1 research on trip chain characteristics of round-trip car ...docs.trb.org/prp/17-03059.pdf · 8...
TRANSCRIPT
Research on trip chain characteristics of round-trip car-sharing users in China: A case study 1
in Hangzhou City 2
3
4
5
Ying Hui, Ph.D. Corresponding Author 6
Assistant Professor, 7
Key Laboratory of Road and Traffic Engineering of the Ministry of Education, 8
School of Transportation Engineering, 9
Tongji University 10
4800 Cao’an Road, Shanghai, 201804, China 11
Tel: +86-13636458669 12
E-mail: [email protected] 13
14
Mengtao Ding, Master, Candidate 15
Graduate Research Assistant, 16
Key Laboratory of Road and Traffic Engineering of the Ministry of Education, 17
School of Transportation Engineering, 18
Tongji University 19
4800 Cao’an Road, Shanghai, 201804, China 20
Tel: +86-18701752860 21
E-mail: [email protected] 22
23
Kun Zheng, Senior Student 24
Graduate Research Assistant 25
Key Laboratory of Road and Traffic Engineering of the Ministry of Education, 26
School of Transportation Engineering, 27
Tongji University 28
4800 Cao’an Road, Shanghai, 201804, China 29
Tel: +86-18817599398 30
E-mail: [email protected] 31
32
33
Yi’an Liu, Ph.D. 34
Research Assistant 35
Chelizi Intelligent Technology Company 36
Tel: +86-13634198360 37
E-mail: [email protected] 38
39
40
41
Word count: 4,739 words text + 10 tables/figures x 250 words (each) = 7,239 words 42
43
44
45
Submission Date:August 1st, 2016 46
Y.Hui, M.Ding, K.Zheng, Y.Liu 2
ABSTRACT 1
Car sharing as a means to provide individuals with a short-term access to automobiles to complete 2
a personal trip has been well received in Europe and North America but is still in its infancy in 3
China. Currently, there are few case studies based on operation data to support the importance of 4
car sharing system in urban comprehensive transportation system. In this study, vehicle GPS data 5
of a round-trip car-sharing system in Hangzhou was used to derive trip information and classify 6
the trip chain pattern. The rental behaviors, of which time is within 24 h, are chosen as study 7
samples. Through data preprocessing, 26,085 valid behavior samples are obtained. Several trip 8
chain characteristic indices are selected to describe the characteristics. The Clustering LARge 9
Applications algorithm was used to analyze the clustering of such sized samples. Next, 26,085 10
samples were classified into five clusters as nonactivity point, multiactivity point, high stop-rate, 11
long-distance, and common patterns; each cluster has significant differences in one or two trip 12
chain characteristics indices. The cluster result reflected that different use patterns exist, and at 13
least in three typical use patterns, shared vehicles are used as private cars to satisfy continuous 14
daily trip demands or even commute demands. The car-sharing system in Hangzhou still presents 15
a multipattern condition. Therefore, the proposed method could facilitate companies in formulating 16
a flexible price strategy and determining valued customers. Moreover, the traffic administration 17
could more deeply understand the position and function of car-sharing mode in an urban 18
comprehensive transportation system, and plan scientific policies. 19
Keywords: car sharing; GPS data; CLARA clustering algorithm; trip chain characteristics; use 20
pattern; 21
22
Y.Hui, M.Ding, K.Zheng, Y.Liu 3
INTRODUCTION 1
With the development of the Chinese economy, the popularization of private cars is increasing 2
rapidly along with the growth of urban mobility demand. This has resulted in associated problems 3
such as traffic congestion and parking dilemmas. In some Chinese cities, the government has 4
formulated a policy to control the growth of private cars. Car sharing, which offers mobility and 5
flexibility of private cars and simultaneously eases pressure on public transportation, can reduce 6
private car ownership to a certain extent (1). Based on the experience in North America, a share 7
car can substitute for six to seven private cars on average and reduce members’ total trip mileage. 8
(2). However, car sharing is still a new transportation mode for Chinese people. Therefore, 9
reasonable positions and development policies for car sharing must be formulated. Analyses based 10
on operation data can directly describe the usage and deduce the trip demand. This research focuses 11
on the actual trip chain characteristics based on GPS data to provide reference for subsequent 12
researches. 13
Introduction of the Chefenxiang car sharing system 14
“Chefenxiang” is the first operational EV car sharing system in China operable since 2010. This 15
is a round-trip car sharing system, implying that the rental vehicle must be returned to the same 16
service station where it was rented. By 2015, the system had more than 1000 users and 80 stations. 17
The rent fare can be charged by hour, mile, or contract server hours. In addition, there is a 18
preferential price policy if the contract hours were from 5:00 p.m. to 9:00 a.m. The users must 19
register and obtain a smart card before using this service. 20
The GPS data facilitated a more comprehensive understanding of a trip’s spatial-temporal 21
characteristics. The following are the main objectives. 22
Identifying the typical trip chain characteristics and analyzing their usage patterns. 23
Proposing probable trip purposes and usage preference in each use pattern. 24
The first objective is discussed for the overall understanding of trip chain characteristics of round-25
