A ship trajectory prediction method based on GAT and LSTM

Ship trajectory prediction plays an important role in ship route planning and collision avoidance in the development of autonomous ships. Previous models related to ship trajectory prediction have mainly focused on exploiting spatial and temporal correlation, but the accuracy and reliability of their predictions may be limited. To address this issue, this study introduces a graph attention network (GAT) and long short-term memory (LSTM) to predict ship trajectories. First, a graph network of ship trajectories is constructed based on the dependency relationship between ship trajectory data. GAT-LSTM uses GAT to extract the spatial features of ship trajectory data, while LSTM is introduced to learn the temporal features of ship trajectory data; finally, the prediction results are obtained. In this study, three real ship trajectory datasets from the AIS are used to verify the effectiveness of the proposed model and compare it with other prediction models. The experiments show that GATLSTM always obtains better values of the evaluation metrics than other prediction models. The model proposed in this study has high robustness in ship trajectory prediction, and the accurate prediction of ship trajectories has positive significance for maritime traffic control and safe navigation of ships.