A Charging Location Choice Model for Plug-In Hybrid Electric Vehicle Users

Abstract

Electric vehicles (EVs) are promising alternatives to replace traditional gasoline vehicles. The relationship between available charging stations and electric vehicles has to be precisely coordinated to facilitate the increasing promotion and usage of EVs. This paper aims to investigate the choice of the charging location with global positioning system (GPS) trajectories of 700 Plug-in Hybrid Electric Vehicle (PHEV) users as well as the charging facility data in Shanghai. First, the recharge accessibility of each PHEV user was investigated, and 9% rely solely on public charging networks. Then, we explored the relationship between fuel consumption and the average distance between charging to analyze the environmental benefits of PHEVs. It was found that 16% PHEVs are similar to EVs, and 9% whose drivers rely solely on public charging stations are similar to internal combustion engine (ICE) vehicles. PHEV users were divided into four types based on the actual recharge access: home and workplace-based user (private + workplace + public), the home-based user (private + public), the workplace-based user (workplace + public), and the public-based user (public). Models were developed to identify and compare the factors that influence PHEV user’s charging location choices (home, workplace, and public stations). The modeling and results interpretation were carried out for all PHEV users, home and workplace-based users, home-based users, and workplace-based users, respectively. The estimation results demonstrated that PHEV users tended to charge at home or workplace rather than public charging stations. Charging price, charging price tariff, the initial state of charge (SOC), dwell time, charging power, the density and size of public charging stations, the total number of public charging, vehicle kilometer travel (VKT) of the current trip and current day are the main predictors when choosing the charging location. Findings of this study may provide new insights into the operational strategies of the public charging station as well as the deployment of public charging facilities in urban cities.

1. Introduction

Electric vehicles (EVs), including both battery electric vehicles (BEVs) and plug-in hybrid electric vehicles (PHEVs), are regarded as permanent and feasible alternatives to the traditional internal combustion engine (ICE) vehicles, thereby reducing the noise and air pollution and improving urban air quality [1,2,3]. Over the past decades, many countries and regions have committed to vehicle electrification through considerable investments in consumer incentives, charging infrastructures as well as technology research and development [4,5]. In order to promote the usage and adoption of EVs, Shanghai had made great investments to subsidize EV consumers and manufacturers, offered free and specific license plate, built public charging stations, and offered purchasing tax deductions and exemptions and other non-monetary incentives (e.g., free usage restriction policy during peak hours) [6]. In 2016, the EV ownership in Shanghai has surpassed the 100,000, ranking the first in the world. By 2018, the number had increased to attain 239,000, accounting for 6% of the total vehicle ownership in Shanghai [7], in which percentages of BEV and PHEV were around 30% and 70%, respectively.

One of the most pressing challenges is to deploy charging infrastructures in public locations considering the travel characteristics of EV drivers, so as to promote EVs and maximize the overall travel electrification [8,9,10]. Mismatch and disparity of charging demand and charging facility supply may lead to underutilized charging stations and points [11,12,13]. To this end, the EV industry falls into a kind of “egg-chicken” paradox in promoting EVs and charging infrastructures [14,15]. Thus, the relationship between charging stations and electric vehicles needs to be precisely coordinated.

In Shanghai, the average public charger occupancy rate in 2018 was just 1.3% which shows that the public charging stations and chargers are widely underutilized [16]. As the public charging network is generally considered to be insufficient for the overall 239,000 EV ownership in Shanghai, the low utilization rate is rather confusing and surprising and a reasonable explanation is the misalignment of charging supply due to the poorly sited location of the public charging stations. Given the limited knowledge and experience, one effective practice is to analyze the travel pattern and charging behavior of PHEV users, as well as the utilization of charging networks through the available massive data.

Research on EV charging behavior has been carried out during recent years with the growing global usage of EVs. Jabeen et al. [17] explored the PHEV drivers’ charging preference at home, workplace and public stations in Western Australian EV trial, and indicated that PHEV drivers tended to charge PHEV at home or work rather than public charging stations. Moreover, they were price-sensitive and preferred to perform low-cost PHEV charging. Zoepf et al. [18] used a mixed logit model to estimate the choice to charge at the end of each trip and suggested that charging is usually conducted after the day’s last trip when returning home. Sun et al. [19] examined the charge timing choice behavior of BEV users using a mixed logit model and suggested that the state of charge (SOC), vehicle kilometer travel (VKT) on the next travel day, and interval in the days before the next travel day, are the main attributes to decide whether the BEV should be charged or not. Meanwhile, most users prefer to charge during midnight. Xu et al. [20] developed a mixed logit model to explore the choice for charging location and mode of BEV users with the revealed preference data in Japan. It was concluded that the main predictors for choosing the charging location and mode are the battery capacity, midnight indicator, SOC, and the number of past fast charging events.

Little previous research has investigated the choice of charging stations (private/workplace or public) and the charging preference of public stations of PHEV users, especially with massive multi-source global positioning system (GPS) data. In consideration of the leading position of the promotion and usage of EVs in Shanghai, it is more instructive and representative to explore the charging choice behavior of PHEV users with the GPS trajectories and the charging facility data in Shanghai.

In general, EVs can be recharged at home (private), working places, and public charging stations. Public charging stations are accessible to all EVs, mainly installed in locations, such as public parking lots and shopping malls. Charging stations are partially provided at workplaces, such as government agencies, companies, and the private charging points are mainly installed at home for personal private use only.

BEV users recharge their vehicles less than PHEV users from the U.S. Department of Energy’s EV Project (an average of 1.1 charging events per day driven for BEVs, versus 1.4 events per day for PHEV users) [21]. Users for whom recharging EV is necessary to perform charging less frequently than those for whom it is optional and is paradoxical. Motivated by minimizing vehicle operating cost, PHEV users seek to avoid consuming gasoline through more frequent charging. BEVs’ limited driving range causes BEV users to be more likely to perform public charging, especially for fast charging (an average of 18.4 fast public charging events and 11.9 normal public charging events per year for BEV users, versus 17.4 normal public charging events for PHEV users) [20]. Therefore, the deployment of public charging stations needs to consider the charging behavior of PHEV and BEV users separately.

PHEVs combining an ICE with a grid-connected battery, allow the vehicles to be powered by both gasoline and electricity, and thus, potentially reduce gasoline usage and the emission of air pollution [22,23,24]. Offering more charging opportunities for PHEVs could help to promote the adoption and market and extend the charging-depleting (CD) range [25]. Moreover, public charging stations bring more convenience to users and have the potential to reduce fuel cost. Understanding the charging behavior of PHEV drivers has significant implications for pricing strategies and deployment of charging facilities, and predicting charging demand spatially and temporally.

