Spatio-Temporal Traffic Forecasting with Uncertainty Quantification via Bayesian Graph Convolution

Traffic prediction plays an important role in urban traffic management and signal control optimization. As research in this area advances, traffic prediction has become increasingly accurate. However, the complexity of the traffic system makes the quantification of uncertainty particularly important, as it is influenced by various factors such as weather changes, emergencies and road construction, which lead to the fluctuation and uncertainty of the traffic state. Although some progress has been made in traffic uncertainty quantification, most methods remain primarily focused on individual traffic observation points, with little exploration of the complex spatiotemporal dependencies at the road network level. In light of this situation, this paper proposes a spatiotemporal traffic prediction model based on Bayesian graph convolutional network, which can effectively capture the spatiotemporal dependence in traffic data, facilitating accurate predictions and comprehensive uncertainty quantification. Through the design of a mixed loss function, the model achieves a good balance between the accuracy of point estimation and the effectiveness of uncertainty quantification. The experimental results show that the proposed model has high efficiency and stability in dealing with complex traffic conditions, providing new insights for research in traffic prediction and uncertainty quantification at the road network level.