Analysis of urban subway passengers' detention behavior based on big data

[Objective] As an irreplaceable green travel mode, urban subway involves considerable daily travel. However, the passenger flow of subway systems has recently increased remarkably, resulting in overcrowding in space and time. Many trains operate at the highest frequency and maximum capacity during peak hours and still cannot meet passenger demand. Thus, detention is becoming increasingly severe, affecting the basic performance indicators of the subway systems while reducing the service quality and reliability perceived by subway passengers. However, research on the behavioral mechanism of detention is currently inadequate. Based on bus smart card data, supplemented by data mining technology, this study aims to quantify the difference between the rational travel time and the actual travel time of passengers. It reproduces the waiting process of each passenger as realistically as possible and predicts the occurrence and duration of the detention. The study provides a reference with application value for improving the future subway operation status evaluation index system and lays a research foundation for refined passenger flow distribution simulation, depending on the individual level of passengers. [Methods] The bus smart card data record the spatiotemporal information of each subway passenger's travel start and end station at a low cost. Furthermore, data mining technology is used to mine the travel information of passengers from the bus smart card data. This study uses the Shenzhen Tong data from september 2019. First, a suitable pretreatment method is adopted for the problems existing in the original data. Through an analysis of the fluctuation of each time component of the travel time, the travel time is redivided into the entry time, waiting time, in-car time, and outbound time. Second, the rational travel time value of the passenger is obtained by estimating the time of each part individually. Then the passenger's detention is identified based on the maximum rational travel time of the passenger. The influencing factors of detention are excavated from the three levels of individuals, stations, and lines. Frequent passengers are selected as research objects, and nonequilibrium panel travel data are used. Finally, a panel logistic regression model and panel data random effect regression model are established to achieve a prediction of detention and detention time. [Results] The study results, based on Shenzhen Metro Line 1, showed that the average detention rate of each station was 12.0%, the detention time was concentrated in 1.00 8.00 min, and the average detention time was 4.20 min. The travel time in the downward direction of the evening peak and the upward direction of the morning peak increased by 0.61 and 0.88 min, respectively, while the travel time in the downward direction decreased. The study also found that long-distance passengers had longer detention time than others, and they were willing to sacrifice their time in exchange for a comfortable ride. They had more choices at the starting station and were willing to wait for one or more trains to obtain seats. However, the passengers who took the subway less had cognitive bias, leading to the detention time increased. [Conclusions] Detention is a common problem during the morning and evening rush hours, which affects a passenger's travel time. This research clarifies the mechanism of detention behavior, provides a basis for generating operation management plans and issuing customized guidance strategies and is conducive to maximizing the operational efficiency of a subway transit system. © 2023 Press of Tsinghua University. All rights reserved.