Generalization of Dominance Relation-Based Replacement Rules for Memetic EMO Algorithms

In this paper, we generalize the replacement rules based on the dominance relation in multiobjective optimization. Ordinary two replacement rules based on the dominance relation are usually employed in a local search (LS) for multiobjective optimization. One is to replace a current solution with a solution which dominates it. The other is to replace the solution with a solution which is not dominated by it. The movable area in the LS with the first rule is very small when the number of objectives is large. On the other hand, it is too huge to move efficiently with the latter. We generalize these extreme rules by counting the number of improved objectives in a candidate solution for LS. We propose a LS with the generalized replacement rule for existing EMO algorithms. Its effectiveness is shown on knapsack problems with two, three, and four objectives.