摘要
从生物学和逼近论出发 ,将任意一组正交基函数作为三层前向神经网络各隐含神经元的活跃函数 ,再以其加权和作为网络的输出特性 ,构成一种新型的神经网络模型 ;从理论上解决了单输入神经网络隐含层数及隐含单元个数难以确定的问题。仿真实验表明 ,该网络具有优良的逼近任意非线性对象的特性 ,且收敛速度远远快于 BP网络。
A new neural network is constructed in this paper based on biology and approximation theories by noting any class of orthogonal basis functions as actuation functions of the hidden neurons of the three layer forward network and the sum of weighted mean functions as the output of the network, which solves theoretically the problem on determining the numbers of the hidden layer. The analogue emulation presented in the paper shows that the model has excellent characteristics of performing approximately any nonlinear object and higher convergence speed.
关键词
神经网络
最佳逼近
基函数
basis function
\ best approximation
\ neural network
