The Research Of Particle Swarm Optimization Based On Classification

Particle Swarm Optimization (PSO), a novel evolutionary computation technique originally inspired by certain social behaviors of bird flocking and fish schooling, is an adaptive stochastic algorithm based on swarm searching strategies. Due to its simplicity of implementation, little number of parameters and fast convergence, PSO has been approved to be a good global optimization algorithm and has won more and more attention. However, there are some drawbacks, such as premature convergence, poor diversity of optimization and so on.To overcome the shortcomings of PSO algorithm, this paper gives a brief introduction of PSO algorithm and some related background. The PSO algorithm theory, the algorithm framework model and the mechanism model are also deeply studied. Based on these theoretical investigations, this paper presents two improved algorithms of PSO. The experiments show that the improved algorithm is feasible and effective. This paper is carried out as follows:Firstly, this paper studies and analyzes PSO algorithm. The principles, implementation and developing causes of PSO algorithm are elaborated here. Then the author probes into the information exchanging model of PSO and studies several classical topology structure of particle swarm. In addition, some typical strategies of perfecting PSO algorithm are also introduced and the principles and methods of those algorithms are elaborated, all of which help the understanding of the meanings of PSO research and development.Secondly, this paper presents a PSO based on Classification, into which the idea of classification is introduced. According to particle fitness value, the particle swarms are divided into three groups, the superior, the common and the worse. Different particle swarm executes different optimization method, through which the diversities of particle swarms are guaranteed in different classifications and the optimization ability of PSO algorithm is improved.Thirdly, based on the previous study, Hybrid Particle Swarm Optimization based on classification is further presented. This algorithm adopts dynamic method to divide the particle swarms into three groups based on the above classification, and the amounts of particle in different classifications are changed adaptively. The blending of improving Simplex Method (SM) and half stochastic optimization method into different level particle swarms optimization greatly ensures the diversity and convergence of the improving PSO algorithm, improving the optimization ability and efficiency of PSO algorithm.Finally, the author summarizes the study as well as puts forward further research expectation.