Reinforcement Learning for Dynamic Traffic Management: A Scalable Approach to Congestion Reduction

Large-scale urban events with high population density often lead to severe traffic congestion around the event venue, disrupting normal traffic flow and reducing commuter efficiency. This study proposes a reinforcement learning (RL)-based traffic management framework to dynamically optimize vehicle routing in such high-density scenarios. By leveraging a predefined shortest path as a baseline, the system intelligently reroutes origin-destination (OD) vehicles to mitigate congestion near the event area while allowing partial access when necessary. Through reinforcement learning, the agent learns adaptive movement strategies that balance shortest-path efficiency with congestion avoidance, effectively reducing bottlenecks in real time. The proposed approach is implemented and evaluated using a high-fidelity SUMO simulation environment, where its performance is compared against traditional traffic assignment models. Experimental results demonstrate that our RL-driven traffic management strategy significantly improves traffic flow efficiency, highlighting its potential as a scalable and adaptive solution for congestion mitigation in dynamic urban settings.