ABSTRACT
From the experiment on ZDT series problems through NSGA-II, we found that: (1) based on the condition of constant evolution generation, with the increase of population size, NSGA-II still has good robustness and its relationship curve of IGD population is monotonic and similar to Aa–Bx + C; (2) NSGA-II is influenced by evolution generation, showing two-stage characteristics on ZDT series problems: the first-stage characteristic exists in the stability of a normal evolutionary period, and the second-stage characteristic is volatility shown under the big evolution generation, whose variation curve of the IGD-generation is not monotonic and similar to Ae–Bx sin(Cx + D) + E. In addition, compared with the mixed NSGA-II and RM-MEDA algorithm, we found that the hybrid algorithm can produce fine-adjustment, thus ensuring the stability of NSGA-II.
