Hybrid Optimization Strategy Based On Partical Swarm Optimization And Application Research

Particle Swarm Optimization(PSO) was inspired by the prey behavior of bird and originally introduced by Kennedy and Eberbart in 1995.It is a new kind of evolutionary computation in the swarm intelligence theory.PSO guides the optimal searching process based on the swarm intelligence which produced by the particles'cooperation and competition.As a kind of swarm intelligence,PSO is charactered by stong universal property and has proven to be a powerful global optimization method.lt has been widely applied in function optimization and shows great potential in particle.The dissertation focuses on the principles,theory,and application of PSO.The main achievements of this dissertation include:1. A novel Cultural Particle Swarm Optimization with Adaptive Guidance (AG-CPSO) algorithm is proposed. This algorithm model consists of a PSO-based main population space and a belief space,which repectively has its own population to evolve independently and parallel.Expecially, colony fitness variance is introduced to population space. By calculating the colony fitness variance,decisions are made whether to have mutate operation on population space, which is prone to get into local best position in the last period of evolution.The improved algorithm can make better use of mechanism of dual evolution and dual promotion in AG-CPSO,in which the mutaion operator adopted by influence function may disturb the structure and convergence of PSO algorithm in population space.Experimental results prove that the performance of new algorithm gets ahead of PSO and CPSO.2.A novel Bee Evolution Particle Swarm Optimization(BEPSO) algorithm is proposed. The algorithm combines the genetic operation of Bee Evolution Genetic Algorithm(BEGA) such as elite reservation,dual selection,crossover and mutation with the update rules of velocity and situation of PSO. It combines the advantage of BEGA's global search and divers population,PSO's strong evolutionary direction and convergence. The new method can deal with the problem of premature convergence and slow search speed effectively. Computational results prove that the performance of new algorithm gets ahead of PSO and BEGA.Finally, new algorithms have been applied to train neural network.Results show that both the two proposed algorithms can improve the classification accuracy while speeding up the convergence process and avoiding premature effective at the same iterations.