Abstract
Accurate and real-time traffic forecasting plays an important role in theIntelligent Traffic System (ITS) and is of great significance for urban trafficplanning, traffic management, and traffic control. However, traffic forecastinghas always been considered an open scientific issue, owing to the constraintsof urban road network topological structure and the law of dynamic change withtime, namely, spatial dependence and temporal dependence. To capture thespatial and temporal dependences simultaneously, we propose a novel neuralnetwork-based traffic forecasting method, the temporal graph convolutionalnetwork (T-GCN) model, which is in combination with the graph convolutionalnetwork (GCN) and gated recurrent unit (GRU). Specifically, the graphconvolutional network is used to learn complex topological structures tocapture the spatial dependence and the gated recurrent unit is used to learndynamic changes of traffic flow to capture the temporal dependence. Then, theT-GCN model is employed to forecast traffic based on the urban road network.Experiments demonstrate that our T-GCN model can obtain the spatiotemporalcorrelation from traffic data and the predictions outperform state-of-artbaselines on real-world traffic datasets. Our tensorflow implementation ofT-GCN is available at https://github.com/lehaifeng/T-GCN.