Investigating the charging behavior of PHEV users, especially the charging frequency and location choice, may assist in understanding some critical issues as follows:

(1)

Where did the PHEV users charge their vehicles, at home, workplace, or public stations? How many charging locations that a PHEV user can choose?

(2)

For a PHEV, is it more similar to BEV or to ICE in terms of charging behavior and fuel consumption? How about the percentage?

(3)

What factors may influence the choice of charging locations (e.g., home, workplace, or public) of different users considering the actual recharge access?

(4)

Is it likely to encourage PHEV users to recharge the battery during off-peak periods by a load-shift-incentivizing electricity tariff?

The main contribution of this study is as follows. First, this study attempts to match the charging events with the corresponding charging stations, thus to find out the charging stations where the charging events occur, with the corresponding charging price. The procedure may be applicable to any EV GPS data to access home/workplace recharge access. Second, this study attempts to classify PHEV users according to the number and type of charging locations and explores who regularly recharge a PHEV at home or workplace. For each PHEV, it is similar to BEVs according to driving and charging styles or similar to ICEs? Third, this study establishes separate models to explore the factors that influence the charging location choice considering different types of users and provides suggestions for the operation and deployment of public charging stations.

For the remainder, Section 2 introduces the PHEV GPS data and the pre-process procedure, and discusses the driving and charging patterns of PHEV drivers. Section 3 expounds the method that mining charging facility data, so as to match the charging events with the corresponding charging stations. The result assists in classifying PHEV users into five types according to the numbers of accessible charging locations. Section 4 presents the charging location choice model for the PHEV users, with the resulting interpretations and discussions. Section 5 summarizes the paper and the outlooks into future research.

2. PHEV Data Reduction and Analysis

In Shanghai, each EV has been installed with an on-board GPS terminal, recording the vehicle identification, and real-time trajectories, including the time, position, speed (km/h), sum mileage (km), voltage (V), current (A), and working and charging status with a sampling interval between 10 and 30 s. These property data were centralized and stored in Shanghai’s EV public data collection, monitoring, and research center (hereinafter referred to as “SHEVDC”), established in 2013.

2.1. Data Pre-Processing Procedure

Data used in this study is from SHEVDC. The GPS data for 700 private PHEVs in 2018 (from 1 January to 31 December) were obtained from SHEVDC in the present study.

Table 1. GPS data for a particular PHEV.

Vehicle ID	Date Time	Vehicle Status	Charge Status	Speed	Sum Mileage	Voltage	Current	SOC	Latitude	Longitude
SHEVDC_795G081B	2018/1/1 16:42:33	1	3	0.0	12,068	737.0	11.1	80	31.1707260	121.7856531
SHEVDC_795G081B	2018/1/1 16:43:03	1	3	16.6	12,608	729.0	26.8	80	31.1707729	121.7856831
SHEVDC_795G081B	…	…	…	…	…	…	…	…	…	…
SHEVDC_795G081B	2018/1/1 17:27:53	1	3	0.0	12,103	678.0	2.7	25	31.0042109	121.7220360
SHEVDC_795G081B	2018/1/1 17:28:23	2	3	0.0	12,103	678.0	−0.1	25	31.0042109	121.7220360
SHEVDC_795G081B	2018/1/1 17:29:53	2	1	0.0	12,103	682.0	−4.6	25	31.0042109	121.7220360
SHEVDC_795G081B	2018/1/1 17:30:23	2	1	0.0	12,103	684.0	−4.6	25	31.0042109	121.7220360
SHEVDC_795G081B	…	…	…	…	…	…	…	…	…	…
SHEVDC_795G081B	2018/1/1 23:08:57	2	1	0.0	12,103	784.0	−4.0	99	31.0042109	121.7220360
SHEVDC_795G081B	2018/1/1 23:09:27	2	4	0.0	12,103	784.0	0.0	100	31.0042109	121.7220360

presents the GPS data of a particular PHEV. We can easily figure out that this PHEV had made one trip on the 1st of January, 2018, and was then charged to SOC of 100% from the afternoon (17:29:53) to midnight (23:09:27). In particular, the vehicle status is a binary variable indicating whether it is moving or not, (1 if the EV is moving, and 2 otherwise). The charge status is a variable indicating the charging states of EV, 1 for the EV is parking and charging, 2 for the EV is moving and charging, 3 for the EV is not charging, and 4 for the EV is fully charged. The data processing includes a two-step procedure: dataset elimination at the vehicle level and the pre-processing at the event level, as presented in Figure 1.

The first step includes three pre-process principles. At the vehicle level, two PHEVs were with empty vehicular datasets, and the accumulated mileage of another three PHEVs had decreased at certain points. A sudden increase in sum mileage indicates that the loss of vehicle data occurred and 27 vehicle datasets were eliminated for this issue. The raw PHEV GPS data were first processed to obtain the charging and traveling events of each PHEV. The PHEV event data may also have potential problems as follows. 0.62% of the traveling event data with zero trip distance were removed. 0.01% of the PHEV traveling events with SOC greater than 100% or less than 0% were obviously illogical. 0.2% were the charging events with plug time less than two minutes or SOC difference less than 2%. These are considered as error-prone event data and consequently were removed. After applying the PHEV data pre-processing, a total of 118155 charging events from 668 PHEVs were obtained.

2.2. Driving and Charging Behavior of PHEV Users

To understand PHEV users’ driving and charging patterns, 10 driving, and charging behavior metrics at the vehicle level were chosen with the corresponding descriptive statistics shown in

Table 2. Descriptive statistics of driving and charging behavior metrics.

Metrics		Min	25%	Median	75%	Mean	Max
Driving behavior	Total Distance (km)	229	7333	12,648	19,373	14,774.95	78,782
	Total Driving Day (days)	6	165	268	308	232.12	352
	Aver Distance Per Day Driven (km/day)	12.60	40.27	57.08	79.49	65.01	397.25
	Trip Times Per Day Driven (times/day)	1.85	3.13	3.78	4.62	4.12	14.54
	Aver Distance Per Trip (km/time)	0.20	10.89	14.20	18.90	16.27	122.23
	Aver Distance Between Charging (km)	2.72	58.38	76.54	111.57	156.99	5971
Charging behavior	Total charge Num.	1	77	158	259	176.87	650
	Aver Charge Num. Per Day Driven	0.005	0.48	0.71	0.94	0.73	4.05
	Aver Initial SOC (%)	11.09	27.46	34.67	41.83	35.35	71.28
	Aver SOC (%)	20	90.21	94.13	96.88	91.93	100

[20].