trip car-sharing users. The exact knowledge of trip-behavior segmentation explicates the different 26
types of use patterns. It is quite helpful for researchers and government to clearly understand the 27
importance of car sharing in urban transportation system. By analyzing the correlation of the time-28
dimension feature and use pattern, the time-based accounting strategy or preferential policy can 29
be formulated and the vehicle use efficiency can be enhanced. 30
31
LITERATURE REVIEW 32
Car-sharing system and use pattern 33
As car sharing have become a widely used transportation mode over the decades with over a 34
million users worldwide, an increasing number of scholars and institutions are focusing on car 35
sharing. 36
37
Studies in Europe and North America 38
Over the years, there have been several case studies and model validations on car-sharing systems 39
in Europe and North America. 40
A study by Kim focused on the car-sharing usage-pattern in New York City, particularly in 41
marginalized neighborhoods. A linear regression models fitted with collected vehicle inventory 42
data and the results revealed that there is high demand on weekdays and weeknights, and car-43
sharing usage is highly correlated with the total number of vehicles available (3). Morency et al. 44
analyzed the data issued after one year of operation of a car-sharing system in Montreal, and then 45
described and clustered the usage patterns by using a K-means cluster tool (4). Febbraro et al. 46
Y.Hui, M.Ding, K.Zheng, Y.Liu 4
analyzed the state of stations to determine if they were oversupplied, undersupplied, or balanced. 1
Further, they grouped stations with similar behavior into zones and charged prices per pair of zones 2
(5). 3
Modeling studies by investigations or transaction data on car-sharing user’s behaviors are 4
common topic to researches. De modeled the propensity in adhering to a car-sharing system within 5
the random utility framework through starting from a stated preferences survey (6). Cervero used 6
the method of Logit model and Regression models and concluded that 29% of the member gave 7
up using one or more vehicles while they chose bus more frequently in daily trips (7, 8). Habib 8
presented an economic model to jointly forecast membership duration, the decision to become an 9
active member in a particular month, and the frequency of monthly usage of active members based 10
on the membership directory and monthly transaction data of Communauto Inc (9). 11
12
Researches in China 13
In China, the lack of practical support contributes to the relatively less research on car sharing. 14
Currently, there are two main research aspects of car sharing by Chinese scholars. One aspect 15
focuses on the willingness of Chinese consumers to accept such a system; the Logit model and 16
theory of planned behavior have been used to analyze this, and its impact factors have been 17
discussed (10–12). The other aspect mainly discusses the development prospect and operation 18
strategy of the system (13, 14). 19
20
In summary, European and North American countries have abundant research foundation on 21
analyzing characteristics, modeling research, and operation optimization regarding car-sharing 22
systems. Obviously, in many cities, car sharing is an important transportation mode that alleviates 23
the pressure of parking and reducing car ownership. However, in China, car-sharing studies remain 24
at the developmental level and willingness analysis, and lack details of trip characteristics and 25
user’s usage patterns, which are based on operation data. 26
27
GPS data preprocessing and deriving trip information 28
Currently, three methods are used to collect trip data in major studies: resident trip survey, revealed 29
preference and stated preference survey and GPS data. Among the three methods, GPS data has 30
higher accuracy and spatial-temporal continuity. Therefore, we used GPS data in this study to 31
collect trip chain characteristics. The method and technology of deriving trip information from 32
GPS data mainly refer to the existing studies and are appropriately modified according to data 33
analysis and experience. 34
Entrance into poor-signal buildings, tunnels, or districts with exuberant plants or some 35
operation error may cause vehicle GPS data to deviate and show Data Missing and Data Noise 36
(15). Before the research, data preprocessing is necessary. A method based on calculating the 37
average speed of adjacent records has been recommended to inspect the error trajectory point. The 38
speed threshold for distinguishing noise points was set at 200 km/h (16). This method performs 39
well considering large deviation coordinate points but is inefficient in small offset deviations. An 40
improved preprocessing method based on the average peed rate of change is considered. According 41
to the method, the average speed rate of change of adjacent records must satisfy a constraint of 42
mathematical inequality (17). 43
The previous research on trip identification can be split into two main methods. 44
One method is based on record gap of flameout. An early research set the record gap threshold 45
to 120, 90, and 60 s; the analysis result indicates that 120s is a reasonable threshold (18). Some 46
other researchers have adopted similar methods with different thresholds (19–21). 47
Y.Hui, M.Ding, K.Zheng, Y.Liu 5
The other method is based on still points. By setting a threshold speed that indicates the still 1
point, when the total time added by continuous still points exceeds the threshold, a trip stop point 2
