Research On K-center Combination Optimization Clustering Algorithm Based Onartificial Fish Swarm

The data sets are divided into similar classes by Cluster Analysis without any priorknowledge, the useful characteristics can be selected to distinguish the different groups,Cluster Analysis plays a significant role in knowledge classification and informationdiscovery. After researching the clustering methods and swarm intelligence algorithm,it is found that using swarm intelligence technology to solve clustering problem is veryeffective, so K-center combination optimization clustering algorithm based on artificialfish swarm is proposed by this paper, the research results are as follows:(1) Based on study of previous clustering method, K-center combinationoptimization clustering method is proposed based on combination optimization thought,the combination function and criterion function which based on combination functionare proposed, be constrained by the combination function, the optimal value ofclustering criterion function is obtained, attribute weight and dynamic class centers isintroduced and the complete design of clustering model is given in this paper, themethod is efficient in handling large data sets.(2) After studying the basic Artificial Fish Swarm Algorithm(AFSA), in order toimprove the convergence speed and the accuracy of the algorithm, an improved AFSA isproposed in this paper, artificial fish vision threshold is reseted, the state increment isadded in artificial fish praying behavior, crowding factor is improved in swarmingbehavior and following behavior，using the attenuation factor to constraint step, at last using Matlab simulation to verify the algorithm is improved in stability, convergencerate and the correct rate.(3) It is found that K-center combination optimization clustering algorithm may beaffected by the initial parameters and difficult to achieve global optimization. Based onthe shortcoming of this clustering method, K-center combination optimizationclustering algorithm based on artificial fish swarm is proposed by this paper, fish codingand food concentration function are designed, the food source as the cluster center,infeasible solutions is discarded adaptively, location of artificial fish information isclustering results, it is shown by experiment that the improved algorithm which forsolving the problem is adaptive and has the good ability to overcome the problem ofgetting into local extremum, at last it is shown in comparative experiment that thisdesign has better accuracy and clustering effects in UCI datasets.