Professional Documents
Culture Documents
Transportation Science
Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org
. https://doi.org/10.1287/trsc.2021.0494
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.
With 12,500 members from nearly 90 countries, INFORMS is the largest international association of operations research (O.R.)
and analytics professionals and students. INFORMS provides unique networking and learning opportunities for individual
professionals, and organizations of all types and sizes, to better understand and use O.R. and analytics tools and methods to
transform strategic visions and achieve better outcomes.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
TRANSPORTATION SCIENCE
Articles in Advance, pp. 1–19
https://pubsonline.informs.org/journal/trsc ISSN 0041-1655 (print), ISSN 1526-5447 (online)
a
Department of Management Sciences, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada; b Amazon.com, Seattle, Washington 98109
*Corresponding author
Contact: obkinay@uwaterloo.ca (ÖBK); fgzara@uwaterloo.ca, https://orcid.org/0000-0003-2893-8597 (FG); sibel.alumur@uwaterloo.ca,
https://orcid.org/0000-0003-1550-8053 (SAA)
Received: November 26, 2021 Abstract. This paper studies the problem of determining the strategic location of charging
Revised: October 31, 2022; March 10, 2023 stations and their capacity levels under stochastic electric vehicle flows and charging times
Accepted: May 8, 2023 taking into account the route choice response of users. The problem is modeled using bilevel
Published Online in Articles in Advance: optimization, where the network planner or leader minimizes the total infrastructure cost of
June 16, 2023 locating and sizing charging stations while ensuring a probabilistic service requirement on
the waiting time to charge. Electric vehicle users or followers, on the other hand, minimize
https://doi.org/10.1287/trsc.2021.0494 route length and may be cooperative or noncooperative. Their choice of route in turn deter
mines the charging demand and waiting times at the charging stations and hence, the need
Copyright: © 2023 INFORMS
to account for their decisions by the leader. The bilevel problem reduces to a single-level
mixed-integer model using the optimality conditions of the follower’s problem when the
charging stations operate as M/M/c queues and the followers are cooperative. To solve the
bilevel model, a decomposition-based solution methodology is developed that uses a new
logic-based Benders algorithm for the location-only problem. Computational experiments
are performed on benchmark and real-life highway networks, including a new eastern U.S.
network. The impact of route choice response, service requirements, and deviation tolerance
on the location and sizing decisions are analyzed. The analysis demonstrates that stringent
service requirements increase the capacity levels at open charging stations rather than their
number and that solutions allowing higher deviations are less costly. Moreover, the differ
ence between solutions under cooperative and uncooperative route choices is more signifi
cant when the deviation tolerance is lower.
History: This paper has been accepted for the Transportation Science Special Issue on 2021 TSL Work
shop: Supply and Demand Interplay in Transport and Logistics.
Funding: This research was supported by the Ontario Graduate Scholarship when Ö. B. Kınay was a
PhD candidate at the University of Waterloo, and this support is acknowledged.
Supplemental Material: The online appendix is available at https://doi.org/10.1287/trsc.2021.0494.
Keywords: facility location • capacity allocation • charging station • electric vehicles • bilevel optimization
1. Introduction
The transportation sector contributes to approximately passenger cars and light trucks sold in 2030 to be ZEVs
one quarter of global direct CO2 emissions, where pas (The White House 2021). The majority of the European
senger and freight road vehicles account for nearly 74% Union countries and the United Kingdom aim by
of that share (IEA 2020). The widespread adoption of 2025–2035 and Canada aims by 2040 to reach 100% ZEV
zero-emission vehicles (ZEVs), which consist of battery sales (IEA 2021). As for the private sector commitments
electric vehicles (EVs) and fuel cell electric vehicles, is toward electric mobility, the Climate Group’s EV100 Ini
considered to be one of the most effective ways to miti tiative brings together over 100 companies dedicated to
gate the negative environmental impacts of transporta making electric transport the new normal by 2030 (The
tion and to promote environmental sustainability (Zeng Climate Group 2021a). Many corporations are aiming to
et al. 2021). speed up the transition to electric mobility by converting
The Electric Vehicles Initiative, which was established their fleets to EVs and designing private networks by
by 15 countries including Canada, China, Germany, locating charging stations.
and the United Kingdom, has set an aspirational goal of Despite these global initiatives and endeavors, signifi
reaching a 30% sales share of EVs by 2030. The United cant adoption barriers remain. In particular, the EV100
States declared their goal of achieving 50% of all new members reported the lack of charging infrastructure
1
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
2 Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS
as their top concern (The Climate Group 2021b). To ad decisions are made. We use a bilevel structure where
dress this challenge and improve spatial distribution of the leader is a regulatory body that seeks to establish the
charging stations, many countries have set investment charging infrastructure by determining the strategic
goals for the deployment of publicly accessible charging locations and sizes of the charging stations (e.g., govern
infrastructure. For example, EU countries are mandated ments or private EV network operators) and the follower
to set deployment targets for publicly accessible EV is the set of EV users who seek to complete their long-
chargers for the decade ahead, with a suggested ratio of distance EV trips.
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
one charging stall per 10 EVs (European Commission The leader’s problem minimizes the total cost of locat
2019). ing and sizing charging stations while ensuring a con
The majority of shorter-distance trips using EVs typ venient service level taking follower’s response into
ically rely on the private slow chargers readily avail account. The follower’s problem, on the other hand, is
able at homes and workplaces. However, private slow defined to incorporate the route and charging station
charging has to be complemented by publicly acces choice of EV users while minimizing the length of their
sible en route fast-charging stations to support EV trips. The choice of facilities by the EV users (i.e., the
charging on the go for long-distance trips. Further allocation of recharging demand to charging stations) is
more, recent advancements in battery and charging crucial for the leader because the congestion level at sta
technologies for passenger EVs are paving the way tions will directly impact the required capacity.
toward long-haul freight distribution with heavy-duty The generic model that we propose assumes a stochas
EVs (i.e., electric trucks (ETs)). Lately, manufacturers, tic stream of EV flow, stochastic service times at charging
such as Tesla, Daimler AG, and Volvo, have unveiled stations, and a homogeneous fleet of EVs with a deter
their plans for mass production of ETs within this ministic battery range. The EVs energy consumption is
decade. The International Energy Agency states that assumed to be proportional to the travel distance. There
planning needs to start now for megachargers to enable is a shortest path deviation tolerance for EV users, and
long-distance electric trucking as there is currently no they cannot exceed this tolerance while completing their
infrastructure to support it (IEA 2021). Therefore, the fea OD trips. In order to take range anxiety and energy con
sibility and convenience of long-distance trips with electric sumption variability into account in our computational
vehicles depend on the spatial distribution of adequately experiments, without loss of generality, we assume that
capacitated Direct Current (DC) fast-charging stations. EV users start their journey with 80% of the available bat
The capacity of a DC fast-charging station is defined tery range, and they may charge up to 80% at a charging
as the number of charging stalls, also known as outlets, station visit. Charging stalls at stations are considered to
chargers, or electric vehicle supply equipment. Consid be homogeneous, and their count at a single station can
ering the relatively long recharging times for EVs even not exceed a predefined limit. There is a fixed cost of
using fast chargers, waiting times may become a major locating a charging station and a variable cost of capacity
inconvenience. Moreover, the demand for charging at a allocation, which is a nondecreasing function of the num
charging station is hardly known in advance. Hence, ber of charging stalls installed at a station. As a special
the stochasticity of demand and service times should case of this generic setting, we additionally introduce a
not be neglected when determining the locations and model that assumes an M/M/c queuing system for
capacities of the charging stations. charging stations and configures the problem on an open
In this work, we seek to find the optimal locations of multiserver Jackson network (i.e., with Poisson arrivals,
DC fast-charging stations and their respective number of exponential service times, and first come, first served and
chargers to serve a given set of origin-destination (OD) single-line queues at stations). The cooperative and unco
EV trips on a transportation network. To ensure quality operative user responses further the analysis of the bile
of service considering stochasticity in charging demand vel modeling. Under a cooperative response, EV users
and service times, probabilistic service-level requirements make the minimum number of stops required to com
are defined that make sure that the expected waiting plete their trips, and the model is converted into a single-
times at a charging station do not exceed inconvenience level formulation. Under the uncooperative response, we
limits. To provide cost-effective solutions, the framework assume EV users prefer to make the minimum number of
of this problem allows the possibility of detours from stops as well.