The distributions of the six metrics are presented in Figure 2. The total distance and the average travel distance per day driven are the two indicators for travel demand and represent the charging demand to a certain extent. It could be figured out that PHEVs in Shanghai were driven 65.01 km per day and 16.27 km per trip. This indicates that PHEV users have a longer commute distance and higher travel demand than traditional ICE vehicle users [26]. The average value and distribution of the total driving days and trip times per day driven show that PHEVs average more journeys in total than the traditional ICE vehicles.

Figure 2e displays the average travel distance between charging, with a maximum value of 5971 km and a minimum value of 2.73 km. It is apparent that 91% of the PHEVs were plugged in after driven 200 km or less, which are charged 156.99 km on average. In general, PHEV users have good charging performance. It is obvious from the table and the histogram that there remains considerable differentiation in the charging behavior among PHEV users.

The histogram and CDF in Figure 2e show that 74% of private PHEVs are charged 0.25–1 time per day driven and 8% are charged less than 0.25 times per day driving. Moreover, Figure 2f shows that the distribution of initial SOC before PHEV charging approximates to normally distributed and the CDF indicates that 10% of PHEVs would not be plugged in only when the SOC falls below 23%. As for the average SOC, it has a mean value of 91.91 and a median value 94.93, indicating that PHEVs users prefer to charge the PHEVs fully. Considering that PHEVs require 5 to 10 h to get a full charge, we can infer that PHEVs are usually replenished at home/workplace.

On the basis of the four metrics, namely the total charge number, the charging number per day driven, the average distance between charging, and the initial SOC, about 9% PHEV users were identified with poor charging performance, who were further analyzed to identify the accessibility to a fixed charging station.

3. User Classification Based on Accessible Charging Location

To access home/workplace recharge access, a reasonable procedure is proposed to match the charging events with the corresponding charging stations in order to classify PHEV users according to accessible charging locations and explore those that could recharge a PHEV at home or workplace.

3.1. Matching Charging Events with Corresponding Charging Stations

Matching charging events with corresponding charging stations include three steps: first, a web crawler with the python programming language was used to collect charging facility data. Then, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm was introduced to explore the accessible charging locations of each PHEV users. Finally, the accessible charging locations were matched with the corresponding charging stations. Figure 3 presents the entire matching procedure.

3.1.1. Charging Facility Data Collection

The charging facility data were obtained mainly from Shanghai’s charging facilities public data collection, monitoring, and research center (hereinafter referred to as “SHCFDC”). By the end of 2018, a total of 210,650 charging points has been installed in Shanghai, including 141,786 private charging points, 36,233 public charging points, and 32,631 workplace charging points [16]. To locate the charging station that the charging events of PHEV users occur, and get the corresponding charging rate and the number of chargers provided, all public and workplace charging stations in Shanghai were collected and matched with charging events through coordinates.

In this study, a web crawler with the python programming language was used to obtain data from the website of the SHCFDC and the Mobile applications of seven charging station operators which account for more than 90 percent of the total charging points in Shanghai including Teld, State Grid Corporation of China, Star Charging, Anyo Charging, Potevi, EV power, and ‘Let’s Charge’ Charging [27,28,29,30,31,32,33].

Table 3. Selected data from charging stations in Shanghai.

Station Name	Longitude	Latitude	Station Type	POI Type	Operator	Parking Price (yuan/h)	Electricity Price (yuan/kWh)	Service Price (yuan/kWh)	Fast_num	Normal_num
The bund hotel	121.4814900	31.2319930	Public	Hotel	Potevio	10	1.5/0.99	0.99/0.99	5	20
Zhongfang East China Building	121.4517557	31.2469587	Public	Office building	Star	10	0.87/0.35	0.59/0.64	0	3
Leatop Plaza	121.4841500	31.2530900	Public	Shopping Plaza	SGCC	8	0.68/0.68	0.57/0.57	7	8
Yangpu Technology Commission	121.5342957	31.2836227	Workplace	Government	Teld	0	1.0/1.0	0.8/0.8	1	4
Nanhui middle school	121.7824402	31.0587433	Workplace	Primary School	Anyo	0	0.63/0.63	0.6/0.6	0	6

As for the electricity/service price, the former refers to the price from 8:00–22:00, the latter refers to the price from 22:00–8:00.

shows the data of five particular charging stations, including the station name, longitude, latitude, Station Type, POI type, Service operator, parking price (yuan/h), electricity price (day/night, yuan/kWh), service price (day/night, yuan/kWh), and the number of fast and normal charging points.

3.1.2. Classification of the Accessible Charging Locations of Each PHEV User

To better understand PHEV users’ charging behavior, the charging events of each PHEV user were classified into a specified number of clusters. Charging events within the same cluster means the PHEV has been recharged at the same charging station. The Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm [34,35] was introduced to cluster the charging locations of each PHEV user. It is a spatial density-based clustering method and directly divides all point densities reachable from different points into clusters.

Compared with other clustering algorithms [36], DBSCAN possesses superior performance, which is insensitive to the order of the points in the dataset and does not require the number of the clusters a priori. Two important parameters for the algorithm include the number of occurrences and the radius: MinPts and Dist, which were chosen as 10 and 200 m in this study. That means if a PHEV is charged more than 10 times within the studying period and the distance between any two charging locations is less than 200 m, we could conclude that the charging events occurred in the same charging station, labeled as a frequently-used station. This algorithm explains how many charging locations a PHEV driver can choose, which also simplifies the implementation in matching the charging location with the given stations.

3.1.3. Matching of the Accessible Charging Locations with the Corresponding Charging Stations

The “Near Analysis” function in ArcGIS was used to find the nearest charging station of each charging event, with the distance threshold set as 200 m, while the address and point of interest (POI) type of the charging locations and stations were also used as references. For instance, a PHEV user parks and recharges the vehicle in the Leatop Plaza, the nearest station is in the Leatop Plaza, and the distance is less than 200 m, which means the charging event occurs in the corresponding charging station.

