Lamarckian Repair and Darwinian Repair in EMO Algorithms for Multiobjective 0/1 Knapsack Problems

Multiobjective 0/1 knapsack problems have been frequently used as test problems to examine the performance of evolutionary multiobjective optimization algorithms in the literature. It has been reported that their performance strongly depends on the choice of a constraint handling method. In this paper, we examine two implementation schemes of greedy repair: Lamarckian and Darwinian. In the Lamarckian implementation of greedy repair, a feasible solution is generated from an unfeasible one by removing items until all the constraint conditions are satisfied. That is, the genetic information of the unfeasible solution is modified. On the other hand, the genetic information of the unfeasible solution is not changed in the Darwinian implementation where greedy repair is used only to evaluate the fitness value of each solution. We compare these two implementation schemes with each other through computational experiments. We also compare greedy repair-based methods with a penalty function approach.