Attention-based Bicomponent Synchronous Graph Convolutional Network for traffic flow prediction

Cheng Shen, Kai Han, Tianyuan Bi

Abstract: Traffic flow prediction is of great importance for city risk assessment and traffic management, which profoundly impacts people’s lives and property. However, the traffic forecasting task is difficult due to the complex interactions and spatial-temporal characteristics. Previous studies usually focus on capturing the spatial correlations and temporal dependencies separately, meanwhile, off-the-shelf studies neglect the effect of explicit differential information. What’s more, there is a lack of effective methods for capturing potential interactions. In this paper, we propose a novel model for traffic flow prediction, named as Attention-based Bicomponent Synchronous Graph Convolutional Network (ABSGCN). This model is able to capture the synchronous spatial-temporal information with a fused signal matrix and potential interactions by constructing a novel edgewise graph, which can remedy the shortcomings in traditional approaches. Extensive experiments are implemented on two real-world datasets and the results demonstrate that our model outperforms other baselines by a margin.