Urban-Road Average-Speed Prediction Method Based on Graph Convolutional Networks

Abstract

Prediction of urban-road average speed is an essential part of an intelligent traffic-information-service system and provides important support for an intelligent traffic-control and -management system. This paper takes an actual urban regional road network as the research object, constructs the road-network spatial weight matrix, and uses the cross-correlation function to analyze the temporal and spatial correlation of the average speed of urban roads. Then, the study builds an average-speed prediction model based on graph convolutional network and convolutional neural network (GCN-CNN). The average speed for every 5 min in the next 5 days is predicted. Finally, the proposed GCN-CNN model is compared against autoregressive integrated moving average (ARIMA), back-propagation neural network, CNN, and CNN with long short-term memory. The root mean square error, mean absolute error, and weighted mean absolute percentage error are used to evaluate the prediction accuracy. The evaluation results confirm the superior prediction accuracy and applicability of the proposed GCN-CNN model. This study provides traffic managers with a decision-making basis for predicting traffic accidents and alleviating traffic congestion. Also, it is a useful supplement to the technology of urban-road traffic-congestion control.