is marked. Stopher et al. combined speed, acceleration, and direction to confirm a still point where 3
the speed is near 0 km/h and the direction is constant, then the time threshold was set to 120 s (22). 4
Du used 1.15 mile/h as the speed threshold (23). 5
In summary, mining technology has been studied deeply and could provide methodology for 6
this research. 7
8
DATASET 9
Data overview 10
The data used in this research is provided by Hangzhou EVnet Co.: the operator of the 11
“Chefenxiang” car sharing system in Hangzhou City. The data contains 31,446 rental records with 12
the corresponding vehicle GPS data from December 2013 to June 2015. The GPS data has 13
abundant spatiotemporal information of shared vehicles. The data return interval is approximately 14
30 s, even when the vehicle was not rented,the GPS data was also returned to the database. The 15
moment when a user rents a car and opens its door, a unique order ID is automatically generated. 16
The data review confirmed that it is sufficiently accurate for the research. Table 1 lists the GPS 17
data frame, with the implications shown in the Explanation column. 18
19
TABLE 1 GPS Data Frame 20
21
Column Data type Explanation
ORDERID CHAR Identification of one rent process, unique
CARNUMBER VARCHAR License number of vehicle
ADD_TIME DATETIME Return time of data
RENTED BOOLEANS 1=’been rented’ 0=’for rented’
SPEED INT Instantaneous speed of return time
longitude FLOAT Longitude of vehicle position
latitude FLOAT Latitude of vehicle position
22
Data preprocessing 23
As described by previous studies, the GPS data usually shows different deviations. To increase the 24
accuracy of this research, the data was preprocessed. By referring to the GPS preprocessing method 25
in a previous study, we completed preprocessing based on average velocity and average speed rate 26
of change as follows: 27
Step1 28
The first record must be determined to be correct by calculating the distance (D) from the service 29
station. The occupied parking space was considered as the dilemma, and its value was set at D < 30
1 km. Tab the accepted order set {S}. Further, the orders not in {S} were excluded and tagged as 31
invalid orders. 32
Step2 33
For all the orders in {S}, the average velocity between adjacent records are calculated as follows. 34
∆𝑇(𝑖,𝑖+1) = 𝑎𝑑𝑑_𝑡𝑖𝑚𝑒𝑖+1 − 𝑎𝑑𝑑_𝑡𝑖𝑚𝑒𝑖 (1) 35
�̅�(𝑖,𝑖+1)
=D(𝑖,𝑖+1)
∆𝑇(𝑖,𝑖+1)
(2) 36
Y.Hui, M.Ding, K.Zheng, Y.Liu 6
where 1
D(i,i+1)is the distance between two adjacent records 2
If �̅�(𝑖,𝑖+1) > 200 km/h(Speed threshold), then the i + 1 record is considered as the noise point. 3
Next, remove the i + 1 record, use the i + 2 record, and recalculate. 4
Step3 5
Ensure that ∀𝑖,�̅�(i,i + 1) is correct, then test whether �̅�(i,i + 1) satisfies the inequality (17) as follows: 6
7
|�̅�(𝑖+1,𝑖+2)−�̅�(𝑖,𝑖+1)|
(𝑇(𝑖+1,𝑖+2)+𝑇(𝑖,𝑖+1))
2
≤ 10 (3) 8
After preprocessing, the total number of valid orders was 28,553. 9
To emphasize the feature as an urban traffic mode for daily trips and to distinguish it with the 10
conventional car rental mode, we chose the orders with time less than 24 h for the research. In 11
summary, we obtained a sample set of 26,085 orders. 12
13
CONCEPTS AND METHODOLOGY 14
15
Concepts of trip chain index 16
Trip chain usually reflects the continuous trip features within a period (usually an active day is 17
chosen between leaving and returning home), including spatial-temporal feature as trip time, trip 18
distance and traffic feature as traffic mode, and trip purpose. The vehicle GPS data has highly 19
accurate trip trajectory; however, only the information of the trips in which shared vehicles were 20
used is available, and is thus inaccurate for analyzing the trip purpose. In this research, several 21
available trip chain characteristic indices are chosen to describe and analyze users’ trip chain 22
constitution: 23
Number of activity points: The number of activity points in a rent cycle. 24
Trip chain radius: The linear distance between a service station and the farthest activity point 25
from the station 26
R = max {𝑟𝑠𝑡𝑜𝑝𝑖|𝑖 = 1,2,…,𝑛|} (4) 27
Trip chain start time: The start time of the first trip 28
Trip chain end time: The end time of the last trip 29
Longest stop time: 30
Tl = max {𝛥𝑇𝑠𝑡𝑜𝑝𝑖|𝑖 = 1,2,…,𝑛|} (5) 31
Total trip time: The total trip time in a rent cycle. 32
𝑇𝑡𝑟𝑖𝑝=∑ 𝛥𝑇𝑡𝑟𝑖𝑝 𝑖𝑖 (6) 33
Stop-trip time rate (t): The ratio of the longest stop time and the total trip time. This index reflects 34
the time relative proportion of the main activity and trip time. 35
𝑡 =𝑇𝑙
𝑇𝑡𝑟𝑖𝑝 (7) 36
37
Algorithm of deriving trip information 38
39
The process of algorithm 40
The algorithm is based on the still point-based method. According to the experience, we set the 41
Y.Hui, M.Ding, K.Zheng, Y.Liu 7
stop-time threshold to 120 s and the speed threshold to 2 km/h. Fig. 1 presents the flow diagram 1
of the algorithm. 2
3 Fig. 1 The flow diagram of algorithm. 4
The stop-time threshold of 120 s works well in some American and European cities; however, the 5
traffic environment may be different in Chinese cities such as Hangzhou. The stop-time threshold 6
must be tested using a density plot, which is presented with a semilogarithmic coordinate system, 7
as shown in Fig. 3. 8
Y.Hui, M.Ding, K.Zheng, Y.Liu 8
1 Fig. 2 The density plot of stop time (round-off logarithmic abscissa; unit: hour) 2
Fig. 2 shows that the peak value corresponds to 120 s. A reasonable speculation about this value is 3
that a part of stop points resulted from traffic factors, such as traffic congestion and queuing up, 4
