Short-term Trajectory Prediction of USV Based on Multi-source Data

In recent years, research on Unmanned Surface Vessel (USV) trajectory prediction has gained attention due to its potential to reduce voyage time and eliminate collision hazards in maritime transportation. Deep learning, particularly Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM), has shown promise in this field. CNN is effective at extracting spatial features from time series data, while LSTM is adept at handling long-term dependencies. The combination of these models can enhance prediction efficiency and accuracy. This paper puts forward a data-driven deep learning approach to USV trajectory prediction, utilizing a CNN+LSTM model. Tests conducted with simulated unmanned vessel data from three trajectories revealed a reduction in RMSE errors by 60.08% and 37.15% in comparison to LSTM and CNN methods, respectively, which indicates that this model can fully consider the influence of external environmental factors on the system and verifies the effectiveness of the USV trajectory prediction model.