Adaptive Signal Control for Multiple Urban Road Intersections

The group signal control at intersections is proposed based on the multi-intersection signal control method, which divides multiple interconnected intersections into the same control area and adopts a control strategy. The adaptive control at intersections automatically adjusts the signal control parameters and generates timing plans based on the dynamic random changes of traffic flow, making it the most popular signal control method at present. Existing adaptive control methods based on multi-agent reinforcement learning have not referenced any methods in setting the reward function. They linearly combine parameters such as queue length, accumulated delay, and waiting time and adjust coefficients to achieve optimization. To address this issue, this paper proposes an adaptive signal control method, MPA2C, based on multi-agent reinforcement learning and transportation theory. The max pressure control theory is applied to the multi-agent reinforcement learning algorithm MA2C to provide a rational basis for the setting of the reward function and state, Furthermore, considering the coordination relationships between neighboring intersections, an spatial discount factor is introduced to take the influence of adjacent intersections into account when calculating the pressure of intersections, which enables the intelligent agent to adopt more coordinated strategies. The case study selected the road network in the Huangshan Road area of the High-tech Zone of Hefei City. The proposed method was validated using traffic flow data obtained through SUMO simulation based on intersection information in the real road network. The experimental results show that the proposed method is effective in reducing the average queue length, delay, waiting time, and total travel time at intersections in the area.