Learning Transformer-based Cooperation for Networked Traffic Signal Control

Networked traffic signal control (NTSC) is essential for intelligent transportation systems. How to control multiple intersections in a cooperative way based on traffic conditions is critical for the success of NTSC. This paper proposes a Transformer-based cooperation mechanism (TCM) with the consideration of dynamic modeling and scale requirements simultaneously for large-scale traffic network control. Considering the physical constraints in traffic scenarios, a relative position encoding is designed to embed into TCM to characterize traffic conditions better. With the shared TCM module, intersection controllers could adequately exploit spatial-temporal correlations and adaptively capture global traffic dynamics, guiding them to explore collaborative traffic strategies more efficiently. Experimental results on two realworld datasets demonstrate that the suggested strategy greatly outperforms the state-of-the-art methods.

Automated summary

This paper presents a Transformer-based cooperation mechanism for controlling large-scale traffic networks. By considering dynamic modeling and scale requirements, as well as designing relative position encoding, this mechanism can better describe traffic conditions and effectively utilize spatial-temporal correlations to improve traffic control efficiency.