A Novel Ramp Metering Algorithm based on Deep Reinforcement Learning

In order to solve the problems of slow response and inaccurate control of the traditional ramp metering algorithm, we propose an intelligent ramp metering model based on deep reinforcement learning. By building a neural network and designing a reinforcement learning model based on the Double-Deep Q Network, the intelligent control of on-ramp traffic flow can be realized and the real-time of ramp metering is ensured. The reinforcement learning environment state space is designed based on key traffic flow parameters, and the incentive mechanism of the ramp metering system is established based on part of the traffic data to measure the value of the action. The JavaScript simulation platform is built, the update mode of the detectors is reset to real-time update, and new training rules are defined. The experimental results show that the proposed method can effectively improve the mainline service level. Compared with the case of no-control, the throughput of the system is improved by 52.67%, and the average travel time of the system is reduced by 58.21%, superior to ALINEA and another machine learning method. In addition, the ramp throughput is similar to the comparison method, and the average travel time of ramp vehicles is further shortened.