Multi-Objective Evolutionary Algorithm Fully Based on Unbounded Archive for Problems Requiring Very Expensive Solution Evaluations

We propose a new MOEA framework, unbounded archive-based multi-objective evolutionary algorithm (UABMOEA), in which the parent population is selected from the unbounded archive in every generation update. We apply it to the well-known and frequently used MOEA, NSGA-II to examine the effect of our proposed UABMOEA framework. In our computational experiments, we indicate that environmental selection from the unbounded archive in every generation update clearly outperforms NSGA-II and NSGA-II with periodical parent population selection from the unbounded archive, especially in cases with 4 or more objectives. We also show the increase in computation time by UABMOEA is negligible which is an important consideration given the high cost of the long evaluation time for real-world design problems.