小波框架支持向量机的模糊小波网络分类方法

针对现有的支持向量机(SVM)不具有多分辨率学习的特点,提出一种新的小波框架的多尺度支持向量机(SVM)的模糊小波网络(FWN)算法.将小波多尺度学习和模糊推理方法相结合,由于FWN对应着多个模糊规则,而每个模糊规则的后件对应一个小波网络,解决了模糊规则后件难以描述的问题;对高维输入的小波网络的初始参数和网络结构的确定困难问题,用基于正交小波框架的支持向量机代替小波网络的方法,使FWN模型具有更好的泛化性能;为了提高FWN模型的逼近精度,使用梯度下降方法调节FWN参数.仿真结果表明,与传统的模糊神经网络(FNN)相比,该方法能显著地提高分类精度.

A classification method of fuzzy wavelet network based on wavelet framework SVM

For the lack of multi-resolution learning feature for existing support vector machine(SVM),a new algorithm about fuzzy wavelet network(FWN) based on wavelet frame multi-scale SVM is put forward.This algorithm combines wavelet multi-scale learning with fuzzy reasoning.As a result,FWN corresponds to many fuzzy rules,and the consequence of each fuzzy rule corresponds to a wavelet network,which makes it easy to describe the consequence of fuzzy rule.Aimed at the difficulty in ascertaining initial parameters and network structure of wavelet network with high-dimension inputs,a method of using SVM based on orthogonal wavelets framework to replace wavelet network is adopted,which makes the generalization performance of FWN better.In order to improve the approximation precision of FWN,the gradient descent algorithm is applied to the adjustment of parameters.Simulation result shows that the method can improve the classification precision greatly comparing with traditional fuzzy neural network(FNN).