Coordinated Ramp Metering Control Based on Multi-Agent Reinforcement Learning

In view of the high nonlinearity, fuzziness, randomness and uncertainty of expressway system, it is suitable to use the reinforcement learning method of "no model, self-learning, data-driven" to model, which can solve the problem that the traditional control method relies on prior knowledge and model parameter calibration. Based on the SARSA learning algorithm, this paper proposes a coordinated control strategy based on Multi-Agent Reinforcement Learning, which aims at maintaining the main line occupancy near the critical value and keeping the queue length of each ramp within the critical value. The online simulation platform is built by MATLAB and VISSIM, and the control model proposed in this paper is compared with ALINEA, Bottleneck and other traditional control methods. The results show that the model can not only maintain the stability of the main line traffic flow, but also balance the traffic pressure between adjacent ramps, and effectively improve the overall traffic condition of the expressway.