SIMO傅里叶三角基神经网络的权值直接确定法和结构自确定算法  

A Weights-Direct-Determination Method and Structure-Automatic-Determination Algorithm for SIMO Trigonometrically-Activated Fourier Neural Networks

作　　者：张雨浓[1] 李钧[1] 张智军[1] 阮恭勤[1] 姜孝华[1] 

机构地区：[1]中山大学信息科学与技术学院,广东广州510006

出　　处：《信息与控制》2011年第4期507-513,524,共8页Information and Control

基　　金：教育部"新世纪优秀人才支持计划"资助项目(NCET-07-0887)

摘　　要：根据傅里叶级数逼近理论,将正交三角函数系作为隐层神经元激励函数,合理选取这些激励函数的周期参数,构造单输入多输出(SIMO)傅里叶三角基神经网络模型.根据该网络的特点,推导出一种基于伪逆的权值直接确定法,从而1步计算出网络最优权值,并在此基础上设计出隐层结构自确定算法.仿真结果表明,与传统BP(反向传播)神经网络及基于最小二乘法的SIMO傅里叶神经网络模型相比,本网络模型具有更高的计算精度和更快的计算速度.According to Fourier series approximation theory,a single-input multiple-output（SIMO） trigonometrically-activated Fourier neural network model is constructed by setting the hidden-layer neuron activation function as orthogonal trigonometric function series and selecting the periodical parameter of these activation functions properly.In light of the characteristics of the presented network,a pseudo-inverse based weights-direct-determination method is derived to determine the optimal weights of the network with one step,and a structure-automatic-determination algorithm is designed.Simulation results substantiate that,compared with the traditional BP（backpropogation） neural network and the SIMO Fourier neural network model based on least square method,this model has higher accuracy and faster computing speed.

关 键 词：SIMO傅里叶三角基神经网络 权值直接确定 隐层结构自确定 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]