Study In Adaptive Particle Swarm Optimization With Crossover And Mutation Operator

Cluster analysis is an important task in data mining. The so-called "feather flock together". In order to optimize queries and find useful information in the large-scale data, cluster analysis in search for similarities between data attributes by cluster analysis, and then classifies the data objects.This article focuses on in-depth study the traditional partitioning method - K-means clustering method, which is used widely in the clustering analysis. The advantages of K-means algorithm are simple ideas and structure, efficient execution, and so on. However, K-means algorithm has two shortages which there are sensitive to the initial cluster centers and easily converge to local optimum. And these two main drawbacks are the reasons for restricted application of the algorithm. In order to remedy the drawback of K-means clustering method, this paper proposes hybrid K-means and the improved particle swarm optimization, which has global search capability and higher accuracy of classification. The main work includes:(1) The study of inertia weight factor of the standard PSO. Because of the particle has dependence on inertia weight w in flight direction and speed. Therefore, this paper proposed an improvement scheme on inertia weight, which is with the particle swarm algorithm implementation process dynamics Adaptive inertia weight.(2) Designed based on fitness crossover operator, thus ensuring the diversity of the population of particles. This operator not only enhances the global search ability of particles, but also speeds up the convergence rate of PSO. This paper proposed based on the standard deviation of cluster fitness mutation operator. The mutation probability is controlled by the standard deviation of cluster fitnessσ² and optimal value of the current theory (f_gbest). Meanwhile, construct the straightforward fitness function. Fitness function is crucial in particle swarm algorithm. Especially in this paper, it not only directly affects the crossover operation, but also determines the mutation probability. So that it will affects the optimization of the cluster center and the situation of the cluster. In this paper according to the principle of total cluster dispersion build the fitness function(3) Integrated the improved particle swarm optimization and K-means clustering algorithm. This article gives the encoding scheme of the particle and the process of the blend algorithm. Through a combination of the two algorithms to remedies the drawback of the traditional K-means algorithm. Finally, the based on crossover and mutation operator adaptive particle swarm optimization used to the UCI database, and compared the K-means clustering methods, GA, CKPSO. The results of the experiment can be seen the based on crossover and mutation operator adaptive particle swarm optimization than these three algorithms has a better fitness value and the classification accuracy and the algorithm is more stable, therefore, this method remedy the shortage of the traditional K-means clustering algorithm is sensitive to the initial cluster centers, and easily converge to local optimal.