STGFP: information enhanced spatio-temporal graph neural network for traffic flow prediction

Abstract

Accurate traffic flow prediction is crucial for the development of intelligent transportation systems aimed at preventing and mitigating traffic issues. We present an information-enhanced spatio-temporal graph neural network model to predict traffic flow, addressing the inefficient utilization of non-Euclidean structured traffic data. Firstly, we employ a multivariate temporal attention mechanism to capture dynamic temporal correlations across different time intervals, while a second-order graph attention network identifies spatial correlations within the network. Secondly, we construct two types of traffic topology graphs that comprehensively describe traffic flow features by integrating non-Euclidean traffic flow data, regional traffic status information, and node features. Finally, a multi-graph convolution neural network is designed to extract long-range spatial features from these traffic topology graphs. The spatio-temporal feature extraction module then combines these long-range spatial features with spatio-temporal features to fuse multiple features and improve prediction accuracy. Experimental results demonstrate that the proposed approach outperforms state-of-the-art baseline methods in predicting traffic flow performance.