A Multi-Channel Spatial-Temporal Transformer Model for Traffic Flow Forecasting

Jianli Xiao,Baichao Long

DOI: https://doi.org/10.1016/j.ins.2024.120648

2024-05-10

Abstract:Traffic flow forecasting is a crucial task in transportation management and planning. The main challenges for traffic flow forecasting are that (1) as the length of prediction time increases, the accuracy of prediction will decrease; (2) the predicted results greatly rely on the extraction of temporal and spatial dependencies from the road networks. To overcome the challenges mentioned above, we propose a multi-channel spatial-temporal transformer model for traffic flow forecasting, which improves the accuracy of the prediction by fusing results from different channels of traffic data. Our approach leverages graph convolutional network to extract spatial features from each channel while using a transformer-based architecture to capture temporal dependencies across channels. We introduce an adaptive adjacency matrix to overcome limitations in feature extraction from fixed topological structures. Experimental results on six real-world datasets demonstrate that introducing a multi-channel mechanism into the temporal model enhances performance and our proposed model outperforms state-of-the-art models in terms of accuracy.

Artificial Intelligence

What problem does this paper attempt to address?

This paper presents a new approach to solving the traffic flow prediction problem. Traffic flow prediction is a critical task in urban traffic management, but it faces challenges of lower accuracy as the prediction time lengthens and relies on the extraction of spatio-temporal dependencies in road networks. To address this, the paper proposes a Multi-Channel Spatial-Temporal Transformer Model that improves prediction accuracy by integrating results from different traffic data channels.

The model utilizes graph convolutional networks to extract spatial features from each channel and adopts a transformer-based architecture to capture temporal dependencies between channels. To overcome the limitation of fixed topology, the paper introduces an adaptive adjacency matrix. The experiments conducted on six real-world datasets validate the effectiveness of the proposed model, demonstrating that introducing multi-channel mechanisms to the temporal model can enhance performance and outperform existing state-of-the-art models in terms of accuracy.

In summary, the paper aims to address the problem of predicting traffic flow more accurately, especially in dealing with the challenges of temporal sequence growth and complex spatio-temporal dependencies. By incorporating multi-channel and adaptive spatio-temporal modeling, this model better captures the randomness, uncertainty, and periodicity of traffic flow, thereby improving the accuracy and reliability of predictions.