Content uploaded by Mohammad Sadra Rajabi
Author content
All content in this area was uploaded by Mohammad Sadra Rajabi on Sep 10, 2024
Content may be subject to copyright.
Applications of Machine Learning in the
Planning of Electric Vehicle Charging Stations
and Charging Infrastructure: A Review
Bhagyashree Panda, Mohammad Sadra Rajabi,
and Alimohammad Rajaee
Contents
1 Introduction............................................................... 2
2 Review of Different Machine Learning Used for EV Charging Station
Location Problem.......................................................... 3
3 Present Research on Charging Station Network Design............................ 5
3.1 Traditional Mathematical and Probabilistic Models........................... 6
3.2 Heuristic and Data-Driven Algorithms..................................... 8
3.3 Optimization Models................................................... 11
4 Conclusion and Future Directions............................................. 15
5 Cross-References........................................................... 17
References................................................................... 17
Abstract
While electric vehicles (EV) and plug-in hybrid electric vehicles (PHEV) have
the potential solution from an environmental perspective, they face an obstacle in
accessing charging systems. Moreover, the charging system offers its own chal-
lenges compared to petrol stations due to the participation of different charging
options. Researchers have been studying the optimization of PHEV/EV charging
B. Panda ()
Department of Civil and Environmental Engineering, The George Washington University,
Washington, DC, USA
e-mail: bpanda90@gwu.edu
M. S. Rajabi ()
School of Civil Engineering, Faculty of Engineering, University of Tehran, Tehran, Iran
e-mail: msadra.rajabi@gmail.com
A. Rajaee
Department of Maritime Engineering, Amirkabir University of Technology, Tehran, Iran
e-mail: Arajaee.10@aut.ac.ir
© Springer Nature Switzerland AG 2022
M. Fathi et al. (eds.), Handbook of Smart Energy Systems,
https://doi.org/10.1007/978-3-030-72322-4_202-1
1
2B. Panda et al.
infrastructure for the past few years. Introducing electric vehicle charging infras-
tructure services creates new challenges and opportunities for the development
of smart grid technologies. In this study, an extensive literature review has been
carried out regarding the use of several optimizations and machine learning
models for determining the optimal location of EV charging stations (EVCS)
and infrastructure. Previous literature has also proposed different model-solving
algorithms or techniques to solve the complex and dynamic nature of EVCS
location problems, suggesting that the research on EVCSs has recently gained
popularity. Although research seems to have advanced, findings indicate to
incorporate of real-time EV user behavior for optimal geographical placement
of new charging stations, satisfying the transportation demand, randomness, and
variability in space as well as time. Coupled EV networks will be cost-effective
from the power grid perspective. Identification of factors resulting in spatial
inequities for EVCS location across different cities based on socioeconomic
characteristics needs to be addressed for robust EV charging infrastructure.
Keywords
Electric vehicle · Charging stations · Charging infrastructure · Machine
learning
1Introduction
Fossil fuel-dependent transportation systems have aggravated environmental and
energy challenges (Aghakhani et al. 2022; Beigi et al. 2022c; Rajabi et al. 2022a).
Therefore, a growing number of countries around the world are promoting emerging
vehicle technologies, such as battery electric vehicles (BEVs), plug-in hybrid
electric vehicles (PHEVs), and fuel-cell electric vehicles (FCEVs) (Beigi et al.
2022a,b). The environmental implications of electric vehicles (EVs) depend largely
on the fuel mix of electricity generation but can replace vehicular tailpipe emissions
using liquid fossil-based fuels for power plants which are concentrated and easier
to control (Moeinifard et al. 2022; Mudiyanselage et al. 2021). On the other
hand, the rapid growth of electric vehicles can cause several problems, such as an
insufficient number of charging stations, uneven distribution, excessive cost, etc.
(Rajabi et al. 2022b). The unplanned installation of EVCS can adversely affect the
voltage stability, reliability, andother operating parameters of the power distribution
network. Further, charging opportunities should be abundant, fast, and inexpensive
to expand EV adoption. Moreover, understanding complex interactions of social-
economic and demographic factors leading to the inequitable placement of EVs
charging stations (EVCS) is crucial to mitigating accessibility issues and improving
EV usage among all people (Lotfi et al. 2022; Shakerian et al. 2022). Public charging
infrastructure must be able to simultaneously support diverse types of demand from
local users, taxis, city logistics, and long-distance drivers visiting the city, which are
expected to increase in the future. This demands for well-designed EVCS network
Applications of Machine Learning in the Planning of Electric Vehicle Charging... 3
and infrastructure (Amir Davatgari 2021; Kavianipour et al. 2022; Niroumand et al.
2018).
Currently, three types of Electric Vehicle Supply Equipment (EVSE) are used
for charging BEVs. Level-1 EVSE with 110 V/15 A connection typically takes
10–20 h to charge, restricting to a vehicle’s home base. Level-2 (220–240 V/15–
30 A) chargers require about 4–8 h and are used in both commercial and home
charging settings. Level-3 constitutes a “supercharging station” that uses high-
voltage (often 400–500 V) DC fast charging and takes as little as 20 min to charge
to 170 miles, making in-route/mid-trip charging feasible for most travelers. But
the cost of setting up Level-2 charging stations is almost doubled and goes up to
20 times for Level-3 charging stations than Level-1. Though energy requirements
of EVs can be met through single nighttime in-home charging or route to public
EVSE, to use these static facilities, EV drivers must stop and wait for charging.
Thus, dynamic wireless charging technologies have been developed, evaluated, and
commercialized to enable charging while driving.
The EV charging station location is basically a facility location problem focusing
mostly on associated costs of charging stations, station-user factors, station power
supplier factors, and environmental factors. Previous literature has paid attention
to the optimal layout of charging stations and proposed different model-solving
algorithms, which points out that the research on EVCSs has recently drawn
extensive interest. Studies on the planning of electric vehicle charging infrastructure
(EVCI) involved charging station placement, charging demand prediction, charging
scheduling and pricing, charger utilization computation, etc.
2 Review of Different Machine Learning Used for EV
Charging Station Location Problem
Investigating and dealing with the EVCI planning and modeling problem can be
grouped into three broad categories such as optimal placement of charging stations
based only on the transportation network or based only on the distribution network,
or based on the superposition of transport and distribution networks (Fig. 1).
According to the above figure, the modeling approach considering only transport
networks can be sub-divided into a node-, tour-, and path-based approaches.
