Forecast on Bus Trip Demand Based on ARIMA Models and Gated Recurrent Unit Neural Networks

摘要

Bus is the most basic trip mode in public transport system. Precise bus trip generation forecast indicates the short-term number of passengers in each bus station, providing passengers information to optimize trip strategy. The ARIMA models and gated recurrent unit neural networks were developed to predict bus trip demand in different intervals including the 30-minute, 1-hour, 2-hour and 4-hour intervals based on the I C card data in Shenzhen, China. Time series data was structured by the IC card record, where one record was viewed as a passenger in this station. The comparison results suggested that the GRU NN models provide better prediction accuracy than the ARIMA model for different time intervals. The developed GRU NNs can be further used to guide the bus network optimization and urban planning.