Vol 77 No 3 77309

Abstract

Accurate prediction of vessel traffic flow is crucial for ensuring the safety of inland river shipping and enhancing the efficiency of traffic operations. Inland vessel traffic flow typically exhibits significant complexity and spatio-temporal dynamic characteristics. To address these challenges, this paper proposes a Global-Local Spatiotemporal Transformer (GL-STFormer) deep learning model. The Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) algorithm is utilized to decompose the original data into multi-feature inputs, effectively mitigating data non-stationarity. The model integrates Gated Recurrent Units (GRU) with a self-attention mechanism to extract temporal features of traffic patterns. The multi-head attention and local masking mechanisms of the Transformer model are employed to extract global and local spatial dependencies. Furthermore, the Whale Optimization Algorithm (WOA) is applied to optimize the model’s hyperparameters. This study employs real-world Automatic Identification System (AIS) data from the Nantong waters of the Yangtze River for experimental validation. The results show that the proposed method significantly outperforms various baseline models in inland vessel traffic flow prediction. This study provides scientific support for precise traffic prediction and offers novel insights for the intelligent development of dynamic waterway traffic management.