摘要
通过把对向传播(CP)神经网络的竞争层神经元的输出函数定义为模糊隶属度函数,提出了模糊对向传播(FCP)神经网络.该网络是CP网络的推广,它不仅能有效克服CP存在的问题,而且具有全局函数逼近能力.在结构上,FCP网络同径向基函数(RBF)网络是等价的.实际上,它是一种RBF网络,而且还是一种模糊基函数网络.FCP在时间序列预测中的应用表明,FCP不仅在学习精度上,而且在泛化能力方面较之CP和RBF均有较大的改善.
A fuzzy counter propagation(FCP) neural network,which is a generalized model of the counter propagaton(CP) network,is proposed in this paper by defining output of the competitive unit of CP network as a fuzzy membership function. FCP not only is able to overcome the shortcomings of CP,but has the ability of universal function approximation as well.In view of network structure,FCP is equivalent to the radial basis function(RBF) network.In fact,FCP is an RBF network,also a fuzzy basis function network.In the end of this paper,the experiment to apply FCP to time series prediction shows that FCP outperforms CP and RBF in learning precision and generalizaton ability.
出处
《自动化学报》
EI
CSCD
北大核心
2000年第1期56-60,共5页
Acta Automatica Sinica
基金
国家自然科学基金
关键词
神经网络
CP神经网络
模糊隶属度函数
时间序列
Counter propagation network, fuzzy membership function, time series prediction.
