KGCN‐LSTM: A graph convolutional network considering knowledge fusion of point of interest for vehicle trajectory prediction

Abstract

Abstract Urban vehicle trajectory prediction positively alleviates traffic congestion, avoids traffic accidents, and optimizes the urban transportation system. Since taxi trajectories are influenced by the driving intention, it is significant to consider the Points of Interest (POI) as the spatial features for trajectory prediction. A Knowledge Graph Convolutional Network Long Short‐Term Memory (KGCN‐LSTM) model is proposed here to improve the accuracy and robustness of trajectory prediction. POI information is considered as the prior‐knowledge of the trajectory by the Graph Convolutional Network (GCN). Under multiple comparison experiments, Shopping POI gains the highest positive effect weight of 15% in holidays, and Hospital POI gains the highest weight of 16% in working days. In holidays, higher accuracy and robustness are achieved compared with benchmarks when performing the KGCN‐LSTM model with POI of shopping, food, life service, scenic spots, and entertainment classes, while the performance is not improved with the rest of the POI classes. In working days, higher accuracy and stronger robustness are achieved compared with benchmarks when performing the KGCN‐LSTM model with POI of hospital, life service, and exercise. While the performance is not improved with the rest of the POI classes.

Keywords

• feature fusion
• graph convolutional network
• map‐matching
• prior knowledge
• trajectory prediction