再励学习在交通信号控制中的应用

Reinforcement learning applied in traffic signal control

再励学习是一种利用评价信息 (而不是网络实际输出与期望输出之差 )来改善行为的神经模糊算法 ,采用“奖”“罚”信号训练控制器 .用再励学习的目的建立一个可调的模糊交通信号控制器 ,它能在不同交通情况下修改隶属函数参数 ,以达到较好的控制效果 .其评价指标是车辆延误 .仿真结果表明 。

Reinforcement learning is a neurofuzzy algorithm which improves action by using criterion rather than the difference of the network output and the desired output and trains controller by using credit and punishment signals. The objective of the reinforcement learning is to create an adjustable fuzzy traffic signal controller that can modify its parameters in different traffic situations, and then reach a better control result. The criterion is the delay of vehicles. Simulation shows that the learning algorithm is found successfully at stable traffic volumes.