Utilizing a data-driven methodology to resolve the passenger-to-train assignment problem

Abstract

Understanding the passenger distribution within the metro system is a prerequisite for metro network planning and operation. However, as automatic fare collection (AFC) data records only entry and exit information, directly obtaining passenger distribution through AFC data and established timetables remains challenging. Although many studies have explored passenger distribution in metro systems based on accurate timetable inputs, parameter collection and calibration are challenging due to the spatiotemporal dynamics of both passenger demand and headway. This study proposes a data-driven passenger-to-train assignment model (PTAM). The posterior probability of passengers boarding the train is computed using a two-stage Gaussian mixture model (GMM). This method does not require precise timetable inputs, and both the initial parameter collection and final estimation processes are automated, eliminating the need for manual calibration. Using the Nanjing metro as a case study, the effectiveness of the PTAM is demonstrated. Additionally, the study computes in-vehicle passengers, left-behind passengers, and passengers’ willingness to pay (WTP) using PTAM. The results demonstrate significant differences in crowding level and left-behind at different stations on the same line. During the evening peak, passengers bear about 50% of welfare costs. The findings can provide managers with a basis for passenger flow organization and guidance for passenger’s travel decision.