Proactive Traffic Light Signal Control using Attention Transformer

Abstract

Urban traffic congestion challenges existing fixed-time and reactive controllers. This paper proposes ProACT, a proactive traffic signal control framework that utilizes an attention-transformer architecture. By integrating real-time IoT traffic data with historical patterns, ProACT constructs spatio-temporal representations to enable early congestion detection. Furthermore, the framework captures long-range dependencies across adjacent intersections to proactively optimize signal phases. Evaluated using a traffic prediction dataset and the SUMO simulator across multiple intersections, ProACT reduced delays by over 50% under normal conditions and by more than 65% during rush hours compared to traditional controllers. These results demonstrate the potential of ProACT to improve intersection throughput, alleviate congestion, and support sustainable smart city development.