Therefore, modeling approaches for transport and distribution networks individually
can be combined and will be compatible for modeling coupled networks. However,
combining three approaches at the same time may not be practical and has not been
acknowledged in any existing research (Deb et al. 2018).
Popular optimization methods include conventional optimization approaches
(such as linear, non-linear, integer, and mixed-integer programming), nature-
inspired optimization (NIO) methods (such as genetic algorithm, particle swarm
optimization, or evolutionary algorithm), or hybrid approaches. Optimal EVCI is
based on various key performance indicators, including proximity to chargers (slow
charging EVCI), waiting for time minimization (fast charging EVCI), placement
and sizing of charging stations, charging demand, etc. Models are classified as
4B. Panda et al.
Fig. 1 Schematic overview of the classification of charging infrastructure planning problem (Deb
et al. 2018)
either node-based such as set covering location models, node-based capacitated
vehicle routing problem, maximum covering location model, uncertainty modeling
and random sampling, zone-based demand formulation, etc., or flow-based, such
as flow-capturing location model, user equilibrium traffic assignment model, flow-
refueling location model, etc. or agent-based as multi-agent simulation models,
parametric Markov queues (M/M/s), non-parametric queuing models following
Kendall’s notation, etc. (Unterluggauer et al. 2022).
Geospatial and statistical approaches focus on density calculation and clus-
tering methods by calculating the charging demand at individual points, thereby
aggregating to certain demand areas using conversion tools, the sum of block
statistics, k-means fuzzy clustering method, etc. Explicit spatial location methods
use statistical data such as census data, EV travel data, EVCS data, fossil-fueled
vehicles travel data, questionnaire surveys, and simulation or test data (Fig. 2).
Agent-based models can define exact geographical allocations for EVCS using
GPS-tracked user behavior of EVs or conventional vehicles (Pagany et al. 2019).
Data partitioning can be employed by techniques such as hash partitioning, list
partitioning, and composite partitioning. Semi-supervised machine learning can
combine a small amount of labeled data and a large amount of unlabeled data during
training. Gaussian mixture model (GMM) is a probabilistic learning model that con-
Applications of Machine Learning in the Planning of Electric Vehicle Charging... 5
Fig. 2 Three levels of categorization on a structural and a content level (Pagany et al. 2019)
siders multiple normal distributions of the dataset to represent normally distributed
subpopulations. Hierarchal clustering, cellular automaton agent-based model, and
kernel density estimator (KDE) were used in the nonparametric probability density
function. Integration of genetic algorithm (GA) with reinforcement learning (RL)
improved model performance by not being stuck in local optima (Deb 2021).
The EVCS location problem was divided into the P-median problem (Lagrangian
relaxation algorithm, etc.), P-center problem (Drezner-Wesolowsky method), and
Maximum Coverage problem (Branch and Bound Method, Genetic Algorithm, etc.).
The traditional EVCS location problem is divided into Point Demand (charging
demand generated at a fixed point, such as residence, workplace, etc.), Flow
Demand/Traffic Demand (also known as the Flow Capturing Location Model,
assuming that charging demand is generated during driving) and Mixed Model
(also known as Flow Refueling Location Model that maximized capture flow and
minimized the sum of weighted distance). Multilevel charging station location
model and algorithm is based on TABU search algorithm and finding solution using
2-opt domain strategy. A combination of decision-making models with traditional
EVCS location models, such as game theory, queuing theory, grey hierarchical
analysis (GAHP), etc., is used for optimal EVCS location. Real-time site selection
requirements can be integrated into the Voronoi diagram in computational geometry
for a dynamic assignment using advanced hybrid algorithms such as chaos theory,
particle swarm optimization algorithm, NIO algorithms, etc. (Wang et al. 2019).
3 Present Research on Charging Station Network Design
Researchers have utilized a variety of approaches, objective functions, and opti-
mization algorithms to solve the complex and dynamic nature of the EVCS
6B. Panda et al.
location problem. In the following sections, the latest findings, developments, and
literature related to different machine learning approach for the location of charging
stations and charging infrastructure planning for electric vehicles are reviewed and
presented.
3.1 Traditional Mathematical and Probabilistic Models
A location model for EVCS was proposed considering the EVs charging require-
ment, economy, and power grid safety along with the satisfaction of EV traffic. The
objective was to minimize the cost, with the waiting time of the EV users and the
ability of the distribution network to run safely as the constraint condition. Lastly,
the EVCS service area was divided by the Voronoi diagram, modeling a 25-node
traffic network with a 24-node underground distribution network, and the proposed
method was demonstrated using the MATLAB environment (Zhu et al. 2017).
A multidisciplinary approach in the form of a two-stage stochastic programming
model was proposed for the location and size of coupled EVCS and distributed
photovoltaic (PV) power plants. The comprehensive model considered explicit EV
driving range constraints in transportation networks, probability-based quality of
service constraints of EVCS, PV power generation with reactive power control, and
alternating current distribution power flow. First, a mixed-integer second-order cone
program was generated, and then a generalized Benders Decomposition Algorithm
was developed for the solution. Results showed positive effects of the combined
design of PV power plants with EVCS, such as reducing social costs, promoting
renewable energy integration, and relieving power congestion (Zhang et al. 2018a).
Considering heterogeneous PHEVs driving ranges and charging needs, a closed-
form model quantifying the service capabilities of fast-charging stations was
developed by modifying capacitated flow refueling location model based on sub-
paths (CFRLM SP). A stochastic mixed-integer second-order cone programming
(SOCP) model was proposed for planning PEV fast-charging stations. In the model,
CFRLM SP was used to consider the transportation network constraints as well
as the power network constraints to enhance the model’s computational efficiency
and accuracy and was demonstrated numerically through simulations (Zhang et al.
2018b).
A GIS-based multi-objective particle swarm optimization model was developed
for optimal placement of EVCS and rapid EV production in Changping district,
Beijing. Based on the dual benefit of public service and investment in EVCI,
two objectives were identified as minimizing the total costs and maximizing the
coverage. Results suggested that EVCS scale and expansion cost limited the type
of station that could be built in some nodes. The Pareto curve between costs and
coverage indicated a change in scale economies (diseconomies of scale) in the
process of designing and constructing EVCS (Zhang et al. 2019).
A novel and simple approach were formed to determine the best site for charging
stations, where the integrated cost of installation of charging stations was considered
along with the penalty for violating grid constraints. Teaching-Learning Based
Applications of Machine Learning in the Planning of Electric Vehicle Charging... 7
Optimization (TLBO) and Chicken Swarm Optimization (CSO) were both efficient
evolutionary algorithms that were combined to achieve an optimal solution to this
problem. Several benchmark and charging station placement problems were used to
assess the effectiveness of the proposed algorithm by comparing the results of the
hybrid algorithm with other algorithms (Deb et al. 2021a).
