Journal of Cleaner Production 232 (2019) 531e541

Contents lists available at ScienceDirect

Journal of Cleaner Production

journal homepage: www.elsevier.com/locate/jclepro

Data-driven intelligent location of public charging stations for electric

vehicles
Qi Liu a, Jiahao Liu a, Weiwei Le a, Zhaoxia Guo a, *, Zhenggang He b
a
Business School, Sichuan University, Chengdu, 610065, PR China
b
School of Transportation and Logistics, Southwest Jiaotong University, Chengdu, 610031, PR China

a r t i c l e i n f o a b s t r a c t

Article history: This paper investigates a location problem of public charging stations for electric vehicles with the
Received 17 September 2018 objective of CO2 emissions minimization through massive GPS-enabled trajectory data. The problem
Received in revised form considers two distinct features, including CO2 emissions generated in round trips to charging stations
21 April 2019
and remaining electricity restrictions on charging decisions. A data-driven and particle swarm
Accepted 30 May 2019
Available online 31 May 2019
optimization-based intelligent optimization approach is developed to handle this problem. We then
present how to implement this approach by using taxi trip data in Chengdu, China as case data and
explore how much data could reflect effectively the travel patterns of an area. The results of case study
Keywords:
Electric vehicles
show that one-week taxi trip data are sufficient to handle the investigated problem. The results also
Charging station location validate the necessity of considering two realistic features, including CO2 emissions in round trips to
CO2 emissions charging stations and remaining electricity restrictions on charging decisions, in charging station loca-
GPS-Enabled trajectory data tion problems. It can lead to (1) the reduction of daily CO2 emissions captured by about 0.14e0.37 ha of
forests in one year, and (2) 0.85%e2.64% more charging demands being satisfied per day.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction Previous studies investigate the location problem of public

charging stations for EVs from different perspectives. A brief
Many countries in the world have made tremendous efforts in summary of the representative literature is presented in Table 1.
energy saving and greenhouse gas (GHG) emission reduction in Some researchers consider service or supply capability-related
recent years. As a major consumer of petroleum energy, trans- constraints and objectives (Shao et al., 2016; Alhazmi et al., 2017;
portation sector accounts for about 20%e30% of total GHG emis- Kong et al., 2017). For instance, Frade et al. (2011) used a maximal
sions (IEA, 2015). Low CO2 emissions and sustainable coverage location model to decide the minimum number of EV
transportation initiatives are increasingly important around the charging stations needed within an acceptable level of service for
world (Labib et al., 2018). Developing new energy vehicles (NEVs) an area. Xi et al. (2013) developed a model to maximize the amount
and electric vehicles (EVs) is an inevitable trend (Choma and Ugaya, of energy and the expected number of EVs recharged. Asamer et al.
2017; Garcia et al., 2017; Guo et al., 2017). Chinese government has (2016) proposed a decision support system for the placement of CSs
made national strategy of developing NEVs and EVs with the target aiming to satisfy as much charging demands of battery electric
of at least 5 million NEVs and EVs on the road by 2020 vehicles (BEVs) as possible with a limited number of CSs.
(Administration, 2012), which leads to a natural demand to Some researchers considered cost-related constraints and ob-
construct public EV charging stations (CSs). It brings out an jectives in EV charging station location problems (Escudero-Garzas
important decision-making problem, the location problem of and Seco-Granados, 2012; Zhang et al. 2015, 2016; Ma and Zhang,
public charging stations for EVs, which is concerned with where 2018). He et al. (2013) put forward a modeling framework to decide
CSs should be built in a certain area. an optimal distribution of a given number of public charging sta-
tions aiming to maximize the total social welfare. He et al. (2015)
investigated the BEV drivers’ spontaneous adjustments and the
* Corresponding author. interactions of travel and recharging decisions, which aims to
E-mail addresses: [email protected] (Q. Liu), [email protected]. minimize the total driving and recharging time cost. A multi-period
cn (J. Liu), [email protected] (W. Le), [email protected] optimization model for the deployment of public EV charging
(Z. Guo), [email protected] (Z. He).

https://doi.org/10.1016/j.jclepro.2019.05.388
0959-6526/© 2019 Elsevier Ltd. All rights reserved.
532 Q. Liu et al. / Journal of Cleaner Production 232 (2019) 531e541

Table 1
A summary of the representative literature.

Objectives Data used Problem characteristics Representative references

Service-related: maximizing charging Population data; reality vehicles data 1 Neglecting environmental factors; Frade et al. (2011), Xi et al. (2013), Tu
demands, service levels, etc. about several weeks. 2 Neglecting the effects of using et al. (2016), Alhazmi et al. (2017) etc.
different data volumes on results.
Cost-related: minimizing time, Benchmark networks; simulated travel 1 Simulating travel patterns indirectly; He et al. (2013), He et al. (2015), Li et al.
construction cost, cost loss etc. data. 2 Neglecting environmental factors. (2016), Tian et al. (2018) etc.
Environmental-related: maximizing the Simulated data; reality vehicles data 1 Neglecting the effects of considering Liang et al. (2012), Sathaye and Kelley
recharged electricity; minimizing about several weeks or several million energy consumptions of both trips (2013), Dong et al. (2014), Shahraki
energy losses, emissions, etc. trips and round trips to CSs and remaining et al. (2015), Andrenacci et al. (2016)
electricity restrictions; etc.
2 Neglecting the effects of using
different data volumes.
2.2 billion raw data records from over 1 Considering various energy This paper
10,000 taxis about 1 month. consumptions;
2 Exploring the effects of using
different data volumes;
3 Simulating travel patterns by real
data directly.

