Spatial-Temporal Deep Learning Attention-based Traffic Prediction Model

Abstract

Accurate and timely short-term prediction of traffic states has become a key element in most of the intelligent transport systems. This research investigated a new attention-based deep learning model for traffic state prediction. The spatial and temporal attentions in the model are used to exploit the spatial dependencies between road segments and temporaldependencies between time steps respectively. The proposed model has been demonstrated to have potential for improving both the accuracy and the understanding of spatial-temporal correlations in a traffic network, which contributes to better traffic state prediction.