TrAISformer: Spatio-Temporal Ship Trajectory Prediction Based on Transformer

Spatio-temporal ship trajectory prediction is an important issue for maritime traffic management and safety. This research introduces a pioneering technique predicated on a Transformer architecture, specifically designed for the precise forecasting of maritime vessels' paths in forthcoming intervals. Utilizing archival navigational data as the foundational input, this study's model integrates both chronological sequences and spatial-temporal interrelations by employing an advanced multi-head self-attention mechanism paired with positional encoding strategies. The empirical findings indicate a marked elevation in predictive accuracy for vessel course forecasting, surpassing conventional methodologies, especially in response to dynamic maritime route alterations and the intricacies of marine settings. This investigation contributes a robust analytical framework to the domain of maritime spatio-temporal trajectory forecasting, with anticipated superior utility in empirical seafaring contexts.