Representation learning and Graph Convolutional Networks for short-term vehicle trajectory prediction

Trajectory prediction is an important topic in intelligent transportation research. This study considers in-depth, the diversity and complexity of spatial–temporal trajectory, proposes a feature extraction method based on representation learning, and further designs and implements a short-term spatial–temporal vehicle trajectory prediction model on this basis. First, we propose a feature representation method aiming at the potential preference features of spatial–temporal vehicle trajectory space, so as to accurately represent the vehicle preference features. Second, considering the complexity of the topology of the road Network of vehicles and the fact that (Graph Convolutional Network, GCN) can learn the node features of the network structure, a feature extraction method based on GCN is proposed, which achieves the goal of vectorizing the road Network structure into a low-dimensional dense vector. Finally, a feature fusion method is proposed to comprehensively predict trajectory. Experiments show that this model not only effectively and comprehensively represent the trajectory features, but also synthesize the spatial features, temporal features and trajectory preference features, and finally obtains improved prediction results.