Comparative study of the use of fuzzy logic in improving particle swarm optimization variants for mathematical functions using co-evolution

Abstract

Particle Swarm Optimization (PSO) is a nature-inspired computation technique based on the social behavior of birds flocking or fish schooling, and can be also viewed as a biologically inspired computational search and optimization method. Since its introduction by Kennedy and Eberhart in 1995, several variants of the original Particle Swarm Optimization technique have been created to improve the quality of the solutions, to improve the speed of convergence, and to avoid getting trapped in local minima. The main proposal is to design a new approach to dynamically adapt some parameters of the PSO algorithm, such as the weight of inertia and learning factors. In this regard, fuzzy logic is applied to the design of the proposed approach. To measure performance, a comparison of different approaches for the inertia weight is performed; these approaches are: constant, random adjustments, linear decreasing, nonlinear decreasing and fuzzy adaptive inertia. The co-evolution concept is also applied to improve the performance of the PSO algorithm for problems of higher degree of complexity. A set of 11 mathematical functions is used to validate the proposed approach. The benchmark functions that are considered are widely used in this field of study.