and the other stop points resulted from non-traffic factors in the neighborhood (120 s). To balance 5
this, we reset the stop-time threshold to 180 s, and all the stop points below 180 s will be considered 6
in the trip process. 7
The density plot of stop time shows an inflection point near the point of 720 s. This may be because 8
the proportion of a stop type mode increases. Thus, we set 720 as the threshold of the activity point. 9
Then, we have two concepts: 10
Short stay point 11
This type of stop point usually occurs because of meetings, transfers, goods transportations, 12
refueling, or traffic factors such as queuing up and traffic control. In this research, the stop time of 13
a short stay point is 180–720 s. 14
Activity point 15
Usually, stop-time length of the stop points based on work, activities, leisure differ but are 16
generally longer than a short stay point. In this research, the stop time of an activity point is longer 17
than 720 s. Here, “activity” refers to a generalized stop, that is, if the vehicle is stationary for an 18
entire night or a long time, it is also considered as an activity point. 19
20
Combination of short stay point 21
The short stay point is likely to be an additional purpose for a certain travel activity. This activity 22
may be an initiative choice, such as meeting and refueling, and is likely to be passive, for example, 23
queuing up for waiting. Further, being stationary in some additional situations can affect the 24
calculated time of trips. 25
To reduce the effect of some short stay points, such as queuing and meeting, the combination 26
of short-stay point must be studied. According to the research, there are several trips whose main 27
purpose include transfers, meeting, or some short activities. Thus, we cannot simply merge all the 28
short stay points. We introduce a combination rule, that is, all the short stay points that meet the 29
following two characteristics should be included in the travel rather than a travel destination: 30
Y.Hui, M.Ding, K.Zheng, Y.Liu 9
Rule 1:The short stay point whose distance from the last or next stop point is less than 1 km. 1
Rule 2: The trip time between the short stay point and the next or last stop point is less than 500 s. 2
Fig. 3 shows the schematic of the short stay point combination. 3
4 Fig 3. Schematic of merging rule of a short stay point 5
After combining the eligible short stay points, we acquired the detailed trip feature including the 6
longitude and latitude of each stop point, trip start and end times, and trip distance. 7
8
Clustering tool: CLARA Algorithm 9
To conclude the typical constitution of a trip chain, a clustering method is used. Clustering analysis 10
has a variety of algorithms. Kaufman and Rousseeuw (24) developed an algorithm especially 11
adapted to large data sets: Clustering LARge Applications (CLARA). In this algorithm, 12
partitioning around medoid (PAM) clustering is applied on data subsets of fixed sizes to convert 13
the overall time and storage requirements into a linear rather than a quadratic function of the total 14
number of objects, thus economizing on the computational time. A standard partitioning method 15
directs its main computational effort for searching among a large number of subsets of k objects 16
(𝐶𝑛𝑘 possible subsets) for a subset yielding a satisfactory, locally optimal clustering. With the 17
increase in the value of n, the number of subsets increases dramatically; for a fixed k, the rate of 18
increase is in the order of the kth power of n. Another factor with the same effect is the storage 19
requirement, which makes the number of memory locations less dependent on the number of 20
objects, of which it is a quadratic function in the PAM algorithm. 21
Considering the date scale of this research, the CLARA algorithm is applicable. All 22
multivariate analyses were conducted using R language for statistical computation. 23
24
RESULTS AND DISCUSSION 25
26
Overview of the trip chain model 27
As the trip information was mined from the GPS data, all trip chain samples are classified into 28
three models according to the characteristics of the activity point. Fig. 5 shows the model 29
Y.Hui, M.Ding, K.Zheng, Y.Liu 10
expressions and statistical results. 1
2 Fig. 4 Trip chain model description and statistics 3
According to Fig. 4, 91% of car-sharing trip behaviors comprise at least one activity point, and 4
only 8% comprise short stay points. The remaining 1% has no identified stop points. As 5
aforementioned, several short stay points are due to some trip purpose such as meeting, picking up 6
someone, or goods transportation. If these are the main trip purposes, the corresponding trip chain 7
model is likely to be Model I or II. As the same pattern is reflected for users whose main trip 8
purpose does not involve remaining at a place for a long time, Models I and II are merged into one 9
type of trip chain called type I, and another model, that is, Model III, is marked as type II. 10
Analysis of type I trip chain: Nonactivity point pattern 11
It is easy to infer that the trip time will account for most of the proportion of the total usage time 12
in type I trip chain. Fig. 5 reflects the probability density distribution of trip chain radius, trip time, 13
and total trip distance. 14
15 Fig. 5 Density plots of (a) Trip chain radius, (b) Total trip distance, and (c) Trip time (type 16
I) 17
The plot shows that the maximum proportion of trip chain radius is within 15 km, total trip distance 18
is within 30 km, and trip time is within 2 h. After analyzing the distribution characteristics, we can 19
