Journal of Geo-information Science ›› 2021, Vol. 23 ›› Issue (7): 1185-1195.doi: 10.12082/dqxxkx.2021.200334

Previous Articles     Next Articles

Big-data Oriented Commuting Distribution Model and Application in Large Cities

LIU Yunshu1,2,3(), ZHAO Pengjun1,2,3,*(), LV Di1,2,3   

  1. 1. Shenzhen Graduate School, Peking University, Shenzhen 518055, China
    2. College of Urban and Environmental Sciences, Peking University, Beijing 100871, China
    3. Laboratory for Earth Surface Processes(LESP) Ministry of Education, Peking University, Beijing 100871, China
  • Received:2020-06-29 Revised:2020-07-26 Online:2021-07-25 Published:2021-09-25
  • Contact: ZHAO Pengjun E-mail:yunshu.liu@pku.edu.cn;pengjun.zhao@pku.edu.cn
  • Supported by:
    National Natural Science Foundation of China(41925003)

Abstract:

In recent years, big data has been widely applied in traffic analysis. However, they are mostly used for data visualization and phenomenon description. There is a lack of big-data oriented transport modeling, which leads to limited application of big-data in transportation planning. In this study, we propose a Location-Space Dependent Indicator (LSDI) based on the time-space interaction between transportation and land use. Based on this indicator, the urban commuting distribution model is developed, which improves the traditional gravity model. Taking Beijing as a study case, the developed model is applied and verified using mobile phone signaling big data derived from the communication service of an operator in September 2017. Travel generation and distribution models are constructed and verified respectively. Our results show that: (1) For the travel generation model simulations, commuter population and resident population show a good linear relationship. This model generates a significant prediction with a goodness of fit of 0.84; (2) For the travel distribution model simulations, a comparison analysis is conducted between gravity model, radiation model, and modified model with LSDI. The gravity model corrected by real commuting data performs best in regression analysis with a goodness of fit of 0.94. But large errors occur in the probability density distribution. The radiation model performs normal in regression analysis with a goodness of fit of 0.37. It has a better accuracy in the probability density distribution. The modified gravity model with LSDI has the best overall performance. The underestimation phenomenon is optimized in the commuter population distribution with a highest goodness of fit (0.85). Our findings provide new insights in developing big-data oriented transport prediction models and contribute to promote the application of big data in transport planning.

Key words: big-data oriented model, commuting transport model, trip distribution, location-space dependence indicator, mobile phone signaling data, transport planning, commuting OD, model simulation