In this paper, we propose a reinforcement learning–based traffic signal control method and a real-time analysis simulation system (RTASS: Real-Time Analysis Simulation System) designed to simultaneously mitigate urban traffic congestion and reduce c...
In this paper, we propose a reinforcement learning–based traffic signal control method and a real-time analysis simulation system (RTASS: Real-Time Analysis Simulation System) designed to simultaneously mitigate urban traffic congestion and reduce carbon dioxide (CO2) emissions. The proposed RTASS is integrated with a SUMO-based simulation to collect and visualize data such as traffic volume, waiting time, and carbon emissions in real time, enabling comparative performance evaluation among various control algorithms. A comparison between real-world and simulated traffic data showed an average error rate of approximately 6%, demonstrating that the proposed system effectively reflects actual traffic conditions. In a single intersection environment, a Deep Q-Network (DQN) based reinforcement learning algorithm was applied and evaluated against conventional fixed-time and actuated control methods. As a result, CO2 emissions were reduced by 54%. Furthermore, the study extended its scope to a multi-intersection network by designing a centralized DQN structure in which a central agent integrates the learning outcomes of local agents. Simulation results confirmed that the proposed method achieves global
optimization, improving both traffic efficiency and environmental performance. As a result, we provide an experimental foundation for quantitatively analyzing the performance of traffic signal control algorithms and suggest the potential applicability of the proposed approach to smart city traffic management and carbon-neutral transportation policies.