T-GCN: A Temporal Graph Convolutional Network for Traffic Prediction

Abstract

Accurate and real-time traffic forecasting plays an important role in the\nIntelligent Traffic System and is of great significance for urban traffic\nplanning, traffic management, and traffic control. However, traffic forecasting\nhas always been considered an open scientific issue, owing to the constraints\nof urban road network topological structure and the law of dynamic change with\ntime, namely, spatial dependence and temporal dependence. To capture the\nspatial and temporal dependence simultaneously, we propose a novel neural\nnetwork-based traffic forecasting method, the temporal graph convolutional\nnetwork (T-GCN) model, which is in combination with the graph convolutional\nnetwork (GCN) and gated recurrent unit (GRU). Specifically, the GCN is used to\nlearn complex topological structures to capture spatial dependence and the\ngated recurrent unit is used to learn dynamic changes of traffic data to\ncapture temporal dependence. Then, the T-GCN model is employed to traffic\nforecasting based on the urban road network. Experiments demonstrate that our\nT-GCN model can obtain the spatio-temporal correlation from traffic data and\nthe predictions outperform state-of-art baselines on real-world traffic\ndatasets. Our tensorflow implementation of the T-GCN is available at\nhttps://github.com/lehaifeng/T-GCN.\n

Keywords

Affiliated Institutions

• Georgia Institute of Technology US
• Central South University CN
• Zhejiang University CN
• Peking University CN

Related Publications