A robust evolutionary framework for multi-objective optimization

Abstract

Evolutionary multi-objective optimization (EMO) methodologies, suggested in the beginning of Nineties, focussed on the task of finding a set of well-converged and well-distributed set of solutions using evolutionary optimization principles. Of the EMO methodologies, the elitist non-dominated sorting genetic algorithm or NSGA-II, suggested in 2000, is now probably the most popularly used EMO procedure. NSGA-II follows three independent principles-domination principle, diversity preservation principle and elite preserving principle-which make NSGA-II a flexible and robust EMO procedure in the sense of solving various multi-objective optimization problems using a common framework. In this paper, we describe NSGA-II through a functional decomposition following the implementation of these three principles and demonstrate how various multi-objective optimization tasks can be achieved by simply modifying one of the three principles. We argue that such a functionally decomposed and modular implementation of NSGA-II is probably the reason for it's popularity and robustness in solving various types of multi-objective optimization problems.