shortest paths within a threshold, referred to as the devia To the extent of our knowledge, this study is the
tion tolerance (Arslan et al. 2019). first to model a chance-constrained bilevel optimization
Given the possibility of detours, when there exist framework and that makes charging station sizing deci
alternative routing options, the decision makers need to sions based on stochastic queuing models focusing on
consider the choice of charging stations to be visited long-distance transportation with EVs. The main contri
by rational service seekers who act with self-interest. butions of our work are (i) introducing a new bilevel
This necessitates the route choice response of EV users mathematical model that allows shortest path devia
to be taken into account when the location and sizing tions without path pregeneration or enumeration of
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS 3
alternative routes, (ii) merging the bilevel nature of the problem where the coverage is in terms of capturing the
problem with realistic queuing models to incorporate vehicle flow associated with the OD trips. The main
stochastic travel demand, (iii) developing an exact solu assumptions of this formulation includes uncapacitated
tion algorithm for the uncapacitated version of the stations, deterministic demand, identical vehicles, and
problem, and (iv) developing a novel decomposition- no deviation from the shortest path. FRLM received
based algorithm (DA) that is capable of rapidly finding many modeling improvements and extensions over the
high-quality solutions and also able to handle both last decade, including but not limited to Kuby and Lim
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
cooperative and uncooperative responses of EV users. (2007), Kuby et al. (2009), Upchurch, Kuby, and Lim
Through computational experimentation, we evaluate (2009), Capar and Kuby (2012), Capar et al. (2013), Mir
the performance of the solution algorithm and show Hassani and Ebrazi (2013), Arslan and Karaşan (2016),
how the follower’s position, deviation tolerance, and and Hong and Kuby (2016). The reader can refer to Ko,
service-level parameters affect the obtained solutions Gim, and Guensler (2017) and Shen et al. (2019) for a
on benchmark instances and real-life California and comprehensive review on these studies and a synthesis
eastern U.S. highway networks. of the most relevant works that adopt variations, such as
The next section of the paper provides the literature the set-covering approach (e.g., Wang and Lin 2009,
review of the location problems regarding refueling Wang and Wang 2010, Li and Huang 2014).
and charging stations while emphasizing our contribu A key scope of extension to FRLM involves the con
tions. Section 3 gives the problem statement, defines the sideration of allowing shortest path deviations, which
underlying expanded network and its reduction proce incorporates EV users’ routing choices in the mathemati
dure, and presents the general bilevel modeling frame cal models (Kim and Kuby 2012, 2013). Given the addi
work. Section 4 provides the analysis of the bilevel tional computational complexity this extension brings,
optimization model under cooperative and uncoopera the initial computational experiments were performed
tive EV user response and M/M/c queuing systems. on small-sized benchmark networks with 25 nodes. The
Section 5 presents the logic-based Benders decomposi computational challenges posed by this extension are
tion algorithm developed for the uncapacitated version addressed by Yıldız, Arslan, and Karaşan (2016), Arslan
of the problem and a decomposition-based algorithm et al. (2019), Göpfert and Bock (2019), and Kınay, Gzara,
that can account for the cooperative or uncooperative and Alumur (2021), and the size of the solvable instances
increased significantly to over 500-node networks. All of
user response for the original problem. Section 6 intro
these studies, except Kınay, Gzara, and Alumur (2021),
duces the data sets and test instances and presents the
focus on the traditional max-cover objective and put a
results of extensive computational experiments per
limit on the extent of allowed deviation to maintain a
formed to derive managerial insights as well as to
tractable problem size and to obtain applicable solutions
assess the performance of the algorithm. Lastly, Sec
within the max-cover framework. Kınay, Gzara, and
tion 7 includes a summary of the insights, concluding
Alumur (2021) proposes a new full-cover approach and
remarks, and key takeaways.
studies optimal routing of all OD trips by minimizing
the total en route recharging as well as minimizing total
2. Literature Review cost of locating charging stations. The authors show that
Many researchers focused on the spatial dimension of this approach is superior to the max-cover one in terms
the charging infrastructure problem within an uncapaci of routing solution quality even when a deviation toler
tated framework excluding nonspatial attributes and ance is not involved.
capacity limitations. These problems can be considered None of these formerly mentioned studies consider
as extensions of a general facility location problem with a capacitated framework; thus, they neglect EV users’
additional scopes of EV-specific characteristics, such as route choice response to charging station locations or
limited vehicle range. Alternative research streams con congestion. On the other hand, the studies that model
cerning the node-based and flow-based models emerged the capacitated extensions of the charging station loca
to cover different perspectives of the problem (Upchurch tion problem (e.g., Upchurch, Kuby, and Lim 2009;
and Kuby 2010). The former approach, which consists of Jiang et al. 2012; Hosseini, MirHassani, and Hoosh
p-median and node coverage-based location models, is mand 2017) adopt approaches that impose hard limits
more suitable for applications in an urban setting (Good on the number of vehicles that can be served by each
child and Noronha 1987, Frade et al. 2011, Tu et al. 2016). open station instead of assuming that all flow passing
On the other hand, the flow-based models, led by the through a station can be served regardless of their
pioneering study by Kuby and Lim (2005), received volume. Additionally, there are studies that consider
more attention as they are applicable for enabling longer capacity-level expansions under a multiperiod frame
EV trips without running out of battery. work. Among these, Zhang, Kang, and Kwon (2017)
The flow-refueling location model (FRLM) proposed extends the capacitated FRLM considering the demand
by Kuby and Lim (2005) is a maximum covering location dynamics (increasing EV market share) throughout the
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
4 Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS
planning horizon. Anjos, Gendron, and Joyce-Moniz station location and sizing. Their model allows for incor
(2020) uses a hybrid approach that takes both node- porating charging price elasticity and station congestion
based (urban) and flow-based (long-distance) travel into a discrete set of route choices for EV users under
demands into account. elastic demand. However, station capacity deployment
There are stochastic modeling approaches that use is still modeled in a deterministic way. The numerical
queuing theory to determine station capacity levels. Xie experiments are performed on an urban network with
et al. (2018) and Xie and Lin (2021) use M/M/c queuing 74 nodes. A genetic algorithm is used to find solutions
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
systems to model facility congestion under the context to the proposed model. This algorithm uses a queuing
of multiperiod capacity expansion and charging station averse logit choice model that determines users’ alloca
location problem. The former proposes a stochastic tions to charging stations. Makhlouf, Kchaou-Boujelben,
optimization model that minimizes total system cost and Gicquel (2019) also propose a bilevel problem where
defined as the sum of the capital cost of locating and the upper-level problem is a max cover-type station loca
expanding charging stations and the penalty cost of not tion and sizing problem and the lower-level problem
satisfying (covering) an OD trip. The deviations from represents EV-user behavior in terms of making the min
the shortest paths are taken into account by means of imum number of stops to reach their destination. The
pregenerating a number of alternative paths for each solution approach is based on a single-level reduction
OD trip. Even though deviations are considered, all assuming cooperative response, and they solve the re
alternative paths are assumed to be equivalent for the sulting reformulation using commercial solvers on ran
EV users (i.e., EV users’ responses to charging station domly generated network instances of 100 nodes.
