Controlled elitist non-dominated sorting genetic algorithms for better convergence

Abstract

Preserving elitism is found to be an important issue in the study of evolutionary multi-objective optimization (EMO). Although there exists a number of new elitist algorithms, where elitism is intro- duced in different ways, the extent of elitism is likely to be an important matter. The desired extent of elitism is directly related to the so-called exploitation-exploration issue of an evolutionary algorithm (EA). For a particular recombination and mutation operators, there may exist a selection operator with a particular extent of elitism that will cause a smooth working of an EA. In this paper, we suggest an approach where the extent of elitism can be controlled by fixing a user-defined parameter. By applying an elitist multi-objective EA (NSGA-II) to a number of dificult test problems, we show that the NSGA-II with controlled elitism has much better convergence property than the original NSGA-II. The need for a controlled elitism in evolutionary multi-objective optimization, demonstrated in this paper should encourage similar or other ways of implementing controlled elitism in other multi-objective evolutionary algorithms.