A Multi-Agent Reinforcement Learning Framework for Adaptive Traffic Signal Control in Real Urban Networks: Simulation, Performance Analysis, and Statistical Evaluation

Urban traffic congestion continues to pose a significant challenge for modern cities, calling for intelligent traffic signal control (TSC) strategies able to coordinate multiple interconnected intersections under dynamic traffic conditions. To address this challenge, we propose a Hybrid Q-learning–SARSA–Monte Carlo Multi-Agent Reinforcement Learning framework (H-QSMC-MARL), which combines the complementary learning mechanisms of Q-learning, SARSA, and Monte Carlo algorithms within a multi-agent architecture through a fixed-weight aggregation scheme. This approach was implemented and evaluated in the Simulation of Urban MObility (SUMO) environment, using a realistic road network based on the Saïss District in Fès, Morocco, where eight autonomous agents independently control eight signalized intersections. The three learning mechanisms were combined through fixed weighting, and the resulting hybrid framework was benchmarked against the three individual multi-agent reinforcement learning algorithms under identical experimental conditions. Results show that the hybrid approach consistently outperforms the individual algorithms, achieving more efficient traffic management, improved learning performance, and greater stability. Overall, these findings demonstrate the potential of hybrid multi-agent reinforcement learning as an effective solution for adaptive traffic signal control in realistic urban networks.