Multi-Ship Trajectory Prediction Framework for High-Density Maritime Traffic Areas based on BI-GRU and GAT

The accelerated growth of the maritime industry has led to an increasing number of collision incidents, not only resulting in human casualties, but also giving rise to severe issues like environmental pollution. In light of this, the prediction of vessel trajectories based on Automatic Identification System (AIS) data has emerged as a pivotal element in maritime traffic services and averting maritime navigation collisions. Current research on ship trajectory prediction primarily relies on the individual vessel's historical trajectory data to determine its future trajectories. failing to adequately account for the dynamic interaction of ships. Consequently, this will result in a decline in the accuracy of the predicted outcomes. Because, in regions of dense maritime traffic, ship interaction behavior is a key factor influencing changes in ship trajectories. To address this problem, we propose a sequence-to-sequence vessel trajectory prediction framework, named BIGRU-GAT, based on Graph Attention Neural Networks (GAT) and Bidirectional Gated Recurrent Units (BIGRU). This model utilizes a spatiotemporal encoder module to capture and encode the spatiotemporal information of the surrounding vessels intricately. Subsequently, the decoder processes these captured spatiotemporal hidden states and maps them to the predicted trajectory points. The evaluation results, derived from datasets harvested from two maritime areas characterized by intense traffic in the real world, substantiate that the predictive precision of our proposed BIGRU-GAT model outperforms that of other existing multi-vessel trajectory prediction models. To best of our knowledge, this is the first attempt to integrate GAT with BIGRU for ship trajectory prediction in densely crowded marine traffic zones.