Journal of Harbin Institute of Technology (New Series)  2019, Vol. 26 Issue (1): 17-29  DOI: 10.11916/j.issn.1005-9113.18001
0

Citation 

Xinwei Ma, Yanjie Ji, Yao Fan, Chenyu Yi. Exploring the Evolution of Passenger Flow and Travel Time Reliability with the Expanding Process of Metro System Using Smartcard Data[J]. Journal of Harbin Institute of Technology (New Series), 2019, 26(1): 17-29. DOI: 10.11916/j.issn.1005-9113.18001.

Fund

Sponsored by Projects of International Cooperation and Exchange of the National Natural Science Foundation of China (Grant No. 51561135003) and Key Project of National Natural Science Foundation of China (Grant No. 51338003)

Corresponding author

Yanjie Ji, E-mail: jiyanjie@seu.edu.cn

Article history

Received: 2018-01-02
Exploring the Evolution of Passenger Flow and Travel Time Reliability with the Expanding Process of Metro System Using Smartcard Data
Xinwei Ma, Yanjie Ji, Yao Fan, Chenyu Yi     
School of Transportation, Southeast University, Nanjing 210096, China
Abstract: Metro system has experienced the global rapid rise over the past decades. However, few studies have paid attention to the evolution in system usage with the network expanding. The paper's main objectives are to analyze passenger flow characteristics and evaluate travel time reliability for the Nanjing Metro network by visualizing the smart card data of April 2014, April 2015 and April 2016. We performed visualization techniques and comparative analyses to examine the changes in system usage between before and after the system expansion. Specifically, workdays, holidays and weekends were specially segmented for analysis. Results showed that workdays had obvious morning and evening peak hours due to daily commuting, while no obvious peak hours existed in weekends and holidays and the daily traffic was evenly distributed. Besides, some metro stations had a serious directional imbalance, especially during the morning and evening peak hours of workdays. Serious unreliability occurred in morning peaks on workdays and the reliability of new lines was relatively low, meanwhile, new stations had negative effects on exiting stations in terms of reliability. Monitoring the evolution of system usage over years enables the identification of system performance and can serve as an input for improving the metro system quality.
Keywords: metro expansion     smart card data     passenger flow characteristics     travel time reliability     visualization    
1 Introduction

Many cities in China are experiencing unprecedented traffic pressures as a result of urban expansion and population growth. Meanwhile, the development of China's economic and urbanization has allowed more and more cities to meet the requirements of building metro system[1]. Metro systems in China have experienced more than 40 years of development, as for June 2017, the number of cities that have metro on operation is 32, the total mileage of metro reaches 4 400 km[2]. Metro is a fast, effective and environment-friendly urban infrastructure system with high capacity and plays an important role in reducing traffic pressures[3-4]. What's more, with city expansion, sub-center or satellite cities appeared, therefore, metro systems play a role of linkage between main urban and sub-center zone[5].

Contactless smart card systems have gained universal prevalence in modern metro systems[4], Chinese cities with rail transit systems have promoted automatic fare collection based on smart card[6]. Compared with manual survey used in some papers, smart card is a cost-saving alternative to the manual fare collection method[6-7]. In addition to initial purpose of revenue, smarts card can also provide planners and operators with management information at a low marginal cost, at the same time, passengers' trip details also be recorded[8-10]. Smart card can record trip details[10]. With these data, more and more researches had been performed to mine the smart card data as estimation of alighting point to build origin-destination matrix[9], and some concentrates on trip chains[11], forecasting short-term passenger flow and travel time[12-13]. Although there are many researches based on smart card data, few pays attention to the visualization of passenger flow characteristics at the network level. In some way, the lack of effective exhibition will hold back the development of metro service improvement plans.

The objective of this study is to study the evolution regulation of passenger flow characteristics and time reliability based on the massive smart card data with the metro expanding. This paper comparatively studied the variation of Nanjing metro in April 2014, April 2015 and April 2016, in which years Nanjing metro has expanded a lot both in lines and miles, to achieve this, a few visualization techniques are used to demonstrate the spatial-temporal characteristics of passenger flow.

The following sections include review of relevant literature on passenger flow characteristics, visualization and reliability. And then the third section will present the metro network process of Nanjing, the data and the methods used for the study, and detailed presentation of the research findings. The last section will draw a conclusion and provide recommendations for future research.

