基于参数融合的Q学习交通信号控制方法

A Control Method of Traffic Signals Based on Parameter Fusion of Q-learning

传统Q学习交通控制方法经常因城市道路交通流的随机性和不确定性而产生维数问题,导致控制系统的学习效率和响应速度降低。针对该问题进行分析,提出了一种基于Q学习的改进路口交通信号自适应控制方法。其中,采用模糊技术与Q学习算法相结合的控制策略,改进了Q学习的奖惩机制;同时在Q学习中引入基于经验的状态划分对状态空间进行了优化,并通过建立交通参数融合函数的方式在保持多参数评价交通状态的前提下降低了状态空间存储及更新的复杂度;针对不定周期Q学习配时方案状态空间过于庞大的问题,给出基于相位的绿灯配时方案,从而最终达到对交通流的实时响应控制。使用仿真软件对控制方法进行了仿真研究,仿真结果表明该方法的控制效果优于传统控制。

Traditional Q-learning traffic control methods often generate dimension problem due to the randomness and uncertainty of traffic flow which will result in the reduction of learning efficiency and response speed of the control system.For this,we introduce an improved traffic signal adaptive control method based on Q-learning.A fuzzy technology and Q-learning algorithm is combined to improve the reward and punishment mechanism of Q-learning.At the same time,the empirical state partition is introduced in Q-learning to optimize the state space,and the complexity of state space storage and update is reduced on the premise of maintaining multi-parameter evaluation of traffic state by establishing the traffic parameter fusion function.Aiming at the problem that the state space of the unsteady period Q-learning timing scheme is too large,the phase-based green light timing scheme is presented,so as to achieve the real-time response control of traffic flow.At last,simulation software is used for the control method research,which shows that the control effect of this method is better than traditional control.