摘要
模糊C均值聚类算法(FCM)由于样本模糊隶属度归一性的约束,导致FCM算法对噪声数据敏感。提出松弛模糊C均值聚类算法(RFCM),RFCM算法在可能性C均值聚类算法(PCM)目标函数的基础上,放弃了FCM算法单个样本模糊隶属度归一化约束,转为n个样本模糊隶属度之和为n的约束;并利用粒子群算法对样本模糊隶属度进行优化估计,使得模糊指标可拓展为m>0的情况,同时采用梯度法得到RFCM算法聚类中心迭代公式。RFCM理论分析了算法对噪声数据抗噪的原理,解释了RFCM算法模糊指标m>0的合理性,讨论了RFCM算法的收敛性。基于Gauss数据集和UCI数据集的仿真测试验证了所提出算法的有效性。
The FCM algorithm is sensitive to noise data due to the normalized constraint of fuzzy membership.A novel clustering algorithm is proposed and named as relaxed fuzzy C means clustering(RFCM),the objective function of PCM is utilized as the objective function of RFCM,and RFCM loosens the normalized constraint and only requests the whole summation of n samples'fuzzy memberships equal to n,particle swarm optimization algorithms(PSO)are optimally used to select the fuzzy memberships of RFCM,and the value scope of fuzzy index m is extended to m>0,The iterative formula of clustering centers are derived by gradient method for RFCM.The anti noise performance of RFCM is analysed theoretically,and the rationality of new value scope of m>0is explained for RFCM,and the convergence of RFCM is discussed simultaneously.The effectiveness of RFCM are proved through simulation experiments.
作者
文传军
詹永照
WEN Chuan-jun;ZHAN Yong-zhao(School of Mathematical Sciences and Chemical Engineering, Changzhou Institute of Technology, Changzhou 213002, P.R.China;School of Computer Science and Communication Engineering, Jiangsu University, Zhenjiang 212013, P.R.China)
出处
《科学技术与工程》
北大核心
2017年第36期96-104,共9页
Science Technology and Engineering
基金
国家自然科学基金(61170126)
常州工学院校级课题(YN1305)资助
关键词
模糊聚类
归一化约束
模糊指标
粒子群算法
噪声数据
fuzzy clustering
normalized constraint
fuzzy index
particle swarm optimization
noise data
