Short-Term Railway Passenger Flow Prediction Using Deep Learning: Model Construction and Empirical Study

Abstract

Aiming at the problems that short-term railway passenger flow is random and sudden, and easily disturbed by multiple external factors, traditional prediction methods have insufficient accuracy and poor multi-source feature fusion effect, this paper carries out research on short-term passenger flow prediction of high-speed railway. Based on the multi-source operation data of a high-speed railway hub station in eastern China, a Transformer-GRU hybrid model fused with multi-source features is constructed, and data anomaly processing module is designed to optimize data quality. The GRU is used to extract the short-term time series dependent features of passenger flow, and the self-attention mechanism of Transformer is used to realize the adaptive fusion of multi-source features. In addition, the lightweight design of the model is carried out. The experimental results show that the MAPE of the model on the test set is 6.24%, which is significantly better than the AutoRegressive Integrated Moving Average Model (ARIMA) and the single LSTM/GRU model. It has prominent prediction advantages in the passenger flow mutation scenario, good robustness under data missing conditions, and the single-step prediction time is ≼ 0.08s, which can meet the real-time and accurate requirements of railway operation and provide technical support for railway intelligent dispatching.