A Dynamic Clustering Algorithm By ASA And Its Validity

Clustering analysis is a kind of multivariate statistical method of sample classification and data analysis method without training sample set. When we utilize this method to classify the given data, though we don't know the type of adopted sample, it will classify the samples automatically according to their similar stage. The mainly aim of clustering analysis is determined by dividing the given data set into a series of meaningful subset (or cluster). On the basis of certain rules properly, it can be used each data of cluster to be similar as fully as possible (or close), and the different data has considerable different as fully as possible. A good result of clustering, on the one hand, we can grasp the each group of given data's characteristics according to their inherent nature, so it comes up to the aim of concentrating the former data scale, on the other hand, we can obtain the data which has simple and object structure according to the relatively complicated initial data set, in order to benefit further discussion and research to the given problem.Generally,when we discuss the actual problem, we often do not grasp the information of the structure of given data and concrete the number of cluster fully (in advance,sometimes it's even close to know nothing), so it's usually a very complicated problem of defining clustering objectively and exactly.meanwhile,a lot of dynamic clustering algorithm generally exist the problem which clustering result greatly depends on initial classification. It causes the algorithm only converge to local optimal solution probably. This thesis mainly provides an improvement scheme and appropriate algorithm on the basis of discussing and analyzing ordinary simulated annealing, and applying its theory and method to clustering analysis. I advance a new cluster algorithm that is accelerated simulated annealing (ASA), in order to overcome some problems of K-means algorithm.And I advance a tunable dynamic algorithm of clustering numbers,it may provide some consideration for solving the problem of clustering and defining suitable number of clusters conveniently, on thebasis of simulated researching and discussing and analyzing the actual problems, I compare classification results of the two kinds of clustering algorithm and choose the convergent rate of optimal solution and so on, so that I test and verify validity of the given clustering algorithm.