自适应模糊神经网络构建方法  

An Adaptive Construction of Fuzzy Neural Network

作　　者：马波[1] 王正群[1] 沈杰[1] 邹军[1] 

机构地区：[1]扬州大学信息工程学院,江苏扬州225009

出　　处：《江南大学学报（自然科学版）》2007年第6期821-824,共4页Joural of Jiangnan University (Natural Science Edition)　

基　　金：江苏省高校自然科学基金项目(05KJB5201);扬州大学自然科学基金项目(KK0413160)

摘　　要：模糊极大极小神经网络的结构与超盒形状系数有关,该神经网络的性能取决于超盒形状系数的选择.在构建该神经网络时,最优超盒形状系数的确定比较困难,故提出了一种自适应的模糊极大极小神经网络构建方法,取消了超盒形状系数对扩张过程的限制,以是否包含其他类样本为超盒扩张条件.实验结果表明,使用这种模糊神经网络方法生成的神经网络结构更简单,对模式分类的效果更好.Structure of fuzzy max-rain neural network is related to hyper-box＇s shape factor, which also decides its performance. But It is difficult to get the optimum hyper-box＇s shape factor. A method of constructing adaptive fuzzy max-rain neural network is given in this paper. The method cancels the restriction of hyper-box＇s shape factor for expanding process, and take containing other kinds of samples or not for hyper-box expanding condition. The experiment results indicate that the structure of fuzzy network constructed by this method is simpler and its performance is more efficient for pattern classification.

关 键 词：神经网络 模糊神经网络 模式识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]