Vessel Trajectory Prediction Based on Context-Assisted Information

In this paper, aiming at the problem of vessel trajectory prediction, a Context-Assisted Long Short-Term Memory Network (CA-LSTM) is proposed to process historical trajectories of vessel navigation as well as contextual information such as water depth, water temperature, wind speed, wind direction, wave height, and wave direction to solve the problem of high accumulated errors and low accuracy of ship trajectory long-term prediction results caused by differences in navigation strategies. The method first utilizes the water depth data, which directly affects vessel navigation, to establish a depth penalty function that constrains the area of vessel trajectory prediction. Subsequently, for other contextual information like water temperature et al that indirectly affects vessel navigation, an encoder-decoder architecture is constructed to extract the implicit features that influence the vessel's trajectory. Finally, experiments conducted on actual AIS datasets have demonstrated that the proposed method possesses superior predictive capabilities compared to other representative vessel trajectory prediction methods.