Max-min K-means clustering algorithm and application in response signal feature extraction

Abstract

In order to overcome the problem of classical K-means clustering algorithm, such as sensitiveness to its initial cluster centers selection and easiness of falling into local optimization, this paper proposed max-min K-means clustering algorithm. The algorithm adopts max-min principle to determine initial cluster centers of K-means clustering algorithm. Comparison of clustering effects of the new method with that of muilt-restart K-means clustering algorithm on artificial data set showed the better effect and the greater excellence of max-min principle. Then, the max-min K-means clustering algorithm was applied to analyze vibration response signal of cylindrical shell automatic in order to discover space distribution feature of dynamics response. Clustering results tallied with feature of dynamic response of cylindrical shell, which validated the effectiveness and reasonableness of the new clustering method itself.