The POI type and address are queried through Baidu map API, in which the ‘poiRegions’ request was used to obtain the address and POI type of any coordinates of charging events [37,38]. The correlation between the properties of each charging station/point (private, workplace, and public) according to the POI type obtained from Baidu map API, are shown in

Table 4. The correlation between Baidu POI type and the Charging Station/Point Property.

Baidu POI Type	Baidu POI Subtype	Charging Station/Point Property
Residential district	-	Private
Office Building	-	Workplace/Public
Enterprise	enterprise, industrial park	Workplace/Public
Government	government, public security, administrative units, etc.	Workplace
Education	primary school, high school, scientific research institution	Workplace
	library, university, education, and training council	Public
Cultural media	TV and broadcast station, press and publication	Workplace
	exhibition hall, art gallery	Public
Shopping	grocery, shopping plaza, furniture market, etc.	Public
Transportation facility	airport, railway station, long-coach station, etc.	Public
Financial service	bank, investment, and financing	Public
Hotel	star hotel, budget hotel	Public
Tourism	museum, scenic spot, park, etc.	Public
Catering	bakery, cafe, restaurant, etc.	Public
Auto	car sales, car repair, car wash, etc.	Public
Living service	estate agent, housekeeping, photo studio, etc.	Public
Recreation & Entertainment	KTV, cinema, theatre, etc.	Public
Health	dentist, health, hospital, etc.	Public
Sports and Fitness	gym, fitness center	Public

.

3.2. PHEV User Classification According to the Accessible Charging Location

Having matched charging locations and charging stations, we have a clear understanding of whether PHEV users charge at home or workplace and how many charging location choices a PHEV user has. The number of charging in the private, workplace, and public stations/points were figured out as 87,228, 19,115, and 11,812, respectively. The results also provide insights into the drivers’ charging behavior as follows:

(1) The proportion of total numbers of users for the private, workplace and public stations is about 8.5:2:1. Users generally tend to recharge PHEV at home or workplace instead of at public charging stations. One reasonable explanation is that most PHEVs are not compatible with DC fast charging and consequently has a lower charging power, requiring 5 to 10 h to get a full charge. Home and workplace charging stations and points offer great flexibility and convenience to recharge the battery when PHEVs are parked all day in working places or overnight at home. What is more, the advantage of the charging cost has also contributed to this phenomenon.

(2) According to the number of charging location choices, PHEV users could be divided into five types. As shown in

Table 5. PHEV user classification according to charging location.

Num of Charging Location	Subtype	# of Vehicles	Num of Charging	Aver Initial SOC (%)	Aver Distance Between Charge (km)	Fuel Consumption (L/100 km)
0		31	12.035	26.376	1495.549	9.479
1	Overall	405	153.488	35.743	106.446	4.424
	Home	338	158.204	36.020	99.567	4.109
	Workplace	42	137.619	37.102	127.294	5.608
	Public	25	116.400	29.724	164.433	6.696
2	Overall	175	212.394	35.359	87.417	4.259
	Home + Home	39	190.769	32.723	94.782	4.439
	Home + Workplace	74	207.946	37.084	75.779	3.873
	Home + Public	59	237.305	34.975	96.997	4.572
	Workplace + Workplace	3	113.333	34.630	90.343	5.331
3		28	240.392	36.722	77.120	3.895
>=4		29	347.241	39.895	67.510	3.599

, the majority of PHEV users has one or two fixed charging location choices. As the number of charging location choices increases, the average charge time increases and PHEV have a higher average initial SOC, and the average distance between charging and fuel consumption decreases, which may stimulate petroleum replacement and bring additional environmental benefits. Increasing charging options also contributes to improving the charging behavior of users. The deployment of charging facilities should obey the rule “taking the private stations as a principle, and the public stations as the auxiliary”, so as to ensure that the users can charge either at home or at their workplaces.

(3) An overall of 565 (85%) PHEV users have owned charging facility at home was found from Table 5. In fact, the Shanghai government requires purchasers of EVs to have their own private charging facilities. With the implementation of the policy, although 6% of PHEV users have no charging access at home, the workplace charging stations in their workplaces were used as a surrogate.

(4) As mentioned above, on the basis of four metrics, about 9% of PHEV users have poor charging performance. Through identifying charging locations of PHEV users, users who lack a fixed charging station and who only have one charging station belonging to a public charge the battery less times. Home charging and workplace charging may not be accessible, and they have to rely solely on the public charging network. The private charging infrastructure availability actually plays a major role in impacting, whether charging or not.

As shown in Figure 4, PHEV users show a strong preference for charging stations in office buildings, shopping, enterprises, auto and education land use areas. An interesting phenomenon is that PHEV users favor auto places, which may be explained by the fact that they can plug in their PHEVs in 4S/Dealer shops for free.

3.3. Environmental Benefit Assessment

Is a PHEV more of a BEV according to driving and charging behavior, or is it more similar to an ICE? The key to this question lies more with the charging behavior of the users and less with the technology of PHEV. The main factors in determining the ratio of gasoline to electric miles driven are driving and charging behavior. The environmental benefits of PHEVs depend on how these vehicles are used, especially the average distance between charging. In previous studies, it is common that PHEV users are generally assumed to maximize the battery utility and minimize the travel cost by matching the battery to driving needs and charging their vehicles as much as possible. But the reality is that charging behavior has been limited to the accessibility of existing charging infrastructure.

An average 91% of the PHEVs were charged after driven less than 200 km, indicating that the supply of charging station in Shanghai can meet the charging demand of PHEV users but could still be improved. Pearson correlation analysis (${\mathsf{\rho }}_{X,Y}=0.741$) indicates that there is a positive significant correlation between the average distance between charging and the average fuel consumption. As shown in Figure 5, the curve fit indicates there is a strong linear relationship between them when the average distance between charging is less than 200 km, while there is a quadrative curve and the average fuel consumption maintains at a higher interval when the average distance between charging exceeds 200 km.

The average fuel consumption of PHEV is 4.56 L/km compared to ICE; the environmental benefit of PHEVs is significant, particularly when the average distance between charging is less than 200 km. Considering the electric driving range is 50 km, 16% of the PHEVs are charged after driven 50 km or less, and the average fuel consumption is 0.933 L/100 km, it is concluded that these PHEVs are similar to EVs. While 9% of the PHEVs are charged after driven 200 km or more, and the average fuel consumption at 9.451 L/100 km, these PHEVs are similar to ICEs. The average fuel consumption increases rapidly with the increasing of the average distance between charging when the average distance between charging is less than 200 km. Therefore, it is of significance to ensure that each user has one accessible and convenient charging station such that the average distance between charging does not exceed 200 km and provide a workplace and public charging to users in order to minimize the average distance between charging and vehicle operating cost.

