Ship trajectory prediction based on time-aware graph convolutional network
According to the characteristics of irregular time intervals of ship track data,a ship trajectory prediction method based on time-aware graph convolutional network(T-GCN)was proposed.This method converted the time-series trajectory data of ships into a graph structure,and combined an improved Gaussian kernel function to construct a spatiotemporal weighted adjacency matrix based on the time distance and spatial Euclidean distance between graph nodes.It explicitly modeled the spatiotemporal correlation strength between non equidistant time interval sampling points in order to capture the spatiotemporal dynamic relationship of ship trajectory data.By using graph convolutional networks and self-attention mechanisms,the spatiotemporal features of nodes in the graph were extracted and weighted based on the spatiotemporal weighted adjacency matrix,enabling the model to focus on key spatiotemporal information of ship trajectories.Simulation experiments were conducted on publicly available data collected by the ship automatic identification system(AIS),and the results show that compared with the method based on recurrent neural networks,the proposed method significantly improves prediction accuracy.
trajectory predictionirregular time intervalweighted adjacency matrixgraph convolutional networkself-attention mechanism
万方数据
维普