stations was developed by Li et al. (2016) to minimize the total cost demands based on 255 million raw data records from Beijing and
of new stations’ installations and existing stations’ relocations. Tian proposed an optimization model to maximize the amount of
et al. (2018) proposed an optimization model of CSs based on vehicle-miles traveled being electrified. Asamer et al. (2016) used
waiting time to minimize the time cost to EV drivers. Xie et al. the operational taxi data from about 800 vehicles to identify and
(2018) minimized the total systems cost for deploying CSs when estimate the charging demands for electric taxis based on frequent
BEVs driving across the entire network, which included fixed and origins and destinations of trips. Tu et al. (2016) employed 7-day
variable cost of CSs and penalty cost of where trips cannot be taxi trip data from Shenzhen to extract taxi demands, with the
satisfied. objective of maximizing the level of electric taxis service in the road
With the increasing environmental awareness and sustainable network and the level of CSs service. Andrenacci et al. (2016)
development trend, it is critical to consider environmental-related analyzed 57,890 trips extracted from one-week vehicular travel
constraints and objectives in EV charging station location problems. data in the urban area of Rome to simulate driving and charging
Liang et al. (2012) proposed a model to minimize vehicles’ energy behaviors. Yang et al. (2017) investigated the siting and sizing of CSs
losses on the way to the charging stations through the partition of by using a typical queue model and dwell patterns extracted from
the network. Sathaye and Kelley (2013) utilized continuous facility one-week taxis GPS data in Changsha, China. However, previous
location models to obtain the minimum total distances or energy studies have not discussed the effects of using different data vol-
losses on the way to the CSs. Dong et al. (2014) explored the effects umes on location solutions. It is still open how much data should be
of public charging facilities deployment on electric miles traveled utilized to describe effectively the travel patterns of a region. It is
and aimed to reduce users’ range anxiety by increasing electric worthwhile to discuss this issue due to the computational
miles. Shahraki et al. (2015) proposed an optimization model for complexity of the huge amount of data.
plug-in hybrid electric vehicles (PHEVs) based on energy con- Thus, this paper aims to investigate a location problem of public
sumptions of travel trips, which aims at maximizing the amount of charging stations for electric vehicles based on GPS-enabled taxi
vehicle-miles-traveled being electrified. Andrenacci et al. (2016) data with the objective of minimizing the total CO2 emissions. The
minimized the sum of the distances between all the destination most notable problem features that differentiate this paper from
points and charging facilities. Vazifeh et al. (2019) minimized the previous studies include (1) the consideration of the overall emis-
number of CSs and the energy consumed by drivers from the end of sion minimization objective, (2) the identification on the effects of
trips to closest available CSs. However, energy consumptions of different amount of raw data on the results, and (3) the observation
both travel trips and round trips to CSs has not been considered in on the impacts of round trips to CSs and remaining electricity re-
previous studies, which indicates that existing research cannot lead strictions on charging decisions in terms of the daily CO2 emission
to the minimization of the total GHG emissions of EVs in a region. It reduction and charging demand satisfaction rate.
is thus worthwhile to minimize the total GHG emissions in The paper is organized as follows. Section 2 presents the
handling the location problem of charging stations. Besides, pre- mathematical model of the investigated location problem of public
vious studies assume that each vehicle needs to recharge once it is CSs for EVs. Section 3 introduces the data-driven intelligent loca-
parked without the consideration of remaining electricity re- tion approach proposed. The data processing and results of the case
strictions on charging decisions. It is illogical in the real world study are described in Section 4. Section 5 presents the analysis of
because recharging is not required when the remaining electricity the corresponding results. Finally, we summarize this paper and
is sufficient. suggest the future research directions in Section 6.
Nowadays, GPS-enabled trajectory data played an important
role in investigating the location problem of charging stations since 2. Problem statement
these data are able to depict effectively the daily travel patterns of
vehicles in a region. Dong et al. (2014) used 18-month travel data 2.1. Problem description
from 445 private gasoline vehicles in Seattle metropolitan area to
simulate regional travel and charging behavior of BEVs, minimizing Plug-in hybrid electric vehicle (PHEV) is one type of electric
the number of trips which cannot be completed by electricity. Cai vehicles between traditional vehicles and BEVs. PHEVs allow
et al. (2014) used the mobility data of 11,000 taxis over a three- drivers finishing trips by gasoline when batteries are depleted. By
week period in Beijing to estimate a public charging demand (CD) so doing, the mileage can be increased without recharging, and
of PHEVs. Similarly, Shahraki et al. (2015) captured public charging drivers’ range anxiety about the fear that an EV has insufficient
 Q. Liu et al. / Journal of Cleaner Production 232 (2019) 531e541 533

electricity to reach its destination decreases effectively. The energy

n o
consumption of PHEV includes electricity power consumption and b ¼ max P ; 0 ;
P jk jk cj; k (2)
gasoline consumption. Electricity can be generated by renewable
clean energy such as wind and solar energy (Fathabadi, 2018), or by
thermal power traditionally. If it is generated by clean energy, EVs ( )
X
I
b þZ
Ejk ¼ min Ej  P ijk1 dijk1 Wa ; M Zijk ; ci; j; k
produce no pollution in driving. However, CO2 emissions are not jk (3)
negligible in view of the fact that the widespread use of thermal i¼1
power in most areas of China (Fang et al., 2018). This paper uses
Let Cijk (unit: kg) denote the CO2 emission generated by vehicle j
PHEVs as an example to the placement of CSs and considers CO2
on round trips to CS i if it goes to CS i after trip k, which is calculated
emissions generated indirectly by the consumption of electricity.
by formula (4). Let l (unit: kg/kWh) denote the CO2 emission rate of
We consider the public charging station location problem in a
using electricity, and g’ (unit: kg/km) the CO2 emission rate of using
city. There are I candidate stations (indexed by i) available, which
gasoline. If dijk is no longer than the distance afforded by the real
are dispersedly distributed in the city. The investigated problem b , the CO2 emission in the round-trip is
remaining electricity P jk
needs to select out s stations from these candidate stations so that
generated by electricity. Otherwise, it is generated by both gasoline
the total CO2 emissions, generated by PHEVs in the city, can be
and electricity.
minimized. s is a given constant. There are J PHEVs (indexed by j) in

8
>
<      
Cijk ¼ b
2ldijk Wa ; if dijk  P Wa l Pb þ d Wa þ g’ d  Pb Wa ; otherwise; ci; j; k (4)
>
:
jk jk ijk ijk jk

total travelling in the city. Each vehicle produces Kj (indexed by k)