Previous studies emphasized the mileage anxiety of EV users but ignored their
competitive and strategic charging behavior. Therefore, depending on the choice of
other EV users, an EV user’s charging cost was calculated consisting of the travel
cost to access the charging station and the queueing cost at the charging stations.
Firstly, a Charging Station Placement Problem (CSPP) was formulated as a bi-level
optimization problem, which was then converted into a single level by using the
equilibrium of the EV charging game. To compute the optimal allocation of electric
charging stations, the properties of CSPP were analyzed, and an optimization
algorithm, OCEAN (Optimizing electric vehicle charging stations) was introduced.
To address the scalability issue of OCEAN, an algorithm using continuous variables
was developed to manage large-scale real-world problems. Results from the analysis
of extensive experimental data showed that the above approach outperformed the
baseline method significantly (Xiong et al. 2018).
A game theory-based method for determining the optimal location of electric
vehicle charging stations was developed by analyzing the traditional site-selection
method. Detailed steps were given for the game experiment and its algorithm.
According to the results, this optimization method was able to make charging station
locations more rational and scientific using game theory (Meng and Kai 2011).
The impact of EV charging station loads on voltage stability, reliability, and
harmonics of the IEEE 69-bus distribution network was analyzed and was used for
optimal placement of EVCS in the distribution network using a Genetic Algorithm
without disturbing the operating parameters of the network. Results showed the
degradation of operating parameters following the addition of EV charging load
(Deb et al. 2019).
Point spatial patterns identified by G-estimation can be compared with complete
spatial randomness (CSR). A generalized log-linear model (GLM) or a Poisson
lognormal spatial model using Besag York Mollié Model (BYM2) method addresses
any residual clustering issues using the Bayesian method. Integrated Nested Laplace
Approximations (INLA) combines Laplace approximations and numerical integra-
tion and enables Generalized Linear Mixed Models (GLMMs) to address temporal
and spatial error terms. The intrinsic Conditional Auto-Regressive (ICAR) model
focuses on spatial autocorrelation with K-means clustering analysis identifying
similar regions in terms of socioeconomic and housing patterns. A geographically
weighted regression (GWR) model identifies sensitivities of concerned predictors
in EV charger installations to census tracks. Empirical analyses of residential EV
chargers in Seattle, WA, included equitable spatial distribution analysis of EV
charger-installed buildings with respect to housing and socioeconomic character-
istics by advanced data mining techniques. Results revealed social equity and
economic status issues based on the uneven or clustered distribution of residential
8B. Panda et al.
EV charger installations, i.e., certain communities don’t deserve clean energy
technologies (Min and Lee 2020).
Traditional adaptive algorithms for EV charging station location problems
mainly include Genetic Algorithm (GA), Particle Swarm Optimization (PSO),
Tabu Search (TS), Simulated Annealing (SA), etc. To improve efficiency, various
combination of algorithms was used, such as quantum particle swarm optimization
algorithm, Fuzzy multi-objective-based grasshopper optimization algorithm, multi-
group hybrid genetic algorithm (MPHGA) combining standard genetic algorithm
(SGA) with the Alternate Location Allocation Algorithm (ALA), two-step optimiza-
tion model combining immune algorithm and fuzzy analytic hierarchy process, etc.
(Wang et al. 2019).
3.2 Heuristic and Data-Driven Algorithms
A methodology was proposed for decision-makers in the field of developing
EVCI, which combined with multiple heterogeneous real-world data sources, which
includes business data describing charging infrastructure, historical data about
charging transactions, information about competitors, geological information, data
about places of interest located near chargers (e.g., hospitals, restaurants, and shops),
and driving distances between charges. The study suggested the optimal location of
a new EVCS by incorporating the proposed methodology into decision support tools
and using historical data on EVCS utilization. The methodology was illustrated by
using Dutch EVCI data from 2013 to 2016, based on various objectives, such as
increasing the number of chargers in sparsely populated areas or maximizing the
overall charging network utilization (Pevec et al. 2018).
A hybrid approach incorporating geographical information system (GIS) and
Bayesian network (BN) was developed to solve the location selection problem of
EVCS in Singapore. GIS integrated spatial and geographical data, whereas the BN
model demonstrated the cause-effect relationships of nine criteria when selecting
suitable alternative sites. Conditions such as the number of rapid mass transit
(MRT) stations, the number of household units, transportation efficiency, and the
charging efficiency were identified to be the most crucial factors in influencing
location selection over social and economic factors. Compared to the traditional
decision-making method (TOPSIS, etc.), the hybrid GIS-based BN approach was
more accurate and stable in the presence of noise interruption (Zhang et al. 2022).
Four solution strategies were explored for the location of charging stations, and
a heuristic solution for fleet routing was established. In addressing the routing
problem, the location strategy was applied at the client site without considering
displacements for the recharges. While the other three proposals performed poorly
when it came to locating the charging stations at the center of the cluster, the K-
means approach performed the best (Gatica et al. 2018).
A heuristic methodology was presented for an extensive urban transportation
network that considered the deployment of EVCS for coverage and the fulfillment
of user preferences and constraints as two separate processes. This methodology
Applications of Machine Learning in the Planning of Electric Vehicle Charging... 9
proposed a reallocation algorithm to prioritize the selection of Locations of Interest
and to reduce the number of stations with overlying coverage areas. The results
were compared to those derived from a Greedy Algorithm based on a multipath
consideration, and the proposed methodology was able to significantly reduce the
computational time required for the solution of the location problem (Torres Franco
et al. 2021).
An agent-based traffic simulation-based approach was presented for EVCS
placement using heuristic objectives to achieve sufficient network coverage to keep
charging-related inconvenience within an acceptable range while minimizing the
overall number of EVCS in Singapore. The algorithm identified locations where
the charging procedure was seamlessly integrated into driving routes, thus reducing
detours and waiting times. Results showed that the proposed methodology was able
to cover the charging needs of 20,000 EVs with approximately 2500 EVCS by
accepting average detours of 410 m and waiting times of less than 10 min. Based
on the results, the algorithm was able to converge toward an EVCS distribution that
was effective at satiating charging demand within the constraints of inconvenience
(Bietal.2017).
