Dynamic Collaborative Optimization Framework for Ship Trajectory Prediction: Dynamic Noise Suppression and Spatiotemporal Modeling

Aiming at the problems of significant sensor noise interference and complex spatiotemporal feature correlations in ship trajectory prediction tasks, this study proposes an attention mechanism-enhanced Kalman filter-gated recurrent unit (GRU) fusion prediction model (AKalman-GRU). The model constructs a multi-stage collaborative processing framework: first, the Kalman filter algorithm performs dynamic optimal estimation on the original latitude and longitude sequences to suppress GPS positioning noise; subsequently, a deep GRU network with attention mechanism models temporal features to adaptively capture key state changes in navigation patterns. To verify the model effectiveness, this paper designs a hierarchical ablation experiment system to evaluate performance differences among the basic GRU model, Kalman filter-enhanced version, and complete model. Experimental results show that the Kalman filter module significantly reduces high-frequency noise interference in raw data through state-space modeling, while the attention mechanism effectively enhances the model adaptability to complex waterway scenarios, particularly demonstrating outstanding prediction accuracy at key nodes such as port areas. This study innovatively combines Kalman filter theory from signal processing with deep learning models, achieving cross-domain upgrade from basic temporal prediction models to intelligent prediction systems through triple optimization mechanisms of feature enhancement, noise suppression, and attention focusing. Actual deployment tests demonstrate that the model achieves better spatial coincidence between predicted trajectories and real records in complex waterway scenarios compared with traditional temporal prediction methods, providing reliable technical support for situational awareness modules in intelligent navigation systems.