Study Of Particle Swarm Optimization

Optimization algorithm is a method that solves problems related to optimal issues. Traditional engineering optimization algorithms have been applied widely and achieved a big success across many fields. While, as the complexity a nd scale of optimal issues raise sharply, traditional engineering optimization algorithms may fail sometimes. Optimization algorithms via swarm intelligence develop ed soon these years. As a member of swarm intelligence algorithms, PSO(Particle Swarm Optimization) has attracted more and more attention and acceptances as the following advantages: few parameters, simple procedures, relatively excellent performance, and so on. Many improvements and similar swarm algorithms are proposed based on PSO in these years, but it can be hard for these algorithms to balance convergence precision and efficiency with some of them keep ing low precision or have few ways in improve the local search abilities of PSO. So it is essential to research PSO further.LHS(Latin Hypercube Sample) is a sampling experiment design method with a good performance. It is used in sampling experiment as usual, while few use to optimization algorithms are reported. The feasibility of appling sampling methods to optimal processes is studied in this paper, and PSO via sampling strategy(SSPSO) is proposed. Firstly, update of the particles’ speed and location via LHS is proposed, middle swarm and middle particles are produced by LHS, and the best particles are sent to the next iteration for speeding up the convergence process. Then, correction of the global best location via random sampling is proposed so that better locations in searching spaces can be found in time. Finally, “double sampling” LHS is proposed, and the “rough sampling and fine sampling” are used for local search to improve the convergence precision. The experimental results show that SS-PSO can improve the PSO’s convergence precision and speed.Two parameters in basic PSO, inertia weight and acceleration coefficients, lack of self-adaption when search the space. The superior information of every dimension can’t be saved properly and is easy to lose. The basic PSO is hard to jump out of local optima as well. Based on the studying of searching process in PSO, particle swarm optimization combining rapid information communication with local search(CRLPSO) is proposed.Firstly, new self-adaptive inertia weight and acceleration coefficients are designed so that particle can adjust its velocity in every period of the algorithm according to the population’s status. Then, benefiting from the exchange process of genes, the rapid particle information communication mechanism is presented to speed up the convergence process. At last, traditional Hooke-Jeeves search method is improved and applied to global best particle. The new algorithm can imp rove the basic PSO’s convergence precision and speed effectively.The main work of this paper is to study the improvement of PSO, but with the limitation of the randomness of improvement tools and the amount of computation, the two algorithms we proposed still have a large amount of computatio n. The optimal results of some benchmarks are not so satisfactory and the computation time is too long especially when solving the high dimensions problems. The convergence precision and the efficiency of algorithms still have a large space to improve. The algorithms proposed will be improved further and applied to practical engineering issues to reflect the practical value of the algorithms in future work.