A data-driven approach was developed for optimizing the layout of existing
electric taxis (ET) charging stations in a more realistic manner by using four
different types of data: ET trajectory, points of interest, station data, and road
network information. Citywide charging behavior was modeled using a 3D tensor,
filling in the missing entries of times and stations with sparse data using contextual-
aware tensor collaborative decomposition techniques to estimate the popularity of
charging stations and develop queueing systems to estimate the frequency of visits
among stations. A spatial-temporal demand coverage method, Bass model or Bass
diffusion model, and Dynamic traffic simulation-based optimization were proposed
to predict the permeability and optimize the layout of charging stations for electric
taxis. The search and navigation behavior and charging patterns of EV users were
analyzed, and a Bayesian reasoning-based method was developed for evaluating
charging demand. A charging spot model known as a single output multiple cables
(SOMC) charging spot was proposed to increase the utilization of charging stations
and decrease investment costs associated with coordinated charging. Propose (Yang
et al. 2020).
Gray Analytic Hierarchy Process (GAHP), comprising of AHP and gray statistics
decision-making along with the Delphi technique, were combined to form a new
comprehensive evaluation method to solve the personal one-sidedness problem of
expert judgments and to deal with ambiguous gray factors. The new method was
applied to produce the comprehensive evaluation index system of EVCS and to
determine the weights of its indicators. Based on this, the rationality comprehensive
evaluation value for each optional hail was defined and used to formulate an optimal
decision regarding the location of charging stations (Liu et al. 2012).
Maximum Covering Location Problem (MCLP) was formulated for optimally
allocating PHEV Charging Stations (CSS) in a highway network. The model
considered the Trip Success Ratio (TSR) to estimate Charging Station Service
Range (CSSR) and enhance CS accessibility for PHEV drivers, allowing for
10 B. Panda et al.
different driving habits and trip types. Essentially, the allocation model had two
stages. In the first stage, the CSSR was estimated utilizing TSR, considering the
uncertainty of the trip distances (city, highway) and the uncertainty associated with
the remaining electric range (RER) of PHEVs. As part of the CS allocation process,
the estimated CSSR was used to select the optimal EVCS locations covering the
network with a certain guaranteed TSR level (Alhazmi et al. 2017).
The location model was formulated to minimize total social cost (direct and
indirect costs covering construction, operation, charging, and the wastage cost), and
a genetic algorithm using a moderate value function was used to solve for quantity
and location of charging in the urban area of Nanjing. The evaluation index was
based on five location influencing factors: land cost, construction costs, road traffic
flow, power grid conditions, and the surrounding environment. Numerical results
showed that both grey correlated scheme decision-making and grey target theory
could conveniently conduct quantification treatment to the qualitative problem
while selecting the optimal location. This model was advantageous as it has less
data collection requirement and treatment, a simple calculation process, presents
quantified results, and is more credible than a qualitative description. Multi-expert
scoring and selection or repeated testing was conducted during the evaluation (Ren
et al. 2019).
A multi-period multipath refueling location model capturing the dynamic inter-
city origin-destination (O-D) trips on both spatial and temporal dimensions was
developed to expand the public electric vehicle (PEV) charging network in Sioux
Falls Road in South Carolina. The objective of the model was the minimization
of the installation cost of new stations and relocation of existing stations to satisfy
every O–D trip or at least one path between an O–D pairs called the deviation path.
The multi-period or dynamic facility location problem was formulated as a mixed-
integer linear program and solved by a heuristic-based genetic algorithm using a
CPLEX solver. Results indicated that the geographic distributions of cities, vehicle
range, deviation choice, and the types of charging station sites are the major factors
impacting charging station location. Heuristics and anticipation of future demands
can yield high-quality solutions, especially when the problem is getting complex
with deviations, and help reduce the overall cost of EVCI (Li et al. 2016).
The machine learning frame, work along with quantitative spatial analysis,
examined spatial disparities and barriers in EVCS placements by combining social,
economic, and demographic factors with field data for predicting future ECVS
density and identifying the optimal spatial resolution for Orange County, California.
Finally, the optimal EVCS placement density was compared with a spatial Electric
Vehicle Charging Inequity index, developed using a multicriteria decision analysis
approach to quantify how equitable these placements would be. The existing EVCS
location was evaluated using kernel density estimation (KDE), and supervised
machine learning models with repeated “k” fold cross-validation predicted the
EVCS grids for further spatial analysis using constructed raster layers. Random
Forests achieved the highest predictive accuracy of EVCS placement density at a
spatial resolution of 3 km using. Results indicate that a total of 74.18% of predicted
EVCS placements will lie within a low spatial equity zone, indicating less accessible
Applications of Machine Learning in the Planning of Electric Vehicle Charging... 11
populations require the highest investments in EVCS placements (Roy and Law
2022).
Neural network and multivariate optimization conditions using historical or
experimental data were used to create the Ethe VCI network model (Wang et al.
2019).
3.3 Optimization Models
Access to the mobile charging station (MCS) for electric vehicles was examined,
where charging station owners were the EV parking lots, proposing a new self-
scheduling model for EV smart parking lots (SPLs) that optimized SPL energy
generation and storage schedule while scheduling MCSs as temporary charging
infrastructures. By modeling prioritized demand of the prioritized events based
on several indices, the MCSs accessibility measures, the equity impacts of MCSs
locations, the optimal set of SPL components (such as coupled heat and power,
photovoltaic system, and electric and heat energy storage) to manage electrical peak
demand and the economic benefits of SPLs were accessed. Results showed that the
proposed demand prioritization function model was able to meet the required EV
charging demands for prioritized events, while the self-scheduling model for SPLs
was able to meet the changing demands of EVs located at SPL locations (Nazari-
Heris et al. 2022).
Two optimization models were investigated to determine the locations of public
EVCS using fast and charging modes in Greater Toronto and Hamilton Area
(GTHA) as well as Downtown Toronto. The objective was to minimize the overall
cost while satisfying coverage needs. Geometric objects were used to represent
charging demands instead of discrete points, using the polygon overlay method,
which was extended to split the demand on the road network to resolve the partial
coverage problem (PCP). The methods offered better accuracy than complementary
partial coverage (CP) models and were able to eliminate PCP (Huang et al. 2016).
Bi-level programming model addressed the planning of fast-charging stations
located in electrified transportation networks with uncertainty in charging demand.
