The Multi-Task Time-Series Graph Network for Traffic Congestion Prediction

Abstract

Accurate prediction of traffic congestion is an important for people's travel and the building of smart city. However, the inherent non-linear relationships and spatiotemporal autocorrelation remain big challenges. To overcome these issues, we propose a Multi-Task Time-Series Graph Network (MTG-Net) framework, which uses a Temporal Convolutional Network (TCN) to capture the temporal relationships and models the correlations between regions dynamically with graph attention network (GAT). Further we achieve collaborative prediction of congestion on elevated and ground road with multi-task training and incorporate the external factors from different domains. Experiments on real traffic congestion data demonstrate effectiveness of our approach over state-of-the-art methods.