Spatial linear transformer and temporal convolution network for traffic flow prediction.

Area of Science:

• Transportation Engineering
• Artificial Intelligence
• Data Science

Background:

• Accurate traffic flow prediction is vital for effective traffic management and control.
• Existing models face challenges in capturing dynamic global spatial correlations and long-term temporal dependencies in traffic data.

Purpose of the Study:

• To propose a novel model, the spatial linear transformer and temporal convolution network (SLTTCN), for accurate multi-step traffic flow prediction.
• To address the limitations of current methods in handling spatial correlations and temporal dependencies.

Main Methods:

• The SLTTCN model integrates spatial linear transformers for spatial information aggregation and bidirectional temporal convolution networks for temporal dependency modeling.
• Spatial linear transformers reduce computational complexity while capturing spatial dependencies.
• Bidirectional temporal convolution networks with gate fusion mechanisms mitigate gradient vanishing and high computational costs over long time intervals.

Main Results:

• Extensive experiments on two large-scale public traffic datasets demonstrate SLTTCN's superior predictive performance across various error metrics.
• Attention visualization analysis confirmed the spatial linear transformer's effectiveness in capturing dynamic global spatial dependencies.

Conclusions:

• The proposed SLTTCN model significantly enhances traffic flow prediction accuracy.
• SLTTCN offers an efficient and effective solution for complex traffic flow forecasting challenges in transportation networks.