locations are neglected). The resulting mathematical To sum up, none of the studies consider a bilevel
model is solved by a genetic algorithm under the framework that simultaneously takes the infrastructure
assumption of last-minute charging, which may cause cost of locating and sizing charging stations, route choice
suboptimal demand allocation to charging stations. The response of EV users, and stochastic waiting times at
computational experiments are based on a case study in facilities into account. Moreover, the existing studies
California. Xie and Lin (2021) extends this framework that use a bilevel model assume only a cooperative user
by introducing an inconvenience cost function of trave response, whereas we develop a solution methodology
lers in their system cost definition. This model is also that can be tailored for both the cooperative and unco
solved by a genetic algorithm, but it no longer allows operative responses of the follower. Our algorithm is
deviations from the shortest paths. capable of solving large-scale transportation networks
There are several studies that take the route choice with up to 420 nodes, which is the largest network size
behavior of EV users into account using bilevel location that has been solved within a bilevel framework in the
models under an uncapacitated framework (e.g., Jing
literature.
et al. 2017; Zheng et al. 2017; Guo, Yang, and Lu 2018;
He et al. 2018; Tran et al. 2021). These studies are more
suitable for an urban setting as they consider traffic
3. The Bilevel Charging Station Location
congestion using stochastic or deterministic user equi and Sizing Problem
librium. Hence, the case studies and numerical tests fea Let the underlying transportation network be defined
tured in these works cover smaller geographical regions. as Go � (N, Ao ), where N denotes the set of nodes (e.g.,
These mathematical models rely on path pregeneration highway intersections and population centers) and Ao
or enumeration, except for Zheng et al. (2017), in which denotes the set of directed arcs (e.g., highway segments).
the state of charge (SoC) is tracked in the model to Consider K as the set of OD pairs, where Ok and Dk
ensure path feasibility. Bilevel optimization problems denote the origin and destination nodes for OD pair
are known to be intrinsically hard to solve. Even the k ∈ K. Each OD pair represents a stochastic stream of EV
models with both linear leader and follower’s problems, flow between two population centers for which the
which are generally the simplest to solve, are shown to travel distance exceeds the EV battery range R, and thus,
be strongly NP hard (Labbé and Marcotte 2021). Typi recharging en route is necessary.
cally, solution methods used for these studies are meta A charging station may host up to C charging stalls,
heuristics, such as a genetic algorithm or large-scale and each stall offers stochastic service times. As a result
neighborhood search, or a single-level reformulation of of limited capacity, stochastic arrivals, and stochastic ser
the bilevel problems is proposed to be able to solve the vice times, charging stations may experience congestion.
models using commercial solvers. Consequently, EV users may have to enter a service
To the best of our knowledge, there are only two queue when the system is busy (i.e., when all charging
studies to date that make charging station location and stalls are occupied).
sizing decisions while taking EV users’ response into The waiting time at a charging station is a function of
account. Huang and Kockelman (2020) use a bilevel its capacity given by the number of charging stalls and
profit maximization problem for multiperiod charging the total allocated EV charging demand. In order to
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS 5
ensure service convenience, a service-level requirement may be used when shortest path deviations are allowed.
is introduced to guarantee that the probability of wait Accordingly, we define G � (N, A) to be the expanded
ing at most α minutes is greater than or equal to β% at network of Go for a given range R. We now show how
each open station. For example, a service-level require to further reduce the expanded network by defining a
ment is for the waiting time to be less than or equal to permissible arc for OD pair k ∈ K as follows.
α � 10 minutes, β � 90% of the time.
Definition 1. An expanded network arc (i, j) ∈ A is
A deviation tolerance is defined to model EV users’
called a permissible arc for OD k ∈ K if it connects Ok to
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
Figure 1. (Color online) An Illustration of Permissable Arc Expanded Network Generation with R � 200 and τ � 10%
(a) (b)
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
6 Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS
the expanded network arc costs. To achieve this, the context, the leader makes a decision, and the follower
cost of expanded network arcs that do not enter the des responds. The framework of such a problem is based
tination node would be increased by the fixed cost of on the fact that the leader anticipates the follower’s
charging. optimal reaction to their decisions.
To formulate the bilevel charging station location and The objective function of the leader (1) minimizes
sizing problem under congestion, we introduce the fol total infrastructure cost composed of the cost of locat
lowing decision variables: ing charging stations and installing charging stalls.
8
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
4. Analysis of the Bilevel the integrality property of the follower’s problem under
Modeling Framework cooperative response.
Let Ψ(y) define the reaction set of the follower based on Proposition 1. The follower’s problem of BLP always has
the leader’s decisions y � {yj , j ∈ N} as an integer optimal solution when feasible.
(
Proof. Assume that there exists an OD pair k ∈ K with
Ψ(y) � x � {xijk , (i, j) ∈ Ak , k ∈ K} : fractional xijk in an optimal solution. Without loss of
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
4.2. Case of Poisson Arrivals and Exponential Recall that the expanded network ensures that xijk � 1 of
Service Times: M/M/c Queuing System OD k ∈ K charges at j ∈ N. Moreover, the underlying
DC fast-charging stations for EVs typically include open multiserver Jackson network ensures that the
multiple charging stalls that can deliver 80–400 kW at arrival rate of an OD to a station is equal to their depar
50–1,000 V (IEA 2021). Each charging station provides ture rate as long as all queues on the network are stable.
simultaneous service to multiple EVs, where each stall The queue stability requirement is defined as
acts as a server utilized by service seekers on a first
Λj ≤ µzj � ɛ j ∈ N, (24)
come, first served basis. Moreover, common courtesy
dictates that there is a single queue for each charging where µ is the exponential service rate of each server of
station in case of concurrent utilization of the servers. a station and ɛ is an infinitesimal constant. This con
This single-queue, multiple-parallel server structure straint ensures that the Poisson flow rate faced by a sta
constitutes the main framework of our queuing system tion is strictly less than its processing rate in order to
models of the DC fast-charging stations. have a finite queue length.
