An Active Elitism Mechanism for Multi-objective Evolutionary Algorithms

Classical (or passive) elitism mechanisms in the MOEA (Multi-objective Evolutionary Algorithm)literature have a holding/sending back structure. In this paper, an active elitism mechanism formulti-objective evolutionary algorithms is proposed. In the active elitism mechanism, a set of elite(or non-dominated) individuals is excited by genetic operators (crossover/mutation) in the archivein the hope of generating better and more diverse individuals than themselves. If a set of excitedelites are any better than originals, then archive can be viewed as a place of active solutionprovider rather than a static storage place. The main motivation behind this approach is that eliteindividuals are inherently the closest individuals to the solution (of any optimization problem onhand) and exciting those individuals can likely generate more significant outcomes than a faraway one. The proposed active elitism mechanism is embedded into well-known multi-objectiveSPEA and SPEA2 methods (named ACE_SPEA and ACE_SPEA2 respectively) and compared tothe original methods using four unconstrained test problems. The active elitist versions of SPEAand SPEA2 maintain better spread and convergence properties than the original methods on alltest problems. The proposed active elitism mechanism can easily be integrated into existingmulti-objective evolutionary algorithms to improve their performance.