4. PHEV Charging Location Choice Model

In order to capture the charging location choice of PHEV users, three alternatives were defined as: (i) private (home), (ii) workplace, and (iii) public charging stations. However, three alternatives are not accessible to each PHEV and PHEV drivers could be categorized into four types: home and workplace-based users (private + workplace + public), home-based users (private + public), workplace-based users (workplace + public), and public-based users (public). Figure 6 depicts the charging alternatives and the number of charging events for each type. Separate models were proposed to identify and compare the factors that influence a respondent’s charging location choice considering different types of users. In this study, the general charging location choice behavior of all PHEV users was explored. Then, different empirical models were examined by considering the charging behavior of home and workplace-based users, home-based users, and workplace-based users of PHEV.

4.1. Explanatory Variables Identification

It is reasonable that PHEV users are assumed to follow the principle of utility-maximization when choosing among the three alternatives. Here, the independent variables which may affect the charging location choice could be categorized into six groups: time-related, charging cost-related, charging power-related, SOC-related, public charging station-related, and travel pattern-related [17,18,19,20]. The definitions and source of the variables were provided in

Table 6. Definitions and source of the explanatory variables.

Group	Variable Name	Variable	Description	Source or Calculation
Time-related	Dwell time (h)	$t_{dwell}$	The time duration for which respondent stayed at the station	Shanghai GPS trajectory data
	Working day	$I_{working}$	Whether the day belongs to working day	Shanghai GPS trajectory data
Charging cost-related	Parking price (yuan per kWh)	$p_{parkingprcie}$	The parking price per kWh for the current charging event	Equation (1)
	Charging price (yuan per kWh)	$p_{chargingprcie}$	The parking price per kWh for the current charging event	Equation (2)
	Charging price tariff	$I_{tariff}$	Whether there is a charging price tariff for the current charging event	Equation (2)
Charging power-related	Charging power (kw)	$Power$	the charging speed for the current charging event	Shanghai GPS trajectory data
SOC-related	SOC (%)	$SOC$	initial state of charge for the current charging event	Shanghai GPS trajectory data
Public charging station-related	Dense stations	$P_{dense}$	the numbers of public charging stations within 1.6 km of the parked location.	Shanghai GPS trajectory data and Charging facility location data
	Average number of public chargers	$P_{size}$	the average number of chargers of public charging stations within 1.6 km of the parked location.	Shanghai GPS trajectory data and Charging facility location data
	Number of public charging events	$P_{num}$	The total number of public charging stations before current charging event	Shanghai GPS trajectory data and Charging facility location data
Travel pattern-related	VKT of current trip (km)	$V_{ct}$	Vehicle-kilometers of the current trip	Shanghai GPS trajectory data
	VKT of next trip (km)	$V_{nt}$	Vehicle-kilometers of the next trip	Shanghai GPS trajectory data
	VKT on travel day (km)	$V_{cd}$	Vehicle-kilometers of travel on current travel day	Shanghai GPS trajectory data
	VKT on former travel day (km)	$V_{fd}$	Vehicle-kilometers of travel on the former travel day	Shanghai GPS trajectory data
	VKT on next travel day (km)	$V_{nd}$	Vehicle-kilometers of travel on the next travel day	Shanghai GPS trajectory data

as a reference list for the variables.

Time-Related

Variable 1: Dwell time (unit: hr). PHEVs generally take 5–10 h to get a depleted battery fully charged. When PHEV users plan to initiate charging for their vehicles, they may estimate the available duration for charging. PHEV users will perform charging only if parking duration permits.

Variable 2: Working day indicator. The traveling pattern of PHEVs is supposed to be different for working days and non-working days, especially for workplace stations.

Charging Cost-Related

Variable 3: Parking price (unit: RMB per kWh). For the charging of PHEVs, the most significant difference among home, workplace, and public station charging is parking cost and charging cost. The average parking price of the public charging station is 3.94 yuan/h, and 68.2% of public station collects parking fee. In this study, whether the parking price belongs to charging cost is an important issue, and the variable contributes to justifying whether users accommodate for travel demand or charging demand.

$$p_{parkingprcie}=\frac{p_{parking}\times t_{dwell}}{P_{plug}}$$

(1)

in Ref. [39].

Variable 4: Charging price (unit: RMB per kWh). Previous researches have proved that the probability of home charging at midnight is relatively high and drivers are price-sensitive and tends] to perform low-cost charging in general [17,19]. Compared to home charging, the charging cost at the workplace and public charging stations for users includes electricity price and service price. A low electricity tariff begins from 22:00 to 8:00 in Shanghai for home charging, and the charging price (per kWh) fall from 0.64 yuan/kWh to 0.331 yuan/kWh. The average charging price of public charging station is 1.83 yuan/kWh, while the average of the workplace is 0.651 yuan/kWh. In order to increase usage, 4.6% public charging station adopt cost tariff. Moreover, plug time is used to calculate the total charging cost, including electricity price and service price for each charging event.

$$p_{chargingprcie}=\frac{(p_{electricity1}+p_{service1})\times t_{plug1}+\left(p_{electricity2}+p_{service2}\right)\times t_{plug2}}{t_{plug}}$$

(2)

in Ref. [39].

Variable 5: An indicator for charging price tariff. The charging price tariff is coded in the ordinal form to represent non-tariff, partial-tariff, and full-tariff as 0, 1, and 2, respectively.

Charging Power-Related

Variable 6: Charging power (unit: kW), indicating the charging speed of the vehicle battery. Power refers to the mean value in the process of charging. Generally speaking, charging speed at a public charging station is relatively higher than that of home and workplace.

SOC-Related

Variable 7: Initial SOC before charging (unit: %), visually represents the need for charging. As suggested by existing studies, charging PHEVs is, to a certain extent, a kind of demand-based behavior, which means that PHEVs are less likely to be recharged when the SOC is high [18].

Public Charging Station-Related

Variable 8: The density of public charging stations, which is expected to influence PHEV user’ charging behavior. It represents the accessibility of public charging stations. The service radius of 1.6 km is used in this study for PHEV drivers to change their behavior to accommodate charging needs [9]. The density refers to the numbers of public charging stations within 1.6 km of the parked location.

Variable 9: The size of public charging stations. It refers to the average number of chargers of public charging stations within 1.6 km of the parked location.

