Coordinated Control Method of Urban Expressway and Surface Road Traffic Flow Based on Multi-Agent Reinforcement Learning

Based on multi-agent deep reinforcement learning technology, this paper studies the coordinated control method of urban expressways and surface roads. Expressways and surface roads near Saihongqiao Interchange in Nanjing are selected as the experimental scene. A multi-agent coordinated control model for expressways and surface roads is constructed. The core of the approach is to create the right set of states and an excellent reward function that yields reasonable actions. It greatly improves the efficiency of the freeway merging area while suppressing long queues at surface intersections. Then, the reinforcement learning control method is compared with the classical signal control methods in the literature. The results show that the reinforcement learning control method is significantly better than the classical signal control method in alleviating the congestion of the expressway and improving the overall traffic efficiency of the road network.