Existing public charging stations are often located We define the average load of a stall at a charging sta
near discretionary service facilities (e.g., restaurants, tion by ρj � Λj =µzj and rewrite the left-hand side of Con
stores, plazas, or shopping centers) in order to mitigate straint (3) using a modification of the Erlang-C formula
the relatively longer recharging times for EVs. As a (Chromy, Misuth, and Weber 2012; Xie et al. 2018):
result, not every customer unplugs their EV right after z
(zj ρj ) j
getting up to their desired SoC. This is one of the main
zj !(1�ρj )
reasons for existing studies to consider stochastic and P(Wj ≤ α) � 1 � z
Pzj �1 zj ρj t e�µ(zj �ρj zj )α j ∈ N:
(zj ρj ) j
typically, exponential service times (Xie et al. 2018, Xie
zj !(1�ρj ) + t�1 t!
and Lin 2021). The stochastic flow rate of an OD is
assumed to have Poisson distribution. The rate of the (25)
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS 9
Although this substitution in (3) involves a quadratic DLP may be solved using off-the-shelf solvers, our pre
function, it may be rewritten using a step function on liminary computational experiments revealed that these
integer zj: solvers are not able to handle the problem complexity
for large-sized networks. In order to solve DLP for real-
8
>
> 0, Λj � 0 life instances, we develop a decomposition-based algo
>
> rithm that exploits the problem structure as described
> ɛ ≤ Λj < b1
< 1,
>
in the next section.
zj � 2, b1 ≤ Λj < b2 if P(Wj ≤ α) ≥ β
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
>
>
>
> ::: :::
>
>
:
5. Solution Methodology
C, bC�1 ≤ Λj < bC : There are three decisions to make in our problem set
(26) ting: the locations of the charging stations, the number
of stalls at open stations, and the OD routes. We pro
Service-level parameters α and β of Constraint (3) deter pose a solution method that decomposes these decisions
mine the values of the break points (b1 , b2 , : : : , bC ) of this and iteratively solves two single-level problems: a loca
step function by a preprocessing procedure based on tion problem and a sizing problem. As the location
(25). Such a function can be easily incorporated within problem turns out to be computationally demanding,
any MILP model using special ordered set of type 1 vari we develop a tailored logic-based Benders algorithm
ables or using the following set of additional constraints (LBBA) that solves all instances within seconds.
to substitute (3): The proposed decomposition structure first solves a
location problem taking into account OD flows to obtain
X
C
the charging station location vector y and the corre
sjm � 1 j∈N (27)
m�1
sponding optimal reaction set of the follower defined
on y:
X
C
( (
Λj ≤ bm sjm j∈N (28) X X
m�1 Ψ(y) � x : x ∈ argmin δij xijk :
k∈K (i, j)∈Ak
X
C ))
Λj ≥ bm�1 sjm j ∈ N (29) (8), (10), (11), (32) :
m�2
X
C
X
zj ≥ msjm j∈N (30) where xijk ≤ y j j ∈ Nk , j ≠ Dk , k ∈ K: (32)
m�1
i:(i, j)∈Ak
sjm ∈ {0, 1} j ∈ N, m ∈ {1, 2, : : : , C}: (31) For given y and Ψ(y), the sizing problem reduces to
determining the number of stalls at open stations (z)
Constraints (28)–(29) determine the two consecutive and allocating charging demand of OD flows to open
break points of the step function that Λj value lies stations (x). The sizing problem accounts for route
between. Constraints (30) exploit the fact that both m choice response by considering all alternative optimal
and zj are integer and determine the minimum zj value paths on the network defined by y and modeling both
in conjunction with the minimization objective function cooperative and uncooperative response cases. The siz
(1) of BLP. ing problem determines a partial solution (z, x) that is
The number of sjm variables increases with C, so optimal for given y and a complete solution (y, z, x)
choosing the right value of C is important. In this study, that is bilevel feasible. The flowchart of the solution
we determine this value so that it is large enough to methodology is shown in Figure 2. At each iteration,
keep Constraint (2) nonbinding for the optimal solution an elimination cut is generated after solving the sizing
but small enough so that the number of binary sjm vari
ables is not excessively many.
Substituting a step function instead of the chance Figure 2. The Flowchart of the Decomposition-Based Solu
Constraint (3) by pregenerating its break points based tion Method for BLP
on the Erlang-C formula and linearizing this step func
tion with the constraints, the single-level reduction of
the bilevel model is provided in Online Appendix A.
To summarize, when OD flows are Poisson distrib
uted and service rate at a stall is exponential, SLP is
reformulated as an MILP denoted as DLP. Even though
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
10 Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS
problem and is fed back into the location problem. Sim redundant to say that this does not impact the correctness
ply put, this algorithm performs a smart enumeration of the logic-based Benders algorithm, even though it may
by iteratively constructing bilevel feasible solutions and lead to an alternative optimal solution for PL. However,
tracking the incumbent. When stopped prematurely, it the objective function of the follower becomes critical
has no guarantee of optimality. within the solution methodology of BLP, where it is nec
The rest of the section is organized as follows. Section essary to account for their minimum-length path choice.
5.1 describes the location problem and the suggested The subproblem SPk for OD k ∈ K is given by
X
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
ing problem reduces to a single-level location problem xijk ∈ {0, 1} (i, j) ∈ Ak : (38)
and determines the minimum cost charging station
locations subject to flow balance and deviation tolerance When a solution y of MP is infeasible to SPk, the aim is
constraints. The location problem PL is to derive valid cut(s) that (i) remove y from the feasible
X region of MP and (ii) do not eliminate any feasible solu
[PL ] minimize f (yj ) (33) tion of PL. The valid cuts are added to MP, and the latter
j∈N is solved repeatedly to find a new set of station loca
tions. When the valid cuts satisfy the two conditions,
s:t: (4), (8)–(11): this approach is guaranteed to converge to an optimal
solution (Hooker and Ottosson 2003). We next discuss
PL may be solved directly by a solver or using Benders the derivation of the valid cuts and prove the correct
decomposition, where the master problem determines ness of the algorithm.
charging station locations and the subproblems check Once a solution y is obtained, the node restrictions
path feasibility for each OD. Recall that the subpro enforced by Constraint (37) are used to form a reduced
blems are linear because of Proposition 1. graph Grk � (Nkr , Ark ) for each OD pair k ∈ K such that
Despite being a relaxation of BLP, preliminary testing Nkr � {j ∈ Nk |y j � 1} ∪ {Ok , Dk } and Ark � {(i, j) ∈ Ak |i, j ∈
showed that PL is still challenging to solve in reasonable
Nkr }. Subproblems SPk, ∀k ∈ K, are solved on their re
times both using the solver and using Benders decom
spective reduced graphs Grk using a labeling algorithm.
position. Because PL is solved repeatedly within the
A given subproblem SPk may be feasible, yet the opti
solution method of BLP, we develop a logic-based
mal path length may not satisfy the deviation tolerance
Benders algorithm that is able to solve PL optimally
constraints of the original problem. Alternatively, a
within a few seconds for the majority of the large-scale
given subproblem SPk may be infeasible if there is no
instances. The logic-based Benders master problem
path that connects OD in the reduced graph. In both
(MP) is given by
X cases, feasibility cuts are generated and added to MP.
[MP] minimize f (yj ) (34) We first discuss the case where SPk is infeasible for
j∈N some OD k̂ ∈ K. In this case, the corresponding graph
Grk̂ has at least two connected components that sepa
s:t: yj ∈ {0, 1} j∈N (4) rately include Ok̂ and Dk̂ . Let Gok̂ be the connected com
+ logic-based Benders cuts, ponent of Grk̂ that includes Ok̂ and Gdk̂ be the reverse
graph of the component of Grk̂ that includes Dk̂ . Let Arev k
and the subproblem is a path feasibility problem with denote the reverse Ak. We define N f as the set of nodes
deviation constraints on the network defined by y, the
that connect Gok̂ to Gdk̂ and N b as the set of nodes that
solution of MP.