An upper-level model for locating refueling stations was used to minimize the
cost of planning fast-charging stations, whereas a lower-level traffic assignment
model was used to estimate the location and timing of plug-in electric vehicle flow
on whole transportation networks. Two-level models reveal a correlation between
charging demands, electrical demands, and the spatial and temporal distribution of
plug-in electric vehicle traffic. Under distribution-free uncertain charging demands,
robust chance constraints were formulated to describe the service capability of fast-
charging stations, where the ambiguity set was constructed to estimate the potential
values of the uncertainties based on their moment-based information, thus reducing
the robust chance constraints exactly to mixed-integer linear constraints. Adding
new variables converted the bi-level model to a single-level mixed-integer second-
order cone programming model that can be solved with off-the-shelf solvers, which
guarantee the optimality of the solution. The effectiveness of the proposed planning
12 B. Panda et al.
model was illustrated by a case study that revealed the significant impact of the three
critical factors on the planning outcomes (Zhou et al. 2020).
A review of the literature was performed to examine the impact of electric vehicle
charging stations on the power distribution network. As a result of the findings, 34%,
34%, 15%, and 19% of the literature have examined voltage stability, power quality,
peak load, and transformer performance associated with EV charging stations,
respectively. Although there were fewer publications analyzing the effect of EV
charging stations on peak loads, it was observed several researchers were becoming
intrigued by this paradigm. It has been documented in all research studies that the
addition of EV charging load degrades the power grid’s operating parameters. While
EVs may suffer negative impacts from the charging load, the vehicle-to-grid (V2G)
scheme has several positive impacts which should not be ignored (Deb et al. 2017).
A comprehensive review of the Nature Inspired Optimization (NIO) algorithms
for solving the charging station placement problem was presented. The goal of this
work was to provide the research community with a deep understanding of the key
features, advantages, and disadvantages of the various NIO algorithms for solving
the charging station placement problem. In other words, this review would help
researchers not only in selecting suitable algorithms but would also serve as a model
for developing efficient algorithms to solve the charging station placement problem.
Also, a general classification of NIO Algorithms was proposed, as depicted in the
Fig. 3below.
Furthermore, the performance of seven NIOs in solving four variants of the
charging station placement problem was compared. The hybrid algorithms involving
Genetic Algorithm, Particle Swarm Optimization, Chicken Swarm Optimization,
and Teaching Learning Based Optimization were found to be more efficient in
solving the charger placement problem than the standalone algorithms. However,
hybrid algorithms run slower than stand-alone algorithms (Deb et al. 2021b).
Fig. 3 Classification of NIO Algorithms (Dhal et al. 2019)
Applications of Machine Learning in the Planning of Electric Vehicle Charging... 13
Long-distance travel data was used to place charging stations with the objective
of maximizing long-distance trip completions. To reduce the dimensionality of the
problem, K-means clustering was used to reduce the area centroids down to 200
cluster centers with the highest total density (sum of households and jobs per square
mile), and only heavily used OD paths were tracked between them. A mixed-
integer problem scenario assuming 50–250 charging stations and an all-electric
range (AER) of 60–250 miles, based on a modified flow-refueling location model
(FRLM) was formulated and solved via a branch-and-bound optimization algorithm.
The final proposed model was a facility location and network expansion problem to
maximally serve long-distance demands between clusters’ origins and destinations.
Balance constraints are used to expand the scope to all potential paths and guarantee
that EVs can pass a path smoothly when there is a charging station. Restrictions
on the total number of stations installed were introduced, along with power and
budget constraints. Results reveal that at least 100-mile-range EVs were needed to
avoid long-distance-trip issues for U.S. households, whereas 200-mile-range EVs
can serve nearly all long-distance trips with just 100 charging locations in eastern
and central U.S. (He et al. 2019).
A simulation-optimization model was used to locate Level-1 and Level-2 EV
charging infrastructure for maximizing EV service levels in the central-Ohio region,
utilizing Mid-Ohio Regional Planning Commission (MORPC) data. The volume of
EV flows was determined by assigning an EV adoption probability, which depends
on demographic and macroeconomic data. Simulation of expected service levels
with different numbers of chargers was conducted by bootstrapping EV trips from
the tour record data by randomly generating Bernoulli trials. Finally, the linear
integer programming (IP) model was used to determine the location and size of
charging stations, with the objective of maximizing fleet-wide EV charging and
the amount of battery energy recharged. Constraints include higher distribution-
level transformer capacity of chargers and general budget limits. Sensitivity analysis
was conducted through simulation to account for interdependencies such as the
possibility of an EV being charged earlier at another charging station and EV driving
patterns. Results showed that with two chargers, 87% of EVs at a university parking
location and around 45% at workplace and shopping locations could charge up to
90%. Moreover, the station locations were “robust” when the budget constraint was
relaxed, proving that the optimal location is sensitive to the specific optimization
criterion, irrespective of overall service levels (Xi et al. 2013).
Mixed Integer Linear Programming (MILP) optimization model to maximize the
electrified fleet vehicle-miles-traveled (VMT), minimize the total travel distances
without electrification, and select the locations of public charging stations based
on real-world vehicle travel patterns and public charging demand from large-scale
individual vehicle trajectory data. Vehicle trajectory data of 11,880 taxis over a
period of 3 weeks in Beijing, China, was used in the model to capture the variation
in vehicle travel behavior. The mathematical problem was formulated in the GAMS
modeling environment, and the CPLEX optimizer was adjusted to find the optimal
solutions using a branch and cut algorithm. Along with 40 existing public charging
stations, the 40 optimal ones selected by the model were able to increase the
14 B. Panda et al.
electrified fleet VMT by 59% and 88% for slow and fast charging, respectively.
With the increase in the number of charging stations, the optimal station locations
expand outward from the inner city. Results indicate that the optimal slow charging
stations have a higher retention rate than the fast-charging ones and can be used for
long-term planning. Additionally, more charging stations can cover the variation of
travel patterns among different weeks (Shahraki et al. 2015).
A mixed-integer bi-level program was formulated for an optimal deployment
plan of static and dynamic charging infrastructure allowing for interdependency
between transportation and power networks, with an objective to minimize the total
social cost of the coupled networks. First, a variational inequality (VI) network
equilibrium model was devised and solved by converting it to a nonlinear program to
describe the correlation among BEVs route choices, charging plans, and electricity
cost. Active-set algorithm was used for the solution to the model. Sensitivity
analysis was conducted to study the impact of the variations in power rate, charging
efficiency, and battery size on the numerical equilibrium results. Results on three
networks suggest that for individual BEV drivers, the choice between usingcharging
lanes and charging stations is more sensitive to the value of travel time, service fee
markup, and battery size than to the charging rates and travel demand. Charging
lanes are more beneficial for transportation networks with sparser topology, while
charging stations is desirable in denser networks (Sun et al. 2020).
