基于Gaussian型RBF神经网络的一元函数逼近性能研究

为了研究Gaussian型RBF神经网络对于一元非线性函数的逼近能力，编程建立了Gaussian型RBF神经网络和BP神经网络，并以正弦函数、指数函数、阶跃函数三种典型的一元非线性函数为例，分别用两种神经网络对其进行逼近．仿真结果表明，相对于传统BP神经网络而言，Gaussian型RBF神经网络对于一元非线性函数的逼近精度更高、收敛速度更快，具有良好的逼近能力，为解决一元非线性函数的逼近问题提供了良好的解决手段．

Study of approximation performance of simple function based on Gaussian- RBF neural networks

In order to study Gaussian - RBF neural networksj approximation ability of single - variable non- linearity function, Gaussian - RBF neural networks and BP neural networks are designed. And three typical sin- gle- variable nonlinearity functions, namely sine function, exponential function and step function are taken as examples to be approximated via two kinds of neutral networks. Simulation results show that for single - variable nonlinearity functions, Gaussian - RBF neural networks are superior to BP neural networks in approximation pre- cision, convergence rate as well as approximation performance. Thus they provide an ideal method for the solu- tion of single -variable nonlinearity function approximation.