基于改进遗传算法的小波神经网络控制器的研究  

Research of Wavelet Network Controller based on Improved Genetic Algorithm

作　　者：宋清昆[1] 高健凯[1] 丁然[1] 王宏伟[1] 

机构地区：[1]哈尔滨理工大学自动化学院,黑龙江哈尔滨150080

出　　处：《自动化技术与应用》2012年第2期1-5,10,共6页Techniques of Automation and Applications

基　　金：黑龙江省自然科学基金(编号F200830);国家重点基础研究发展计划(973计划)(编号2009CB220107)

摘　　要：小波神经网络是一种引入小波分析理论的前馈型神经网络,其与遗传算法的结合可以得到一种拥有良好全局优化搜索和良好局部时频特性的学习训练途径。本文提出了一种基于改进遗传算法的小波神经网络控制器,此方法可以克服基本遗传算法收敛速度慢,容易陷入"早熟"收敛,计算稳定性不好等一系列问题,进一步提高了小波神经网络控制器的性能。最后通过二级倒立摆仿真和实物控制,证明了控制器的有效性。Wavelet neural network is a kind of feedforward neural network introducing wavelet analysis theory.The combination with genetic algorithm can get the way of learning and training including the good global optimization and the local performance in both time and frequency fields.A wavelet neural network controller based on improved genetic algorithm is proposed in this paper.This method can overcome the series of problems of the basic genetic algorithm such as slow convergence speed,easy to get into ＂premature＂convergence and the poor calculation stability etc,then the performance of the wavelet neural network controller have further improved.Finally,the controller’s effectiveness is demonstrated through the simulation and the real control with the double inverted pendulum.

关 键 词：小波神经网络 改进遗传算法 二级倒立摆 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]