有教师的线性基本函数前向三层神经网络结构研究  

ON STRUCTURES OF SUPERVISED LINEAR BASIS FUNCTION FEEDFORWARD THREE-LAYERED NEURAL NETWORKS

作　　者：高大启[1] 

机构地区：[1]江苏理工大学信息科学研究所,镇江212013

出　　处：《计算机学报》1998年第1期80-86,共7页Chinese Journal of Computers

基　　金：江苏省自然科学基金;国家教委博士后基金

摘　　要：优化选择隐节点数是人们在应用基于误差反传（BP）算法的有教师的线性基本函数（LBF）前向三层神经网络过程中首先遇到的一个十分重要而又困难的问题．本文从国内外大量应用实例中总结归纳出了一个初定这种网络隐节点数的经验公式，提出了一种判断所选隐节点数是否多余的具体方法，并从理论上做了详细的推导．两个应用实例表明，本文提出的经验公式是可靠的，判断隐节点数是否多余的新方法是简便的．How to optimally select the number of hidden nodes of a supervisedfeedforward three-layered neural network using linear basis function (LBF) basedon back-propagation (BP) algorithm is a very important and difficult problem forone to first surmount in applications. To solve it, an empirical formula for initiallyselecting the number of hidden nodes of this kind of networks is summarized and in-duced according to many application examples at home and abroad, and an elaboratemethod for judging whether there exist redundant hidden nodes in the hidden layerof such a selected network is proposed, and further, a detailed deduction in theoryis given. Two examples show that the empirical formula is reliable and that the newjudging method is simple.

关 键 词：线性基本函数 神经网络 拓扑结构 LBP网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]