Expressway traffic flow prediction based on wavelet noise reduction and time graph convolution

Accurate traffic flow prediction is crucial for making informed decisions regarding travel route selection and mitigating traffic congestion. This paper introduces a WGCG model that addresses the impact of traffic flow noise by combining wavelet transform with GCN and GRU models. The Sym6 wavelet basis is utilized to decompose traffic flow into two layers, effectively reducing noise. The road network's topological structure features are described using an undirected graph and adjacency matrix, with GCN extracting spatial rules and GRU mining hidden time correlation information from historical traffic flow data. The WGCG model integrates these modules to capture the dynamic patterns of traffic flow comprehensively. The model's performance is evaluated on a real dataset from Beijing's Second Ring Road, comparing prediction results with baseline models like HA, SVR, GRU, and GCN. Experimental findings indicate that the WGCG model achieves a significant increase in prediction accuracy, reducing errors by 27.9%, 22.3%, and 21.7% respectively, compared to the second-best model SVR. Ablation experiments further demonstrate that the WGCG model outperforms combined models utilizing only specific modules, confirming its feasibility and superiority.