Variable 10: The total number of public charging stations before the current charging event. It is a good indicator for PHEV users’ familiarity with the existing situation and technology of public charging stations. Moreover, users have formed a habit, and there exists inertia so they may stay with the choices they previously made. If the experienced public charging is convenient and better aware of the distribution of accessible public stations to find a suitable station, the probability of public charging could be expected to increase.

Travel pattern related

Variable 11: Vehicle-kilometers of the current trip. The length of the completed trip is included because a longer trip may make drivers more aware that the battery of PHEV is depleted and likely to plugin.

Variable 12: Vehicle-kilometers of the next trip. The length of the next trip is included representing immediate charging demand.

Variable 13: Vehicle-kilometers of travel (VKT) on current travel day. The change of VKT of the current day usually means that the change of travel purposes.

Variable 14: Vehicle-kilometers of travel (VKT) on the former travel day. The former travel day refers to the day when PHEV user makes the last trip.

Variable 15: Vehicle-kilometers of travel (VKT) on the next travel day, which reflects the demand for electricity. Generally speaking, the possibility of charging may increase with the extension of the planned driving distance.

4.2. Modeling PHEV Charging Location Choice

To model PHEV users’ charging location choice, considering the actual recharge access among different types of users, a multinomial and binary logit model were employed. Four models predicting selection of charging location of different types of users are run: (1) Multinomial logit model (Model 1) for all PHEV users; (2) Multinomial logit model (Model 2) for home and workplace-based users; (3) Binary logit model (Model 3) for home-based users; (4) Binary logit model (Model 4) for workplace-based users. As PHEV charging location choice model were conducted for users who may be available to home or workplace charging, public-based users who rely solely on public charging station were removed from all PHEV users.

This section proposes four multinomial/binary logistic regression models, with the results to validate that PHEV users tended to charge at home or the workplace rather than at public charging stations. The charging location choice model is formulated as follows:

$$U_j=a_j+{\displaystyle \sum }{\beta }_i{x}_i+{ϵ}_j$$

(3)

in Ref. [40].

The descriptive statistics of the selected variables of each model in terms of the alternatives at the charging event level are summarized in

Table 7. The descriptive statistics of the variable.

Variable	MODEL 1			MODEL 2			MODEL 3		MODEL 4
	Private (n = 87,228)	Workplace (n = 19,115)	Public (n = 8766)	Private (n = 17,013)	Workplace (n = 12,205)	Public (n = 2049)	Private (n = 70,215)	Public (n = 6570)	Workplace (n = 6910)	Public (n = 147)
$t_{dwell}\left(\mathrm{h}\right)$	14.52 (2.931)	9.53 (21.632)	8.55 (30.787)	12.64 (16.012)	8.37 (12.884)	7.37 (9.225)	14.96 (31.646)	8.94 (35.098)	10.84 (26.541)	7.37 (17.02)
$I_{working}$	70.45	81.26	79.83	69.31	84.05	72.82	70.70	82.40	79.33	62.58
$p_{parkingprcie}$ $\left(\mathrm{yuan}\mathrm{per}\mathrm{kWh}\right)$	0	0	3.10 (2.901)	0	0	2.82 (2.852)	0	0	0	2.41 (3.283)
$p_{chargingprcie}$ $\left(\mathrm{yuan}\mathrm{per}\mathrm{kWh}\right)$	0.49 (0.135)	0.60 (0.830)	1.92 (0.308)	0.47 (0.133)	0.52 (0.805)	1.86 (0.302)	0.49 (0.134)	1.94 (0.308)	0.78 (0.873)	1.95 (0.271)
$I_{tariff}$	0.95 (0.787)	0	0.009 (0.104)	0.98 (0.777)	0	0.027 (0.173)	0.94 (0.789)	0.0036 (0.067)	0	0.013 (0.165)
$Power \left(\mathrm{kW}\right)$	2.39 (0.640)	2.41 (0.628)	2.60 (0.639)	2.28 (0.650)	2.41 (0.640)	2.63 (0.506)	2.41 (0.635)	2.59 (0.667)	2.42 (0.667)	2.77 (0.387)
$SOC(\%)$	35.59 (21.237)	40.53 (21.668)	40.41 (21.551)	35.18 (20.914)	41.34 (21.231)	43.27 (22.386)	35.69 (21.309)	39.64 (21.178)	40.18 (22.400)	35.12 (22.45)
$P_{dense}$	7.52 (5.961)	4.55 (5.147)	10.77 (9.319)	6.06 (5.530)	4.34 (5.339)	11.17 (9.442)	7.87 (6.007)	10.79 (9.305)	13.18 (38.730)	4.59 (5.230)
$P_{size}$	7.16 (7.103)	8.07 (23.222)	11.20 (30.970)	7.12 (8.467)	5.78 (6.967)	6.40 (6.219)	7.17 (6.727)	12.83 (35.441)	12.83 (35.441)	5.26 (6.986)
$P_{num}$	6.77 (23.323)	6.61 (24.822)	73.05 (75.865)	8.27 (20.802)	7.72 (29.717)	74.36 (80.237)	6.413 (23.893)	74.17 (74.602)	2.09 (4.225)	4.59 (5.231)
$V_{ct} \left(\mathrm{km}\right)$	19.15 (29.830)	17.75 (23.034)	18.25 (25.642)	19.30 (26.693)	18.85 (21.625)	20.404 (32.716)	19.12 (30.553)	17.66 (23.152)	15.55 (23.727)	14.52 (15.868)
$V_{nt} \left(\mathrm{km}\right)$	20.87 (29.999)	18.04 (24.139)	19.356 (27.796)	22.43 (26.618)	18.38 (21.93)	21.11 (31.569)	20.50 (30.759)	18.80 (26.334)	16.74 (25.968)	20.01 (34.227)
$V_{cd}\left(\mathrm{km}\right)$	65.25 (52.976)	63.14 (51.696)	65.35 (47.059)	72.022 (51.816)	63.92 (50.513)	78.81 (53.729)	63.64 (53.129)	61.29 (43.838)	59.51 (53.310)	59.72 (52.271)
$V_{fd} \left(\mathrm{km}\right)$	57.97 (52.350)	56.58 (52.410)	58.98 (48.756)	64.17 (50.776)	58.29 (52.752)	69.67 (54.253)	56.51 (52.639)	56.02 (46.646)	52.79 (52.460)	41.57 (34.880)
$V_{nd} \left(\mathrm{km}\right)$	60.47 (53.228)	57.77 (52.920)	59.416 (47.609)	67.38 (52.213)	58.99 (51.462)	69.71 (51.981)	58.80 (53.332)	56.43 (45.739)	53.67 (54.683)	49.48 (45.563)

Values in the parentheses are the standard deviation (SD) of the parameters.

as below.