trips in an examined period based on GPS-based trajectory data. We This research makes the following reasonable assumptions: (1)
set a current trip’s end position to be the start position of its next PHEVs prefer to use electricity power and use gasoline only when
trip. When a trip is completed, the driver needs to decide whether the electricity is exhausted and have enough gasoline to support
the vehicle needs to be recharged. If relevant charging conditions the vehicle to finish the trip and find an available CS; (2) each
for a vehicle are not satisfied, the vehicle starts the next trip vehicle is fully charged before starting its first trip each day; (3) it is
directly; otherwise it goes to recharge to full capacity. The energy fully charged using fast charging mode once the vehicle goes to
consumption during the whole driving process and recharging charge,1 which is consistent with the setting in (Tu et al., 2016; Xie
process are recorded and the total CO2 emissions thus can be ob- et al., 2018; Vazifeh et al., 2019) and (4) queue waiting in recharging
tained when all trips are completed. is not considered, which is consistent with the setting in (Cai et al.,
Let Pjk (unit: kWh) denote the theoretical remaining electricity 2014; Shahraki et al., 2015; He et al., 2019).
of vehicle j at the end of trip k before it goes to charge under the
assumption that the remaining electricity can be less than 0. Pjk is
less than zero, if the electricity of battery is insufficient for
b (P b
0) denote the real remaining 2.2. Mathematical model
completing trip k. Let P jk jk
electricity of vehicle j at the end of trip k. By defining Pjk and P b , we
jk
The mathematical model of the investigated problem is
can identify easily how many trip mileages are driven actually by
formulated as follows.
gasoline or electricity respectively, which is critical to calculate the
emissions generated by gasoline or electricity and the electricity
X K 
J X  X K 
J X
recharged at CSs. Pjk and P b are formulated in equations (1) and (2). b b lþg b
jk minFðyi Þ ¼ P jk1 þ Ejk1  P jk P jk
Ejk (unit: kWh) is the electricity recharged for vehicle j when j¼1 k¼2 j¼1 k¼1
completing trip k, which is formulated in formula (3). If vehicle j can 
go to charge at CS i at the end of trip k, Zijk is 1; otherwise, it is 0. djk X
I X
J X
K
 Pjk Wb =Wa þ Zijk Cijk (5)
(unit: km) is the actual travel distance of trip k of vehicle j, which
i¼1 j¼1 k¼1
can be calculated based on the interval time between two adjacent
trajectory points in the trip and their corresponding travel speeds. Subject to
dijk1 (unit: km) is the distance between CS i and trip k  1’s end
position of vehicle j. For simplicity, we set dijk1 as straight-line X
I
distance since the CS service radius is usually small (e.g., 2 km) yi ¼ s (6)
i¼1
and the difference between the straight-line distance and actual
travel distance is thus negligible in this case. Wa (unit: kWh/km) is
PHEVs’ the electricity consumption rate. Ej (unit: kWh) is battery
capacity of vehicle j, and M is a sufficiently large number greater
than Ej . 1
This assumption is reasonable mainly because (1) it is not time- and cost-
effective if a vehicle waits for a relatively long time but only gets partially
recharged in one recharging process, especially when the time required for a full
charge is not long (only 15 min or less in the near future), and (2) we have randomly
b
Pjk ¼ P jk1  2Zijk1 dijk1 Wa þ Ejk1  djk Wa ; ci; j; k (1) consulted some electric taxis drivers in Shenzhen and Chengdu, China, all of them
confirmed that they usually got vehicles fully charged after they queued in the
charging stations.
534 Q. Liu et al. / Journal of Cleaner Production 232 (2019) 531e541

This proposed location approach consists of two modules, a data

b
Pjk ¼ P collection and processing (DCP) module and a particle swarm
jk1  2Zijk1 dijk1 Wa þ Ejk1  djk Wa ; ci; j; k (7)
optimization-based location (PSOL) module. The DCP module is
n o applied to collect travel trips data and generate CS candidates.
b ¼ max P ; 0 ;
P cj; k (8) Based on the data obtained in the first module, the PSOL module
jk jk
generates the best CS location solution based on particle swarm
( ) optimization (PSO) algorithm.
X
I In the DCP module, we need to extract travel trips data from raw
b þZ
Ejk ¼ min Ej  P jk ijk1 dijk1 Wa ; M Zijk ; ci; j; k (9) GPS-based trajectory data. In taxis operations, GPS recorders are
i¼1
used to record taxi trajectory data (e.g., time, speed, positions in
longitude and latitude) according to a specific frequency (e.g., once

8
>
<      
Cijk ¼ b
2ldijk Wa ; if dijk  P Wa l Pb þ d Wa þ g’ d  Pb Wa ; otherwise; ci; j; k (10)
>
:
jk jk ijk ijk jk

per 10 s). Based on trajectory data and the interval time between
records, each vehicle’s daily continuous driving activities and
parking events at various locations can be obtained. Travel trips of
Zijk ¼ yi Dijk rjk ; ci; j; k (11) each taxi are generated by connecting all trajectory points between
any two continuous parking events with sufficient parking time.

1; if dijk  r The parking events with insufficient parking time are ignored.
Dijk ¼ ; ci; j; k (12) Trajectory data are cleaned by deleting the irrational trips and GPS-
0; otherwise
trajectory data-based travel trips are obtained finally. Then, CS
( candidates are selected based on the number of parking events
b  mE
1; if P occurring around each gasoline station. The gasoline station,
rjk ¼ jk j; cj; k (13)
0; otherwise around which there are the more parking demands, has the higher
priority of being CS candidates. Virtual charging demands can be
X
I
Zijk  1; cj; k (14)
i¼1

yi ; Dijk ; Zijk ; rjk 2f0; 1g; ci; j; k (15)

The objective, as shown in formula (5), is to minimize the total

CO2 emissions generated by gasoline directly and by electricity
indirectly of all PHEVs. We define that yi is 1 if candidate location i
is used to build a charging station; otherwise it is 0. Let g (unit: kg/
L) denote the CO2 emission rate of gasoline, Wb (unit: L/km) the
gasoline consumption rate consumed by PHEVs. Constraint (6) in-
dicates that the number of CSs to be built should be equal to s.
Constraints (7)e(10) have been introduced in equations (1)e(4).
Constraint (11) indicates that vehicle j can be recharged at CS i at
the end of trip k if and only if constraints (12) and (13) are satisfied
simultaneously. Constraint (12) shows that Dijk is 1 if dijk is less than
the service radius r of the CS; otherwise Dijk is 0. Constraint (13)
indicates the remaining electricity restrictions on charging de-
cisions. We set rjk ¼ 1 if the remaining electricity is below a certain
threshold (0  m  1); otherwise rjk ¼ 0. Constraint (14) ensures
that vehicle j can only be recharged at one CS after trip k. Constraint
(15) defines yi , Dijk , Zijk and rjk as binary variables.

3. Solution approach

This research proposes a data-driven intelligent location

approach, which is a particle swarm optimization-based intelligent
approach based on GPS-enabled trajectory data to solve this
problem since the latter is able to provide effective solutions to
complex discrete optimization problems. The flowchart of this
approach is shown in Fig. 1. Fig. 1. Flowchart of the proposed location approach.
 Q. Liu et al. / Journal of Cleaner Production 232 (2019) 531e541 535

derived from the interaction between travel trips and the place- Table 2
ment of CSs. Required fields of taxi trajectory data in Chengdu city, China.