A social total cost model based on economic cost (construction costs and fees)
and environmental cost (electricity consumption and carbon dioxide emissions)
calculates the total operating cost of charging stations under various distribution
conditions. Secondly, a genetic algorithm-based EVCS location optimization model
provides the solution to an operating cost minimization problem and distribution of
charging stations. Data from five major cities in Ireland was used to establish EVCS
and distribution optimization model based on reality, simulate the optimal results,
and conduct sensitivity analysis. Sensitivity analysis showed that the total cost is
extremely sensitive to the number of charging stations and the probability of EV
charging per day. Instead of traditional Euclidean distance, this model calculates
the distance and demand between the electric vehicle and the charging station by
introducing a parameter known as the road bending coefficient for realistic model
results (Zhou et al. 2022).
A multi-objective optimization based on data envelopment analysis for the
optimal placement of slow EVCI in a combined transportation and distribution
network, solved by a cross-entropy algorithm. Game-theoretical optimization,
Harris hawk’s optimization, and the Differential evolution approach were useful for
the placement of public EVCI in the urban area for plug-in hybrid EVs, solved by an
active-set algorithm. Decomposition-based multi-objective evolutionary algorithm,
Nature-inspires optimization (NIO) algorithm (binary lightning search algorithm,
hybrid chicken swarm optimization, GIS-based particle swarm optimization, Ant
colony optimization, etc.), mixed-integer non-linear programming problem solved
by a genetic algorithm and commercial solvers, bi-level planning model based on
dynamic real-time data solved by a surrogate-based optimization algorithm were
used for deployment of fast EVCI in urban areas. Chance-constrained programming
Applications of Machine Learning in the Planning of Electric Vehicle Charging... 15
approach, teaching learning-based optimization approach, multi-objective grey
wolf optimization with the fuzzy satisfaction-based decision, reliability-oriented
multi-objective planning model solved by simplified reliability correlation analysis
were used in the deployment of fast EVCI in residential areas. Mixed-integer
linear program solved by a branch-and-bound method, stochastic mixed-integer
second-order cone programming model solved by the branch-and-cut method,
iterative column generation algorithm, branch-and-reduce optimization navigator in
GAMS, fuzzy multi-objective optimization approach, solved by a cooperative co-
evolutionary genetic algorithm, decomposition-based multi-objective evolutionary
algorithm, multi-stage charging placement strategy based on a Bayesian game were
used for allocation of fast EVCI for long-distance travel along highway and for non-
private EVs (electric taxis, electric buses, and freight EVs). Mixed-integer convex
programming approach was used for optimal siting and sizing of EVCI, power
generation, new transportation lanes, and distribution feeders (Unterluggauer et al.
2022).
A five-stage multicriteria- and GIS-based EVCS location methodology (5MAGI-
SEV) was formulated for designing the EVCS network for personal and commercial
vehicles, considering the existing locations in urban areas of Poznan, Poland, and the
interaction between every EVCS in the network. A DBSCAN clustering algorithm
was used to the reduction in the number of possible EVCS locations. The design
stages include defining potential EVCS locations, constructing evaluation criteria,
generating alternatives, selecting appropriate multiple criteria decision aid, wherein
light beam searching or LBS solution method was selected, and conducting the
multicriteria evaluation of alternatives. LBS method was able to manage a set of
numerous alternatives and criteria and make comparisons among alternatives using
the natural scale of each criterion (Schmidt et al. 2021).
An optimization-based Flow Refueling Location Model (FRLM) was formulated
for EVs fast-charging stations along designated EV corridors to improve EVs
adoption. The objective was to maximize Corridor-Miles Traveled (CMT) in
EV corridors with high density based on corridor-utilizing and corridor-weighted
traffic flow concepts. EV corridors selected from numerical experiments in the
Maryland highway network and major population centers were inputs to the model.
Results showed that corridor-focused objective function and corridor prioritization
is essential for deciding on a solution with more stations on corridors (Erdo˘
gan et al.
2022).
4 Conclusion and Future Directions
Following are the observations and future scope of work for improving EVCS
location problem and setting up an efficient EVCI according to the researcher’s
opinions and findings from various literature.
1. Faster computers or smarter algorithms such as greedy algorithms, heuristics
and/or simulated annealing, etc. can allow enormous OD pairs, more network
16 B. Panda et al.
details, charging station capacity, trip scheduling information, charging time
and speed, and demand of all EVs to reflect the city level public charging
infrastructure for mixed uses. Deep learning-based location algorithms can attain
multi-factor coverage ability, independent feature selection ability of the model,
evolvability over time, and simulation of large-scale vehicle movement scenarios.
2. Route choice for EV travelers, congestion feedbacks mostly arising from local-
trip sources at peak times of day, driver’s willingness to wait, shorter local travel,
queuing delays (queuing models), travel behavior before and after EV adoption
and limitations on EV range could have a strong influence in charging station
location modeling. Monitoring user behavior in a real-world environment is
resource-intensive and costly. Optimal location decisions can be automatically
updated based on observations from GIS. Further, a bi-level optimization
framework can be utilized to incorporate traffic flow between O–D pairs at the
lower level and locational decisions at the upper level. A variety of data-based
driving behavior and travel behavior analysis methods can be used to improve
the positioning strategy.
3. Not only charger station location but the number or type of chargers to install
at each location must be determined for developing static slow or fast charging
infrastructure for EV. Large-scale deployment of public charging infrastructure
will lead to more regular and profound interactions between the transportation,
charging infrastructure, and power networks increasing the risk of supply bottle-
necks in the power system, hence requiring integrated planning and operation
of coupled networks. Moreover, photovoltaic power generation coupled with
reactive power control can enhance the quality of power supply to EVCI. Most
studies considered energy consumption to be directly proportional to the distance
traveled; rather, more energy is used going uphill or higher speed compared to
going downhill or lower speed. Future charging infrastructure planning would
include finding an optimal geographical placement and new charging station
sizing to satisfy the transportation demand, considering a mix of several types
of charging options to account for randomness and variability in space and time,
as well as being cost-effective from the power grid perspective. Elastic demand
for EVs should be considered in establishing optimal charging infrastructure
deployment plans and changes in electricity locational marginal price (LMP).
4. Future models can expand to optimize the environmental benefits of EVs fleet,
which depends on the grid fuel mix, charging time, and individual driving
conditions.
