Two-Stream Spatial-Temporal Attention Network for Inland Multi-Bridge Waterway Traffic Prediction

Ship traffic volume reflects inland river waterway's busyness and safety. Accurate prediction aids China's inland marine equipment industry in consignment, enabling departments to arrange resources for smooth shipping. However, existing methods oversimplify, ignoring waterway facility diversity and only considering temporal correlation, resulting in low accuracy. We propose a two-stream spatial-temporal attention network (TS-STAN) for inland multi-bridge waterway traffic prediction, considering traffic speed. We subdivide the waterway into regions, build a spatiotemporal matrix, and design a network to learn traffic characteristics and correlations. Our TS-STAN model, tested on Yangtze River data, achieves lower MAE and RMSE compared to existing studies.