2 Literature Review

Smart card data has been widely studied in recent years and various range of substantial articles relating metro have been available. This literature review emphases on passenger flow characteristics, visualization and travel time reliability.

2.1 Passenger Flow Characteristics

From the perspective of operation management and passenger experience, passenger flow characteristics are a key factor. Li[14] et al. regarded the analysis of characteristics of passenger flow as a need for the safe operation of the metro system, they analyzed basic data analysis, including time and space distribution characteristics, meanwhile, the relevant indicators such as passenger load intensity and peak hour factor were also analyzed. Many articles applied analysis of passenger flow characteristics to forecast metro passenger flow[15]. Zhu[16] introduced a 'Seven-day' average volumes to predict the daily volumes, and the error of the prediction was proved to be within 2%. Some researchers carried out the application of passenger flow from the perspective of the operators, Zhang et al.[17] proposed a model of passenger flow management and organization at metro station, Sun et al.[18] studied the assignment of passenger flow in metro networks.

In conclusion, researches of passenger flow characteristics remain popular, however, the data used in most papers was limited in short periods[14, 18], none of the aforementioned researches concentrate on the change of passenger flow with metro system's expansion. Zhang[19] made use of Zhongshan's bicycle-sharing system data from 2012 to 2014 and studied the impact of the expansion of the bicycle-sharing system on the system usage. It's meaningful to both understand the time-based characteristics and offer recommendations for the further expansion of metro system, as China is undertaking the most ambitious urban rail expansions to meet the growing travel demand of urbanization.

2.2 Visual Analytics from Smart Card Data

Data visualization can help researchers approach large amount of data in a comprehensive way and enable experts to synthesize and explore data. As it is shown in the presentation of Shneiderman et al.[20], network visualization was detailed introduced. Due to the space limitation, this paper concentrates on visualization applied from transit data. Chu et al.[21] demonstrated a method to augment transit trip and activity with a mixed use of GIS, visualizations, and data mining. Tao et al.[22] created flow-maps to visually show the spatial trajectories of BRT. They developed a geo-visualization methodology to reveal the spatial-temporal dynamics of BRT trips. Various software tools also have been introduced to visualize the transit data. Circos was initially used to analysis similarities and differences of genomes. Then, researchers applied it to show passenger trip distributions[23]. Liao and Liu[24] developed a data-processing framework, which can compare transit running time and scheduled time. Kwan[25] described several GIS-based 3D visualization methods for dealing with the spatial and temporal dimensions of passengers' travel pattern.

All the studies reviewed above offer some software tools for visualization, these tools utilize modern technologies to reveal passenger's travel patterns. More novel visualization designs are necessary for revealing the information hidden behind data analysis.

2.3 Travel Time Reliability

Reliability is a desired attribute and an important evaluation of public transportation systems[23]. Travel time reliability can be defined as consistency or dependability in travel times[26]. There are many studies focused on reliability in traffic area, some concerned about freeway[27-28], some about bus reliability[29]. Many metrics for measuring travel time reliability exist. According to FHWA[30], measures of travel time reliability include 90th or 95th percentile travel time, buffer index, planning time index, frequency that congestion exceeds some expected threshold. With the application of smart card, the detailed information of trips can be obtained, including travel time and origin-destination matrix[6]. Joanne Chan[32] proposed a metric for journey time reliability of London underground based on smart card data. Sun and Xu[33] presented a travel time reliability study for Beijing Metro, they decomposed the rail journey time and then inferred platform elapsed time, platform elapsed time-transfer by using smart card data. A good metric for measuring travel time reliability should involve variables of operators as well as passengers. Normalized and standardized metrics also needs to be included[23].

In conclusion, passenger flow of metro system is a focused topic. Since application of smart card in entry and exit station, details of passenger trips can be obtained, passenger travel behavior could be explored in more detailed ways. Except for statistical approaches, vivid and effective visualization technology is powerful to analyze travel patterns. To analyze the change of the metro system, reliability, as a comprehensive indicator of the performance, is introduced in this paper. Lots of researches aimed at different characteristics with different environmental conditions, Li[34] studied the bicycle mode share changes in the past 15 years, Bayesian before-after method could be used to study the effects of speed limit[35]. However, little time-based researches exist. Changes of metro operation environment and the metro expansion will certainly result in the shift of people's travel behavior. Dynamic study of passenger flow characteristics with metro expansion can benefit in adjusting systematic schedule, planning new lines and fare policy.

