Evaluation of the Performance of an Intelligent Traffic Flow Management Model Based on Reinforcement Learning in Reducing Pollutant Emissions

In today’s world, environmental concerns and the adverse impacts of air pollution on public health have made reducing traffic-related pollutant emissions a fundamental priority in major cities. In Tehran, the pollution levels generated by key traffic-related pollutants, such as hydrocarbons, carbon monoxide, carbon dioxide, nitrogen oxides, and particulate matter, frequently exceed permissible standards, thereby posing serious health risks to residents. This study presents an intelligent approach to pollutant emission management by introducing a reinforcement learning-based model employing Proximal Policy Optimization. By dynamically adjusting permissible speed limits and the ramp metering signals, the model aims to reduce key pollutant emissions and fuel consumption without negatively affecting traffic flow indices. The use of detailed traffic simulation via SUMO software and model training under various traffic volumes has led to robust performance across different network conditions. In implementation, the model is executed over a one-hour simulation, during which control parameters related to speed limits and ramp metering signals are continuously updated through parallel environment training. The agent, upon receiving the traffic state of the network—which includes density and average speed across different road segments—selects the appropriate control actions to minimize pollutant emissions. The resulting outputs, comprising indicators of pollutant emissions and traffic flow performance, demonstrate that the model is capable of achieving a significant reduction in emissions without disrupting traffic flow.