Adaptive traffic signal control model for a single isolated intersection using deep reinforcement learning
Year: 2026
Type: article
Abstract: Traffic congestion has major social, environmental, and economic consequences, and adaptive traffic signal control is one approach for improving intersection performance under changing demand. This study presents a reinforcement-learning-based controller for a single isolated intersection and compares its performance with a firefly-based heuristic. The main contribution is a normalized composite reward that combines queue length and vehicle waiting time. These measures are correlated, but they describe different operational aspects of intersection performance: queue length reflects spatial occupancy and potential spillback, whereas waiting time reflects the temporal delay experienced by vehicles. A weighting parameter (β) is used to vary the relative emphasis assigned to the two measures. The proposed reward was compared with queue-only and waiting-time-only rewards under the tested simulation conditions. The results indicate that the composite reward can provide a tunable compromise between the two measures, although it does not consistently outperform the single-metric rewards for every performance indicator and traffic-demand level.
Source: Scientific Reports
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7226018026