5. The dynamic programming method is suitable for the multi-period charging
location problems (two-phase approach), wherein the first phase can utilize
integer programming to identify the locations to be placed over time, and the
second phase solves a linear program to identify optimal routes between O–D
pairs. Multistage stochastic program using nested decomposition solution can
address the stochasticity of multiple trajectories embedded in future demand,
though challenging in both modeling and solution. Combined geospatial analyses
of the EV transport and charging station location planning were found in a few
studies.
Applications of Machine Learning in the Planning of Electric Vehicle Charging... 17
6. Studies reveal that EV charger infrastructures are clustered in neighbors with
higher income levels, single housing ownership rates, and urban areas; how-
ever, further research is required to identify factors contributing to spatial
inequities across different cities based on socioeconomic characteristics. The
rapid transition to clean energy technologies can impact community response and
create inequalities resulting in uneven distribution of the charging infrastructure.
The inequity index levels can identify regions with limited access to charging
infrastructure and propose prospective charging station locations.
5 Cross-References
A Novel Mathematical Model for Infrastructure Planning of Dynamic Wireless
Power Transfer Systems for Electric Vehicles
Analysis of the Renewable Energy Generation Capability for Attending a National
Renovation Fleet Through Ethanol-Cell Electric Vehicles in a South American
Market
Applications of Machine Learning for Renewable Energy: Issues, Challenges, and
Future Directions
References
S. Aghakhani, B. Mohammadi, M.S. Rajabi, A new hybrid multi-objective scheduling model for
hierarchical hub and flexible flow shop problems. arXiv (2022). http://arXiv:2205.06465
Y.A. Alhazmi, H.A. Mostafa, M.M.A. Salama, Optimal allocation for electric vehicle charging
stations using Trip Success Ratio. Int. J. Electr. Power Energy Syst. 91, 101–116 (2017)
P. Beigi, M. Haque, M.S. Rajabi, S. Hamdar, Bike share’s impact on COVID-19 transmission and
bike share’s responses to COVID-19: a case study of Washington DC. arXiv (2022a). http://
arXiv:2205.05011v1
P. Beigi, M. Khoueiry, M.S. Rajabi, S. Hamdar, Station reallocation and rebalancing strategy for
bike-sharing systems: a case study of Washington DC. arXiv (2022b). http://arXiv:2204.07875
P. Beigi, M.S. Rajabi, S. Aghakhani, An overview of drone energy consumption factors and
models. Arxiv (2022c). http://arXiv:2206.10775
R. Bi, J. Xiao, D. Pelzer, D. Ciechanowicz, D. Eckhoff, A. Knoll, in A Simulation-Based Heuristic
for City-Scale Electric Vehicle Charging Station Placement. 2017 IEEE 20th International
Conference on Intelligent Transportation Systems (ITSC) (IEEE, New York, 2017), pp. 1–7
A. Davatgari, Location Planning for Electric Charging Stations and Wireless Facilities in the Era
of Autonomous Vehicle Operations (Purdue University, West Lafayette, 2021)
S. Deb, Machine learning for solving charging infrastructure planning problems: a comprehensive
review. Energies 14(23), 7833 (2021)
S. Deb, K. Kalita, P. Mahanta, in Review of Impact of Electric Vehicle Charging Station on the
Power Grid. 2017 International Conference on Technological Advancements in Power and
Energy (TAP Energy) (IEEE, New York, 2017), pp. 1–6
S. Deb, K. Tammi, K. Kalita, P. Mahanta, Review of recent trends in charging
infrastructure planning for electric vehicles. WIREs Energy Environ. 7(6) (2018).
https://doi.org/10.1002/wene.306
S. Deb, K. Kalita, P. Mahanta, Distribution network planning considering the impact of electric
vehicle charging station load, in Smart Power Distribution Systems, (Elsevier, Amsterdam,
2019), pp. 529–553
18 B. Panda et al.
S. Deb, X.-Z. Gao, K. Tammi, K. Kalita, P. Mahanta, A novel chicken swarm and teaching learning
based algorithm for electric vehicle charging station placement problem. Energy 220, 119645
(2021a)
S. Deb, X.-Z. Gao, K. Tammi, K. Kalita, P. Mahanta, Nature-inspired optimization algorithms
applied for solving charging station placement problem: overview and comparison. Arch.
Comput. Methods Eng. 28(1), 91–106 (2021b)
K.G. Dhal, S. Ray, A. Das, S. Das, A survey on nature-inspired optimization algorithms and their
application in image enhancement domain. Arch. Comput. Methods Eng. 26(5), 1607–1638
(2019)
S. Erdo˘
gan, ˙
I. Çapar, ˙
I. Çapar, M.M. Nejad, Establishing a statewide electric vehicle charging
station network in Maryland: a corridor-based station location problem. Socio Econ. Plan. Sci.
79, 101127 (2022)
G. Gatica, G. Ahumada, J.W. Escobar, R. Linfati, Efficient heuristic algorithms for location of
charging stations in electric vehicle routing problems. Stud. Inform. Control 27(1) (2018).
https://doi.org/10.24846/v27i1y201808
Y. He, K.M. Kockelman, K.A. Perrine, Optimal locations of U.S. fast charging stations for long-
distance trip completion by battery electric vehicles. J. Clean. Prod. 214, 452–461 (2019)
K. Huang, P. Kanaroglou, X. Zhang, The design of electric vehicle charging network. Transp. Res.
Part D Transp. Environ. 49, 1–17 (2016)
M. Kavianipour, H. Mozafari, M. Ghamami, A. Zockaie, R. Jackson, Effects of electric vehicle
adoption for state-wide intercity trips on emission saving and energy consumption. Int. J.
Sustain. Transp. (2022)
S. Li, Y. Huang, S.J. Mason, A multi-period optimization model for the deployment of public
electric vehicle charging stations on network. Transp. Res. Part C Emerg. Technol. 65, 128–143
(2016)
Y. Liu, B. Zhou, C. Feng, S. Pu, in Application of Comprehensive Evaluation Method Integrated
by Delphi and GAHP in Optimal Siting of Electric Vehicle Charging Station. 2012 International
Conference on Control Engineering and Communication Technology (IEEE, New York, 2012),
pp. 88–91
R. Lotfi, B. Kargar, A. Gharehbaghi, M. Afshar, M.S. Rajabi, N. Mardani, A data-driven robust
optimization for multi-objective renewable energy location by considering risk. Environ. Dev.
