Abstract
Predicting traffic speed accurately is a very challenging task of the intelligent traffic system (ITS), due to the complex and dynamic spatial-temporal dependencies from both temporal and spatial aspects. There not only exits short-term local neighboring fluctuation and long-term global trend in temporal aspect, but also local and global correlations in spatial aspect. Most existing work focus on the local spatial-temporal dependencies, ignoring the global dynamic spatial-temporal corrections, which is comparably critical for traffic speed prediction. To address this problem, we propose a novel Dynamic Global-Local Spatial-Temporal Network(DGLSTNet) for traffic speed prediction, which consists of multiple spatial-temporal module considering the local and global information simultaneously from both temporal and spatial perspective. Each temporal module applies stacked dilated convolution block to exploit multi-scale local temporal information. Moreover, we empoly a global temporal attention block to capture global dependencies of temporal domain in an attention mechanism. In each spatial module, we not only learn the local but also focus on dynamic global spatial information learned by dymamic graph learning block. Combining the feature results from local and global perspective, the capability and expressiveness of traffic predicting model is improved. Experiment results on two real-world traffic datasets have demonstrated that our proposed model can effectively capture the comprehensive spatial-temporal dependencies and can achieve state-of-the-art prediction performance compared with the existing works.
Highlights
Predicting large-scale network-wide traffic becomes increasingly popular in the intelligent transportation systems due to its application and research significance
There are two mainly issues should be considered in traffic prediction: (1) there exits short-term local neighboring fluctuation and long-term global trend in temporal aspect and traffic conditions are different at various times. (2) the saptial dependence is often found to exist in a wider range of the traffic networks, for instance, congestion can effect the neighboring regions and reachable far distant regions
We develop a dynamic temporal module which considers the short-term local neighboring and the longterm global trend dependencies. It consists of a global temporal attention block(GTAB) and a stacked dilated convolution block (SDCB), where the former is used to extract whole-range global temproal features in an attention mechanism and the latter is used to capture the multi-scale local temporal features
Summary
Predicting large-scale network-wide traffic becomes increasingly popular in the intelligent transportation systems due to its application and research significance. We develop a dynamic temporal module which considers the short-term local neighboring and the longterm global trend dependencies It consists of a global temporal attention block(GTAB) and a stacked dilated convolution block (SDCB), where the former is used to extract whole-range global temproal features in an attention mechanism and the latter is used to capture the multi-scale local temporal features. In this way, dynamic temporal dependencies can be captured effectively. By integrating the pre-defined and the adaptively learned global adjacency matrices into graph convolution operation to capture both local and global spatial dependencies simultaneously, the capacity and expressiveness of capturing saptial dependencies are enhanced. We conduct extensive experiments on two real-world traffic datasets, METR-LA and PEMS-BAY, and the proposed model achieves the state-of-the-art results
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
