E-RL:Vehicle Driving Strategy Learning with Environmental Reinforcement Learning Framework

To address urban traffic congestion simulation after Traffic Management Initiatives (TMI) implementation, a method combining deep reinforcement learning with environmental optimization is proposed, optimizing the environmental strategy through a fixed agent policy. To tackle the challenges posed by road network topology changes post-TMI, the method designs an environmental Markov reinforcement learning framework and introduces graph neural networks to dynamically calculate road resistance in urban areas. Combined with the Dijkstra algorithm, the vehicle routing strategy is determined. This strategy is then applied to the road network after TMI to predict road resistance post-TMI. Experiments validate that this method achieves good congestion replication in the original road network environment and effectively predicts congestion after TMI implementation.