Given the set of CS candidates, we use the PSOL module to Data Fields Description
obtain the optimal CS location solution. Related PSO procedures are Vehicle ID License plate number.
described as follows. Parameters used in the PSOL module are Position Latitude and longitude.
initialized in step 1, including the population size N, the number C Speed Speed of current record.
of CS candidates, the number s of CSs to be built and the maximum Record time Date and time of current record.

number gmax of iterations. The solution individuals of initial pop-

ulation are initialized in step 2. Every individual is encoded by a
position string ½y1 ; y2 ; :::; yI  and a velocity string ½v1 ; v2 ; :::; vI . The
data records are shown in Table 2.
position string represents a location solution of charging stations.
These raw data are processed to obtain the trips data required
The velocity string is used to reflect the individual’s moving situ-
for this problem. We prescribe that the driver considers whether
ation and applied in the optimization process. The length of them is
to recharge only if the parking time (speed equals to 0) of the
equal to C. Each element in a string corresponds to a candidate
vehicle reaches 30 min, which usually means the driver needs to
station. If its element value in position string is 1, the corresponding
have a rest and there are no passengers on the vehicle, thus this is
candidate station is selected as a CS. The speed value in velocity
a good timing for recharging. The parking with less than 30 min is
string is limited in [-10, 10]. The position and velocity of each in-
ignored. If the parking time between any two consecutive trips is
dividual in the first population are initialized randomly. Then, in
less than 30 min, these consecutive travel trips of a vehicle are
step 3, the objective function of each solution individual is evalu-
consolidated into a large trip. Then, the irrational trips that over
ated. The lower the value of the objective function, the better the
350 km and less than 1 km are removed. Finally, over 700,000
individual’s performance. Next in step 4, the best position Pbestn of
valid trips are retained. Each trip contains information as shown
individual n in the previous optimization process and the best
in Table 3.
position Gbest of all individuals are obtained based on the objective
function evaluation in step 3.
Afterwards, in step 5, new individuals are generated by updating 4.1.2. Candidate charging stations
their speeds and positions, and these new individuals constitute a Some gasoline stations are selected out and set as CS candidates
new population. The speed of each binary bit of solution individual in case study. There are more than 300 gas stations in Chengdu city
is updated according to formula (16), where w is the weight of currently. To simplify the problem, this research selects some of
inertia and equal to 0.8, c1 and c2 are learning factors and equal to 2, existing gasoline stations as CS candidates according to parking
r1 and r2 are random numbers. We set r1 ¼ 0:6 and r2 ¼ 0:3. Then events of taxis. The reasons why we select gasoline stations as CS
we update the position based on the updated speed above ac- candidates are straightforward. First, with the fast development of
cording to formulae (17) and (18). Let rand represent the random EVs, the demands for gasoline stations will decrease inevitably and
number between [0, 1]. some gasoline stations can be reconstructed into CSs. Second, we
set an appropriate service radius for each CS candidate so that most
vi ¼ wvi þ c1 r1 ðPbestn  yi Þ þ c2 r2 ðGbest  yi Þ; ci (16) of parking points are within the service range of CS candidates
selected. Service radius r of CSs and the number s of CSs to be built
1 should be determined before selecting candidates. In accordance
sðvi Þ ¼ ; ci (17) with the construction plan of EV charging infrastructures issued by
1 þ expð  vi Þ
the local government of Chengdu city, the public charging service
 radius should be less than 1 km in the downtown area of the city
1; if rand  sðvi Þ
yi ¼ ; ci (18) and less than 1.5 km in the non-core area, while in the suburb area
0; otherwise service radius should be less than 2 km. Due to our data covers both
Finally, the termination condition is checked in step 6. If the the downtown and suburb areas of Chengdu, we set r ¼ 2 km. Ac-
maximum number gmax of iterations is reached, the optimization cording to the policy of EV Charging Infrastructure Development
process is terminated and the best objective value and the best Guide issued by National Development and Reform Commission
solution are returned. If the termination condition is not satisfied, and National Energy Administration, the number of various types of
the process returns to step 3 for a new iteration. EVs will be up to 96,000 and every 2000 EVs need to be equipped
with one public CS at least (Commission, 2016). Consequently, s is
4. Case study and results set as 40 in experiments.
Afterwards, we identify all parking events within each gasoline
4.1. Data collection and processing station’s service radius based on the taxi trajectory data and sort all
gasoline stations by the number of parking events in each station’s
This section describes how relevant data in DCP module are service range. We find that more than 99% trips end within the
collected and processed, including the generation of taxi trips from service ranges of at least one CS candidate. Thus, these 100 gasoline
raw taxi trajectory data and the selection of CS candidates. stations are selected as CS candidates. The distributions of existing
gasoline stations and CS candidates are shown in Fig. 2 and Fig. 3.
4.1.1. Taxi trip data
The GPS-based trajectory data of all taxis from 30 days (1st- 30th
Table 3
Jun 2015) in Chengdu are collected from the taxi management The data fields of trip records.
authority of Chengdu Municipal Transportation Commission to
simulate travel trips of PHEVs. The data contain a total of 2.2 billion Data Fields Description

original data records by recording once every 5 or 10 s, which Vehicle ID j License plate number.
reflect the real-time operation statuses of more than 10,000 taxis Trip Number k Unique number of a trip.
Parking Position Parking location in latitude and longitude.
per day. They are collected by a GPS-enabled system equipped in
Travel Distance djk Length of the current trip, Unit: km.
each taxi. The required fields in this research from the trajectory
536 Q. Liu et al. / Journal of Cleaner Production 232 (2019) 531e541

Due to the page limit, this section presents the results of 4 typical
cases:
Case 1. it uses one-day data (30th June),
Case 2. it uses four-day data (20th 23rd June),
Case 3. it uses one-week (1st-7th June) data,
Case 4. it uses two-week (1st 14th June) data.
The optimal location solutions to cases 1e4, generated by the
proposed approach, are demonstrated in Fig. 4. The solutions to
cases 3 and 4 are the same, which are different with the solutions to
cases 1 and 2. The solutions to cases 1 and 2 are different as well.

