Spatiotemporal Attention Fusion Network for Short-Term Passenger Flow Prediction on New Year’s Day Holiday in Urban Rail Transit System

The short-term passenger flow prediction of the urban rail transit (URT) system is of great significance for traffic operation and management. Emerging deep learning-based models provide effective methods to improve prediction accuracy. However, most of the existing models mainly predict the passenger flow on general weekdays and weekends. Only a few studies focus on predicting the passenger flow on holidays, which is a significantly challenging task for traffic management because of its suddenness and irregularity. To this end, we take passenger flow prediction in the URT system during the New Year’s Day holiday as an example to study passenger flow prediction on holidays in depth. We propose a deep learning-based model, Spatial–Temporal Attention Fusion Network (STAFN), for short-term passenger flow prediction in the URT system during New Year’s Day, which includes a novel multigraph attention network (MGATN), convolution–attention (conv–attention) block, and feature fusion block. The MGATN is applied to extract the complex spatial dependencies of passenger flow dynamically, and the conv–attention block is applied to extract the temporal dependencies of passenger flow from global and local perspectives. Moreover, in addition to historical passenger flow data, social media data, which have proved that they can effectively reflect the evolution trend of passenger flow during events, are fused into the feature fusion block of STAFN. STAFN is tested on two large-scale URT automatic fare collection system datasets from Nanning, China, on New Year’s Day, and the prediction performance of the model is compared with that of several basic and advanced prediction models. The results demonstrate better robustness and advantages of STAFN among benchmark methods, which can provide overwhelming support for practical applications of short-term passenger flow prediction on New Year’s Day.