EMO algorithms on correlated many-objective problems with different correlation strength

It has been pointed out in many studies that standard Pareto dominance-based EMO (evolutionary multi-objective optimization) algorithms do not work well on many-objective problems with four or more objectives. However, it has also been demonstrated in some studies that many-objective problems are not always difficult for such an EMO algorithm when many objectives are highly correlated or dependent. In this paper, we examine the performance of well-known EMO algorithms on many-objective problems with weakly correlated objectives as well as those with highly correlated objectives. As test problems, we generate many-objective problems with 4–10 objectives from a two-objective knapsack problem with randomly generated two objectives. Each of the other objectives is correlated to one of the two objectives. The strength of the correlation can be arbitrarily specified by a parameter value used for generating correlated objectives. Performance of well-known EMO algorithms such as NSGA-II and MOEA/D is examined by applying them to our test problems with different correlation strength.