Traffic Flow Prediction Based on Multi‐View Fusion Graph Convolutional Network

Le résumé fourni par la source

ABSTRACT In urban rail transit systems, due to their efficiency and punctuality, metro systems have become the preferred choice for daily commuting. Accurate metro passenger flow prediction is crucial for ensuring stable system operations and optimizing resource allocation. In recent years, Graph Convolutional Networks (GCNs) have been widely adopted to extract spatial features in traffic flow data. However, they still face challenges in capturing complex global spatial dependencies, especially latent associations between different traffic nodes. Moreover, existing methods often focus on local neighborhood information, which makes it difficult to fully model the widespread correlations among regions with similar functional characteristics. To address these issues, this paper proposes a Multi‐View Fusion Graph Convolutional Network (MVFGCN) model, which introduces a multi‐view fusion strategy to capture spatial features of traffic flow from multiple perspectives. This enhances the model's capability to represent global spatial dependencies among different traffic nodes. In addition, a functional‐region‐based hypergraph construction method is designed, which includes node functional region recognition using K‐means and DTW algorithms and the generation of a functional‐region‐based hypergraph. This approach effectively captures correlations among nodes with similar periodic characteristics. By combining multi‐view graph convolution and self‐attention convolution, the proposed method can more effectively capture spatiotemporal features in traffic networks, leading to more precise traffic flow forecasts. Tests on real‐world datasets from metro systems and highways show that the proposed method significantly outperforms several mainstream models in prediction accuracy, validating the effectiveness and robustness of MVFGCN in complex urban traffic scenarios.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.