We further modify the subproblems in two ways. First, connect Gdk̂ to Gok̂ :
we drop the deviation tolerance constraints and deal
with them using feasibility cuts. Second, we keep the N f � {j ∈ Nk̂ \Nk̂r : (i, j) ∈ Ak̂ , i ∈ Nk̂o } (39)
objective function of the follower and solve a minimum
N b � {j ∈ Nk̂ \Nk̂r : (i, j) ∈ Arev , i ∈ Nk̂d }: (40)
cost path problem instead of a feasibility problem. It is k̂
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS 11
Using N f and N b , we introduce the following feasibility The solution of the location problem opens charging
cuts for an infeasible SPk̂ : stations and determines OD flow paths within the devia
X tion tolerance. It remains to determine the number of
yi ≥ 1 (41)
charging stalls at each open location to satisfy the service
i∈N f
X requirement, which we refer to as the sizing problem.
yi ≥ 1: (42) These two problems are solved iteratively to generate
i∈N b solutions for DLP. Next, we present the sizing problem
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
Cuts (41) and (42) are added to MP to cut the current and the complete solution for BLP.
infeasible solution. Note that adding either of them is
sufficient to cut the current infeasible solution. 5.2. The Sizing Problem and DLP Solution
For the case when SPk is feasible and deviation toler When the charging station locations are fixed, one can
P always equip each station with a sufficient number of
ance is exceeded for some OD k̂ ∈ K (i.e., (i, j)∈A δij xijk̂ >
(1 + τ)δOk̂ Dk̂ ), we introduce the following cut:
k̂
stalls to satisfy the service-level Constraints (3), pro
X vided stall capacity C is sufficiently large. To determine
yi ≥ 1: (43) a solution that minimizes the total cost of capacity
i∈Nk̂ \Nr installation, one has to account for the amount of flow
k̂
is used in path r ∈ Ψk (y) of OD pair k ∈ K, and let urk be 5.2.2. The Uncooperative Response. The cooperative
a binary decision variable that takes the value of one if response assumes that EV drivers will use a shortest
OD pair k ∈ K is assigned to path r ∈ Ψk (y). The sizing path that minimizes the leader’s objective. In practice,
problem PS under the cooperative response and M/M/c however, it is difficult to enforce such a choice. Conse
case is quently, the leader may be interested in investigating
X the uncooperative response or the worst possible out
[PS ] minimize g(zj ) (44) come of the follower’s choices. To account for the un
j∈Nk | y j �1
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
networks to assess the proposed solution methodology For this computational analysis, we assume that every
by evaluating its computational performance on larger EV user starts their journey with 80% of the available
instances. Experiments are performed using Gurobi battery range, and they may charge up to 80% SoC level
v.9.1.1 with Python API and NetworkX package v.2.5.1 at a charging station visit. The reason for this assump
(Hagberg, Swart, and Chult 2008) using up to eight tion is twofold. First, the battery charging process at
threads on a Windows OS computer with an Intel public charging stations gets exponentially slower after
i7-9700K 3.60-GHz processor and 32 GB of RAM. 80% SoC, which may cause additional congestion. In
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
Then, we test the computational performance of this algo experiments are summarized in Tables 2 and 3 in
rithm on the large-scale CA339 and US-E420 networks. the online appendix and are presented in the online
For N25, we generate 48 instances for each R value appendix.
we test, which add up to a total of 144 test instances. The performance of DA is evaluated in terms of the
For each instance, the numbers of all possible long- number of logic-based Benders cuts added (i.e., feasibil
distance OD pairs (i.e., the cardinality of set K) are 167, ity cuts (41) and (42) and deviation tolerance cuts (43))
108, and 57 for R � 15, 20, and 25, respectively. Each and the number of iterations completed within the time
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
instance is solved both with the optimization model limit. The averages of these performance metrics are
DLP and with DA, and their solutions are compared to depicted in Figures 4 and 5 in the online appendix for
assess the performance of the latter. The stopping con CA339 and US-E420, respectively.
dition of the algorithm is set to 60 iterations. Our algo It is observed from these results that the number of
rithm is able to find the optimal solutions for all 144 iterations DA can complete in the one-hour time limit
instances. The results of these experiments are detailed tend to decrease slightly for CA339 instances as devia
in Table 1 of the online appendix. We further visualize tion tolerance increases. This is because of the decreas
the average solution time comparisons in Figure 3 in ing number of required feasibility cuts. Figure 4 in
Online Appendix C. Online Appendix C also shows that the deviation toler
The solver finds the optimal solutions quicker only ance cuts (43) are utilized just 16 times throughout the
when the deviation tolerance is zero as the size of the experiments. This is a clear indication that introducing
expanded network is smaller because of our introduc permissible arcs is effective for generating solutions that
tion of the set of permissible arcs. Instances with higher are deviation tolerance feasible.
R values are also relatively quicker to solve by the The expanded networks of US-E420 include over
solver as they include fewer long-distance OD pairs. 700,000 arcs for R � 400 and 1 million arcs for R � 600
On the other hand, time savings by our algorithm are at high levels of deviation tolerances. This results in
drastically higher for R � 15 instances with deviation tens of millions of decision variables and constraints
tolerance greater than 0%. Particularly, the average for the bilevel optimization model. When R is lower
solution time obtained with Gurobi for 50% deviation and the number of OD pairs is higher, the likelihood
tolerance instances of R � 15 is over 1,000 seconds, of obtaining an infeasible set of station locations in the
whereas our algorithm is capable of returning the opti algorithm increases. Consequently, it is inevitable to
mal solution identified by Gurobi even when it runs add thousands of feasibility cuts even with the effi
less than 4 seconds. cient approach that we are proposing. In particular,
Now, we focus on the large-sized CA339 and US- observe from Figure 5(a) in Online Appendix C that
420E instances to further evaluate the computational when R � 400 and the deviation tolerance is 50%, the
performance of the algorithm. We generate 32 instances algorithm adds over 14,000 feasibility cuts, whereas it
on the CA339 network at two different R values, 200 can only complete one iteration. Moreover, the num
and 250 km, with a fixed β of 90%. The R � 200 and ber of times deviation tolerance cuts are utilized is sig
R � 250 instances include 502 and 335 long-distance OD nificantly higher for US-E420 compared with CA339
pairs, respectively. The same 32 configurations as with instances.
CA339 are used to create the US-E420 instances, with
higher values of R (400 and 600 km). For these instances, 6.3. Insights from Bilevel Solutions
we include a trip distance threshold of 1,600 km as those In this section, we derive insights from the solutions
longer than this threshold would require overnight by evaluating the effects of deviation tolerance, vehicle
stays. This results in totals of 469 and 429 long-distance range, and service-level parameters under the cooper
OD pairs defined on the US-E420 network for R � 400 ative response. We first analyze the N25 solutions and
and R � 600 instances, respectively. then move on to the large-scale instances.
We first attempted to solve each instance using Gurobi For all R configurations of the small-sized N25 net
and were unable to obtain even an incumbent solution work, 0% and 10% deviation tolerance solutions turn
for deviation tolerance levels greater than 0% within a out to be identical. Moreover, the number of located
two-hour time limit. When this time limit is increased, stations does not change with varying α or β values for
Gurobi terminates the solution procedure with an out of a fixed deviation tolerance. This indicates that the ser
memory error because of the size of the branch and vice level only affects the capacities of the stations for
bound search tree. Note that these instances require the tested configurations. In all instances, stricter α
and
over 1 billion decision variables. This indicates that β result in installing more capacity and hence, a higher
the state-of-the-art solvers are not able to handle prob objective function value. For example, when R � 15 and
lem instances on real-world networks. Consequently, the deviation tolerance is 25%, the α � 1 minute and β �
we utilize DA to solve these instances with a stopping 95% combination opens 31% more charging stalls com
condition of one hour. The results of these extensive pared with the case where α � 30 minutes and β � 80%.
