A Novel Multi-objective Optimization Framework Combining NSGA-II and MOEA/D

Multi-objective Evolutionary Algorithms (MOEAs) are efficient tools for solving multi-objective problems (MOPs). Current existing algorithms such as Multi-Objective Evolutionary Algorithms based on Decomposition (MOEA/D) and Non-dominated Genetic Algorithm II (NSGA-II) have achieved great success in the field by introducing important concept such as decomposition and non-dominated sorting. It would be interesting to employ these crucial ideas of the two algorithms in a hybrid manner. This paper proposes a new framework combining the key features from MOEA/D and NSGA-II. The new framework is a grouping approach aiming to further improve the performance of the current existing algorithms in terms of overall diversity maintenance. In the new framework, original MOP is decomposed into several scalar subproblems and every group is assigned with two scalar subproblems as their new objectives in the searching process. Non-dominated sorting is conducted within each group respectively at every generation. Experimental results demonstrate that the overall performance of the new framework is competitive when dealing with 2-objective problems.