Improving the NSGA-II performance with an external population

Abstract

The NSGA-II algorithm is among the best performing ones in the area of multiobjective optimization. The classic version of this algorithm does not utilize any external population. In this work several techniques of reintroducing specimens from the external population back to the main one are proposed. These techniques were tested on multiobjective optimization problems named ZDT-1, ZDT-2, ZDT-3, ZDT-4 and ZDT-6. Algorithm performance was evaluated with the hypervolume measure commonly used in the literature. Experiments show that reintroducing specimens from the external population improves the performance of the algorithm.