Evaluation of Three Steady-State NSGA-III Offspring Selection Schemes for Many-Objective Optimization
Courtney Powell, Phyo Thandar Thant, Masaharu Munetomo · 2016
In this study, we implemented three steady-state versions of NSGA-III that differ by the manner in which the offspring combined with the parent population is selected. These three schemes were then evaluated on the standard problem sets DTLZ1-4 in terms of four popular criteria: inverse generational distance (IGD), hypervolume (HV), convergence, and diversity. The results obtained suggest that utilizing a selection scheme in which the offspring is selected from the first non-dominated rank results in better solutions than other steady-state offspring selection schemes.