Research on Offshore Vessel Speed Prediction Based on TCN-Transformer-CrossAttention

Abstract

Accurate speed prediction of maritime vessels is critical for navigation safety, traffic management, and collision avoidance, particularly in busy nearshore areas where complex spatial-temporal dependencies challenge traditional prediction models. To overcome the limitations of conventional single neural networks in feature extraction, this paper proposes a novel hybrid deep learning model that integrates Temporal Convolutional Network (TCN), Transformer, and Cross-Attention mechanisms for enhanced vessel speed prediction using Automatic Identification System (AIS) data. The TCN module captures local temporal patterns, while the Transformer models long-range dependencies, and the Cross-Attention mechanism effectively fuses spatial and temporal features to improve prediction robustness. Furthermore, we introduce an optimized feature fusion strategy to dynamically weigh multi-scale features, ensuring better representation learning. Through extensive experiments and hyperparameter tuning, our model achieves significant improvements over baseline methods in key metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). The proposed framework not only advances the state-of-the-art in maritime speed prediction but also provides a practical solution for enhancing maritime traffic efficiency and safety management.