Traffic Flow Prediction Based on Pivotal Graph Convolutional Network and Transformer

Traffic prediction is a fundamental task in Intelligent Transportation System (ITS), as accurate Traffic Flow Forecasting (TFF) can significantly improve the utilization efficiency of public resources. In traffic flow forecasting, the combination of Graph Neural Networks (GNNs) and Transformers has been widely used to boost prediction accuracy. Nonetheless, most existing approaches fail to consider the importance of pivotal nodes, which have extensive connections with other nodes. Moreover, the self-attention mechanism in traditional Transformers is primarily designed to process discrete tokens and does not account for the inherent local trend information in continuous traffic flow data. To tackle these challenges, we present a new method Transformer-based Spatio-temporal Pivotal Graph Convolutional Networks (TSPGCN). Firstly, the model identifies pivotal nodes and generates a pivotal graph through a pivotal graph learning module. Secondly, it aggregates the spatial information of pivotal and non-pivotal nodes using a pivotal graph convolution module. Finally, a multi-scale temporal attention mechanism is employed to capture local and global temporal dependencies simultaneously. We conducted experiments on two real-world datasets and the results demonstrated that TSPGCN achieves better prediction accuracy compared to conventional traffic flow prediction methods.