Search Pattern Analysis And Application Of The Particle Swarm Optimization

As the better understand of the world, the complexity of the optimization problems that we attempt to solve is ever increasing. The Particle Swarm Optimizer (PSO) was introduced in 1995 and widely used nowadays, yet very few formal analyses of the behavior of the algorithm have been published. Most of the published work was concerned with empirical results obtained by lots of experiments. These results bring valuable understand of the natural of PSO algorithm, but could not explain effectively the reason of premature and could not guide a better improvement. So it is necessary and meaningful to propose a general, theoretical description of the behavior of the algorithm.This thesis starts with the analysis of Benchmark optimization problems. A novel classification method based on the "current domain" is presented. This method could combine with the algorithm analysis effectively, and give an intuitive understanding on how the problems affect the performance of the optimizer. Secondly, based on the published works of the PSO convergence study, another search mode, called intensive search mode, is put forward, which occurs after the convergence mode of PSO. Analyzing intensive search mode could not only figure out the final behavior of particles, but also get an approximate possibility of finding the global best fitness value. Then a new algorithm based on social behavior is presented:Social Particle Swarm Optimizer (SPSO). Each particle in SPSO interacts with others by follow threshold, and this mechanism effectively prevents diversity loss. The Benchmark experiments show that SPSO significantly improves the outcome performance, convergence rate and adaptation of different problems. The last part is application of SPSO algorithm on wireless sensor network placement optimization, and simulation shows that SPSO performs well on part placement problem, complete placement problem and abundance placement problem.