3 Methodology 3.1 Study Area

Located in the Yangtze River Delta economic zone, Nanjing is the capital of the Jiangsu province and has long been the second-largest commercial center in the East China region, after Shanghai. Nanjing covers an area of 6 587 km2, with a total population increasing from 8.23 million by 2016 to 10.6 million by 2020, and 86% of that population being urban[36-37]. Like many other Chinese cities, Nanjing has experienced ongoing and rapid urbanization, economic growth and motorization during the past decades, which results in aggravated transportation problems such as traffic congestion, air pollution, and greenhouse gas emissions in both central and suburban areas[6, 38].

It has been a consensus that public transit is an affordable, clean, convenient and sustainable travel mode to residents and travelers. Nanjing built the first metro line in September, 2005. In recent 10 years, Nanjing metro has experienced rapid development with the evolution of passenger flow and travel time reliability. By the end of April 2014, the metro network had only Line 1 and Line 2 traveling on 85 route kilometers. From April 2014 to April 2016, Line 10, Line S1, Line S8 and Line 3 were launched successively, the total mileage of Nanjing metro system reaches over 224 km. The proportion of smartcard use in Nanjing is 67% for rail, with the rest paid with single-trip (one-way) tickets in 2013[6]. A significant number of occasional rail transit users in Nanjing who do not ride rail transit regularly and only selected metro stations have ticketing offices that can provide smartcards, which may cause the relatively low penetration ratio of smartcard use for rail transit. Consequently, our research can only provide limited insights into the passenger travel patterns of metro users by focusing on smartcard data.

Just as Zhang did[19], who explored the evolution of bikeshare usage with the system expansion based on the bikeshare smartcard data of March 2012, March 2013, and March 2014, we used the metro smartcard data of April 2014, April 2015 and April 2016 provided by Nanjing Metro Company. Fig. 1(a) shows the average daily passenger flow and frequency per year since the first line opened. Fig. 1(b) divides the city into three regions: urban (area within the inner ring road), suburban (area between the inner ring road and the Beltway), and exurban (area outside the Beltway) and shows the weekly number of riders exiting at each station during a typical week in April 2016.

Fig.1 Expansion of Nanjing metro system

It can be seen from Fig. 1.(b), the overall trend is that the metro system was gradually expanding to the suburban even exurban area and the densest site of metro stations was located in the urban area. In April 2014, there were only line 1 and line 2 with large passenger traffic as the two main lines of east-west trunk line and north-south trunk line. After the newly added line 3 and line 10 in April 2015, the passenger traffic of original line 1 and 2 had a significant increase. Compared to the existing metro lines, the passenger traffic of new lines was relatively low. In 2016, line S1 as the airport line and line S8 as the suburban/exurban line were opened. However, the passenger traffic of S1 and S8 was relatively low.

3.2 Datasets

The data for this Nanjing case study was collected inApril 2014, April 2015 and April 2016, provided by Nanjing Citizen Card Company, the organization responsible for maintaining the daily use of smart card. No major disruption was found in transit service period.

The original metro database contains exit events and boarding events. Since the smart card information databases in three years remain the same, Table 1 shows a typical sequence of metro-related smart card transaction record on 1st, April, 2016. For each transaction record, available attributes include trip date and time, card number, card type, entry and exit station numbers. Each transaction contains the following fields:

Table 1 A sequence of metro-relate smart card transaction records

1) Transaction date indicates the day when the transaction associated with a board and exit even occurs.

2) Transaction time represents the time entry and exit metro rail station.

3) Number ID represents the smart card number involved in the transaction.

4) Card type indicates the cardholder group, including the Adult Card (18-60 years old), the Student Card (below high school under 18 years old), the Elderly Card (above 60 years old), and the Disabled Card.

5) Metro station number identifies the boarding and exiting station involved in a transaction.

3.3 Methods

This paper aims to explore how the passenger flow characteristics and travel time reliability change following metro system expansion by utilizing transit data visualization and statistical analyses. To achieve this, we used a large dataset built upon smart card transactions in April 2014, April 2015 and April 2016. There is a traditional festival called Tomb-sweeping Day in April, since 2008, China has officially designated it a continuous 3-day holiday.