5. Comparison and discussions

This section discusses the differences among the results ob-

tained from different data volumes to explore how much data can
effectively capture the travel patterns of a region.
Fig. 2. Distribution of all gasoline stations in Chengdu.
5.1. Comparison of performances resulted by different data volumes

We compare the performance differences of the best location

solutions to problem cases with different-period trip data, found in
cases 1e4, in terms of two performance indicators. The two per-
formance indicators include daily CO2 emissions reduction (DCER,
unit: tons) and charging demand satisfaction rate (CDSR). DCER is
defined as the difference between the total daily CO2 emissions
generated by traditional vehicles and PHEVs, which is calculated by
equation (19). CDSR is the proportion of the trips that the charging
demand is satisfied in all of trips with charging demand and
calculated by formula (20).
!
P
J P
Kj P
I P
J P
Kj
djk þ 2Zijk dijk g  Fðyi Þ
j¼1 k¼1 i¼1 j¼1 k¼1
DCERðyi Þ ¼ (19)
the number of days

P
I P
J P
Kj
Zijk
Fig. 3. Distribution of candidate stations. i¼1 j¼1 k¼1
CDSRðyi Þ ¼ (20)
P
J P
Kj
rjk
j¼1 k¼1
4.2. Results
This research uses the two performance indicators because (1)
This section presents the results of handling various CS location one of main goals of developing EVs is to reduce the environmental
problem cases based on different taxi trips data collecting from pollution; (2) the popularity of EVs depends largely on the conve-
different periods respectively. Parameters in experiments are set as nience of charging and the layout reasonability of public charging
follows: (1) consistent with the setting in the related literature and infrastructures is reflected by the ability to effectively satisfy
EVs’ general characteristics, we set Wa ¼ 0:22kWh=km (Shahraki charging demands. Each CS location solution presented in section
et al., 2015), l ¼ 0:47kg=kWh (Cai et al., 2007), and g’ ¼ 4.2 is used to calculate the values of two performance indicators on
0:22kg=km (Huo et al., 2010); (2) according to the information the basis of other different trips data volumes. The corresponding
provided by the Alternative Fuels Data Center of U.S. Department of comparison results are shown in Table 4.
Energy, we set Wb ¼ 0:17L=km, Ej ¼ 35kWh, b ¼ 14kWh=h, g ¼ From Table 4, it can be found that the DCER ranges from 26.18 to
2:24kg=L, and PHEVs’ gasoline consumption equals 0.1L/km; (3) we 28.81 tons. It is reported that a hectare of trees captures 1e10 tons
set m ¼ 30%, r ¼ 2km, s ¼ 40, C ¼ 100 based on survey and real- of CO2 per year depending on its age, climate zone, type of forest
istic requirements (described in section 4.1.2); and (4) we set M ¼ and soil (Sicirec, 2017), and we assume that 1 ha of forests absorbs 5
100000 since M is a number sufficiently larger than Ej . The setting tons CO2 per year on average. Thus, the DCER of 26.18 tons is
of algorithm parameters could have large effects on the location equivalent to the CO2 absorbed by about 5.24 ha of forests in one
results. We thus conduct experiments, based on 10 different sets of year, which shows the impact of adopting PHEVs on environment is
algorithm parameters (i.e., N, gmax ), to choose the appropriate pa- enormous.
rameters. It is found that the setting of N ¼ 30, gmax ¼ 150 results In terms of DCER, it can be seen that for one-day and four-day
in the best location performance while the optimization process data, the result based on one-week data is not the best. But when
gets converged. This setting is thus used in experiments. data volume increases, one-week data-based solution becomes the
The experiments are carried out on a laptop with Intel Core i5- best one. We conducted a series of experiments based on different-
7200U CPU @2.5 GHz and 8 GB RAM using Python version 3.6.2. period trips data, and the experimental results show that the
(a) Location solution to case 1 (b) Location solution to case 2

(c) Location solution to case 3 (d) Location solution to case 4

(e) Location solutions to cases 1-4

Fig. 4. Best location solutions for different cases.
538 Q. Liu et al. / Journal of Cleaner Production 232 (2019) 531e541

Table 4
Performance comparison of different period lengths-based solutions.

Data Type Performance Indicators Performances of 1- or 2-week Performances of 1-day Performances of 4-day
data-based solution data-based solution data-based solution

One Week DCER 28.13 27.49 27.94

CDSR 81.67% 81.09% 80.71%
Two Weeks DCER 27.31 26.74 26.18
CDSR 81.41% 80.89% 80.91%
One Day DCER 28.65 28.81 28.40
CDSR 81.66% 81.30% 81.26%
Four Days DCER 28.52 28.35 28.61
CDSR 80.27% 80.16% 80.44%

DCER: daily CO2 emissions reduction CDSR: charging demand satisfaction rate.

results based on one-week data are better than those based on less resulting performance values of these solutions on one-week data.
than one-week data. From Table 4, the DCER based on one-week The average values of the 9 sets of results are 27.49 and 81.09% for
data is at least 0.2 tons higher than others, which is equivalent to the two indicators respectively. It can be found from Table 5 that,
the CO2 absorbed by about 0.04 ha of forests in one year. In terms of same to the conclusions from Table 4, the performance of the so-
CDSR, the results based on one-week data are 0.50%e0.96% higher lutions based on one-week data are better than those based on less
than those based on less than one-week data. data.
It seems that these results are very close in terms of the two We further use p-values to measure the performance results
indicators. But it is notable that these results are obtained based on generated based on different time period lengths, by the paired-
only about 10,000 taxis. Due to the large number of vehicles in the samples t-test. Before the t-test, we have checked and validated
case city (more than 4.5 million in 2017 (Williams, 2018)), these firstly that the performance values of the solutions meet the normal
results actually lead to a much larger emissions reduction and distribution based on a normality test. This research uses A ≡(>) B
charging demand satisfaction. If we consider 1 million vehicles on to represent that A generates the same (the better) performances to
average per day running in the city, the CER would be 20 tons per (than) B. By setting A and B as different time periods, we define the
day, equivalent to the CO2 absorbed by about 4 ha of forests in one following two null hypotheses (HA0 and HB0) and their corre-
year. Moreover, the increase of 0.50%e0.96% CDSR can lead to sponding alternative hypotheses.
additional 5000e9600 charging demands are satisfied in the city
every day. HA0: one-week-data ≡ four-day-data; HA1: one-week-data > four-
Next, we further investigate if one-week trip data are sufficient day-data.
to handle the CS location problem effectively based on a paired-
samples t-test. The t-test is performed to explore whether there HB0: one-week-data ≡ four-day-data; HB1: one-week-data > one-
exist significant differences among the solutions obtained based on day-data.
different data volumes. We consider 4 types of trip data with
different time lengths, i.e., one-week, two-week, four-day and one- To validate if the results generated by the one-week data are
day. For each type, we select 3 sets of trip data randomly from significantly superior to the results generated by other data, we
different time periods. For example, three sets of one-day trip data calculate the p-value of each hypothesis by using the average values
are from 5th, 12th, and 18th June respectively. Based on each set of of two performance indicators based on 3 different datasets. The
trip data, we then obtain the resulting CS location solution and two two p-values are 0.002 and 0.006 respectively if the average DCER
values of performance indicators. We then get a performance is used as the performance indicator, while the two p-values are
comparison results shown in Table 5, in terms of two performance 0.001 and 0.000 respectively if the average CDSR is used. All these
indicators. Table 5 is similar to Table 4, but it presents the average p-values are less than 0.05. We thus reject these null hypotheses at
performance of three sets of trip data for each period length. The the significance level of 5%. Take null hypotheses HA0 as an
results generated by one-week data are almost the same to the example. There is a 95% chance of accepting the alternative hy-
results generated by two-week data and we thus put their results pothesis HA1. That is, the one-week data generate the better per-
together in Table 5. Take the solution based on one-day trip data as formance than the other two types of data in terms of both
an example. On the basis of 3 sets of one-day trip data, we obtain 3 performance indicators.
CS location solutions respectively. We then obtain the 9 sets of To further validate if the same conclusions can be made under

