Litcius/Paper detail

Traffic Flow Prediction Using Graph Convolution Neural Networks

Anton Agafonov

202033 citationsDOI

Abstract

Traffic flow prediction using spatial-temporal network data remains one of the most important problems in intelligent transportation systems. Timely and accurate traffic prediction is necessary to provide valuable information for different urban planning, traffic control, and guidance tasks. The complexity of the problem is explained by the fact that traffic flows have high nonlinearity and complex spatial-temporal correlations. The development of mathematical models, and, in particular, the deep learning models, allows to use convolutional neural networks to solve traffic prediction problems. In this article, we analyze the architecture of the graph convolution network for traffic flow prediction. The considered graph convolution network takes into account daily and weekly patterns of traffic flow distributions. Experimental studies of the graph neural network model were carried out on the transportation network of Samara city. Experiments show that the considered model outperforms other baseline forecasting algorithms.

Topics & Concepts

Computer scienceGraphConvolution (computer science)Traffic generation modelTraffic flow (computer networking)Data miningArtificial neural networkFlow networkIntelligent transportation systemArtificial intelligenceMachine learningTheoretical computer scienceReal-time computingMathematical optimizationMathematicsComputer networkTransport engineeringEngineeringTraffic Prediction and Management TechniquesTransportation Systems and LogisticsTraffic control and management