The 22nd COTA International Conference of Transportation Professionals (CICTP 2022)

At present, the congestion of on-ramp regions and the frequent occurrence of traffic accidents easily lead to freeway congestion. The traditional ramp metering   methods have hysteresis in traffic flow adjustment. Therefore, it is of great practical significance to research on the intelligent control methods in on-ramp regions of the freeway. In this work, the mechanism of the active management strategies of the freeway is introduced and the theory of deep reinforcement learning is illustrated. Then, a cooperative control model using a multi-agent deep reinforcement learning algorithm VDN is designed. Simulations are conducted in SUMO compared with other methods. The simulation result shows that the deep reinforcement learning control methods, especially the cooperative control method with VDN algorithm have better performance in the downstream average flow, ramp queue length and total travel time and better prediction ability for traffic conditions, and can make the traffic flow run more smoothly and improve the traffic efficiency.