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS 15
Higher-deviation tolerance solutions require fewer sta Figure 8 in Online Appendix C shows the sensitivity
tions and/or fewer stalls and result in a lower objective of the objective values based on the tested waiting time
function value (e.g., when R � 15, 0% deviation toler threshold and deviation tolerance values for US-E420
ance solutions utilize eight stations and an average of solutions. Similar to the CA339 results, the difference
136 charging stalls in total, whereas 25% deviation toler between the objective values of the solutions with 0%
ance solutions require six stations and 132 charging and 10% deviation tolerances is the highest. On the
stalls on average). The value of using a bilevel frame other hand, the objective function value differences
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
work is higher when solving higher-deviation tolerance between 25% and 50% deviation tolerance instances are
instances because the solution space of the follower’s relatively smaller, particularly for higher R. For both
route choice response grows larger. Within the bilevel R � 400 and R � 600 instances, the α � 1-minute solu
framework, the follower picks the shortest available tions are approximately 5.5%, 10.2%, and 20% more
route, and the average deviation from the shortest paths costly than α � 5, 10, 30-minute solutions, respectively.
never exceeds 5.65% among the optimal solutions of all The average and maximum deviation comparison of
the 50% deviation tolerance instances. US-E420 solutions for α � 10 instances are illustrated in
Next, we analyze CA339 and US-E420 solutions ob Figure 9 in Online Appendix C. For both R � 400 and
tained by DA to derive insights from sensitivity analysis R � 600 solutions, the value of maximum deviation
on the problem parameters. For CA339 solutions, similar closely follows the deviation threshold. On the other
to the N25 results, varying α only changes the capacity hand, the average deviation is only as high as 13% and
levels deployed but not the number of located stations. as low as 1.42% over all solutions.
As the geographical shape of the CA339 network is nar When deviation tolerance is 0% (i.e., every EV user can
row and potential OD nodes are not scattered to the complete their shortest path without running out of
boundaries, between two and seven stations are observed energy) and R � 600 for the instances of the US-E420 net
work, 55 charging stations and over 600 charging stalls
to be able to sufficiently serve various parameter settings
are required, which necessitates an approximately $60
that we tested on this network.
million investment. When the deviation tolerance is set to
In Figure 6 in Online Appendix C, we further analyze
25%, only 31 charging stations and a similar number of
the difference in the objective values based on the wait
stalls are required, and the required investment decreases
ing time threshold and deviation tolerance of CA339
to approximately $50 million. In order to achieve about
solutions. These charts demonstrate how the objective
$10 million in savings in the infrastructure cost, the deci
value decreases with higher deviation tolerance (τ) and
sion maker can allow for a maximum of 25% deviation.
increases with stricter waiting time threshold (α). The
In this case, note that the average deviation per EV user
most significant cost savings are achieved when 0%
will only be 5.92% when α � 10 minutes and β � 90%
deviation tolerance is increased to 10%.
(Table 3 in the online appendix).
Figure 7 in Online Appendix C shows the average and
Figure 10 in Online Appendix C presents the spatial
maximum deviation metrics at 10%, 25%, and 50% devia comparison of charging station locations and sizes on
tion tolerance levels for α � 10 minutes instances of CA339. two representative solutions with two different devia
For R � 250, average deviation does not exceed 0.88% tion tolerances for R � 600 km, α � 10 minutes, and β �
even when the deviation tolerance is set to 50%. However, 90% on US-E420. The sizes of the charging stations are
this performance metric can be as high as 7.07% at 50% depicted using different sizes for the green triangles.
deviation tolerance for R � 200. The maximum deviation When the deviation tolerance is 10% (Figure 10(a) in the
metric follows the deviation tolerance limit closely for this online appendix), there are 39 charging stations. The
range level, whereas it has nearly 30% slack for R � 250 largest stations are located near the center of the net
instances at 50% deviation tolerance. work, whereas numerous smaller stations are scattered
The US-E420 instances require significantly more throughout the region. On the other hand, when the
facilities than the other networks that we tested. The deviation tolerance is 50% (Figure 10(b) in the online
highest number of stations located is 97 for R � 400, appendix), there are only 25 charging stations, and the
whereas the lowest is 25 for R � 600 instances. We also majority are of larger sizes that are located closer to the
observe that R � 400 instances require almost twice as center; the number of smaller charging stations is signif
many stations as their R � 600 counterparts, although icantly fewer.
the number of OD trips is only 9% more for the latter. For a more extensive analysis on how the deviation
As it is the case with the other two networks, lower tolerance and vehicle range affect the size of the facilities,
values of α might require up to 20% additional capacity we provide the histograms in Figure 11 in Online
at the established stations, whereas it does not change Appendix C that compare the solutions with fixed α and
the station count in any of the experiments. The station β values at 10 minutes and 90%, respectively. For both
locations are observed to be insensitive to the changes R � 600 and R � 400, instances with smaller-deviation
in α in the majority of the instances. tolerances require significantly more stations of smaller
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
16 Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS
sizes. Higher deviation tolerances allow us to aggregate 6.5. Analyzing the Types of OD Pairs Affected
the charging demand at fewer stations that are of larger from Deviation Allowance
size. This aggregation not only locates fewer facilities, In order to provide more insights into the trade-off
but it also requires less capacity in total (see Table 3 in between the satisfaction of EV users and infrastructure
the online appendix). This indicates the importance of deployment, we define the shortest path for all OD pairs
determining the right deviation tolerance for the right as a benchmark and then investigate the types of OD
applications. pairs that are mostly affected by deviation allowance.
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
and maximum deviation metrics stay the same when the case of Poisson arrivals and exponential service
exactly the same sets of stations are open in the coopera times. We derive the MILP equivalent of the chance-
tive and uncooperative response solutions. In particu constrained stochastic bilevel optimization formulation
lar, for the solutions of R � 20 and R � 15 instances, the by modeling the underlying charging station network as
set of charging station locations stays the same, whereas an open multiserver Jackson network.
the uncooperative response solutions result in different We develop a decomposition-based algorithm to solve
required capacity levels. On the other hand, the sets of the proposed bilevel optimization model under both
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
open stations are different for all R � 25 instances when cooperative and uncooperative follower responses. The
the deviation tolerance is greater than 0%. This indicates algorithm utilizes a new exact logic-based Benders meth
that the locations of the charging stations may change odology for the uncapacitated version of the problem as
based on the position determined for the follower. The a subroutine. The solution quality and the computational
complete results of this comparison are provided in performance of the algorithm are verified on smaller-
Table 5 in the online appendix. sized network instances, where it finds optimal solutions
Interestingly, smaller differences in the objective func for all of the 144 test instances. The computational per
tion values are observed when the deviation tolerance is formance comparison with a commercial solver reveals
higher. For that matter, the cooperative and uncoopera that our algorithm is much more efficient on instances
tive response solutions turn out to be equivalent when with higher deviation tolerances and lower range values.
the deviation tolerance is 50% for both R � 25 and R � 20. Allowing for higher deviation tolerances (up to 50%) can
This is an insightful result for the decision makers especially be relevant for autonomous long-distance
because it shows that having a higher deviation toler travel with EVs.
