Dual-View Graph Convolutional Neural Networks for Urban Traffic Congestion Level Prediction

Dual-View Graph Convolutional Neural Networks for Urban Traffic Congestion Level Prediction

연구 분야: Networking

학회: 2025 5th International Conference on Neural Networks, Information and Communication Engineering (NNICE)

Accurate and effective traffic prediction is an indispensable part of intelligent transportation systems, as it plays a crucial role in urban management and travel planning. Traffic data typically exhibit spatiotemporal characteristics, meaning that both spatial and temporal dependencies need to be considered. Graph Convolutional Networks (GCNs), due to their ability to handle non-Euclidean data, have developed rapidly in traffic prediction. However, previous works have primarily focused on modeling traffic flow and speed, with limited consideration of congestion index that can intuitively reflect road conditions. Moreover, in capturing temporal dependencies, most methods tend to capture dependencies in adjacent time periods, neglecting the potential periodicity in traffic patterns. We propose a Dual-View Graph Convolutional Neural Network (DV-GCNN) to predict congestion indices, in which the dual-view module aims to capture different temporal dependencies from adjacent and periodic views. The adjacency matrix generation module aims to incorporate congestion index information into the road topology information. Experiments on a self-constructed urban traffic dataset demonstrate that this model performs well compared to baseline methods.