Estimation of Urban Road Network OD Matrix Based on Graph Convolutional Network

Abstract

The Origin-Destination (OD) matrix is important in the intelligent traffic system, but most of the estimators do not predict the spatiotemporal dependence of traffic. In the current state of the art of deep learning, this study utilizes new neural network-based traffic forecasting methods, Graph Convolutional Networks (GCN) and temporal graph convolutional network (T-GCN). This paper uses taxi data from the New York City Yellow Taxi Trip Record in February 2025. Based on the prediction performance of both models, the trends obtained by these models fit the original data, which supports the application of the proposed models to passenger flow prediction. The GCN model is the best of the two available models as well as the best in terms of the spatial and temporal dependence, which shows that both modeling spatial and temporal dependence is necessary for accurate OD flow prediction. This provides a powerful data-driven tool in urban traffic analysis and provides information about travel demand patterns.