0.00
0.01
0.02
0.03
0.04
0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300
trip length(km)
den
sity
0.000
0.025
0.050
0.075
0.100
0.125
0 10 20 30 40 50 60
trip radius(km)
den
sity
(a)
0.00
0.01
0.02
0.03
0.04
0 20 40 60 80 100 120 140
trip length(km)
den
sity
(b)
0.00
0.25
0.50
0.75
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
travel time(hours)
den
sity
(c)
0.000
0.025
0.050
0.075
0.100
0.125
0 10 20 30 40 50 60
trip radius(km)
den
sity
(a)
0.00
0.01
0.02
0.03
0.04
0 20 40 60 80 100 120 140
trip length(km)
den
sity
(b)
0.00
0.25
0.50
0.75
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
travel time(hours)
den
sity
(c)
0.000
0.025
0.050
0.075
0.100
0.125
0 10 20 30 40 50 60
trip radius(km)
den
sity
(a)
0.00
0.01
0.02
0.03
0.04
0 20 40 60 80 100 120 140
trip length(km)
den
sity
(b)
0.00
0.25
0.50
0.75
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
travel time(hours)
den
sity
(c)
0.000
0.025
0.050
0.075
0.100
0.125
0 10 20 30 40
trip radius(km)
den
sity
(a)
0.00
0.01
0.02
0.03
0.04
0 20 40 60 80 100
trip length(km)
den
sity
(b)
0.00
0.25
0.50
0.75
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
travel time(hours)
den
sity
(c)
Y.Hui, M.Ding, K.Zheng, Y.Liu 11
conclude that in this use pattern reflected by type I, the users present the trip-feature type as mid-1
short distance trip and short rent time. Considering that the distribution width of the total trip 2
distance is quite close to the double width of the trip radius, we can infer that probably only one 3
round trip occurred in the rent cycle. 4
Cluster analysis of type II trip chain 5
The constitution of type I is relatively simple because of its existing no-activity point; however, it 6
accounts for a low proportion, indicating that it is not the main use pattern for car-sharing users. 7
As shown in Fig. 4, 91% of rent behavior consists of activity points (Type II), and the constitution 8
is variable. Different trip chain characteristics reflect different use patterns. The method of 9
analyzing clusters is efficient in classifying samples that do not have a clear demarcation of the 10
group border. By mining clusters of samples with similar characteristics, we can describe and 11
analyze the group characteristics but not each sample. In this study, the CLARA algorithm was 12
used to complete the cluster analysis. 13
The first step of cluster analysis involves the determination of variables, and correlation of 14
each variable could influence the cluster result. Therefore, it is preferable to use uncorrelated 15
variables. We choose five indices, that is, the number of activity points, longest stop time, total trip 16
distance, total trip time, and trip chain radius to test the correlation level by using the Pearson 17
correlation coefficient. According to the result of correlation analysis, the total trip distance 18
sufficiently correlates to the total trip time and trip chain radius. By clearly distinguishing each 19
pattern, it is more appropriate to use the ratio of the longest stop time and total trip time to describe 20
the model of trip chain. 21
Therefore, we use the number of activity points, trip chain radius (km), and stop-trip time rate 22
for the clustering analysis. 23
Silhouette coefficients are the testing values in the CLARA algorithm. They are values 24
combined with dissimilarity and isolation, and a higher value signifies a better clustering result. 25
Reflecting by the Silhouette coefficients, the result indicates that clustering into four groups 26
produces the best result. 27
28
The clustering process categorizes 23,657 cases into four clusters. Table 2(a) shows that the 29
largest (cluster 3) and smallest cluster (cluster 4) contain 59.1% and 4.4% of the cases, respectively. 30
Table 2(b) shows the cluster features. As shown, members in clusters 1 and 3 have a higher 31
similarity, indicating that over 80% usage patterns are similar. In contrast, there are outliers in 32
clusters 2 and 4, which match uncommon use patterns. As the isolation reflects, cluster 2 is 33
considerably far away from the other three clusters, implying that the corresponding usage pattern 34
is quite different. 35
36
TABLE 2 (a) Clustering result 37
38
Cluster 1 2 3 4
Cluster size (% of the 23657cases) 5739
(24.3%)
2894
(12.2%)
13993
(59.1%)
1031
(4.4%)
Cluster
centers
Number of activity points 5 2 2 3
Trip chain radius(km) 12 7 9 61
Stop-trip Time rate 1.74 10.89 1.50 0.97
39
TABLE 2 (b) Clustering feature index 40
41
Y.Hui, M.Ding, K.Zheng, Y.Liu 12
Cluster 1 2 3 4
Max dissimilarity 7.22 66.81 3.41 23.73
Average dissimilarity 1.36 2.38 0.96 3.44
Isolation 3.6 21.4 1.7 4.0
1
Cluster profiling and initial interpretation 2
By comparing the distribution of cluster variables, salient characteristics of each cluster could be 3
shown in a plot (Fig. 6(a)). In addition, we illustrated the density plot of the longest stop time and 4
proportion on different days (work day, day off, holiday) for an in-depth analysis. 5
6
7 Fig. 6 (a) Contrastive trip chain characteristics of four clusters. (b) Density plot of the 8
longest stop time. (c) Proportional distribution of four clusters on different days. 9
Y.Hui, M.Ding, K.Zheng, Y.Liu 13
Instructions: In this study, a usage behavior belonging to a day off and holiday implies that 1
the pick time is during 17:00 on Friday or the day before a holiday to 17:00 on Sunday or 2
the final day of a holiday. 3
The profiling of clusters to summarize the usage pattern of each cluster is as follows. 4
