遗传算法的广义回归神经网络建模方法  

Generalized regression neural network modeling method based on genetic algorithm

作　　者：孔国利 张璐璐[1] KONG Guo-li;ZHANG Lu-lu(Information Engineering College, Zhongzhou University, Zhengzhou 450000,China)

机构地区：[1]中州大学信息工程学院,河南郑州450000

出　　处：《计算机工程与设计》2017年第2期488-493,共6页Computer Engineering and Design

基　　金：国家青年基金项目(61405156);国家自然科学基金项目(U1304618)

摘　　要：针对广义回归神经网络中光滑因子难以确定,影响建模精度以及模型泛化能力等问题,提出一种基于遗传算法优化广义回归神经网络的高精度建模方法。以广义回归神经模型为基础构建测试样本,预测误差与光滑因子之间的函数,作为适应度函数;通过遗传算法对光滑因子进行优化,以模型输出值误差达到最小时的光滑因子为最优,提高网络模型精度。测试函数建模实验结果表明,与传统的广义回归神经网络相比,该方法预测值均方根误差下降89.45%,平均绝对误差下降91.53%,平均相对误差下降97.65%,能有效提高建模精度和模型泛化能力,为复杂工业的非线性系统建模提供了有效的方法。Concerning the problems that it is hard to define the smooth factors of generalized regression neural network,which impacts the accuracy and generalization ability of model,a high precision modeling method based on optimizing generalized re-gression neural network using genetic algorithm was proposed.The fitness function based on generalized regress neural network was set to indicate relationship between the error of predicted value and smooth factors.Genetic algorithm was used to optimize the smooth factors to minimize the fitness function,and the accuracy of model was increased Experimental results show that,compared with the traditional generalized regression neural network,the root-mean-square error,mean absolute error and ave-rage relative error decreased by89.45%?91.53%,97.65%using the proposed method respectively.Therefore,the proposed method can improve the accuracy and generalization ability of the model and provides a modeling method for the complex indust-rial nonlinear system.

关 键 词：非线性系统 广义回归神经网络 遗传算法 建模 光滑因子 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]