结合评价模型与多Q值SAC的交叉口信号控制方法OA

To address the challenges of prolonged training time and limited adaptability to dynamic traffic conditions in deep reinforcement learning-based intersection signal control,this study proposes a multi-Q soft actor-critic(Multi-Q SAC)approach integrated with an evaluation model.The evaluation model,constructed from extensive VISSIM simulation data,enables the decoupling of reinforcement learning from microscopic simulations and significantly enhances training efficiency.The model’s evaluation indicators are employed as the state variables of the Multi-Q SAC algorithm,while three representative indicators are weighted to define the reward function.Experimental results demonstrate that,compared with the traditional Webster timing method and other policy-based reinforcement learn-ing algorithms,the proposed method substantially improves intersection performance and maintains fast convergence under complex and varying traffic flows.