Cluster 1: Multi-activity point pattern 5
Cluster 1 contains 24.3% of the entire sample. This type of trip chain comprises more activity 6
points, larger trip chain radius, and stop-trip rate between 1.5 and 3. The most obvious difference 7
is that cluster 1 has the most activity points (basically more than four). In this pattern, users tend 8
to plan several activities; the activity range could be for nearby areas or farther regions. As there 9
are no waiting or transfer times between continuous activities, use of shared cars is more 10
convenient than using transits or taxis. Fig. 7 (b) shows that a larger proportion of the sample 11
reveals the longest stop time within 6 h. This indicates that the vehicle is most likely to be used in 12
the daytime. The slope of the density curve near 7–9 h changes to positive, indicating that the 13
probability of this part of the sample increases. Considering that the stop time is always 14
considerably long, vehicles are probably rented overnight and used for the entire evening or for a 15
second day. The proportion of cluster 1 slightly increased during the holidays and day offs because 16
the trip chain constitutes more multi-activities according to the activity patterns for no workdays 17
such as shopping, leisure, and entertainment. 18
19
Cluster 2: High stop-rate pattern 20
Cluster 2 contains 12.2% of the entire sample. This type of trip chain comprises less activity points, 21
smaller trip chain radius, and large stop-trip time rate. The stop rate significantly differs from other 22
clusters because the longest stop time reaches over 5 times of the total trip time. Therefore, it is 23
inferred that users are perhaps dependent on vehicle usage, convenience, and comfort rather than 24
efficiency. Fig. 7 shows two peaks on the curve that correspond to the two types of behavior that 25
are significantly different: the duration of staying. The first peak indicates a usage pattern in which 26
users employ shared cars as backup vehicles or a transport tool that must not be used for a long 27
time. The second peak indicates a usage pattern in which the shared car is rented for an entire day 28
or night according to the time of commuting trip. The usage pattern of cluster 2 is similar with that 29
of private cars and the use efficiency is not considered. The proportion of cluster 2 significantly 30
increased during workdays. This further supports its correlation with the commuting traffic. 31
32
Cluster 3: Common usage pattern 33
Cluster 3 contains 59.1% of the entire sample. This type of trip chain is the most popular among 34
car sharing users, and comprises less activity points, smaller trip chain radius, and regular stop-35
trip time rate. In this pattern, users tend to use sharing cars to complete 1–3 activity trips, usually 36
not far from the service station. The rental time may not always be extremely long. This illustrates 37
that cluster 3 is mainly based on some definite purpose or social trip. In addition, it is not in 38
accordance with the time characteristics of commuting trip from the viewpoint of longest stop time 39
or stop trip rate. Convenience, efficiency, and cost may affect whether user chooses the car-sharing 40
service. Moreover, the density curve of the longest stop time reflects that the stop time is 41
concentrated on an interval within 3 h. 42
43
Cluster 4: Long-distance pattern 44
Cluster 4 contains 4.4% of the entire sample and accounts for a small proportion but has distinctive 45
features. The number of points is more dispersed, and the trip chain radii are much larger than 46
those of other clusters: mostly more than 20 km. In this pattern, the rental vehicles are likely to be 47
Y.Hui, M.Ding, K.Zheng, Y.Liu 14
used for outings or visiting relatives and friends because most travel distances are beyond the scope 1
of Hangzhou City. The low stop-trip time rate indicates that the vehicle usage efficiency is high. 2
In such a distance range, the cost of employing a taxi is much higher than that of car sharing and 3
may not meet the needs at a continuous travel. The proportion of cluster 4 significantly increases 4
during holidays, supporting the fact that the trip purposes comprise outing or visiting relatives and 5
friends. Its usage is more similar to the characteristics of private cars. 6
7
Distribution characteristics of trip chain start and end time 8
According to the analysis in the previous section, we determined the existence of five typical trip 9
chain constitutions with significantly different characteristics. Furthermore, the corresponding trip 10
purposes and choice preferences are inferred. To further support the aforementioned analysis, we 11
focused on the distribution characteristics of trip chain start and end time. Through statistical 12
analysis of trip chain start and end time distribution, we can focus on whether peak hours for 13
pickup or return exist and the different distribution characteristics among each cluster. 14
Y.Hui, M.Ding, K.Zheng, Y.Liu 15
1 Fig. 7 Pickup and return time distribution of each cluster 2
From Fig. 7, the following conclusions are presented: 3
Types I and II: cluster 3 4
Types I and II of cluster 3 have similar distributions that are quite scattered without obvious peak 5
hours and mainly appear during the daytime. It further infers that the trip purposes are variable in 6
the two usage patterns. 7
Clusters 1 and 2 8
Clusters 1 and 2 have the same pickup and return peak hours, that is, 4–6 p.m. and 8–10 a.m, 9
