摘要
本文介绍了2种应用颇为广泛的神经网络模型,BP及RBF神经网络的基本理论,并从数学角度阐述了2种算法的学习过程,其后简要地阐述了MATLAB神经网络工具箱设计BP和RBF网络的主要函数。为了比较2种网络的性能差异,最后在MATLAB环境下设计了具体的网络来对指定的非线性函数进行函数逼近。仿真结果表明,RBF的泛化能力在多个方面都优于BP网络,但是在解决具有相同精度要求的问题时,BP网络的结构要比RBF网络简单,因此在实际应用中可以此来指导神经网络的设计。
The paper introduced the basic theory of two popular ANN modules, which were BP and RBF networks. It demonstrated the learning procedure of each algorithm from mathematical aspect, and then briefly described main functions of their design in MATLAB neural network toolbox. In order to compare their performance, the article designed a couple of networks, and utilized them to realize the approximation of a particular non-linear function. The simulation result indicates that the generalization capability of RBF is superior to that of BP. However, in settlement of the same problems, the structure of the latter is much simpler, so the research can be used for reference in selecting neural network models in practical application.
出处
《电子测量技术》
2007年第4期77-80,共4页
Electronic Measurement Technology
基金
山西省自然科学基金(20051038)资助项目
关键词
人工神经网络
反向传播算法
径向基网络
训练
函数逼近
artificial neural network
back-propagation algorithm
RBF network
training
function approximation