ance mitigates the difference that might result from an The results demonstrate that the number of located
uncooperative response of the EV users. In other words, charging stations is insensitive to the changes in the
higher deviation tolerances for the cooperative response service-level requirements. Typically, more stringent
case play an important role to provide more robust solu service-level thresholds require more capacity to be
tions for the cases in which the follower may not behave installed at the charging stations. The solutions with
as desired. On the other hand, lower-deviation tolerance higher deviation tolerance are observed to require less
levels tend to yield higher differences in the objective infrastructure cost and are not prone to generate alter
native stopping options for the EV users. This leads to
function values between the cooperative and uncoopera
robust solutions that are equivalent for the cooperative
tive response solutions. Intuitively, one might think that
and uncooperative responses of the EV users and pro
our bilevel model does not make a difference when
tect the decision maker against the worst possible out
the deviation tolerance is set to 0% as the route choices
come. On the other hand, the solutions allowing for
are no longer relevant. However, our computational
lower deviation tolerance tend to increase the number
results reveal that the objective function difference
of required stops and result in higher differences in the
between the cooperative and uncooperative response
optimal objective function values obtained under the
solutions is usually higher for the 0% deviation toler
cooperative and uncooperative responses.
ance instances. This indicates that using lower devia
In this work, the route choice behavior of EV users is
tion tolerances, especially 0%, might generate solutions
driven by shortest path given the locations of charging sta
that may easily be overridden by uncooperative re tions. This allows us to solve our problem on large-scale
sponses of the EV users in terms of choosing at which instances by developing solution methods that exploit
station(s) to recharge. properties of the shortest path problem. Although the
trip distance is the only term that is explicitly considered
7. Conclusion in the objective function of the follower’s problem, to
In this paper, we develop a bilevel optimization model ensure that the charging network offers convenient ser
to determine the strategic locations and capacities of vice, service and waiting times are considered by the
DC fast-charging stations under stochastic demand to probabilistic service-level constraint. Assuming that traf
enable long-distance travel with EVs. The model allows fic congestion is negligible during long-distance trips,
the possibility of detours from shortest paths within a what leads to faster travel are the trip distance, service
deviation tolerance without endogenously determining time, and waiting time. Incorporating features, such as
the deviation paths. Through the bilevel framework, the traffic congestion or dynamic charge pricing and toll
model takes into account the route choice response of pricing, rather than relying on the shortest path route
EV users, as well as their cooperative or uncooperative choice assumption would require new solution algo
charging station selection responses, when alternative rithms. Under our modeling framework, the service-
solutions exist. Besides providing a general formulation level constraint ensures that the levels of congestion
of the problem, we model the M/M/c queuing system experienced at all located facilities are comparable. Since
characterization of the DC fast-charging stations for the actual congestion, hence trip completion time, will be
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
18 Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS
driven by real-time dependencies (e.g., traffic condition), Frade I, Ribeiro A, Gonçalves G, Antunes AP (2011) Optimal loca
this is a feature that may be observed by performing a tion of charging stations for electric vehicles in a neighborhood
in Lisbon, Portugal. Transportation Res. Rec. 2252(1):91–98.
discrete-event simulation. Another interesting extension Goodchild MF, Noronha VT (1987) Location-allocation and impul
of our work is to evaluate various path choice settings, sive shopping: The case of gasoline retailing. Ghosh A, Ruston
such as real-time congestion, toll pricing, and charge G, eds. Spatial Analysis and Location-Allocation Models (van Nos
pricing, through simulation. trand Reinhold, New York), 121–136.
Future work may focus on different types of queuing Göpfert P, Bock S (2019) A branch & cut approach to recharging and
refueling infrastructure planning. Eur. J. Oper. Res. 279(3):808–823.
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
system characterizations for charging stations. For exam Guo F, Yang J, Lu J (2018) The battery charging station location
ple, an extension with M/D/c systems may be a viable problem: Impact of users’ range anxiety and distance conve
option for centralized systems with autonomous vehicles nience. Transportation Res. Part E Logist. Trans. Rev. 114:1–18.
or for commercial applications that can enforce deter Hagberg A, Swart PS, Chult D (2008) Exploring network structure,
ministic service times to the users of a charging station. dynamics, and function using NetworkX. Technical report, Los
Alamos National Laboratory, Los Alamos, NM.
Another research dimension may consider a multiperiod He J, Yang H, Tang TQ, Huang HJ (2018) An optimal charging station
framework to determine infrastructure expansion plans location model with the consideration of electric vehicle’s driving
considering the growth in charging demand because of range. Transportation Res. Part C Emerging Tech. 86:641–654.
the increasing adoption rate of EVs. Hong S, Kuby M (2016) A threshold covering flow-based location
model to build a critical mass of alternative-fuel stations. J.
Transportation Geography 56:128–137.
Acknowledgments
Hooker JN, Ottosson G (2003) Logic-based benders decomposition.
The authors express their appreciation to Research Assistant Math. Programming 96(1):33–60.
Jonathan Wong during the data set generation phase of this Hosseini M, MirHassani SA, Hooshmand F (2017) Deviation-flow refuel
work. The authors also thank the anonymous referees who ing location problem with capacitated facilities: Model and algo
have contributed to the further development of this manu rithm. Transportation Res. Part D Transportation Environ. 54:269–281.
script by their suggestions and comments. Huang Y, Kockelman KM (2020) Electric vehicle charging station loca
tions: Elastic demand, station congestion, and network equilibrium.
Transportation Res. Part D Transportation Environ. 78:102179.
References IEA (2020) Improving the sustainability of passenger and freight
Anjos MF, Gendron B, Joyce-Moniz M (2020) Increasing electric vehicle transport. Tracking Transport 2020. Accessed September 1, 2021,
adoption through the optimal deployment of fast-charging stations https://www.iea.org/topics/transport.
for local and long-distance travel. Eur. J. Oper. Res. 285(1):263–278. IEA (2021) Global EV outlook 2021: Accelerating ambitions despite the
Arslan O, Karaşan OE (2016) A Benders decomposition approach for pandemic. Report, IEA. Accessed September 21, 2021, https://
the charging station location problem with plug-in hybrid electric www.iea.org/reports/global-ev-outlook-2021.
vehicles. Transportation Res. Part B Methodological 93:670–695. Jiang Y, Zhang Y, Zhang C, Fan J (2012) Capacitated deviation-flow fuel
Arslan O, Yıldız B, Karaşan OE (2014) Impacts of battery character ing location model for sitting battery charging stations. CICTP 2012
istics, driver preferences and road network features on travel Multimodal Transportation Systems—Convenient Safe Cost-Effective Effi
costs of a plug-in hybrid electric vehicle for long-distance trips. cient, 2771–2778. https://doi.org/10.1061/9780784412442.282.
Energy Policy 74:168–178. Jing W, An K, Ramezani M, Kim I (2017) Location design of electric vehi
Arslan O, Karaşan OE, Mahjoub AR, Yaman H (2019) A branch- cle charging facilities: A path-distance constrained stochastic user
and-cut algorithm for the alternative fuel refueling station loca equilibrium approach. J. Adv. Transportation 2017:4252946.
tion problem with routing. Transportation Sci. 53(4):1107–1125. Kim JG, Kuby M (2012) The deviation-flow refueling location model
Bard JF (1991) Some properties of the bilevel programming problem. J. for optimizing a network of refueling stations. Internat. J. Hydro
Optim. Theory Appl. 68(2):371–378. gen Energy 37(6):5406–5420.
