Research on Trajectory Prediction Based on Ship Motion Behavior and Temporal Graph Neural Network
With the promotion of the"Marine Power"strategy in China,the shipping industry is rapidly developing.Rapid growth in maritime traffic leads to frequent ship collisions.There is an urgent need for the long-term prediction of ship trajectories based on accumulated ship navigation data to enhance the awareness of maritime traffic conditions and reduce collision rates.Firstly,the paper pre-processes the automatic identification system(AIS)data to eliminate easily removable outliers,and proposes a dynamic trajectory decorrelation method based on ship characteristics to remove decorrelation points.Then,a self-adaptive spatiotemporal traj-ectory clustering method with behavioral semantic constrains based on the characteristics of ship navigation is proposed to mine ship motion patterns and obtain typical ship behavior patterns.Finally,for the ship trajectories and motion patterns,a motion pattern-based temporal graph neural network model is presented.For long-term prediction of ship trajectories and patterns,the Guangdong-Hong Kong-Macao Greater Bay Area is selected as a test region,comparative experiments validate that the proposed model outper-forms the traditional models in long-term prediction.
AIS dataship trajectory predictiontrajectory preprocessingtrajectory clusteringgraph neural networkgated re-current unit
万方数据
维普
