New Learning Automata based Particle Swarm Optimization Algorithms

Particle swarm optimization (PSO) is a population based statistical optimization technique which is inspired by social behavior of bird flocking or fish schooling. The main weakness of PSO especially in multimodal problems is trapping in local minima. Recently a learning automata based PSO called PSO-LA to improve the performance of PSO has been reported. PSO-LA uses one learning automaton for configuring the behavior of particles and also creating a balance between the process of global and local search. Although PSO-LA produces better results than the standard PSO but like standard PSO it may trap into local minima. In this paper four improvements on PSO-LA are proposed. These improvements are proposed to reduce the probability of trapping PSO-LA into local minima. Unlike PSO-LA which uses one learning automaton to guide all particles, in the proposed PSO algorithms one learning automaton is assigned to each particle as the article brain which controls the particle movement in the search space. The proposed algorithms are tested on 8 benchmark functions. The results have shown that the proposed PSO algorithms are superior to standard PSO, PSO with inertia weight (PSOw) and previously reported LA based PSO algorithms.

Keywords:

Particle Swarm Optimization , Learning Automata , PSO-LA , Function Optimization