Towards a Better Diversity of Evolutionary Multi-Criterion Optimization Algorithms using Local Searches
Abstract
In EMO diversity of the obtained solutions is an important factor, particularly for decision makers. NSGA-III is a recently proposed reference direction based algorithm that was shown to be successful up to as many as 15 objectives. In this study, we propose a diversity enhanced version of NSGA-III. Our algorithm augments NSGA-III with two types of local search. The first aims at finding the true extreme points of the Pareto front, while the second targets internal points. The two local search optimizers are carefully weaved into the fabric of NSGA-III niching procedure. The final algorithm maintains the total number of function evaluations to a minimum, enables using small population sizes, and achieves higher diversity without sacrificing convergence on a number of multi and many-objective problems.
Authors: Haitham Seada, Mohamed Abouhawwash, Kalyanmoy Deb