摘要
针对基于粒子群的模糊聚类算法以隶属度编码时对噪音敏感,以及处理样本数小于样本维数的数据集效果较差等问题,通过改进其中的模糊聚类约束方法,提出一种改进的基于粒子群的模糊聚类方法。当样本对各类的隶属度之和不为1时,新方法在粒子群优化得出的隶属度基础上,根据样本与各类之间的距离对隶属度进一步分配,以使隶属度满足模糊聚类约束条件。新方法显著地改善了在隶属度编码下使用粒子群进行模糊聚类的效果,并通过典型的数据集进行了验证。
While particle swarm optimization （PSO） based fuzzy clustering algorithm is encoded by membership, the algorithm is less effective and tenderness to noise when processing the data set that the number of samples is less than the dimensions, so a new constraint strategy for fuzzy clustering is introduced by improving the constraint strategy of fuzzy clustering. When the sum of membership between a sample and all clusters is not one, after considering the membership obtained by PSO, the strategy further distributes the memberships on the basis of the distance between the sample and cluster centers, then making them meet the constraints of fuzzy clustering. The new strategy improves the clustering effect of the PSO based fuzzy clustering algorithm that encoded in membership significantly, and is verified by typical data sets.
出处
《计算机工程与设计》
CSCD
北大核心
2012年第3期1132-1135,共4页
Computer Engineering and Design
关键词
模糊分类
随机优化算法
隶属度改进
约束方法
C均值聚类算法
fuzzy classification
random optimization algorithm membership improvement constraint method c means cluste-ring algorithm