In a statistical model, multicollinearity was checked to avoid skewed estimation of individual independent coefficients. First, the Pearson product-moment correlation coefficients between independent variables in four models were calculated. The results revealed that all of the correlation coefficients were below 0.7. Then, two important factors, the tolerance (TOL) and variance inflation factor (VIF) were used to detect multicollinearity. In a regression model, the variables with TOL less than 0.4 and VIF greater than 2 should be deleted. Therefore, the VIF of the parking price of Model 3 and Model 4 are 4.20 and 2.20, respectively, and these variables were removed. It reveals that multicollinearity does not exist in four models for charging location choice in this study. What is more, workplace charging station do not adopt charging price tariff, and there are only two public charging events are partial-tariff for workplace-based users, and the charging price tariff of Model 4 were also removed.

4.3. Results and Discussion

The charging location choice models of different types of users are established and estimated by SAS statistical software package [41].

Table 8. Results of the charging location choice model.

Variable	Alternative	Model 1 (# of Parameters = 15)		Model 2(# of Parameters =15)		Model 3 (# of Parameters =14)		Model 4 (# of Parameters =13)
		Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value
Specific Constant	private	8.887	0.000 **	6.421	0.000 **	19.070	0.000 **
	workplace	9.267	0.000 **	7.717	0.000 **			8.28	0.000 **
$t_{dwell}$	private	0.075	0.003 **	0.0247	0.000 **	0.0169	0.000 **
	workplace	0.039	0.000 **	0.0103	0.145			0.030	0.037 *
$I_{working}$	private	−0.014	0.836	−0.261	0.028 *	0.073	0.894
	workplace	0.811	0.000 **	0.869	0.000 **			0.872	0.000 **
$p_{parkingprcie}$	private	−101.148	0.993	−91.078	0.991
	workplace	−106.384	0.996	−92.588	0.990
$p_{chargingprcie}$	private	−3.439	0.000 **	−2.384	0.000 **	−17.79	0.000 **
	workplace	−3.543	0.000 **	−2.568	0.000 **			−3.358	0.000 **
$I_{tariff}$	private	3.166	0.000 **	3.920	0.000 **	−0.943	0.283
	workplace	−20.0025	0.994	−18.579	0.974
$Power$	private	−0.412	0.000 **	−0.462	0.000 **	−0.232	0.591
	workplace	−0.411	0.000 **	−0.625	0.000 **			−0.228	0.493
$SOC$	private	0.0062	0.000 **	0.0059	0.014 **	0.0060	0.657
	workplace	0.0011	0.413	0.0033	0.167			−0.0053	0.212
$P_{dense}$	private	−0.056	0.000 **	−0.0173	0.013 *	0.0366	0.374
	workplace	−0.180	0.000 **	−0.0577	0.000 **			−0.0181	0.511
$P_{size}$	private	0.0078	0.000 **	0.0429	0.000 **	0.0605	0.177
	workplace	0.0024	0.000 **	0.0208	0.014 **			0.022	0.017 *
$P_{num}$	private	−0.023	0.000 **	−0.0174	0.009 **	−0.114	0.000 **
	workplace	−0.023	0.000 **	−0.0197	0.000 **			0.0079	0.633
$V_{ct}$	private	0.0049	0.000 **	−0.0057	0.014 **	−0.0075	0.092
	workplace	0.0062	0.000 **	0.0014	0.519			0.0133	0.034 *
$V_{nt}$	private	0.00050	0.690	0.00087	0.681	−0.0094	0.000 **
	workplace	−0.00094	0.467	−0.00028	0.895			−0.0039	0.255
$V_{cd}$	private	−0.0044	0.000 **	−0.0029	0.001 **	−0.0080	0.000 **
	workplace	−0.0012	0.000 **	−0.0062	0.000 **			−0.0012	0.599
$V_{fd}$	private	−0.00012	0.843	0.00075	0.445	−0.00033	0.921
	workplace	−0.00083	0.184	0.00010	0.919			0.0065	0.014 *
$V_{nd}$	private	0.00013	0.837	0.0020	0.063	0.00307	0.417
	workplace	−0.0011	0.098	0.00029	0.793			0.00034	0.873
Number of observations		115,103		31,267		76,785		7057
LR chi2(30/30/14/13)		97,606.01		31,585.37		23,172.41		387.21
Pseudo R2		0.602		0.577		0.524		0.294
Prob > Chi-Sq		0.000		0.000		0.000		0.000

Reference group is: public charging. * Significance at 5% level. ** Significance at 1% level.

displays the results of the logistic regression models of different types of users and compares the results to analyze the respective charging pattern and characteristics.

Here, we observe that the model fit of each model are all significant (p < 0.001), which indicates that all models predict with significantly higher accuracy [42]. Compared to other models, the Pseudo R² of Model 4 equals to 0.294 and is relatively lower, possibly due to the charging behavior of workplace-based users, and it will be explained in detail in Section 4.3.2.

4.3.1. Analysis of Charging Location Choice of PHEV Users

The alternative specific constants with a positive sign for home and workplace of each model indicate PHEV users prefer to charging their PHEV at home or workplace rather than public charging network.

The implication of time-related variables (Variable 1–2)

Compared with public charging, the positive effect of dwell time of home and workplace charging in Model 1 is significant. It reveals that PHEV users tend to charge their battery fully at home or the workplace and the dwell time limits the usage of public charging stations. For PHEVs, the coefficient of a working day for home charging is found insignificant while the indicator for workplace charging is significantly positive in Model 1, implying that drivers tend to perform charging at workplaces on working days.

The implication of cost for charging (Variable 3–5)

For PHEVs, the coefficient of parking price is found insignificant in Model 1 for both alternatives, which shows no difference among parking price for the three alternatives. It indicates that the parking price may not belong to cost for charging, and PHEV users mostly may not change their travel to accommodate for charging needs. On the one hand, PHEV users prefer to perform charging at public stations which provide free parking; on the other hand, PHEV users park their vehicles at a public station due to travel demand rather charging demand. Although PHEV users need to pay parking cost, parking costs are not counted as extra expenses. The negative estimate of the charging price for charging at home and the workplace in Model 1 indicates PHEV users are price-sensitive. The positive coefficient of charging price tariff for charging at home demonstrates that PHEV users tend to initiate home charging from the nighttime for electricity tariff. The electricity tariff incentivizes drivers to postpone charging to off-peak hours, especially for home and workplace-based users. For instance, some users may delay their home charging until 10 pm so as to enjoy the off-peak price.

