The traditional K-means clustering algorithm requires a certain number of clustering

usually the value of clustering number is assumed by priori rules. Besides

the choice of K-means initial clustering centers is also randomized

which made the clustering result instability. In view of the above factors

improved the selection of initial centers of K-means clusters

and a new method to determine the optimal clustering number is proposed based on that. The experimental results show that the proposed method can get better clustering results and have higher accuracy

better stability and better astringency.