Spatio-Temporal Graph Neural Point Process for Traffic Congestion Event Prediction

Traffic congestion event prediction is an important yet challenging task in intelligent transportation systems. Many existing works about traffic integrate various temporal encoders and graph convolution networks (GCNs), called spatio-temporal graph-based neural networks, which focus on predicting dense variables such as flow, speed demand time snapshots, but they can hardly forecast the events that are sparsely distributed continuous axis. In recent years, point process (NPP) has emerged appropriate framework for scenarios. However, most conventional NPP cannot model complex dependencies evolution patterns. To address these limitations, we propose a framework, named STGNPP prediction. Specifically, first design learning module to fully capture long-range from historical state data along with road network. The extracted hidden representation information then fed into gated recurrent unit particular, exploit periodic information, also improve intensity function calculation of mechanism. Finally, our simultaneously predicts occurrence duration next congestion. Extensive experiments two real-world datasets demonstrate method achieves superior performance comparison state-of-the-art approaches.