Bard JF (2013) Practical Bilevel Optimization: Algorithms and Applica Kim JG, Kuby M (2013) A network transformation heuristic approach
tions, vol. 30 (Springer Science & Business Media, New York). for the deviation flow refueling location model. Comput. Oper. Res.
Bose SK (2013) An Introduction to Queueing Systems (Springer Science 40(4):1122–1131.
& Business Media, New York). Kınay ÖB, Gzara F, Alumur SA (2021) Full cover charging station
Burke PJ (1956) The output of a queuing system. Oper. Res. 4(6):699–704. location problem with routing. Transportation Res. Part B Meth
Capar I, Kuby M (2012) An efficient formulation of the flow refueling odological 144:1–22.
location model for alternative-fuel stations. IIE Trans. 44(8):622–636. Ko J, Gim THT, Guensler R (2017) Locating refuelling stations for
Capar I, Kuby M, Leon VJ, Tsai YJ (2013) An arc cover–path-cover for alternative fuel vehicles: A review on models and applications.
mulation and strategic analysis of alternative-fuel station locations. Transportation Rev. 37(5):551–570.
Eur. J. Oper. Res. 227(1):142–151. Kuby M, Lim S (2005) The flow-refueling location problem for
Chromy E, Misuth T, Weber A (2012) Application of erlang formulae in alternative-fuel vehicles. Socio-Economic Planning Sci. 39(2):125–145.
next generation networks. Internat. J. Comput. Network Inform. Secu Kuby M, Lim S (2007) Location of alternative-fuel stations using the
rity 4(1):59–66. flow-refueling location model and dispersion of candidate sites
Electrek (2019) Tesla starts limiting charge to 80% at busy superchargers on arcs. Network Spatial Econom. 7(2):129–152.
to reduce wait times. Accessed June 10, 2021, https://electrek.co/ Kuby M, Lines L, Schultz R, Xie Z, Kim JG, Lim S (2009) Optimiza
2019/05/24/tesla-limiting-supercharger-busy/. tion of hydrogen stations in Florida using the flow-refueling
European Commission (2019) The European Green Deal: Sustain location model. Internat. J. Hydrogen Energy 34(15):6045–6064.
able mobility. Accessed September 24, 2021, https://ec.europa. Labbé M, Marcotte P (2021) Bilevel network design. Crainic TG, Gen
eu/commission/presscorner/api/files/attachment/860070/ dreau M, Gendron B, eds. Network Design with Applications to Trans
Sustainable_mobility_en.pdf.pdf. portation and Logistics (Springer, Cham, Switzerland), 255–281.
Kınay, Gzara, and Alumur: Charging Station Location and Sizing
Transportation Science, Articles in Advance, pp. 1–19, © 2023 INFORMS 19
Li S, Huang Y (2014) Heuristic approaches for the flow-based set heterogeneous vehicles in urban networks. Transportmetrica A
covering problem with deviation paths. Transportation Res. Part Transportation Sci. 17(4):439–461.
E Logist. Trans. Rev. 72:144–158. Tu W, Li Q, Fang Z, Shaw S-l, Zhou B, Chang X (2016) Optimizing the
Makhlouf W, Kchaou-Boujelben M, Gicquel C (2019) A bi-level program locations of electric taxi charging stations: A spatial–temporal
ming approach to locate capacitated electric vehicle charging sta demand coverage approach. Transportation Res. Part C Emerging
tions. 2019 6th Internat. Conf. Control Decision Inform. Tech. (CoDIT) Tech. 65:172–189.
(IEEE, Piscataway, NJ), 133–138. Upchurch C, Kuby M (2010) Comparing the p-median and flow-
MirHassani S, Ebrazi R (2013) A flexible reformulation of the refuel refueling models for locating alternative-fuel stations. J. Trans
ing station location problem. Transportation Sci. 47(4):617–628. portation Geography 18(6):750–758.
Downloaded from informs.org by [131.175.147.3] on 27 June 2023, at 07:22 . For personal use only, all rights reserved.
Nicholas M (2019) Estimating electric vehicle charging infrastructure Upchurch C, Kuby M, Lim S (2009) A model for location of capaci
costs across major us metropolitan areas. Accessed September tated alternative-fuel stations. Geographical Anal. 41(1):85–106.
20, 2021, https://theicct.org/sites/default/files/publications/ Wang YW, Lin CC (2009) Locating road-vehicle refueling stations.
ICCT_EV_Charging_Cost_20190813.pdf. Transportation Res. Part E Logist. Trans. Rev. 45(5):821–829.
Pelletier S, Jabali O, Laporte G, Veneroni M (2017) Battery degrada Wang YW, Wang CR (2010) Locating passenger vehicle refueling sta
tion and behaviour for electric vehicles: Review and numerical tions. Transportation Res. Part E Logist. Trans. Rev. 46(5):791–801.
analyses of several models. Transportation Res. Part B Methodo Xie F, Lin Z (2021) Integrated us nationwide corridor charging infra
logical 103:158–187. structure planning for mass electrification of inter-city trips.
Shen ZJM, Feng B, Mao C, Ran L (2019) Optimization models for Appl. Energy 298:117142.
electric vehicle service operations: A literature review. Transpor Xie F, Liu C, Li S, Lin Z, Huang Y (2018) Long-term strategic planning of
tation Res. Part B Methodological 128:462–477. inter-city fast charging infrastructure for battery electric vehicles.
Simchi-Levi D, Berman O (1988) A heuristic algorithm for the traveling Transportation Res. Part E Logist. Trans. Rev. 109:261–276.
salesman location problem on networks. Oper. Res. 36(3):478–484. Yıldız B, Arslan O, Karaşan OE (2016) A branch and price approach
Sinha A, Malo P, Deb K (2017) A review on bilevel optimization: for routing and refueling station location model. Eur. J. Oper.
From classical to evolutionary approaches and applications. Res. 248(3):815–826.
IEEE Trans. Evolution Comput. 22(2):276–295. Zeng B (2020) A practical scheme to compute the pessimistic bilevel
The Climate Group (2021a) EV100 Initiative. Accessed September optimization problem. INFORMS J. Comput. 32(4):1128–1142.
23, 2021, https://www.theclimategroup.org/ev100. Zeng D, Dong Y, Cao H, Li Y, Wang J, Li Z, Hauschild MZ (2021)
The Climate Group (2021b) EV100 Progress and Insights Report 2021. Are the electric vehicles more sustainable than the conventional
Accessed September 23, 2021, https://www.theclimategroup.org/ ones? Influences of the assumptions and modeling approaches
media/7941/download. in the case of typical cars in China. Resources Conservation Recy
The White House (2021) Executive order on strengthening American cling 167:105210.
leadership in clean cars and trucks. Accessed September 24, 2021, Zhang A, Kang JE, Kwon C (2017) Incorporating demand dynamics in
https://www.whitehouse.gov/briefing-room/presidential- multi-period capacitated fast-charging location planning for electric
actions/2021/08/05/executive-order-on-strengthening-american- vehicles. Transportation Res. Part B Methodological 103:5–29.
leadership-in-clean-cars-and-trucks/. Zheng H, He X, Li Y, Peeta S (2017) Traffic equilibrium and charging
Tran CQ, Ngoduy D, Keyvan-Ekbatani M, Watling D (2021) A user facility locations for electric vehicles. Networks Spatial Econom.
equilibrium-based fast-charging location model considering 17(2):435–457.