Firstly, the paper studies the hourly characteristics, mainly from the following time dimension: years, workdays, weekends, holidays and peak hours. Secondly, in order to learn the direction imbalance, which is an important indicator of the station's properties, we extracted enter and exit transactions from every station from original smart card data on an aggregation level. The directional imbalance A is defined as follows:

When the entry flows (Fen) is greater than exit flows(Fex), A is calculated as:

$ A = \frac{{{F_{{\rm{en}}}}}}{{{F_{{\rm{ex}}}}}} $ (1)

When the exit flows is greater than entry flows

$ A = - \frac{{{F_{{\rm{ex}}}}}}{{{F_{{\rm{en}}}}}} $ (2)

Travel time reliability, as desired attribute and an important measure of the health of public transportation systems, was introduced to show the changes of metro performance. Traditional reliability indicators are mostly operator-oriented, due to the lack of user-related data. To reveal the travel delays experienced by most passengers and the performance of metro system, the coefficient of variation (Cov) was introduced[23, 41]. In order to eliminate the randomness, the sample size was required to be over 200 trips. The reliability is calculated as follows:

$ \hat c = \left( {1 + \frac{1}{{4n}}} \right)\frac{\sigma }{\mu } $ (3)

where ${\hat c} $ is the estimation of population Cov, σ is the standard deviation, μ is the average travel time and n is sample size.

4 Results

This section explores changing patterns of metro-related smart card transactions in Nanjing across different temporal and spatial scales using visualization techniques and statistical method. Travel on weekends and holidays are analyzed separately from those on workdays. The difference between mornings and evening peak hours are also explored.

4.1 Evolution of Time Characteristics

Let us start with transactions characteristics of different hours in one day and different days in a month. The smart card transactions of the network are divided by hour and day. Fig. 2 shows the visualization results of hourly patterns across different days. In morning peak hours (7:00-9:00) and evening peak hours (17:00-19:00) of the workdays, transactions are highly concentrated throughout the morning peaks while mainly concentrated in 17:00-18:00 during the evening peak, which is consistent with Zhao's result[39]. As for the evening peaks of Fridays, transactions gradually increase since 16:00 as people tend to leave work earlier in the last workday in a week. Weekends and holidays record no obvious peak hours, yet transactions during off-peaks are more than that in workdays. For example, April 7th, 2014, April 6th, 2015 and April 4th, 2016, the Tomb-sweeping Day (all are Mondays) witness no obvious rush hours.

Fig.2 Visualization of the entire transaction database of April 2014, April 2015 and April 2016 with aggregation levels of 1 h and one calendar day

During all the three years analyzed, the morning and evening peaks are basically the same on all workdays, weekends and holidays and passenger flow in these days also follow the similar rule, indicating a steady travel pattern in these years. However, daily transactions of April 2014, April 2015 and April 2016 increased on the whole following the system expansion. To mention that the passenger flow in peak hours is larger, which needs to be taken seriously in the operational organization.

4.2 Evolution of Directional Imbalances

Fig. 3 shows that during the morning and evening peak hours of the same year, directional ratios in almost all stations of different regions is quite opposite, but values of entry and exit flows are larger in urban stations than in suburban and exurban, which is especially evident around station A (Xinjiekou). As station A locates in major job centers in Nanjing, it attracts destinations in the morning peaks. Moreover, some urban stations have a ratio of close to 15 in the morning peaks but only close to 10 in the evening peaks, it indicates that passenger flows in the morning peaks are more concentrated while the evening peaks are relatively scattered. This may be because work hour of most companies in China begins in 8 or 9 am in the morning, while in the evening, passengers may need to work late or do some shopping before going home.

Fig.3 Directional imbalance of each station

Directional ratios of peak hours changes little in different years, yet the relative quantity size of entry and exit flows of some stations have a reverse. All stations can be categorized into four groups: the stations that have large directional imbalance; the stations that have small directional imbalance; the stations that the relative quantity in morning and evening peaks has few changes; and the stations that the relative quantity size of entry and exit flows have a reverse in three years. To reveal whether these stations will follow the same rule on weekends and holidays, stations A(Xinjiekou), B (Youfangqiao), C(Minggugong), E(Aotingzhongxin), F(Jinniuhu) are selected for analysis after considering their types, locations and land use shown in Table 2.

