An Urban Traffic Knowledge Graph-Driven Spatial-Temporal Graph Convolutional Network for Traffic Flow Prediction
Chengbiao Yang, Guilin Qi
Abstract
Traffic flow prediction is a critical issue for researchers and practitioners in the field of transportation. Due to the high nonlinearity and complexity of traffic data, deep learning approaches have attracted much interest in recent years. However, existing studies seldom consider the topology of these urban roads and the connectivity of the monitor sensors. As we know, the real cause of the spread of traffic congestion is the connectivity of these road segments, rather than their spatial proximity. But it is challenging to model the dynamic topology of the urban traffic networks for traffic flow prediction. In this vision paper, we present an urban traffic knowledge graph-driven spatial-temporal graph convolutional networks for traffic flow prediction. We first construct an urban traffic knowledge graph that can represent the physical connectivity between roads and monitor sensors. Then, we use the urban traffic knowledge graph to improve the traffic flow networks. Finally, we combine the knowledge graph and traffic flow as the input of a spatial-temporal graph convolutional backbone networks. Experiments on two real-world traffic datasets verify the effectiveness of our approach.