Traffic flow prediction based on spatiotemporal feature fusion and Graph Convolutional Network

Purpose Due to the complex spatiotemporal correlations and heterogeneity inherent in traffic flow data, most existing models exhibit insufficient performance when handling such data, leading to issues such as low accuracy and poor real-time capabilities. Design/methodology/approach In this paper, a novel network model based on Spatio-Temporal Feature Fusion and Graph Convolutional Network (STFGCN) is proposed. First, the High-frequency Filtering Network is employed to extract high-frequency features from the time series. Subsequently, the Multi-graph Fusion Module integrates the road connectivity graph, dynamic speed graph and similarity graph to model spatial correlations. These extracted temporal features and multi-graph spatial features are then fed into the Spatio-Temporal Feature Fusion Module for cross-domain fusion. Finally, the fused features are processed by the Output Layer to generate prediction results. Findings The experimental results indicate that the proposed model achieves a reduction in the Mean Absolute Error for traffic flow prediction in the next hour to 18.37 and 13.72 on the public datasets of PEMS04 and PEMS08, respectively. Originality/value To better capture the high-frequency temporal correlations in traffic flow data, we propose a High-Frequency Filtering Network in the model. This module processes traffic flow data in the frequency domain, effectively handling high-frequency time series. To model the spatial correlations between roads, we design a novel Multi-Graph Fusion Module. This module integrates three graphs: a static road network topology graph, a dynamic road network graph and a similarity graph, thereby capturing both the inherent spatial structure of roads and potential spatial relationships arising from short-term dynamic changes.