Table 2 Directional imbalance of some typical stations

Station A is the business center of Nanjing and has much more exit flows on workdays and at weekends as well as in holidays, which is different from those pure working areas. Entry and exit flows on evening peaks also show no obvious difference, which means that the passengers coming for entertainment are as many as those leave here. Station B, which is occupied by a mix of residential communities and entertainments, has obvious imbalance on morning and evening peak hours on workdays due to commuting activities and in holidays and at weekends, the imbalance is not that obvious. Station C is around a scenic spot, the entry and exit flows are relatively balanced in peak hours and different days and the exit flows are relatively large in the evening peaks of weekends and holidays. It is assumed that there are no obvious morning peaks in a scenic station while many tourists leave the scenic spot in a concentrated period of the evening peaks when the spot is closed. Station E used to have larger entry flows in morning peaks and larger exit flows in the evening peaks, yet by 2015 and 2016, the situation changed to the opposite. Land around station E was little used before 2014, yet it gets rapid development when Nanjing government sped up westward expansion after the Youth Olympic Games was held in Olympic Sports Center in 2014 and now it becomes a mixture of entertainment, education, business and residential area. Station F, located in exurban area, was newly opened in 2016. where the land is little developed and population density is low with only two schools and a few park land around. Thus more exit flows are recorded on workdays, weekends and holidays.

4.3 Evolution of Travel Time Reliability

Cov is introduced to measure the travel time reliability. It is related to the standard deviation σ and the mean μ, therefore, the smaller the sample is, the greater opportunities the errors may occur. The sample size, according to Sun et al.[23], is limited to be over 200 trips for an unbiased estimation and Cov for those eligible ODs for different time periods is calculated.

In Fig. 4, x and y axes present the stations on each line. The segment between original point and "L1" refers to the stations on Line 1, the segment between "L1" and "L2" refers to the stations on Line 2. Blanks in each figure are due to the small sampling size of below 200. The values around diagonal are relatively large, which indicates that Origin-Destination (OD) pairs on the same line record more unreliable travel times. And so are the transfer trips. Take Line S8 as an example, in April 2016, passengers on Line S8 can directly transfer to Line 3, yet to get on Line 1 requires more transfer times. As shown in the Fig. 4, transferring to "L3" from Line S8 is of higher reliability than to "L1". Thus, it is assumed that the shorter the travel distance, the easier the passengers may extend their travel time when affected by some incidents.

Fig.4 Metro network contours of travel time reliability of each line of April 2014, April 2015 and April 2016

In general, Line 1, Line 2 and Line 3 are of the highest unreliability as they are trunk lines with top three passenger flows in Nanjing Metro. And transfer passenger flows related with these three lines will also have rising unreliability. Therefore, to improve the reliability of the whole network, the key is to reduce the delay in these three main lines. In addition, the overall reliability of the network increase as time passes by. In fact, as the metro system attracts more passengers, performs better in the form of increasing number of metros and shorter departure intervals. And the better the network is built, the more choices there are for OD riders and thus higher reliability.

As for morning and evening peak hours, regardless of the year, the workdays, weekends or holidays, morning peak hours experience more unreliability than evening peak hours in line with more passenger flows in morning peaks than in evening peaks. Unreliability of a year is subsequently lower from morning peaks of workdays, evening peaks on workdays and morning peaks on weekends. And reliability is higher in holidays than on workdays and at weekends. However, daily metro passenger flow shows that there is abnormally a small increase in Tomb-sweeping Day. Looking into the operation information of the subway, it is found that Nanjing Metro develops a different operation plan to transport passenger to reduce congestion on holidays. In addition, unlike on workdays, the passenger flow on holidays is more evenly distributed, leading to its' high reliability.

To quantitatively compare the stations before and after the metro system expansion, stations are divided into three groups: (1) 14-station: stations already established in 2014; (2) 15-station: stations along new lines Line 3 and Line 10 opened in 2015; (3) 16-station: stations newly established in 2016. Stations in these three groups number 55, 34 and 26. Table 3 describes average time reliability of different types of stations in different years and different days. The larger the value is, the higher the unreliability is.

Table 3 Average reliability of different categories of stations

14-station all belongs to Line 1 and Line 2, two trunk lines of Nanjing metro with an average reliability of 0.42 in 2014 and lower in 2015 when more lines were opened. 15-station all belongs to Line 3 and Line 10, which are interconnected with some existing stations. They are of high reliability at the beginning and even higher after a year. 16-station, far away from urban area is located in airport and suburban line, thus they are of the highest unreliability in April 2016. On the whole, new stations are not as reliable as existing stations and stations located in suburban are more unreliable.