Sustain. (2022). https://doi.org/10.1007/s10668-022-02448-7
W. Meng, L. Kai, in Optimization of Electric Vehicle Charging Station Location Based on
Game Theory. Proceedings 2011 International Conference on Transportation, Mechanical, and
Electrical Engineering (TMEE) (IEEE, New York, 2011), pp. 809–812
Min, Y. and Lee, H.W., in Social Equity of Clean Energy Policies in Electric-Vehicle Charging
Infrastructure Systems. Construction Research Congress 2020 (American Society of Civil
Engineers, Reston, 2020), pp. 221–229
P. Moeinifard, M.S. Rajabi, M. Bitaraf, Lost vibration test data recovery using convolutional neural
network: a case study (2022). Available at: http://arxiv.org/abs/2204.05440
S.E. Mudiyanselage, P.H.D. Nguyen, M.S. Rajabi, R. Akhavian, Automated workers’ ergonomic
risk assessment in manual material handling using sEMG wearable sensors and machine
learning. Electronics 10(20), 2558 (2021)
M. Nazari-Heris, A. Loni, S. Asadi, B. Mohammadi-ivatloo, Toward social equity access and
mobile charging stations for electric vehicles: a case study in Los Angeles. Appl. Energy 311,
118704 (2022)
R. Niroumand, S. Bahrami, H.Z. Aashtiani, M.J. Roorda, Battery electric vehicles network
equilibrium with flow-dependent energy consumption (No. 18-04519) (2018)
R. Pagany, L. Ramirez Camargo, W. Dorner, A review of spatial localization methodologies for the
electric vehicle charging infrastructure. Int. J. Sustain. Transp. 13(6), 433–449 (2019)
D. Pevec, J. Babic, M.A. Kayser, A. Carvalho, Y. Ghiassi-Farrokhfal, V. Podobnik, A data-driven
statistical approach for extending electric vehicle charging infrastructure. Int. J. Energy Res.
42(9), 3102–3120 (2018)
Applications of Machine Learning in the Planning of Electric Vehicle Charging... 19
M.S. Rajabi, P. Beigi, S. Aghakhani, Drone delivery systems and energy management: a review
and future trends. arXiv preprint (2022a). http://arXiv:2206.10765
M.S. Rajabi, H. Taghaddos, M. Zahrai, Improving emergency training for earthquakes through
immersive virtual environments and anxiety tests: a case study. Buildings 12(11), 1850 (2022b)
X. Ren, H. Zhang, R. Hu, Y. Qiu, Location of electric vehicle charging stations: a perspective using
the grey decision-making model. Energy 173, 548–553 (2019)
A. Roy, M. Law, Examining spatial disparities in electric vehicle charging station placements using
machine learning. Sustain. Cities Soc. 83, 103978 (2022)
M. Schmidt, P. Zmuda-Trzebiatowski, M. Kici´
nski, P. Sawicki, K. Lasak, Multiple-criteria-based
electric vehicle charging infrastructure design problem. Energies 14(11), 3214 (2021)
N. Shahraki, H. Cai, M. Turkay, M. Xu, Optimal locations of electric public charging stations using
real world vehicle travel patterns. Transp. Res. Part D Transp. Environ. 41, 165–176 (2015)
M. Shakerian, M.S. Rajabi, M. Tajik, H. Taghaddos, Hybrid simulation-based resource planning
and constructability analysis of RCC pavement projects. arXiv preprint arXiv:2204.05659
(2022)
X. Sun, Z. Chen, Y. Yin, Integrated planning of static and dynamic charging infrastructure for
electric vehicles. Transp. Res. Part D Transp. Environ. 83, 102331 (2020)
S. Torres Franco, I.C. Durán Tovar, M.M. Suárez Pradilla, A. Marulanda Guerra, Electric vehicle
charging stations’ location in urban transportation networks: a heuristic methodology. IET
Electr. Syst. Transp. 11(2), 134–147 (2021)
T. Unterluggauer, J. Rich, P.B. Andersen, S. Hashemi, Electric vehicle charging infrastructure plan-
ning for integrated transportation and power distribution networks: a review. eTransportation 12,
100163 (2022)
G. Wang, M.Z.A. Bhuiyan, S.D.C. di Vimercati, Y. Ren, in Dependability in Sensor, Cloud, and Big
Data Systems and Applications. 5th International Conference, DependSys 2019, Guangzhou,
China, November 12–15, 2019, Proceedings, vol. 1123 (Springer, New York, 2019)
X. Xi, R. Sioshansi, V. Marano, Simulation–optimization model for location of a public electric
vehicle charging infrastructure. Transp. Res. Part D Transp. Environ. 22, 60–69 (2013)
Y. Xiong, J. Gan, B. An, C. Miao, A.L.C. Bazzan, Optimal electric vehicle fast charging station
placement based on game theoretical framework. IEEE Trans. Intell. Transp. Syst. 19(8), 2493–
2504 (2018)
Y. Yang, Y. Zhang, X. Meng, A data-driven approach for optimizing the EV charging stations
network. IEEE Access 8, 118572–118592 (2020)
H. Zhang, S.J. Moura, Z. Hu, W. Qi, Y. Song, Joint PEV charging network and distributed PV
generation planning based on accelerated generalized benders decomposition. IEEE Trans.
Transp. Electrif. 4(3), 789–803 (2018a)
H. Zhang, S.J. Moura, Z. Hu, W. Qi, Y. Song, A second-order cone programming model for
planning PEV fast-charging stations. IEEE Trans. Power Syst. 33(3), 2763–2777 (2018b)
Y. Zhang, Q. Zhang, A. Farnoosh, S. Chen, Y. Li, GIS-based multi-objective particle swarm
optimization of charging stations for electric vehicles. Energy 169, 844–853 (2019)
Y. Zhang, B.K. Teoh, L. Zhang, Integrated Bayesian networks with GIS for electric vehicles
charging site selection. J. Clean. Prod. 344, 131049 (2022)
B. Zhou, G. Chen, Q. Song, Z.Y. Dong, Robust chance-constrained programming approach for
the planning of fast-charging stations in electrified transportation networks. Appl. Energy 262,
114480 (2020)
G. Zhou, Z. Zhu, S. Luo, Location optimization of electric vehicle charging stations: based on cost
model and genetic algorithm. Energy 247, 123437 (2022)
J. Zhu, Y. Li, J. Yang, X. Li, S. Zeng, Y. Chen, Planning of electric vehicle charging station based
on queuing theory. J. Eng. 2017(13), 1867–1871 (2017)
