Analyze job-housing relationship by using smart card data and household travel survey

City: Beijing, China
Organization: Beijing Institute of City Planning,
Department of Geography and Earth Sciences, The University of North Carolina at Charlotte
Project Start Date: 2015
Project End Date: 4 April, 2015
Reference: Ying Long, Jean-Claude Thill, Combining smart card data and household travel survey to analyze jobs–housing relationships in Beijing,Computers, Environment and Urban Systems, Volume 53, 2015, Pages 19-35, ISSN 0198-9715, https://doi.org/10.1016/j.compenvurbsys.2015.02.005.
Problem: In recent years, urban transportation is becoming a pressing concern in mega-cities. For example, Beijing, one of the international metropolises in China, has over 21 million residents. Due to the huge population in Beijing, city planners and transit system managers face great challenges. They need to understand the pattern of transit usage and to provide a better transportation system. City planners always use conventional travel behavior surveys as a primary source to infer commuting trips. However, smart card data (SCD) have been underestimated for analysis of urban structure. This paper seeks to examine the potential use smart care data(SCD) in Beijing as a case study to answer the following questions:
  1. What is the relationship between job and home location in Beijing?
  2. What kind of information could smart card data provide?
  3. How does the smart card dataset process and generate a story about the job-home relationship?
  4. Could smart card data give us the same value of insight (or better) as conventional travel behavior surveys?
Technical Solution:
  1. Data pre-processing and form
    • Link the bus stop ID in the smart card data with the bus stop layer in GIS
    • Merge the trips from each card ID to retrieve full bus diary
    • Data form 1: TRIP={departure location(OP), time(OT), arrival stop(DP), time(DT)}
    • Data form 2: position-time-duration (PTD) = {P, t, D} Notes: P is the bus stop which the cardholder stays to do activity, t is the start time at the location, D is the temporal duration at this spot.
  2. Training dataset - Identification of home and work location (daily)
    • Using the PTD data
    • Set rules that for non-student cardholder, the work place is been identified as if more than 360 mins is spent at one location other than their first location(Home location)
  3. Training dataset - Identification of home and work location (weekly)
    • Decision tree and clustering algorithm: The decision tree has been applied to combine the one-day results to retrieve one-week results. Clustering also has been applied in this step, and it sets a threshold at 500m to encompass the home locations/job locations that are within the threshold of each other. If a single cluster has the largest number of locations, and then the home location/job location can be indicated as the most frequent location in this cluster.
  4. Testing dataset - Identified home-job locations with extreme commuting times are dropped because these cases may have bias or error.
  5. Python tool bases on ESRI Geoprocessing
Datasets Used:
  • Dataset 1: Bus routes, bus stops and traffic analysis zones (TAZs) of Beijing
  • Dataset 2: One-week smart card dataset (one-week period of 2008 from April 7 to April 13) from Beijing Municipal Administration & Communication Card Co. Ltd
  • Dataset 3: 2005 Beijing travel behavior survey
  • Dataset 4: Land-use pattern in Beijing
Outcome: The raw SCD contains 8,549,072 cardholders. After the model processed, only 221,772 cardholders had been identified both their final home and job location identified with a commuting trip. The average duration has been calculated as 36 minutes and standard deviation is 24.2 minutes. Meanwhile, the average commuting distance is 8.2 km. After comparing the SCD results with the 2005 survey, the two cumulative distribution functions of commuting trips were found to generally overlap. However, the t-test reveals a significant difference. The results indicate that the SCD is able to visualize the commuting trips for the whole region and provide some valuable insights. For example, residents of Tiantongyuan have much shorter commutes than Tongzhou. The test of SCD gives a promising commuting pattern and offers a new approach for monitoring commuting issues in a mega city in addition to conventional travel surveys.
Issues that arose:
  1. This smart card dataset only contains bus riders (other modes such as subway or taxi are excluded).
  2. The smart card data is devoid of information of the cardholder’s socioeconomic and the purpose of individual trips.
  3. Passengers who paid cash for fare are not counted.
Status: In Development
Entered by: Oct 26, 2020: Cheng Wang, andycheng.wang@mail.utoronto.ca


CEM1002,
Civil Engineering, University of Toronto
Contact: msf@eil.utoronto.ca