Generally, the reliability on workdays is significantly greater than on holidays and weekends. Line 2 with little adjustment in route since its opening, is fit for the study of the effects of network expansion except for two transfer stations with Line 1 and Line 10, the average stop interval of other 24 stations of Line 2 is 1.52 kilometers and the annual average daily passenger flow in 2014, 2015 and 2016 are 525 thousand, 627 thousand and 680 thousand respectively.

As is shown in Fig. 5, each 24 points in the vertical and horizontal coordinates represent 24 stations on Line 2, and each intersection on the grid line an OD pair. Much evident difference can be observed from the figure. On the whole, the unreliability in morning peak hours is higher than in evening peak hours each year, which is consistent with network reliability discussed before; high values around diagonal indicate high unreliability between adjacent stations, in line with the conclusion that the shorter the travel distance is, the higher the unreliability is. Reliability in 2014 and 2015, shows no obvious difference, yet in 2016, this index is low, especially in evening peak hours. This is because Line 2 is an east-west trunk line, with passenger flows increasing year by year, it becomes more reliable.

Fig.5 Travel time reliability of Line 2 during AM and PM peaks on workdays of April 2014, April 2015 and April 2016

5 Conclusions and Discussion

This paper is intended to explore characteristics of passenger flow with the metro expanding process based on the smart card data. A few visualization techniques are introduced to show the evolution of temporal and spatial distributions. Reliability of travel time is also calculated. Taking Nanjing as a case study, this paper focus on the following three aspects: evolution of characteristics of travel time, imbalances of station inflow and outflow, and a simple method for evaluating travel time reliability.

Based on the analysis results, this paper gives the following suggestions:

1) From Table 2, we can see that some stations such as Xinjiekou Station and Youfangqiao Station which have a large difference between entry and exit flows especially during the morning and evening peak hours. For these kinds of stations, we suggest that metro operators shall change the access or egress function of a certain number of ticket gates to ease the station flows. One thing to note is that the function change of ticket gates should not affect passengers walking rule and necessary symbols should be placed to remind the passengers the changes. The principle is to transfer passengers quickly, reduce the waiting time of the passengers and improve the service level of the metro.

2) As passenger flow is highly concentrated in the morning and evening peak hours on workdays, some passengers will have to wait for another train when it is too crowed to get on it, thus waste much time. Operators shall take more measures to solve this problem other than current ones to shorten the departure interval and adopt high capacity carriage.

3) From the directional imbalance ratio data in the past three years, changes in the land use around the stations will impact passengers flow in the station and expanding the metro network to newly developed areas could also boost passenger flows of the whole system. Thus, passenger flow is interacted with land utilization, a factor that shall be taken into account while planning new metro lines.

4) According to Nanjing Metro, the daily passenger flow of Line 1, far from hiking like in previous years, turned downward in 2014 due to an increase of the charge. This has important policy implications. Nanjing is in the stage of rapid metro construction, to attract more metro passengers, a scheme for concessionary fare is of great importance.

5) Time unreliability in morning peaks of workdays shows high, concentrated on Line 1, Line 2 and Line 3, which are all trunk lines. And the unreliability of new lines is also high and will increase due to new stations. All these shall be closely attended and well solved. At the same time, for stations with large passenger flow, some reminds and commands are necessary to avoid over-crowdedness in a few entries, especially elevators for both efficiency and safety.

Further work could be extended in the following aspects. Firstly, the factors that may influence passengers' behavior, such as land utilization, population density, GDP, can be included for the analysis of the evolution of passenger flow characteristics. Secondly, other cities could be studied to examine whether the revealed patterns are consistent across regions. Studies on holidays could be extended to other holidays such as Labor Day and National Day to see whether travel behaviors on Tomb-sweeping Day apply to these days. And the difference between month, season also deserves attention. Thirdly, a reliability calculating method combined both passengers and metro operations needs to be proposed. Last but not least, it is found many metro passengers take public bike as a transfer tool, so the visualization techniques can be extended to explore metro-bike transfer behaviors.

