A Method of Vehicle Interactive Information Drive Speed Prediction Based on Temporal Dynamic Graph Convolutional Neural Network

Abstract

Accurate and efficient speed prediction is crucial in autonomous driving or intelligent driving assistance systems to improve vehicle trajectory prediction accuracy, optimal decision planning, and energy management strategies. However, the traffic scene constructed by ego-vehicle and surrounding vehicles is a dynamic process, its complex spatial-temporal characteristics increase the complexity and challenge of predicting vehicle speed. This paper proposes a new method to solve the problem of using spatial-temporal characteristics in the process of vehicle speed prediction. The graph structure is used to describe the vehicle interaction scene and reflect the spatial relationship between vehicles. A temporal dynamic graph convolutional network (TDGCN) is proposed to predict vehicle speed by using temporal and spatial characteristics. The network combines graph convolutional neural network (GCN) and long short-term memory neural network (LSTM). GCN can process the complex topological structure in graph data to extract the spatial characteristics of traffic scenes, and LSTM can process the temporal characteristics of dynamic changes in traffic scenes. Finally, the TDGCN model is used to predict the vehicle speed based on the real driving dataset NGSIM I-80. The simulation results show that when predicting the speed after 1 frame and 10 frames, the root means a square error of the prediction results is 2.051 and 2.086, and the acceleration and deceleration changes of the vehicle can be correctly reflected, which proves the effectiveness of the TDGCN model in predicting the vehicle speed by using the spatial-temporal characteristics.