respectively. This part of samples accords with the commuting trip time and the vehicle rent 10
overnight for a long stop time. The high kurtosis of the distribution of cluster 2 shows that its usage 11
pattern has an obvious time dimension feature performed as follows: “picking at dusk and 12
returning the next morning.” In addition, if there are more activity trips (over four trips) except 13
commuting trips, they are clustered into cluster 1 for the distribution to have similar peak hours. 14
Y.Hui, M.Ding, K.Zheng, Y.Liu 16
Moreover, the distribution of cluster 1 comprises starting peak hours between 7:00–9:00 a.m. 1
Cluster4 2
The distribution shape of cluster 4 is considerably different from the other clusters. It has peak 3
hours but not outstanding. The start time is usually concentrated toward the morning, whereas the 4
end time is always concentrated toward the evening. It is consistent to the previous analysis in 5
which the main trip purpose could be outing or visiting relatives and friends. In addition, there is 6
another peak hour at dusk. In this type of user behavior, the car is probably rented for a long-7
distance trip until the next day. 8
In summary, the analysis of distribution characteristics of the start and end times of trip chain 9
is consistent to the previous clustering result. The usage pattern comprising some special purpose 10
may show a more concentrated characteristic for pickup and return times. 11
12
SUMMARY 13
Table 3 summarizes the main results and findings of the case study. Five trip chain clusters 14
reflecting different usage patterns that express how residents use car sharing or the type of trip 15
pattern that is suitable in China are mined from the GPS data. 16
17
TABLE 3 Summary of characteristics of trip chain types 18
19
Groups TypeⅠ
Non activity
point pattern
TypeⅡ-1 Multi-activity
point pattern
TypeⅡ-2
High stop rate
pattern
TypeⅡ-3
Common use
pattern
TypeⅡ-4
Long-distance
pattern
Proportion 9.0% 22.1% 11.1% 53.8% 4.0%
Number of
activity point
Non Most(almost >
4)
Few(almost <
3)
Few(all < 3) Disperse(main
in 1–7)
Trip chain
radius
Obviously
skew to small
side
Common and
skew to small
side
Common and
skew to small
side
Common and
skew to side
Long(almost >
25 km)
Stop time rate Almost zero Common High Common Common
Longest stop
time
Almost zero Mostly within 3
h and a part of
8–10 h
Two
peaks(More
between 10–14
h and less
between 1–3
h)
Mostly short
stop (< 3 h)
Mostly within 3
hours and quite
smooth
Start and end
time
distribution
Disperse(mai
n in daytime)
Conspicuous
peak hours of
Start time: 16–
18
End time: 8–10
Conspicuous
peak hours of
Start time: 16–
18
End time: 8–10
Disperse(mai
n in daytime)
Inconspicuous
peak hours
Start usually at
forenoon and
end usually in
evening
Probable trip
purpose
Picking up
someone,
goods
transportation
etc.
Various trip
purpose in a
rent cycle
(Maybe
including
commuting)
Commuting
(Working or
attending
school)
Daily life trip
or Business
trip
Outing or
visiting
Usually in off
day and holiday
Substitutability
and inference of
preference
Suit transit or
taxi, depend
on preference
Multi-trip is
more suitable
to vehicle than
Prefer to
choose car in
trip, like a
Suit transit or
taxi, depend
on preference
Similar to
traditional car
rental service
Y.Hui, M.Ding, K.Zheng, Y.Liu 17
and actual
demand
taxi or transit transition
before owing
private car
and actual
demand
1
CONCLUSION AND FUTURE DIRECTIONS 2
This study is based on GPS data and determines the trip chain characteristic of users every 3
time they use the car-sharing system. Five typical trip chains are obtained through statistical cluster 4
analysis. The clusters have constitutions, reflecting different usage patterns for potential trip 5
demands. By analyzing the different trip chain characteristics, we could speculates users’ trip 6
purposes and substitutability. The distribution characteristics of a trip chain’s start and end time 7
can support the speculation of trip purpose and behavioral attitude in different usage patterns. The 8
conclusions of this study are summarized as follows. 9
GPS data, containing detailed trip information, can accurately describe the trip chain 10
characteristics of a car-sharing user’s renting behavior. The clustering methodology can split 11
the trip chain samples into groups that share similar characteristics. 12
According to the trip chain characteristics, a round-trip car-sharing system set in Hangzhou 13
City currently presents a multipattern condition, indicating that its position in the Chinese 14
urban comprehensive transportation system is not distinct. In particular, unlike the experiences 15
in Europe and North America, in addition to the short-term usage pattern for the main purpose 16
of daily or business trips, a nonignorable portion of commuting trip exists. It can be speculated 17
that most car-sharing usage patterns are certainly irreplaceable to other modes of transport. 18
Moreover, Chinese private-car ownership rate is lower than in European and American 19
countries; however, the traffic stress in many big cities is increasing. Under car-ownership 20
restriction policy in some big cities, the cost of owning a private car is quite high. From this 21
research, car sharing can be considered an important supplement of personal motorization 22
level and remitting car ownership demand. 23
It must be noted that whether for an enterprise’s individual operation management or traffic 24
administration’s scientific decision about car sharing, the demand patterns of users differ with their 25