References
[1]
Zhao J, Deng W, Song Y, et al. What influences Metro station ridership in China? Insights from Nanjing. Cities, 2013, 35(4): 114-124. (0)
[2]
[3]
Yang M, Liu X, Wang W, et al. Empirical analysis of a mode shift to using public bicycles to access the suburban metro: Survey of Nanjing, China. Journal of Urban Planning & Development, 2016, 142(2): 05015011. (0)
[4]
Zhang F, Zhao J, Tian C, et al. Spatiotemporal segmentation of metro trips using smart card data. IEEE Transactions on Vehicular Technology, 2016, 65(3): 1137-1149. DOI:10.1109/TVT.2015.2409815 (0)
[5]
Chen Y, Gao F. Non-motorized vehicle parking demand at a suburban railway station. Presented at Fifth International Conference on Transportation Engineering, Dalian, 2015. https://trid.trb.org/view/1369744 (0)
[6]
Song Y, Fan Y, Li X, et al. Multidimensional visualization of transit smartcard data using space-time plots and data cubes. Transportation, 2017(3): 1-23. (0)
[7]
Kim H J, Kim D H, Chung J H. Weekend activity and travel behavior in a developing country: Empirical study using multilevel structural equation Models. Transportation Research Record Journal of the Transportation Research Board, 2004, 1894(1): 99-108. DOI:10.3141/1894-11 (0)
[8]
Tavassoli A, Alsger A, Hickman M, et al. How close the models are to the reality? Comparison of transit origin-destination estimates with automatic fare collection data. Presented at 38th Australasian Transport Research Forum, Melbourne. Victoria, Australia, 2016. (0)
[9]
Li H, Chen X. Understanding Travel Behavior of the Special Populations Using Smart Card Data: A Case Study of Suzhou, China. 2014. (0)
[10]
Huang X, Tan J. Understanding spatio-temporal mobility patterns for seniors, child/student and adult using smart card data. International Archives of the Photogrammetry Remote Sensing & S, 2014, XL-1(1): 167-172. (0)
[11]
Rietveld P. Non-motorised modes in transport systems: A multimodal chain perspective for The Netherlands. Transportation Research Part D Transport & Environment, 2000, 5(1): 31-36. (0)
[12]
Li Y, Wang X, Sun S, et al. Forecasting short-term subway passenger flow under special events scenarios using multiscale radial basis function networks. Transportation Research Part C Emerging Technologies, 2017, 77: 306-328. DOI:10.1016/j.trc.2017.02.005 (0)
[13]
Yin K, Wang W, Wang X B, et al. Link travel time inference using entry/exit information of trips on a network. Transportation Research Part B, 2015, 80: 303-321. DOI:10.1016/j.trb.2015.07.007 (0)
[14]
Li W Y, Wang Z Y, Qin Y, et al. Analysis the characteristics of Beijing Metro Line 4 passenger flow. IEEE International Conference on Information and Automation. IEEE, 2014: 76-80. https://www.researchgate.net/publication/286573640_Analysis_the_characteristics_of_Beijing_Metro_Line_4_passenger_flow (0)
[15]
Smith B L, Williams B M, Oswald R K. Comparison of parametric and nonparametric models for traffic flow forecasting. Transportation Research Part C Emerging Technologies, 2002, 10(4): 303-321. DOI:10.1016/S0968-090X(02)00009-8 (0)
[16]
Zhu H Y. N-day average volume based time-series analysis for passenger flow of metro. International Conference on Multimedia Information NETWORKING and Security. IEEE Computer Society, 2010: 384-387. https://www.researchgate.net/publication/251980437_N-Day_Average_Volume_Based_Time-Series_Analysis_for_Passenger_Flow_of_Metro (0)
[17]
Zhang L. On large passenger flow operating organizations at metro station. Urban Mass Transit, 2011, 24: 47-90. (0)
[18]
Sun L, Lu Y, Jin J G, et al. An integrated Bayesian approach for passenger flow assignment in metro networks. Transportation Research Part C, 2015, 52: 116-131. DOI:10.1016/j.trc.2015.01.001 (0)
[19]
Zhang Y, Thomas T, Brussel M J G, et al. Expanding bicycle-sharing systems: Lessons learnt from an analysis of usage. Plos One, 2016, 11(12): e0168604. DOI:10.1371/journal.pone.0168604 (0)
[20]
Shneiderman B, Aris A. Network visualization by semantic substrates. IEEE Transactions on Visualization & Computer Graphics, 2006, 12(5): 733-740. (0)
[21]
