不确定信息下交通信号灯控制策略的模拟与分析

利用GLD交通模拟软件对交通信号灯控制策略进行模拟试验, 分析了最长等待队列、 相对最长等待队列、 强化学习、 SARSA和遗传算法等交通信号灯控制策略. 针对主干道交通模型, 将道路饱和程度分为欠饱和、 中饱和与过饱和3种状态, 并分别在这3种道路饱和状态下, 研究车流信息确定与不确定时交通信号灯控制策略的性能差异. 实验结果表明, 控制策略的性能曲线能表现出各种程度的偏差, 其中全局控制策略的差异更明显, 即差异程度取决于策略对于交通流信息的依赖与敏感程度.

Simulation and Analysis of Control Strategies of Traffic Lights with Indeterminate Information

Based on the Green Light District (GLD) simulator of traffic, several control strategies of traffic lights were implemented. The analysis was focused on Longest Queue, Relative Longest Queue, Reinforcement Learning, SARSA and GA control strategies. Under three different states of the traffic flow, i.e., undersaturatedtraffic, intermediatesaturated traffic and oversaturated traffic, simulation was completed separately. The different performances between these traffic control strategies were researched for the determinate and indeterminate information of traffic flow, and the data of analysis were given. The experiment data shows, about traffic flow information, the global control strategies are more susceptive than the local ones. They require the veracity of the traffic flow information.