Multi-Intersection Deep Q-Learning for Adaptive Traffic Signal Coordination in Smart Urban Environments

Urbanization and rising vehicle density have worsened traffic congestion, commute times, fuel consumption, and pollution. Traditional traffic control systems—fixed timing and sensor-based fail to adapt to real-time traffic variations, causing inefficient phase switching and bottlenecks. The proposed model treats each intersection as an intelligent agent, using Markov Decision Processes (MDP) and Q-learning to adapt signals based on vehicle density, queue length, and congestion. Integrated with SUMO and trained via Python-based Deep Q-Networks (DQNs), the system uses a reward function that penalises delays and promotes high throughput and low fuel usage. Experimental results show an 80% congestion reduction, outperforming fixed timing (10%) and sensor-based systems (35%). Compared to other AI methods like Genetic Algorithms (70%) and Fuzzy Logic (50%), our model offers better adaptability (9/10 during peak hours) and pedestrian safety (9.5/10), making it ideal for scalable, smart city traffic control.