Table 5
Average performance comparison of solutions generated by 3 sets of data for each time period length.

Data Type Performance Indicators Performance of 1- or 2-week Performance of 1-day Performance of 4-day
data-based solutions data-based solution data-based solution

One Week ADCER 27.96 27.49 27.32

ACDSR 81.47% 81.09% 80.76%
Two Weeks ADCER 27.51 26.60 26.73
ACDSR 81.34% 80.91% 80.91%
One Day ADCER 28.73 28.88 27.94
ACDSR 81.61% 81.44% 81.29%
Four Days ADCER 28.55 28.32 28.62
ACDSR 81.31% 80.72% 81.19%

ADCER: average DCER generated by 3 sets of data with different dates under one type of data volumes (ton).
ACDSR: average CDSR generated by 3 sets of data with different dates under one type of data volumes.
 Q. Liu et al. / Journal of Cleaner Production 232 (2019) 531e541 539

different settings of s, a series of experiments are conducted in electricity restrictions on charging decisions, which have not been
terms of different s and different-period trips data. Table 6 shows considered in the CS location literature. These two features may
the corresponding experimental results. affect final location solutions and drivers’ charging decision-
It can be found from this table that, for any given s, the results making. To validate the necessity of these two features, we
generated by one-week data are the same to those generated by compare the performance differences of solutions to the investi-
two-week data, but better than those generated by one-day data, gated problem and 3 corresponding problems without one or two
which is consistent with the results obtained in Table 4 and of these features in terms of the DCER and the CDSR. These 3
meaning that one-week taxi trip data are sufficient to handle this problems are the same with the investigated problem except for
CS location problem. The values of both indicators increase with s problem features. Problem 1 only considers the emissions pro-
on the whole, because more charging demands are satisfied and duced in round trips to CSs. Problem 2 only considers the remaining
more distances are traveled by electricity when s increases. When s electricity restriction on charging decisions. Neither feature has
increases to 40, the performance improvement of location solutions been considered in problem 3. The corresponding comparison re-
is the most obvious. The improvement is relatively insignificant sults are shown in Table 7.
when s continues to increase. It shows that 40 CSs can well satisfy As shown in Table 7, the investigated problem with the
the charging demands from the about 10,000 taxis in the case data. consideration of two realistic features produces the best perfor-
In summary, the location solution based on one-week or more mance while problem 3 with neither feature generates the worst
data performs more stable than the solution obtained by less than performance. The DCER generated by the investigated problem is
one-week data. The reason maybe is that the travel trips from less about 0.69e1.86 tons greater than those generated by other prob-
than one-week data cannot fully reflect the vehicles’ travel patterns lems and 0.69e1.86 tons of CO2 emissions are equivalent to the CO2
and charging demands in the city. On the other hand, one-week trip absorbed by 0.14e0.37 ha of forests within one year (Sicirec, 2017).
data can reflect the travel patterns of the city well since one-week The CDSR generated by the investigated problem is approximately
data can lead to the same location solution based on two-week or 0.85%e2.64% higher than its counterparts, which is significant in
more data. Therefore, it is not suggested to use less than one-week reality because only about 10,000 taxis’ data are collected in the
data to handle the investigated problem. It is reasonable to
conclude that one-week trip data are sufficient to handle the
investigated problem in the case city. This conclusion also applies Table 7
Performance comparison of solutions to problems with/without realistic features
when data obtained in other regions are similar to the case data.
considered.

Problems Investigated problem Problem 1 Problem 2 Problem 3

5.2. Necessity of the realistic features considered
DCER(t) 28.13 26.79 27.44 26.27
This research investigates two important realistic features, CDSR 81.67% 80% 80.81% 79.03%

including CO2 emissions in round trips to CSs and remaining DCER: daily CO2 emissions reduction CDSR: charging demand satisfaction rate.

Table 6
Performance comparison of solutions based on different data volume and s.

s Data Type Performance Indicators Performance of 1- or 2-week data-based solution Performance of 1-day data-based solution

20 One-week data DCER 12.45 6.15

CDSR 63.44% 57.81%
One-day data DCER 8.74 9.06
CDSR 62.66% 63.38%
Four-day data DCER 9.67 6.10
CDSR 62.87% 56.09%

30 One-week data DCER 21.92 17.48

CDSR 74.65% 73.47%
One-day data DCER 15.14 16.48
CDSR 73.10% 73.45%
Four-day data DCER 18.44 17.93%
CDSR 73.86% 73.54%

40 One-week data DCER 28.13 27.49

CDSR 81.67% 81.09%
One-day data DCER 28.65 28.81
CDSR 81.66% 81.30%
Four-day data DCER 28.52 28.35
CDSR 80.27% 80.16%

50 One-week data DCER 29.03 27.51

CDSR 83.83% 82.01%
One-day data DCER 28.72 28.95
CDSR 82.53% 83.56%
Four-day data DCER 28.81 27.89
CDSR 83.40% 81.66%

60 One-week data DCER 30.47 29.16

CDSR 84.72% 84.02%
One-day data DCER 28.90 29.02
CDSR 83.51% 83.56%
Four-day data DCER 29.44 28.43
CDSR 84.59% 83.10%

