Estimating Left Behind Passengers and Train Loads in Congested Urban Rail Transit Networks: A Data-Driven Passenger-to-Train Assignment Approach

This study presents a data-driven approach for assigning passengers to individual trains. The approach is based on the correlation between the passenger tap-in/out time and train arrival/departure time. We classify passengers into three types according to the number of transfers. First, we propose an inference algorithm to obtain sets of feasible trains that passengers can take on each line. Subsequently, we use the maximum likelihood approach to estimate the egress walking time parameters of each station. We assume that these parameters follow a normal distribution. The estimation depends on the egress walking time of the reference passenger. We assume that passengers take the train with the smallest distance between the egress walking time and mathematical expectation of the destination station. By removing unreasonable trains, this study determines the number of times passengers are left behind owing to in-vehicle congestion. We then propose an assignment approach for transferring passengers based on the reference transfer time. Finally, we apply the approach to a case study of Nanjing, China, during morning peak hours. This study provides a novel method for studying passenger behavior at the microcosmic level.