Traffic Speed Prediction with Missing Data based on TGCN

Traffic speed prediction is an important part of intelligent transportation systems (ITS). This paper proposes a novel approach for traffic speed prediction with missing data. We use Temporal Graph Convolutional Networks (TGCN) which integrates spatio-temporal component and external component to capture the dependencies between traffic speed and various influence factors including road structure, POI and social factors. Meanwhile, there usually exist missing values in the traffic speed data, we use the tensor decomposition method to impute the missing values. Experiments show that the proposed TGCN model outperforms state-of-the-art baselines and tensor decomposition method can improve the prediction performance of TGCN.