A NON-HOMOGENOUS PARTICLE SWARM OPTIMIZATION WITH MULTIPLE SOCIAL STRUCTURES
Pisut Pongchairerks, Voratas Kachitvichyanukul · 2005
Particle Swarm Optimization (PSO) is a population based stochastic optimization technique. One of its most important parameters is the social structure of PSO. Each structure’s performance depends on geography of the search problem. This paper introduces two new versions of particle swarm optimization algorithm. The first proposed version is GLN-PSO. It is based on a structure that is built by combining previously published structures. The second is GLNR-PSO, a non-homogenous PSO algorithm that is the same as the first but allow for some particles to have different parameters. The two proposed algorithms were tested using the benchmark test functions previously published, namely Sphere, Rosenbrock, Rastrigin and Griewank functions. The results of the experiments indicated that the first proposed algorithm GLN-PSO outperformed standard PSO and FDR-PSO on all the test functions. The second proposed algorithm GLNR-PSO further outperformed GLNPSO on the Sphere, Rosenbrock, and Griewank functions.