DCER: daily CO2 emissions reduction CDSR: charging demand satisfaction rate.
540 Q. Liu et al. / Journal of Cleaner Production 232 (2019) 531e541

case study but there are more than 4.5 million vehicles in the case number 2018hhs-37, SKSYL201819, skqx201725].
city in 2017 (Williams, 2018). That is, the DCER will be much greater
and more charging demands will be satisfied per day. These results Acknowledgements
indicate that both features are critical to be considered in the CS
location problem. In addition, the results generated by problems 1, The authors would like to thank the anonymous reviewers and
2 and 3 show that feature 2 has a greater effect than feature 1 in the Editor for their helpful comments.
terms of either performance indicator.
References
6. Conclusions
Administration, N.E., 2012. Energy saving and new energy automotive industry
This paper investigated a location problem for charging stations development plan (2012-2020). http://www.nea.gov.cn/2012-07/10/c_
131705726.htm.
with the objective of minimizing the total CO2 emissions of the
Alhazmi, Y.A., Mostafa, H.A., Salama, M.M.A., 2017. Optimal allocation for electric
whole traffic network in an area, considering two important real- vehicle charging stations using Trip Success Ratio. Int. J. Electr. Power Energy
istic features that have been ignored by previous studies. A data- Syst. 91, 101e116. https://doi.org/10.1016/j.ijepes.2017.03.009.
driven intelligent approach was proposed to handle the location Andrenacci, N., Ragona, R., Valenti, G., 2016. A demand-side approach to the optimal
deployment of electric vehicle charging stations in metropolitan areas. Appl.
problem investigated. Driving trips, charging demands and parking Energy 182, 39e46. https://doi.org/10.1016/j.apenergy.2016.07.137.
events were imitated by utilizing taxis trajectory data. This paper Asamer, J., Reinthaler, M., Ruthmair, M., Straub, M., Puchinger, J., 2016. Optimizing
investigated the impacts of different-period trips data on charging charging station locations for urban taxi providers. Transport. Res. Pol. Pract. 85,
233e246. https://doi.org/10.1016/j.tra.2016.01.014.
station location solutions and the necessity of considering two Cai, H., Jia, X., Chiu, A.S.F., Hu, X., Xu, M., 2014. Siting public electric vehicle charging
realistic features including CO2 emissions in round trips to CSs and stations in Beijing using big-data informed travel patterns of the taxi fleet.
remaining electricity restrictions on charging decisions. We use Transport. Res. Transport Environ. 33, 39e46. https://doi.org/10.1016/
j.trd.2014.09.003.
Chengdu, China as a case city and use the taxi trip data of Chengdu Cai, W., Wang, C., Wang, K., Zhang, Y., Chen, J., 2007. Scenario analysis on CO2
to describe the travel patterns and charging demands in the city. emissions reduction potential in China’s electricity sector. Energy Policy 35 (12),
The DCER produced by PHEVs based on different-period trip data 6445e6456. https://doi.org/10.1016/j.enpol.2007.08.026.
Choma, E.F., Ugaya, C.M.L., 2017. Environmental impact assessment of increasing
ranges from 26.18 to 28.81 tons, which are equivalent to the CO2 electric vehicles in the Brazilian fleet. J. Clean. Prod. 152 (Suppl. C), 497e507.
absorbed by at least 5.24 ha of forest in one year. By comparing the https://doi.org/10.1016/j.jclepro.2015.07.091.
location solutions resulted by different taxi data volumes, it was Commission, C. E. a. I., 2016. Special plan for the construction of electric vehicles in
Chengdu. Retrieved 10 December, 2017, from. http://www.docin.com/p-
found that one-week trip data could reflect the travel patterns of
1852056593.html.
the case city well, which could greatly decrease the computational Dong, J., Liu, C., Lin, Z., 2014. Charging infrastructure planning for promoting battery
complexity of the location problem. The performance comparison electric vehicles: an activity-based approach using multiday travel data.
of solutions to problems with or without realistic features consid- Transport. Res. C Emerg. Technol. 38, 44e55. https://doi.org/10.1016/j.trc.2013.
11.001.
ered showed that these two features cause 0.69e1.86 tons more Escudero-Garzas, J.J., Seco-Granados, G., 2012. Charging Station Selection Optimi-
DCER and 0.85%e2.64% more charging demands being satisfied per zation for Plug-In Electric Vehicles: an Oligopolistic Game-Theoretic Frame-
day, which demonstrates that these features are important and work. https://doi.org/10.1109/isgt.2012.6175791.
Fang, D.B., Zhao, C.Y., Yu, Q., 2018. Government regulation of renewable energy
necessary to be considered in CS location problem. generation and transmission in China’s electricity market. Renew. Sustain. En-
This paper contributes the literature by (1) investigating a ergy Rev. 93, 775e793. https://doi.org/10.1016/j.rser.2018.05.039.
location problem of public charging stations for electric vehicles Fathabadi, H., 2018. Plug-in hybrid electric vehicles: replacing internal combustion
engine with clean and renewable energy based auxiliary power sources. IEEE
with the consideration of two important realistic features, (2) Trans. Power Electron. 33 (11), 9611e9618. https://doi.org/10.1109/
proposing a data-driven PSO-based intelligent approach to handle Tpel.2018.2797250.
this problem, and (3) investigating the effects of different taxi travel Frade, I., Ribeiro, A., Goncalves, G., Antunes, A.P., 2011. Optimal location of charging
stations for electric vehicles in a neighborhood in lisbon, Portugal. Transport.
data volumes on location solutions. Except for these contributions,
Res. Rec. (2252), 91e98. https://doi.org/10.3141/2252-12.
as a side benefit, some lessons can be learned from this research. Garcia, J., Millet, D., Tonnelier, P., Richet, S., Chenouard, R., 2017. A novel approach
First, choosing the right amount of data is important for making for global environmental performance evaluation of electric batteries for hybrid
vehicles. J. Clean. Prod. 156 (Suppl. C), 406e417. https://doi.org/10.1016/j.
reliable data-driven decisions. Second, many realistic features,
jclepro.2017.04.035.
including CO2 emissions in round trips to CSs and remaining elec- Guo, Z.X., Liu, H.T., Zhang, D.Q., Yang, J., 2017. Green supplier evaluation and se-
tricity restrictions, have large effects on charging decision results, lection in apparel manufacturing using a fuzzy multi-criteria decision-making
which need to be considered in public charging station location approach. Sustainability 9 (4). DOI: Artn 65010.3390/Su9040650.
He, F., Wu, D., Yin, Y., Guan, Y., 2013. Optimal deployment of public charging stations
problems. The approach proposed and the research findings are for plug-in hybrid electric vehicles. Transp. Res. Part B Methodol. 47, 87e101.
helpful to make effective decisions on developing public CSs in a https://doi.org/10.1016/j.trb.2012.09.007.
city. Our future work will investigate the waiting time for He, F., Yin, Y., Zhou, J., 2015. Deploying public charging stations for electric vehicles
on urban road networks. Transport. Res. C Emerg. Technol. 60, 227e240.
recharging in a CS as well as the capacity of CSs. https://doi.org/10.1016/j.trc.2015.08.018.
He, Y., Kockelman, K.M., Perrine, K.A., 2019. Optimal locations of US fast charging
Conflicts of interest stations for long-distance trip completion by battery electric vehicles. J. Clean.
Prod. 214, 452e461. https://doi.org/10.1016/j.jclepro.2018.12.188.
Huo, H., Zhang, Q., Wang, M.Q., Streets, D.G., He, K., 2010. Environmental implica-
The authors declare no conflict of interest. The funders had no tion of electric vehicles in China. Environ. Sci. Technol. 44 (13), 4856e4861.
role in the design of the study; in the collection, analyses, or https://doi.org/10.1021/es100520c.
IEA, 2015. Medium-Term renewable energy market report 2015. Retrieved 21 April,
interpretation of data; in the writing of the manuscript, or in the 2017, from. http://www.iea.org/Textbase/npsum/MTrenew2015sum.pdf.
decision to publish the results. Kong, C., Jovanovic, R., Bayram, I.S., Devetsikiotis, M., 2017. A hierarchical optimi-
zation model for a network of electric vehicle charging stations. Energies 10 (5).
https://doi.org/10.3390/en10050675.
Funding
Labib, S.M., Neema, M.N., Rahaman, Z., Patwary, S.H., Shakil, S.H., 2018. Carbon di-
oxide emission and bio-capacity indexing for transportation activities: a
This work was supported by the National Natural Science methodological development in determining the sustainability of vehicular
Foundation of China [grant numbers 71872118]; the MOE (Ministry transportation systems. J. Environ. Manag. 223, 57e73. https://doi.org/10.1016/
j.jenvman.2018.06.010.
of Education in China) Project of Humanities and Social Sciences Li, S., Huang, Y., Mason, S.J., 2016. A multi-period optimization model for the
[grant number 18YJC630045]; and Sichuan University [grant deployment of public electric vehicle charging stations on network. Transport.
 Q. Liu et al. / Journal of Cleaner Production 232 (2019) 531e541 541

