Particle Swarm Optimization Algorithm And Its Applications In Biological Data Clustering

Nowadays, with the rapid development of information technology, massive data have accumulated in many fields, and it has become an urgent demand for the social progress to mine the internal relations and the useful information of the data. As an important tool of data mining technique, clustering analysis has been paid close attention and researched by the majority of scientific researchers.Particle Swarm Optimization (PSO), firstly proposed by Kennedy and Eberhart, is a population-based adaptive stochastic algorithm, and has been practically proved to be an effective global optimization algorithm. Because of its simple principle, few parameters, fast convergence rate and good performances in complex optimization problems, PSO has gained widely applications in areas such as data mining, machine learning, power system and neural network.In this dissertation, through analysis of the mechanism of standard PSO algorithm, a fast chaotic particle swarm optimization algorithm based on K-method (FCPSO-K) and a ring neighborhood based chaotic particle swarm optimization (RCPSO) are proposed, and the single-objective clustering model based on the intra-cluster variation function is improved. Both algorithms are tested on some biological data sets and their performances are compared with other clustering algorithms. The main work is summarized as follows:(1) In order to overcome PSO algorithm’s drawback of sensitivity to initial value and further improve the convergence process of population, a fast clustering algorithm FCPSO-K is proposed. First, based on the concepts of k-distance, core point and density-reachable, a new initialization mechanism K-method is obtained to improve the initial population of clustering. Then, a fast chaotic PSO clustering algorithm(FCPSO-K) is proposed, in which the dynamic-objective constraint-handling method (DOM), K-method and cluster matching mechanism are incorporated into the chaotic PSO. The experimental results show that the FCPSO-K algorithm exhibits a superior performance in terms of global optimization ability, convergence rate, and stability compared to K-means, PSO and other four clustering algorithms based on PSO technology; moreover, FCPSO-K achieves more accurate clustering results.(2) To overcome PSO algorithm’s shortcoming of easily falling into local optima, ring neighborhood and chaotic factors are incorporated into the mechanism of PSO, and a ring neighborhood based chaotic particle swarm optimization (RCPSO) algorithm is proposed. The clustering results of chaotic PSO based on statistic ring structure and random ring structure with different sizes are compared on four test sets, and the neighborhood sizes of the population with the best overall clustering accuracy of RCPSO algorithm are obtained.(3) In order to improve the single-objective optimization clustering model, a novel intra-cluster density function is proposed as an auxiliary index to evaluate the effect of clustering, and then a multi-objective oriented PSO clustering algorithm(MOOPSO), which exploits two criteria indexes to guide the search process of population, is proposed. Experimental results show that, due to the intra-cluster density function, MOOPSO is more accurate than the PSO algorithm, and basically, the error rate of clustering results obtained by MOOPSO in the process of population convergence can be improved continuously. In addition, the simulation results show that the intra-cluster density function can effectively enhance the clustering accuracy of chaotic PSO algorithm.