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

Xu, F; Liu, B; Liu, H; Wang, Y; Wang, J; Wang, C · 2026 · PubMed Central

DOI: 10.1371/journal.pone.0351729

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of accurately predicting traffic conditions in highway work zones where dynamic changes in traffic organization and road geometry create uncertainty. Traditional prediction models often rely on static adjacency matrices that fail to capture the evolving spatial dependencies caused by construction activities such as lane closures and diversions. To mitigate congestion-related environmental impacts and improve traffic management during highway reconstruction, the authors propose a Dynamic Bayesian Graph Convolutional Neural Network (DBGCN). This model integrates physical factors, specifically road alignment metrics and traffic organization schemes, to adaptively learn the dynamic topology of the road network. The methodology employs a two-stage modeling paradigm. First, a Dynamic Bayesian Network (DBN) infers a dynamic adjacency matrix by analyzing statistical dependencies between road geometric parameters and traffic flow data. This process incorporates empirical constraints via whitelist and blacklist mechanisms that reflect permitted and prohibited traffic connections based on construction phases. Second, this dynamic adjacency matrix serves as input for a Graph Convolutional Network (GCN), which fuses spatiotemporal features with real-time traffic flow data to predict traffic conditions. The model uses travel speed as the primary indicator of traffic state and incorporates a time-lag mechanism to capture temporal dependencies. The model was validated using real-world data from the Wuhu–Xuancheng Expressway in China, covering 164 road segments during a reconstruction period from October to November 2024. Traffic flow data were obtained from AutoNavi Maps, while traffic organization plans were derived from official construction documents. The dataset was split temporally, with 24 days used for training and 6 days for testing. The DBGCN was compared against three baseline models: Long Short-Term Memory (LSTM) networks, standard Dynamic Bayesian Networks, and Spatiotemporal Graph Convolutional Networks (STGCN). Results indicated that the DBGCN outperformed all benchmark models in prediction accuracy. Specifically, with a time-lag value of 1, the DBGCN achieved the lowest Mean Absolute Error (2.4112) and Root Mean Square Error (3.8017). The model maintained high accuracy across different prediction horizons, with Mean Absolute Percentage Error values remaining below 0.05, indicating prediction accuracy exceeding 95%. The significance of this research lies in its ability to explicitly model the physical drivers of topological changes in road networks, offering a solution that is both accurate and interpretable. By integrating domain-specific engineering knowledge into the graph learning process, the DBGCN provides clear insights into how construction modifications affect spatial correlations in traffic flow. This approach addresses the limitations of purely data-driven methods that lack physical interpretability, facilitating better integration of prior engineering knowledge. The study demonstrates that incorporating dynamic traffic organization information significantly enhances the ability to predict traffic states in upgraded road sections, supporting more effective traffic management and safety assurance during highway expansion projects.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• work zones