Fitness-Dependent Hybridization of Clonal Selection Algorithm and Random Local Search

Abstract

Artificial immune systems (AIS) and local search algorithms have remarkable differences in the structure of mutation operators. Thus AIS algorithms may be more efficient at the beginning of optimization, while local search algorithms are more efficient in the end, when we need to do small improvements. Our goal is to combine several mutation operators in one algorithm so that the new algorithm will be efficient on fixed budget and will reach optimum within reasonable time bounds. We propose to select mutation operators used in AIS and local search according to a specific exponential probability function which depends on the fitness of the current individual. During the experimental study, we constructed hybrids from AIS mutation operator CLONALG (Clonal Selection Algorithm) and RLS mutation operator (Random Local Search) and used them to solve OneMax problem. We compared the proposed method with a simple hybrid algorithm and empirically confirmed the hypothesis that hybrids are efficient on fixed budget and need only a slightly higher number of iterations to reach the optimum.