A New Fitness Assignment and Parent Selection Strategy within an Evolutionary Algorithm for Constrained Optimization Problems


Abstract

Evolutionary algorithms (EA) are search strategies that mimic the process of natural evolution. All EA's have two fundamental strategies; a selection and a recombination strategy both o f which are known to largely influence the performance of the algorithm. The selection strategy ensures fitter individuals have a greater chance of survival and a greater participation in mating while the recombination strategy aims to inherit meaningful parent properties. In this paper a new fitness assignment scheme and a new parent selection strategy is proposed. The individuals are assigned separate fitness values in the objective a nd the constraint space unlike most EAs that use a single fitness measure for selection. The parent selection mechanism employed in the a lgorithm is both elitist and adaptive. The recombination strategy is based on a parent centric operator that explores the neighborhoods of good parents in search for better ones. In this paper we present t he results obtained b y our algorithm and compare it with the reported results on a suite of six single objective constrained test problems.