Research on ship navigation risk prediction model integrating multi-source data features and self-attention mechanism

This is the authors' abstract. We don't add key points for this paper.

In order to detect potential risks of abnormal AIS data in ship navigation and provide reliable decision-making approaches to relevant management departments, this study combines multi-source data features and self-attention mechanism to predict navigation risk of ship trajectories in five categories of abnormal AIS data. Firstly, AIS data, meteorological data, current data, and topographic conditions of abnormal trajectories are preprocessed to obtain eight Risk Influence Factors (RIFs). These indicators are synthesised to form a risk characteristic dataset. Then, a Multi-Head Self-Attention Mechanism Model (MHSAM) is built to calculate the weights and membership degrees of different RIFs, and the corresponding navigation risk coefficient is obtained. By using different categories of abnormal trajectories for navigation risk prediction, the results are compared with Long-Short Term Memory (LSTM), Recurrent Neural Network (RNN), Transformer and other models. The results show that the Mean-Square Error (MSE) and Mean Absolute Error (MAE) values of the MHSAM model are the lowest, with 0.00126 and 0.0146 respectively, indicating the effectiveness of the proposed model. The results can provide a theoretical basis for maritime regulatory authorities to ensure safe navigation of ships.