The Study Of Particle Swarm Optimization With Varying Population Size

Particle Swarm Optimization (PSO) was originally introduced by Kennedy and Eberhart in 1995 and is inspired by the results obtained from the simulation of social behavior of bird flocking. PSO is a new kind of population-based evolutionary computation algorithm which can be used to solve all the optimization problems and results are not related with the initial population. Otherwise, as a new evolutionary algorithm, PSO is simple in principles, easy to programming and has little parameters to adjust. For the large class of optimization problems, it has been shown through simulation that has good convergence speed and the global optimum. Now, it has been widely applied to many areas, such as nonlinear optimization, artificial neural network and fuzzy system control etc..When PSO is used to solve the high-dimensional and multi-peak function optimization problems, it is often trapped on local optimum, i.e., premature convergence happens. Based on the review and analysis of the intelligent evolutionary algorithm with varying population size, combined with the genetic algorithm, a hybrid PSO with varying population size (VPPSO) is presented in this paper. And the fundamental mechanism and flowchart of VPPSO are given. Meanwhile, the several operators involved in adjusting population size and reservation probability of particles are discussed in detail. Through the experiments of four multi-peak test functions with high dimension,the results show the VPPSO algorithm is effective , and the parameters on the impact of algorithm performance is analyzed. To overcome the deficiency of VPPSO, the VPPSO with cultivated period is proposed to avoid frequently changes of the population size. The cross operator of VPPSO is improved to increase the diversity of population. Finally, the simulation results show the convergence speed and performance of global optimization of VPPSO is improved efficiently.