基于样本曲率的模糊神经网络结构学习算法

提出了一种根据输入与输出样本间映射关系的复杂度确定模糊神经网络的输入隶属函数个数与参数的学习方法 ,采用输入输出样本关系曲率来表示函数的复杂度 ,根据曲率的大小确定隶属函数的中心点与宽度 ,使隶属函数的分布符合映射的变化 ,从而在提高逼近精度的同时减少隶属函数个数的增加 .通过仿真将这一方法与均匀划分方法及自组织聚类方法比较 ,结果表明 ,该方法在学习的快速性与精度方面均具有较优的性能

Samples Curvature Based Algorithm for Fuzzy Neural Network Structure Learning

A novel method based on complexity of input output map for fuzzy neural networks structure learning was presented. The complexity was described by applying curvature of input output samples map. The number and the parameters of membership function were determined according to the volume of the curvature. Thus, the distribution of membership functions is corresponding to the complexity of the map, which improves the accuracy while avoiding the increasing of the number of membership functions. The results of simulation prove that the proposed method shows better performance in the learning speed and accuracy of approximation compared to the uniformly distribution method and self clustering method.