usage patterns. An enterprise can analyze a user’s demand in different patterns to manage strategies 26
accordingly. The traffic administration can analyze users’ usage patterns to understand the 27
functional position of car sharing. Finally, this study can be the basis of follow-up studies. Further, 28
this study merely discusses the classification of a trip chain and several features of time dimension, 29
while the trip information contains abundant spatial dimensions. Follow-up studies must focus on 30
different types of spatial distributions of trip chains, land use characteristics surrounding activity 31
points, and the trip-chain type constitution of service stations. In addition, the combination of 32
station-planning issues and Point Of Interest data must be discussed. Moreover, the individual 33
usage pattern of users and use classification could be studied deeply. 34
35
36
37
ACKNOWLEGEMENT 38
This research is supported by the National nature science foundation of China (51408430) 39
Thanks for data support by EVnet Co. Appreciate for assistance from colleagues: C.Qian (Ph.D.) Q. Sun40
(Master student), C. Liu(Master student) 41
Y.Hui, M.Ding, K.Zheng, Y.Liu 18
REFERENCE 1
1. Meijkamp, R. Changing consumer behavior through eco-efficient services: an empirical study 2
of car sharing in the Netherlands. Business Strategy and the Environment 7 (4), 234–244, 1998 3
2. Martin E W, Shaheen S /. Greenhouse Gas Emission Impacts of Car sharing in North America 4
[J]. Intelligent Transportation Systems IEEE Transactions on, 2011, 12(4):1074 - 1086. 5
3. Kim K. Can carsharing meet the mobility needs for the low-income neighbourhoods? Lessons 6
from carsharing usage patterns in New York City[J]. Transportation Research Part A Policy 7
& Practice, 2015, 77:249–260. 8
4. Morency C, Trepanier M, Agard B, et al. Car sharing system: what transaction datasets reveal 9
on users' behaviors[C]// Intelligent Transportation Systems Conference, 2007. Itsc. IEEE, 10
2007:284 - 289. 11
5. Febbraro A D, Sacco N, Saeednia M. One-Way Carsharing: Solving the Relocation Problem 12
[J]. Transportation Research Record Journal of the Transportation Research Board, 2012. 13
6. De Luca, S., & Di Pace, R. (2014). Modelling the Propensity in Adhering to a Carsharing 14
System: A Behavioral Approach. Transportation Research Procedia, 3, 866-875. 15
7. Cervero Robert, Nina Creedman, Muhammad Pohan, Madhav Pai, Yu-Hsin Tsai. City 16
carshare: assessment of intermediate-term travel-behavior impacts[R]. University of 17
California at Berkeley, Berkeley, 2002 18
8. Cervero Robert, Aaron Golub, Brendan Nee. City carshare: longer-term travel demand and 19
car ownership impacts[C]. 70–80. Transportation Research Board, Washington, D.C., 2007 20
9. Habib, K. M. N., Morency, C., Islam, M. T., & Grasset, V. (2012). Modelling users’ behaviour 21
of a carsharing program: Application of a joint hazard and zero inflated dynamic ordered 22
probability model. Transportation research part A: policy and practice, 46(2), 241-254. 23
10. Ying Hui, Mingquan Wang.,2010. Consuming demand incentive of potential car sharing users 24
and its developing policy:Take Shanghai as a case study (EI). 2010 International Conference 25
on Intelligent Computation Technology and Automation, 1031-1034, Changsha. 26
11. Wang M, Martin E, Shaheen S. 2012. Car sharing in Shanghai, China: analysis of behavior-27
ral response to a local survey and potential competition[C]. Transportation Research Record. 28
Washington D.C., 29
12. Zhang Y, Guo H,et al. Modeling Acceptance of Electric Vehicle Sharing based on Theory of 30
Planned Behavior[J] The Road Traffic and Security, 2015, 4(15): 25-31 31
13. Zhao S. Research on o2o shared economic business model ------ example of car shared 32
business. Master’s thesis, Capital University of Economics and Business, 2015 33
14. Liang Ye, Dongyuan Yang, 2010. The development and application research of car-sharing 34
and motorized management. China Population Research and Environment, 2010(20):243-247. 35
15. Zhang Z. Deriving Trip Information from GPS Trajectories. PH.D’s thesis, East China Normal 36
University, 2010 37
16. Bohte W, Maat K. Deriving and Validating Trip Destinations and Modes for Multiday GPS-38
Based Travel Surveys: Application in the Netherlands [J]. Demographic Research, 2008, 39
31(31):813-860. 40
17. Vidalakis L. Observing travel behaviour from GPS data - A tool comparison survey in the 41
Torino metropolitan area[J]. Gps Travel Suryeys, 2013. 42
18. Wolf J L. Using GPS data loggers to replace travel diaries in the collection of travel data[C]// 43
Dissertation, Georgia Institute of Technology, School of Civil and Environmental Engineering. 44
2000:58--65. 45
19. Pearson D. Global Positioning System(GPS) and travel surveys: Results from the 1997 Austin 46
household survey. 2001 47
Y.Hui, M.Ding, K.Zheng, Y.Liu 19
20. Doherty S T, Miller E J. A computerized household activity scheduling survey[J]. 1
Transportation, 2000, 27(1):75-97. 2
21. Wu L. A GPS-based Research on the Driving Patterns of Private Passenger Vehicle in Beijing. 3
Master’s thesis, Tsinghua University, 2013 4
22. Stopher P, Jiang Q, Fitzgerald C. Processing GPS data from travel surveys[J]. 2005. 5
23. Du J, Aultman-Hall L. Increasing the accuracy of trip rate information from passive multi-6
day GPS travel datasets: Automatic trip end identification issues[J]. Transportation Research 7
Part A Policy & Practice, 2007, 41(3):220-232. 8
24. Kaufman L, Rousseeuw P J. Finding groups in data. an introduction to cluster analysis[J]. 9
Biometrics, 1990. 10
11