基于Q-learning和BP神经元网络的交叉口信号灯控制

解决单交叉口信号灯最优控制问题。提出了基于强化学习的信号灯控制系统结构,应用强化学习中Q学习,将信号灯最优控制问题转变成是否切换运行相位的决策问题,提出了采用BP神经元网络实现Q学习的信号灯控制系统。应用微观交通仿真软件PARAMICS进行仿真分析,结果表明该系统能够感知交通流变化,并能够自适应地调整信号灯切换策略,以达到最优的控制效果,该方法是可行的,与定时控制相比具有明显的优势。

Traffic Signal Control Based on Q-learning and BP Neural Network

The optimal signal control of single intersection is resolved.The control system framework based on reinforcement learning is set up.The Q-learning of reinforcement learning is applied and the optimal control of signal is translated to the decision-making between the phases.A method based on BP neural networks is introduced to realize Q-learning control for traffic signals at an isolated intersection.The performance of the system is evaluated by PARAMICS traffic simulation software.The simulation results indicate that the system is apperceive to the changing of traffic flow.Its control strategy can be adjusted with varying traffic condition in order to obtain the optimal control effect.Compared with conventional fixed-time control,the validity and advantage of the method is proved.