This paper studies the short-term passenger flow forecast of Urban Rail Transit Based on the subway hourly passenger flow data released by the metropolitan transit Administration (MTA) of New York City. Firstly, the missing and abnormal values from February 2022 to September 2023 were processed, and exploratory analysis was carried out from the time and station dimensions to extract the characteristics of passenger flow with significant morning and evening peaks, obvious differences between weekdays and weekends, and high concentration of core hub stations. On this basis, the linear ARIMA model and the deep learning LSTM model are constructed to compare their short-term prediction performance. The results show that LSTM is better than Arima in Mae, RMSE, R² and other indicators, which can better describe the nonlinear fluctuation and periodicity of passenger flow sequence, and provide more reliable data support for intelligent Metro capacity scheduling and passenger flow management. In the future, weather, holidays and other external factors will be further introduced to build a more real-time and transplantable passenger flow forecasting framework for urban rail transit by combining with spatio-temporal modeling methods such as graph neural network.
Zixuan Sun (Mon,) studied this question.