Improving Learning Automata based Particle Swarm: An optimization algorithm

Abstract

Abstract-Numerous variations of Particle SwarmOptimization (PSO) algorithms have been recently developed, with the best aim of escaping from local minima. One of these recent variations is PSO-LA model which employs a Learning Automata (LA) that controls the velocity of the particle. Another variation of PSO enables particles to dynamically search through global and local space. This paper presents a Dynamic Global and Local Combined Particle Swarm Optimization based on a 3-action Learning Automata (DPSOLA). The embedded learning automaton accumulates the information from individuals, local best and global best particles then combines them to navigate the particle through the problem space. The proposed algorithm has been tested on eight benchmark functions with different dimensions. The work is unique from its test bed; evaluations contain large population size (150) and high dimension (150). The results show that, fitness and convergence pace is better than traditional PSO, DGLCPSO and previous PSO based LA algorithms. I.INTRODUCTION One of the common methods for optimization of continuous nonlinear functions is particle swarm optimization (PSO) [1], [2]. This algorithm has two main concepts: artificial life and collective intelligent of brutes. The inertia weight (w) [3][4][5] is one of the PSO parameters to bring about a balance between the exploration and exploitation characteristics of PSO. Cooperative PSO (CPSO) is a variation on standard PSO which used multiple swarms [6]. Recently it is proposed a combined dynamic global and local particle swarm optimization (DGLCPSO) algorithm to improve the performance of original PSO [7].Learning automaton (LA) [8] is a multi-propose casual tool, which is an expansion model for learning machines. PSO, like other stochastic search methods, is highly sensitive to adjustment of affective parameters. Recently a LA based PSO model [9][10][11][12] called PSO-LA has been reported to improve the performance of PSO. In [10] a PSO-LA model is proposed in which the LA is responsible for configuring the behavior of the swarm and also balancing the process of global and local search. In [11] a Cellular LA (CLA) based discrete PSO is introduced as a solution of premature convergence. To reduce the probability of trapping PSO-LA into local minima, in [12] four modifications on PSO-LA model are

An efficient oscillating inertia weight of particle swarm optimisation for tracking optima in dynamic environments

“…Finally,p g is the globally best position found so far by all the particles in the population. Several modifications have been proposed for improving the performance of PSO (Hasanzadeh, Meybodi, & Ebadzadeh, 2013;Hasanzadeh, Meybodi, & Ghidary, 2011;Soleimani-Pouri, Rezvanian, & Meybodi, 2014). One of the most widely used improvements is the introduction of the inertia weight parameter by Shi and Eberhart (Shi & Eberhart, 1998).…”

Adaptive cooperative particle swarm optimizer

“…In order to balance the process of global and local searched in this new PSO-LA algorithm, one learning automaton is assigned to each particle of the swarm. A recent advancement in combining learning automata with PSO was Dynamic PSO-LA (DPSOLA) [63]. The proposed model used three types of existing information, namely individual, neighboring and swarm information.…”

A robust heuristic algorithm for Cooperative Particle Swarm Optimizer: A Learning Automata approach

“…Also a new hybrid method of optimization which called PS�-LA [11][12][13] has been emerged. In PS�-LA algorithms an LA or a group of LAs is assigned to the whole popUlation or each particle of the population.…”