Adaptive Reference Point Generation for Many-Objective Optimization Using NSGA-III

Atiya Masood, Gang Chen, Yi Mei, Mengjie Zhang

Abstract

In Non Dominated Sorting Genetic Algorithm-III (NSGA-III), the diversity of solutions is guided by a set of uniformly distributed reference points in the objective space. However, uniformly distributed reference points may not be efficient for problems with disconnected and non-uniform Pareto-fronts. These kinds of problems may have some reference points that are never associated with any of the Pareto-optimal solutions and will become useless reference points during evaluation. The existence of these useless reference points in NSGA-III significantly affects its performance. To address this issue, a new reference points adaptation mechanism is proposed that generates reference points according to the distribution of the candidate solutions. The use of this proposed adaptation method improves the performance of evolutionary search and promotes population diversity for better exploration. The proposed approach is evaluated on a number of unconstrained benchmark problems and is compared with NSGA-III and other reference point adaptation approaches. Experiment results on several benchmark problems clearly show a prominent improvement in the performance by using the proposed reference point adaptation mechanism in NSGA-III.
BibTeX
@inproceedings{masood2018adaptive,
  author = {Atiya Masood and Gang Chen and Yi Mei and Mengjie Zhang},
  title = {{Adaptive Reference Point Generation for Many-Objective Optimization Using NSGA-III}},
  booktitle = {Australasian Conference on Artificial Intelligence},
  year = {2018},
  doi = {10.1007/978-3-030-03991-2\_34},
}

Loading related content...

Figure