul, South Korea, pp. 460–465.

Kates, R. W., Parris, T. M., Leiserowitz, A. A., 2005. What is sustainable

development? Environment 47 (3), 8.

Khaligh, A., Krishnamurthy, M., October 2012. Special section on sustain-

able transportation systems. IEEE Transactions on Vehicular Technology
61 (8), 3362–3364.

Khalkhali, K., Abapour, S., Moghaddas-Tafreshi, S. M., Abapour, M., 2015.

Application of data envelopment analysis theorem in plug-in hybrid elec-
tric vehicle charging station planning. IET Generation, Transmission &
Distribution 9 (7), 666–676.

Koyanagi, F., Uriu, Y., Yokoyama, R., Sept. 2001. Possibility of fuel cell
fast charger and its arrangement problem for the infrastructure of electric
vehicles. In: Power Tech Proceedings. Vol. 4. IEEE, Porto.

Koyanagi, F., Yokoyama, R., Sept. 2010. A priority order solution of ev

recharger installation by domain division approach. In: 45th International
Universities Power Engineering Conference (UPEC). IEEE, Cardiff, Wales,
pp. 1–8.

Kuby, M., Lim, S., 2005. The flow-refueling location problem for alternative-
fuel vehicles. Socio-Economic Planning Sciences 39 (2), 125–145.

Lam, A., Leung, Y.-W., Chu, X., Oct. 2013. Electric vehicle charging station
placement. In: International Conference on Smart Grid Communications
(SmartGridComm). IEEE, Vancouver, Canada, pp. 510–515.

Lam, A. Y., Leung, Y.-W., Chu, X., 2014. Electric vehicle charging station
placement: Formulation, complexity, and solutions. IEEE Transactions on
Smart Grid 5 (6), 2846–2856.

32
Le Digabel, S., 2011. Algorithm 909: Nomad: Nonlinear optimization with
the mads algorithm. ACM Transactions on Mathematical Software 37 (4),
44.

Lee, Y.-G., Kim, H.-S., Kho, S.-Y., Lee, C., 2014. User equilibrium-based
location model of rapid charging stations for electric vehicles with batter-
ies that have different states of charge. Transportation Research Record:
Journal of the Transportation Research Board (2454), 97–106.

Liu, J., 2012. Electric vehicle charging infrastructure assignment and power
grid impacts assessment in beijing. Energy Policy 51, 544–557.

Liu, Z., Wen, F., Ledwich, G., 2013. Optimal planning of electric-vehicle
charging stations in distribution systems. IEEE Transactions on Power
Delivery 28 (1), 102–110.

Ma, T., Zhao, J., Xiang, S., Zhu, Y., Liu, P., 2014. An agent-based training
system for optimizing the layout of afvs’ initial filling stations. Journal of
Artificial Societies and Social Simulation 17 (4), 1–16.

Momtazpour, M., Butler, P., Hossain, M. S., Bozchalui, M. C., Ramakrish-

nan, N., Sharma, R., August 2012. Coordinated clustering algorithms to
support charging infrastructure design for electric vehicles. In: Proceed-
ings of the ACM SIGKDD International Workshop on Urban Computing.
ACM, Beijing, China, pp. 126–133.

Pashajavid, E., Golkar, M., 2013. Optimal placement and sizing of plug in
electric vehicles charging stations within distribution networks with high
penetration of photovoltaic panels. Journal of Renewable and Sustainable
Energy 5 (5), 053126.

Sadeghi-Barzani, P., Rajabi-Ghahnavieh, A., Kazemi-Karegar, H., 2014. Op-

timal fast charging station placing and sizing. Applied Energy 125, 289–
299.

Sathaye, N., Kelley, S., 2013. An approach for the optimal planning of electric
vehicle infrastructure for highway corridors. Transportation Research Part
E: Logistics and Transportation Review 59, 15–33.

Sebastiani, M. T., Lüders, R., Fonseca, K. V. O., 2014. Allocation of charg-

ing stations in an electric vehicle network using simulation optimization.

33
 In: Proceedings of the 2014 Winter Simulation Conference. IEEE Press,
Piscataway, NJ, USA, pp. 1073–1083.

Smart, J., Schey, S., 2012. Battery electric vehicle driving and charging be-
havior observed early in the ev project. SAE Int. J. Altern. Powertrains
1 (1), 27–33.

