基于统计检验的模糊聚类神经网络  

Fuzzy Clustering Neural Networks Based on Statistical Tests

作　　者：吴斯[1] 程玉林[1] 张诚坚[1] 

机构地区：[1]华中科技大学数学系,湖北武汉430074

出　　处：《计算机工程与科学》2008年第6期76-78,共3页Computer Engineering & Science

摘　　要：针对模糊聚类神经网络FCNN原有学习算法对初值敏感性、吸引域不灵活和稳定点不合理等局限性,本文提出基于统计检验的模糊聚类神经网络FCNN-ST。通过引入T平方抽样的单峰分布模式统计检验逐步调整网络结构,确定最佳聚类数c,并使算法的稳定点趋于合理的聚类中心。仿真结果表明,FCNN-ST具有较好的鲁棒性。In this paper, a fuzzy clustering neural network based on statistical tests （FCNN-ST） is presented to resolve the weaknesses of the current learning algorithm of fuzzy clustering neural networks （FCNN）, such as initial condition sensitiveness, inflexible attraction domain and unreasonable convergence points. Statistical tests of the unimodal distribution model based on the T-square sample are introduced to gradually adjust the structure of networks and determine the optimal number of dusters. The results of simulation demonstrate the effectiveness of FCNN-ST.

关 键 词：模糊聚类 神经网络 统计检验 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]