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

Area of Science:

• Transportation Engineering
• Artificial Intelligence
• Environmental Sustainability

Background:

• Traffic congestion significantly contributes to carbon emissions, energy waste, and noise pollution, hindering sustainable transportation goals.
• Reconstructing and expanding roads introduces dynamic changes, challenging existing traffic prediction models.
• Accurate traffic condition prediction is crucial for mitigating congestion and achieving environmental and social sustainability.

Purpose of the Study:

• To develop an accurate traffic condition prediction model for upgraded road sections with dynamic characteristics.
• To address the limitations of existing methods in predicting traffic flow in modified highway and urban arterial road sections.

Main Methods:

• Proposed a Dynamic Bayesian Graph Convolutional Neural Network (DBGCN) model.
• Incorporated road geometric parameters and dynamic traffic organization changes as inputs.
• Utilized a Dynamic Bayesian Network (DBN) to infer a dynamic adjacency matrix capturing spatiotemporal dependencies.
• Integrated the dynamic adjacency matrix into a Graph Convolutional Network (GCN) for traffic flow prediction.

Main Results:

• The DBGCN model demonstrated superior accuracy in traffic condition prediction for upgraded road sections compared to benchmark models.
• The method successfully generated interpretable traffic conditions propagation diagrams.
• Validation on the Wuxuan highway confirmed the model's effectiveness.

Conclusions:

• The proposed DBGCN model offers a robust solution for accurate traffic condition prediction in dynamic and uncertain road environments.
• This advancement supports sustainable transportation by enabling better traffic management and congestion mitigation.
• The model's interpretability aids in understanding traffic dynamics in modified infrastructure.