Sweda, T., Klabjan, D., Sep. 2011. An agent-based decision support system
for electric vehicle charging infrastructure deployment. In: Vehicle Power
and Propulsion Conference (VPPC). IEEE, Chicago, IL, USA, pp. 1–5.

Sweda, T. M., Klabjan, D., 2015. Agent-based information system for elec-
tric vehicle charging infrastructure deployment. Journal of Infrastructure
Systems 21 (2), 04014043.

UnitedStatesZipCodes.org, 2015. Zip code database.

URL http://www.unitedstateszipcodes.org/zip-code-database/

Wagner, S., Brandt, T., Neumann, D., June 2014. Smart city planning-
developing an urban charging infrastructure for electric vehicles. In: Pro-
ceedings of the Twenty Second European Conference on Information Sys-
tems. Tel Aviv, Israel, pp. 1–15.

Wagner, S., Götzinger, M., Neumann, D., Dec. 2013. Optimal location of
charging stations in smart cities: A points of interest based approach. In:
Proceedings of The 34th International Conference on Information Systems
(ICIS). Milan, Italy.

Wang, G., Xu, Z., Wen, F., Wong, K. P., 2013a. Traffic-constrained multi-
objective planning of electric-vehicle charging stations. IEEE Transactions
on Power Delivery 28 (4), 2363–2372.

Wang, Y.-W., Lin, C.-C., 2013. Locating multiple types of recharging sta-
tions for battery-powered electric vehicle transport. Transportation Re-
search Part E: Logistics and Transportation Review 58, 76–87.

Wang, Z., Liu, P., Cui, J., Xi, Y., Zhang, L., 2013b. Research on quantitative
models of electric vehicle charging stations based on principle of energy
equivalence. Mathematical Problems in Engineering 2013, 1–10.

34
Wikipedia, 2015. Electric car use by country.
URL http://en.wikipedia.org/wiki/Electric_car_use_by_country

Wirges, J., Linder, S., Kessler, A., 2012. Modelling the development of a
regional charging infrastructure for electric vehicles in time and space.
European Journal of Transport and Infrastructure Research 12 (12), 391–
416.

Worley, O., Klabjan, D., Sweda, T. M., March 2012. Simultaneous vehicle
routing and charging station siting for commercial electric vehicles. In:
International Electric Vehicle Conference (IEVC). IEEE, Greenville, SC,
USA, pp. 1–3.

Xi, X., Sioshansi, R., Marano, V., 2013. Simulation–optimization model for
location of a public electric vehicle charging infrastructure. Transportation
Research Part D: Transport and Environment 22, 60–69.

Xu, H., Miao, S., Zhang, C., Shi, D., 2013. Optimal placement of charging
infrastructures for large-scale integration of pure electric vehicles into grid.
International Journal of Electrical Power & Energy Systems 53, 159–165.

Yao, W., Zhao, J., Wen, F., Dong, Z., Xue, Y., Xu, Y., Meng, K., 2014. A
multi-objective collaborative planning strategy for integrated power distri-
bution and electric vehicle charging systems. IEEE Transactions on Power
Systems 29 (4), 1811–1821.

Yi, Z., Bauer, P. H., April 2014a. Energy consumption model and charging
station placement for electric vehicles. In: 3rd International Conference on
Smart Grids and Green IT Systems. Barcelona, Spain, pp. 150–156.

Yi, Z., Bauer, P. H., Oct. 2014b. Spatio-temporal energy demand models for
electric vehicles. In: Vehicle Power and Propulsion Conference (VPPC).
IEEE, Coimbra, Portugal, pp. 1–6.

You, P.-S., Hsieh, Y.-C., 2014. A hybrid heuristic approach to the prob-
lem of the location of vehicle charging stations. Computers & Industrial
Engineering 70, 195–204.

35
 Highlights

l An energy-aware optimization framework is proposed for charging station

placement.

l Multi-objective optimization model with two energy-aware criteria is introduced.

l A detailed EV energy cost model is employed to construct optimization models.

l Mesh adaptive direct search method for solving proposed models is utilized.

l Real world datasets are investigated to perform the case studies.