Data-driven stochastic transit assignment modeling using an automatic fare collection system

摘要

In modern urban transit networks, buses and subways are not distinguished as different modes of transportation; this makes it challenging to analyze travel behaviors with multiple modes for the purpose of developing policies and plans. With the introduction of Automatic Fare Collection (AFC) systems, these modes are operated along a complex of links and nodes that constitute a multimodal transit network. Methods for analyzing travel behaviors in mass transit have been developed, but previous approaches fail to adequately reflect travel behaviors and network features (e.g., transfers, mode and route preferences). To overcome such limitations, this research proposes a smart card data-based analytical method with which travel behaviors can be efficiently and accurately examined. AFC systems provide a tremendous amount of data that contain detailed trip information, and using these data reinforces the reliability of the proposed data driven method. The proposed method of analysis involves four core processes: establishing a scheme for how multiple transit modes can be integrated into one multimodal transit network on the basis of information derived from the AFC system, selecting feasible paths, assigning trips using a stochastic approach, and verifying analytical results by comparing them with findings from trip datasets. This method was used to analyze monthly smart card data collected from the AFC system in 2009 in the greater Seoul area. Multimodal transit networks were constructed from 34,852 bus stops and 539 subway stations using smart card data, and in total, 3,614,875 trips were used in the analysis. The final model for stochastic transit assignment was developed using the proposed method, which was verified by comparing actual and assigned trips. The proposed method exhibits high accuracy (83.93%) and a high R-square value (0.981), which supports the strength of the proposed stochastic transit assignment model. The findings reveal new interesting research directions for exploration, such as developing more disaggregated models (e.g., for specific regions, times, and users), considering detailed transfer features (e.g., transferable boundaries, transfer facilities, and transfer times), confirming the method's applicability by testing it in other cities, and incorporating both multimodal transit and road networks into the proposed model.