Graph Convolutional Networks Combining Dynamic Aggregation of Adjacency Information for Traffic Flow Prediction

Abstract

Accurate traffic flow prediction is crucial for intelligent transportation systems (ITS), especially in traffic management and route planning. Although spatiotemporal graph convolutional networks are widely used for traffic flow prediction, the simple network graph structure is not sufficient to extract periodic temporal features, which limits the accuracy improvement of traffic flow prediction. To address these issues, this paper proposes a convolutional neural network that combines dynamic aggregation of adjacency information (CDAGCN) for traffic. Initially, we designed a graph network generation layer that includes an adaptive dynamic graph generator and a heterogeneous adjacency relationship attention mechanism to effectively capture important features from external factors. To address the extraction of periodic temporal features and the complexity of traffic flow data, we designed a temporal convolutional module (MTCN) comprising multiscale convolutional layers, gate fusion modules, and an efficient pyramid segmentation attention mechanism. These designs significantly enhance the model’s performance and efficiency. The experimental results on multiple real-world datasets demonstrate that, compared with the latest baseline models, the proposed model reduces the RMSE, MAE, and MAPE by 8.82%, 4.43 and 5.67%, respectively. These findings verify the superiority and practicality of the CDAGCN in traffic flow prediction.