基于UD和GA的神经网络最经济参数的研究  

An Research on Most Economical Structure of Neural Network based on UD and GA

作　　者：李冉[1] 陈以[1] 

机构地区：[1]桂林电子科技大学计算机系,广西桂林541004

出　　处：《微计算机信息》2007年第31期295-296,305,共3页Control & Automation

基　　金：广东省自然科学基金项目(011626)

摘　　要：针对已有的控制系统经济目标函数中分母不变的情况,得出了一种用神经网络进行控制,以收敛性为技术约束的最经济控制参数的确定方法。通过该方法,先用均匀设计法安排神经网络参数试验,进行试验得到性能指标数据,及构造多指标的综合值选取神经网络的性能指标值,然后构造网络参数和多指标之间的映射关系,最后以多指标综合值作为适应度函数,用混合遗传算法求出最经济神经网络结构参数。仿真表明,用该法求得的结果和仿真结果误差在3%以内,说明该方法可以为神经网络最经济控制参数的确定提供有效参考。A method of determining most economical control parameters is put forward while designing the neural network structure, with certain denominator economical function, with technology constrains of astringency. Firstly, Uniform design （UD） is used to arrange experiments of neural network parameters and born training data; secondly, building integrated value of multiple index is to select the neural network index, then training data to get a relation between control parameter and control index; finally, using hybrid genetic algorithm （GA） consisted of GA and neural network control computes the optimal index parameters, and the fitness of GA is integrated value. Simulation indicates, the error between the results from the method and the results from the actual control, is below percent 3, that is to say, the method supplies effective reference to determine the neural network parameters of economical control.

关 键 词：神经网络 最经济控制 经济结构 遗传算法(GA) 均匀设计(UD) 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]