Intelligent Traffic Signal Optimization Using Multiagent G - DQN with Adaptive Reward Optimization

Background: in recent years, the Deep Reinforcement Learning (DRL) advanced traffic signal optimization framework is to address the difficulties of inefficient signal control and urban traffic congestion. Challenges: however, existing traffic management system include actuated control methods and fixed-time, which lacks the adaptability and complex traffic conditions, leads to environmental impact, fuel consumption and increased delay. Proposed Methodology: to overcome these challenges, the proposed Multi-Agent Grouping Deep Q Network (GDQN) with Adaptive Reward Optimization (ARO) used the multi agent system where every intersection is demonstrated as the individual and independent agent that learns ideal control policies through environment interaction. Moreover, real time traffic parameters such as traffic density, waiting time, vehicle count and queue length used as input states. This framework uses the GDQN to decrease the state space complexity through the classification of traffic signals into discrete levels, in enhancing the convergence speed and learning efficiency. Further, an adaptive reward optimization adjusts dynamically based on traffic conditions, allows more context aware and responsive decision making. Furthermore, multi-agent coordination improves the global traffic and scalability through the limited information across neighbouring intersections. Results: Experimental results represent that proposed GDQN with adaptive reward mechanism outperformed the existing YOLOv5 method in terms of precision of 99.5%, recall of 98.6% and f1-score of 98.8%. Therefore, this approach is effective and scalable approach for real-time traffic management in smart city environments.