Gaussian型RBF神经网络的函数逼近仿真研究

Simulation Study of Function Approximation Performance Based on Gaussian-RBF Neural Networks

为了研究Gaussian型RBF神经网络的逼近能力,首先介绍了Gaussian型RBF神经网络的结构和算法,然后在MATLAB7.0环境下,编程建立了Gaussian型RBF神经网络和BP神经网络,并以具体的非线性函数为例,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,为解决非线性函数的逼近问题提供了良好的解决手段.

In this paper, in order to study the approximation ability of Gaussian-RBF neural networks,firstly the structure and algorithm of Gaussian-RBF neural networks are introduced. Secondly Gaussian-RBF neural networks and BP neural networks are designed on MATLAB7.0 platform. Then the two kinds of neural networks are used to approximate a certain nonlinear function. Simulation results show that for nonlinear functions, Gaussian-RBF neural networks are superior to BP neural networks in approximation precision, convergence rate as well as approximation performance. Thus they provide an ideal method for the solution of single-variable nonlinearity function approximation.