Chu K K A, Chapleau R. Augmenting transit trip characterization and travel behavior comprehension. Transportation Research Record Journal of the Transportation Research Board, 2010, 2183(1): 29-40. DOI:10.3141/2183-04 (0)
[22]
Tao S, Corcoran J, Mateo-Babiano I, et al. Exploring Bus Rapid Transit passenger travel behaviour using big data. Applied Geography, 2014, 53(53): 90-104. (0)
[23]
Sun Y, Shi J, Schonfeld P M. Identifying passenger flow characteristics and evaluating travel time reliability by visualizing AFC data: A case study of Shanghai Metro. Public Transport, 2016, 8(3): 1-23. (0)
[24]
Liao C F, Liu H X. Development of data-processing framework for transit performance analysis. Transportation Research Record Journal of the Transportation Research Board, 2010, 2143(1): 34-43. DOI:10.3141/2143-05 (0)
[25]
Kwan M P. Interactive geovisualization of activity-travel patterns using three-dimensional geographical information systems: A methodological exploration with a large data set. Transportation Research Part C Emerging Technologies, 2000, 8(1): 185-203. (0)
[26]
Lyman K, Bertini R L. Using travel time reliability measures to improve regional transportation planning and operations. Transportation Research Record Journal of the Transportation Research Board, 2008, 2046: 1-10. DOI:10.3141/2046-01 (0)
[27]
Yang S. Estimating freeway travel time reliability for traffic operations and planning. The University of Arizona, 2016. (0)
[28]
Lint J W C V, Zuylen H J V. Monitoring and predicting freeway travel time reliability: Using width and skew of day-to-day travel time distribution. Transportation Research Record Journal of the Transportation Research Board, 2005, 1917(1): 54-62. DOI:10.1177/0361198105191700107 (0)
[29]
Chow A H F, Li S, Zhong R. Multi-objective optimal control formulations for bus service reliability with traffic signals. Transportation Research Part B Methodological, 2017, 103: 248-268. DOI:10.1016/j.trb.2017.02.006 (0)
[30]
Federal Highway Administration. Travel Time Reliability: Making It There On Time, All the Time. www.ops.fhwa.dot.gov/publications/tt_reliability/TTR_Report.htm. Accessed Jul. 14, 2007. (0)
[31]
Hendren P, Ph D, Antos J, et al. Transit Travel Time Reliability : Shifting the Focus from Vehicles to Customers. Transportation Research Board 94th Annual Meeting. 2015. https://doi.org/10.3141/2535-04. (0)
[32]
Chan, Joanne. Rail Transit OD Matrix Estimation and Journey Time Reliability Metrics Using Automated Fare Data. Massachusetts Institute of Technology, 2007. (0)
[33]
Sun Y, Xu R. Rail transit travel time reliability and estimation of passenger route choice behavior: analysis using automatic fare collection data. Transportation Research Record Journal of the Transportation Research Board, 2012, 2275(1): 58-67. DOI:10.3141/2275-07 (0)
[34]
Li Z, Wang W, Yang C, et al. Bicycle mode share in China: A city-level analysis of long term trends. Transportation, 2017, 44(4): 1-16. (0)
[35]
Pu Z, Li Z, Zhu W, et al. Evaluating safety effects of variable speed limit system using empirical bayesian before-after analysis. Presented at 96th Annual Meeting of the Transportation Research Board. Washington, D.C., 2017. https://trid.trb.org/view/1439331 (0)
[36]
Nanjing Statistics Bureau. Urban Public Transport. http://221.226.86.104/file/nj2004/2016/quxian/index.htm. 2016. (0)
[37]
Nanjing Urban Planning Bureau. http://www.njghj.gov.cn/NGWeb/Page/Index.aspx. 2009. (0)
[38]
Nanjing Urban Planning Bureau. http://www.njghj.gov.cn/NGWeb/Page/Index.aspx. 2012. (0)
[39]
Zhao J, Deng W. Relationship of walk access distance to rapid rail transit stations with personal characteristics and station context. Journal of Urban Planning & Development, 2013, 139(4): 311-321. (0)
[40]
Zhao J, Wang J, Deng W. Exploring bikesharing travel time and trip chain by gender and day of the week. Transportation Research Part C, 2015, 58: 251-264. DOI:10.1016/j.trc.2015.01.030 (0)
[41]
Sokal R R, Braumann C A. Significance tests for coefficients of variation and variability profiles. Systematic Zoology, 1980, 29(1): 50-66. DOI:10.2307/2412626 (0)