Alleviate the Hypervolume Degeneration Problem of NSGA-II

摘要

A number of multiobjective evolutionary algorithms, together with numerous performance measures, have been proposed during past decades. One measure that has been popular recently is the hypervolume measure, which has several theoretical advantages. However, the well-known nondominated sorting genetic algorithm II (NSGA-H) shows a fluctuation or even decline in terms of hypervolume values when applied to many problems. We call it the "hypervolume degeneration problem". In this paper we illustrated the relationship between this problem and the crowding distance selection of NSGA-H, and proposed two methods to solve the problem accordingly. We comprehensively evaluated the new algorithm on four well-known benchmark functions. Empirical results showed that our approach is able to alleviate the hypervolume degeneration problem and also obtain better final solutions.