基于滑模思想和Elman网络的操作条件反射学习控制方法  

Operant conditioning reflex learning control scheme based on SMC and Elman network

作　　者：阮晓钢[1] 陈静[1] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100124

出　　处：《控制与决策》2011年第9期1398-1401,1406,共5页Control and Decision

基　　金：国家863计划项目(2007AA04Z226);国家自然科学基金项目(60774077);北京市自然科学基金项目(4102011);北京市教委重点项目(KZ200810005002)

摘　　要：针对一类单输入单输出高阶非线性控制系统,提出一种基于滑模思想和Elman网络的操作条件反射(OCR)学习控制方法.该方法采用Elman网络构造滑模面-行为对的评价函数,通过滑模面的变化设计奖赏函数,根据奖赏信号更新评价函数,实现行为选择概率的更新.通过每轮次熵的定义,定量分析了所学知识的变化量.针对行走倒立摆系统的仿真实验结果表明,采用该仿生的OCR学习控制方法,可实现行走倒立摆的平衡控制.A bionic operant conditioning refiex（OCR） learning control scheme is proposed based on the thought of sliding model control（SMC） and Elman network for a class of SISO higher-order nonlinear control system. In this method, an Elman network is used as an evaluation function of sliding surface and action in the scheme. Reward signal is designed according to the change of sliding surface, and then the evaluation function is updated through the reward stimulation, while the behavior choice probability is changed. Through the definition of entropy for each round, a quantitative analysis about the knowledge change in the learning process is given. The results of the simulation experiment in the walking inverted pendulum system show that, bionic OCR learning control is used, which realizes the balancing control for the walking inverted pendulum system.

关 键 词：操作条件反射 滑模控制 ELMAN网络 熵 倒立摆 平衡控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]