Vessel trajectory prediction model based on spatio-temporal graph convolutional network with spatio-temporal joint attention mechanism

With the development of vessel intelligence and the complexity of the maritime traffic environment, which brings new safety problems to maritime transport. Deep learning based on AIS data has become a research focus in maritime transportation. In this paper, a vessel trajectory prediction method based on the combination of spatio-temporal graph convolutional network and spatio-temporal joint attention mechanism (STA-GCNN) is proposed. Traditional vessel trajectory prediction is mostly based on single vessel historical data, ignoring the influence of interactions between vessels, so the prediction accuracy is insufficient in complex navigation environments. To improve this problem, we adopt a new method for calculating the interaction force between vessels, and we also use a joint spatio-temporal attention mechanism to make the model pay more attention to more important features, which realises the self-dynamic adjustment of the model’s focus. In this paper, experiments are conducted using AIS data from April 2022 in a US sea area. The results show that the method proposed in this paper has improved prediction accuracy in metrics such as Average Displacement Error (ADE) and Final Displacement Error (FDE) compared to existing LSTM-based and other conventional models.