Fuzzy Clustering Analysis And Application Based On Particle Swarm Optimization

Fuzzy C-means (FCM) clustering algorithm is a common kind of fuzzy clusteringalgorithm. FCM algorithm is based on Euclidean distance but it doesn’t consider the effect ofthe sample. Weighted alternative fuzzy C-means (WAFCM) algorithm can solve the problemof unbalanced sample effectively, but WAFCM is sensitive to the initialization and is easilytrapped in local optimum.Particle swarm optimization (PSO) algorithm updates the velocity and position by simpleformula, and it optimizes the particles through repeated iteration. In order to improve theconvergent performance of the PSO algorithm, particle swarm optimization with aconstriction factor algorithm is adopted in this paper. In the later iteration of this algorithm,particles are accumulated in the direction of the global optimal position, then it will be lack ofthe diversity of the population. By means of ergodicity, randomicity and sensitivity of chaosoptimal algorithm, chaos particle swarm optimization algorithm is adopted, this algorithm notonly maintains the diversity of the population, but also avoids the premature convergenceeffectively, outperforming in precision and efficiency.This paper presents particle swarm optimization with a constriction factor based onchaos (CCFPSO) algorithm to analyze WAFCM clustering. First, CCFPSO algorithm uses theergodicity of the chaos to initialize clustering centers in the domain, evaluates particles byfitness value based on objective function of WAFCM algorithm, and it uses the constrictionfactor to update the velocity. Second, it is easy to judge premature convergence according tothe variance of the population’s fitness or the average of the distance between the particlesand the global optimal position, if premature, then the algorithm updates the position of theparticles by chaos, selects velocity randomly, and updates the global optimal position usingthe small extent of disturbance. The algorithm optimizes the clustering centers andmembership matrix through repeated iteration. Finally, it can determine the optimal number ofclustering by defining a kind of clustering validity function based on data set, clusteringcenters and membership matrix. Two experimental results show that WAFCM clusteringbased on CCFPSO algorithm benefits to overcome the defect of the fuzzy clustering and improves the ability of the global search. This method is effective.