The implication of charging power (Variable 6)

The sign of the estimated coefficient for the charging power in Model 1 and Model 2 interprets that the faster-charging speed contributes to home and workplace-based users to choosing public charging.

The implication of SOC (Variable 7)

The significantly positive effect of SOC on home charging in Model 1 indicates that home charging may not necessarily be driven by demand. It is reasonable that PHEV users get used to recharging their PHEVs at home or workplace even though the remaining SOC is sufficient for the next journey. This charging behavior reveals that PHEVs may be charged more frequently than necessary. As the initial SOC is statistically significant in Model 2, it reveals that home and workplace-based users have a positive attitude towards public charging stations. When the SOC of the battery is lower, the probability of seeking a nearby public charging station in the middle of a journey for emergency use increases.

The implication of the public charging station (Variable 8–10)

The sign of the estimated coefficient for the number of public charging events and the density of public charging stations interprets that these two variables have a significantly positive effect on public charging. Due to familiarity and convenient access to public charging stations and PHEV users averagely favor charging at public stations, and the results are in line with our expectations. Compared with home or work charging, the positive effect of the average number of chargers in Model 1 is statistically significant demonstrates that the large-scale charging station reduces the likelihood of public charging.

Implication of travel pattern (Variable 11–15)

The distance of the current trip depicts the amount of charging demand or electricity offset. It has a positive effect on home or workplace charging, and PHEV users prefer to charge the battery after arriving home or the workplace to replenish electricity. The VKT of the current day reflects the degree of emergence to charge and is a good predictor for immediate charging demand and electricity loss. The increase of VKT of the current day means that the change of travel purposes and the probability of public charging also increases.

4.3.2. Charging Pattern of Different Types of Users

According to results in the charging location choice model, what deserves special mention is that a distinct difference between the charging pattern of PHEV users, especially in usage scenarios of a public charging station. For home and workplace-based users, the charging facilities are complete, and they take the private and the workplace stations as the principle and the public stations as the auxiliary. Compared to other type users, they are more price-sensitive. On the one hand, they extend the charging-depleting (CD) range through charging the battery to reduce travel cost; on the other hand, they prefer to delaying home charging to off-peak hours to enjoy charging price tariff as the coefficient for private charging stations show in Model 2. The most occurred scenario of a public charging station is that home and workplace-based users charge PHEVs when the VKT on the travel day is longer, and the SOC is lower at this parking location. The public charging behavior is demand-based and mostly occurs in leisure and entertainment travels.

For home-based users, the negative effect of VKT of next trip and current day and in Model 3 indicates that the probability of public charging increases when the travel demand and daily commute distance of PHEV users are higher. As the workplace charging may not be accessible to them, the single private charging cannot satisfy the charging demand of 13.9% users whose daily commute distance is longer and they choose one public station as a frequently-used station to serve the function of workplace charging. The coefficient of SOC, the density of public charging stations, and the size of public charging stations is found insignificant, which may prove this judgment. Although the proportion of the group in the home-based users is not high, the total times of public charging for this group equal to 6078 and account for 92.5% of the whole home-based users, and there exists distinct difference for the usage of public station among home-based users.

The average times of public charging for each workplace-based user equal to 3.12 and the charging pattern that they only rely on workplace charging and form a habit to charge on weekdays is single which leads to the relatively low Pseudo R² of Model 4. The coefficient of the density of public charging stations and the number of public charging is found insignificant, which implies the two indicators do not contribute to performing public charging and reveals that workplace-based users are less willing to perform public charging.

5. Conclusions

This study investigated the choice behavior for charging location of PHEV, using PHEV GPS travel trajectory data and charging facility data in Shanghai. Separate models were used to examine the charging location choice of different types of PHEV users by considering the influence of parking price, charging price, and charging price tariff. In addition, we have proposed a feasible method to match charging events with corresponding charging stations, so as to classify PHEV users according to their charging locations The relationship between the average distance between charging and the average fuel consumption has been explored, to reveal how many PHEVs are more similar to BEVs or ICEs in terms of driving and charging styles.

Increasing charging options can contribute to improving the charging behavior. 16% of the PHEVs with the average fuel consumption about 0.933 L/100 km were similar to EVs, while 9% of the PHEVs relied solely on pubic charging were similar to ICEs. It is of rather significance to promote PHEV considering the environmental benefits. PHEV users prefer to perform charging at home or workplace instead of public charging stations. The estimation results show that the charging price, charging price tariff, SOC, dwell time, charging power, the density, and the size of the public charging stations, the total number of public charging, VKT of the current trip and current day are the main influencing factors. There is a distinct difference between the usage scenarios of the public charging station. The usage of public charging for home and workplace-based users is demand-driven in the middle of a trip, and 13.9% of home-based users choose one public station as a frequently-used station to serve the function of workplace charging, while workplace-based users are less willing to recharge PHEVs at public stations.

Not only do the estimation results aim at the PHEV users in Shanghai, but also the findings are instructive for similar issues in the world cities with the booming of EV entering into operation. Encouraging the adoptions of EVs could solve the low utilization of the charging station. First, the charging efficiency and plug time of PHEVs limit the usage of public charging stations for drivers, and it is imperative to improve the charging speed of PHEVs for automobile manufacturers. Second, the deployment of charging stations in specific areas is supposed to coordinate the density and the scale, thus to avoid the excessive size of public charging stations. For example, by improving the density and coverage of public charging stations, more and more PHEV drivers may be attracted to recharge at a public station, especially for home and workplace-based and public-based users. Furthermore, for home-based and workplace-based users, public stations can be deployed to provide easy access to charging when they arrive home or workplace and serve as the substitute for home or workplace charging. Also, operators of public charging stations could provide special incentives and better charging service to attract PHEV users to public charging. For instance, an operator could provide the first five free charging trails to new-comers.

The personal attributes of PHEV users may affect their choice, including income, age, and so on. The stated preference experiment may be carried out to collect such information. Future studies of charging choice may combine GPS trajectory travel data with the data collected from the stated preference experiments.