Res. C Emerg. Technol. 65, 128e143. https://doi.org/10.1016/j.trc.2016.01.008. of electric taxi charging stations: a spatialetemporal demand coverage
Liang, F., Shaoyun, G., Hong, L., 2012. Electric vehicle charging station planning approach. Transport. Res. C Emerg. Technol. 65, 172e189. https://doi.org/10.
based on weighted voronoi diagram. In: 2012 IEEE PES Asia-Pacific Power and 1016/j.trc.2015.10.004.
Energy Engineering Conference. APPEEC, pp. 1e5. https://doi.org/10.1109/ Vazifeh, M.M., Zhang, H., Santi, P., Ratti, C., 2019. Optimizing the deployment of
appeec.2012.6307218. electric vehicle charging stations using pervasive mobility data. Transport. Res.
Ma, J., Zhang, L.Y., 2018. A deploying method for predicting the size and optimizing Pol. Pract. 121, 75e91. https://doi.org/10.1016/j.tra.2019.01.002.
the location of an electric vehicle charging stations. Information 9 (7). DOI: Williams, M., 2018. China’s market opening seen as opportunity for automotive
Unsp 17010.3390/Info9070170. players. Retrieved 8 December 2018, from. http://www.automotiveit.com/
Sathaye, N., Kelley, S., 2013. An approach for the optimal planning of electric vehicle news/chinas-market-opening-seen-as-opportunity-for-automotive-players/.
infrastructure for highway corridors. Transport. Res. E Logist. Transport. Rev. 59, Xi, X., Sioshansi, R., Marano, V., 2013. Simulationeoptimization model for location
15e33. https://doi.org/10.1016/j.tre.2013.08.003. of a public electric vehicle charging infrastructure. Transport. Res. Transport
Shahraki, N., Cai, H., Turkay, M., Xu, M., 2015. Optimal locations of electric public Environ. 22, 60e69. https://doi.org/10.1016/j.trd.2013.02.014.
charging stations using real world vehicle travel patterns. Transport. Res. Xie, F., Liu, C., Li, S., Lin, Z., Huang, Y., 2018. Long-term strategic planning of inter-city
Transport Environ. 41, 165e176. https://doi.org/10.1016/j.trd.2015.09.011. fast charging infrastructure for battery electric vehicles. Transport. Res. E Logist.
Shao, S., Guan, W., Bi, J., 2016. Charging station location problem with queue and Transport. Rev. 109, 261e276. https://doi.org/10.1016/j.tre.2017.11.014.
range in competitive multi-site service system. J. Transp. Sys. Eng. Info. Technol. Yang, J., Dong, J., Hu, L., 2017. A data-driven optimization-based approach for siting
16 (6), 169e175. and sizing of electric taxi charging stations. Transport. Res. C Emerg. Technol.
Sicirec, 2017. Forest and carbon capture. Retrieved 23 August, 2017, from. http:// 77, 462e477. https://doi.org/10.1016/j.trc.2017.02.014.
www.sicirec.org/definitions/carbon-capture. Zhang, L., Shaffer, B., Brown, T., Samuelsen, G.S., 2015. The optimization of DC fast
Tian, Z.H., Hou, W.B., Gu, X.N., Gu, F., Yao, B.Z., 2018. The location optimization of charging deployment in California. Appl. Energy 157, 111e122. https://doi.org/
electric vehicle charging stations considering charging behavior. Simul. Transac. 10.1016/j.apenergy.2015.07.057.
Soc. Model. Simul. Int. 94 (7), 625e636. https://doi.org/10.1177/ Zhang, Y.G., Qi, D.L., Lei, S., Wu, K., Wang, Y.D., 2016. Research on the optimal
0037549717743807. planning of the battery switch station for electric vehicles. IET Intell. Transp.
Tu, W., Li, Q., Fang, Z., Shaw, S.-l., Zhou, B., Chang, X., 2016. Optimizing the locations Syst. 10 (10), 635e641. https://doi.org/10.1049/iet-its.2016.0074.