Traffic condition prediction for highway within work zones under dynamic traffic organization changes

In the context of sustainable transportation development, reducing carbon emissions, energy waste, and noise pollution caused by traffic congestion has become an urgent task for achieving environmental and social sustainability. The key to this goal lies in mitigating and preventing traffic congestion, for which high accuracy traffic condition prediction models serve as essential tools. During the reconstruction and expansion of highways and urban arterial roads, frequent adjustments to traffic organization and changes in geometric alignment introduce dynamic and uncertain characteristics into the traffic system. Existing methods struggle to accurately predict traffic conditions in the modified sections. To address this challenge, this study proposes a Dynamic Bayesian Graph Convolutional Neural Network (DBGCN). The model incorporates road geometric parameters and dynamic traffic organization changes as key inputs. It employs a Dynamic Bayesian Network (DBN) to model multi-source dynamic information and infer a dynamic adjacency matrix that reflects latent spatiotemporal dependencies between nodes. This dynamic adjacency matrix is then input into a Graph Convolutional Network (GCN), which fuses spatiotemporal features with traffic flow data to achieve accurate traffic conditions prediction for upgraded sections. Validation on the Wuxuan highway demonstrates that the proposed method outperforms benchmark models in traffic conditions prediction accuracy and produces traffic conditions propagation diagrams with high interpretability.