Zhao, Yaxin (Chongqing University), Xiang, Pengcheng (Chongqing University), Luan, Xinyu (Chongqing University), Liu, Shengjie (Chongqing university)

A Spatio-Temporal Embedding and Attention Mechanism for Traffic Prediction in Large-Scale Road Networks

Scheduled for presentation during the Regular Session "Traffic prediction and estimation IV" (ThBT7), Thursday, September 26, 2024, 16:10−16:30, Salon 15

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

Various fields have recognized the effectiveness of network embedding for handling large-scale graphs. However, accurate traffic conditions prediction using network embedding on large-scale road networks remains challenging due to the intricate correlations in traffic data. To tackle this challenge, we propose a novel framework called Spatio-Temporal Embedding and Attention Mechanism (STEAM). In the embedding process, we design a novel spatial embedding method to consider both the local structure and global structural role of each node. Concurrently, long-term and short-term temporal dependencies are embedded in the temporal part. During the inference process, an attention mechanism is applied to adaptively capture the nonlinear spatio-temporal correlations. Our model was evaluated by predicting traffic speed on two traffic datasets, demonstrating significantly improved prediction performance and superior generalization capabilities.