基于Q学习算法的变论域模糊控制新算法  

A novel optimal method of variable-universe fuzzy control based on Q-learning algorithms

作　　者：余涛[1] 于文俊[2] 李章文[1] 

机构地区：[1]华南理工大学电力学院,广东广州510640 [2]湖南电力公司永州电业局,湖南永州425000

出　　处：《控制理论与应用》2011年第11期1645-1650,共6页Control Theory & Applications

基　　金：国家自然科学基金面上资助项目(50807016;51177051);广东省自然科学基金资助项目(9151064101000049);清华大学电力系统及发电设备控制与仿真国家重点实验室开放基金资助项目(SKLD10KM01)

摘　　要：变论域模糊控制器的控制函数被"复制"到后代中,往往存在着"失真"现象,这种现象的后果是造成算法本身的误差.针对这一问题,本文提出了一种基于Q学习算法的变论域模糊控制优化设计方法.本算法在变论域模糊控制算法基础上提出了一种利用伸缩因子、等比因子相互协调来调整论域的构想,且通过用Q学习算法来寻优参数使控制器性能指标最小,使其在控制过程中能够降低"失真率",从而进一步提高控制器性能.最后,把算法运用于一个二阶系统与非最小相位系统,实验表明,该算法不但具有很好的鲁棒性及动态性能,且与变论域模糊控制器比较起来,其控制性能也更加提高.When the control function of the variable-universe fuzzy controller is ＂copied＂ to the offspring,there usually exist some ＂distortion＂ phenomena which lead to the error of the algorithms.To deal with this problem,we present a novel optimal method of variable-universe fuzzy control based on Q-learning algorithms.This method adjusts the universe by the contraction-expansion factor and geometric proportional factors,and optimizes the parameters through Q-learning algorithms to minimize the performance index of the controller for reducing the ＂distortion rate＂ in the control process and improve the control performance.This method has been applied to a second-order linear system with non-minimum phase,resulting in desirable robustness and dynamic performance.The control performance is even better than that of the variable universe fuzzy controller.

关 键 词：变论域模糊控制 Q学习算法 伸缩因子 等比因子 

分 类 号：TP273.4[自动化与计算机技术—检测技术与自动化装置]