2014 Volume 64 Issue 4 Pages 510-518
The standard particle swarm optimization (PSO) is an evolutionary algorithm for finding optimal regions of complex search spaces through the interaction of individuals in a population of particles. Many numerical experiments have reported that PSO can provide high-quality solutions. However, PSO cannot yield good solutions for large-scale problems; that is, high-dimensional problems. For this problem, this paper proposes new algorithms using the idea of effective particles moving. The basic idea is that moving of particles is updated according to the Cauchy distribution. The Cauchy distribution has a Gaussian-like peak and Lorentzian wing that imply occasional long jumps among local sampling. This proper trade-off between local search and semi-global search allows effective moving of particles. Moreover, this algorithm has been improved using the new search procedure that assumes a particle swarm is frequently reconstructed by the idea of scatter search. This search procedure operates on a set of good particles, the reference set, by combining these particles to create new ones. It uses strategies for search diversification and intensification by introducing the Cauchy distribution for the process of combining particles. The goal of this methodology is to enable the implementation of solution procedures that can derive new better particle swarms from combined elements. Finally, comparing the results on a set of benchmark multimodal functions, the proposed algorithm shows competitive performance with improved convergence speed and high optimizing precision for difficult problems.
