Guiding Evolutionary Multiobjective Optimization With Generic Front Modeling.

Area of Science:

• Computational Intelligence
• Optimization Theory
• Evolutionary Computation

Background:

• Multiobjective evolutionary algorithms (MOEAs) approximate Pareto fronts (PFs) using candidate solutions.
• Existing MOEAs struggle with approximating PFs exhibiting complex geometries.

Purpose of the Study:

• To propose a generic front modeling method for evolutionary multiobjective optimization.
• To develop a novel MOEA driven by modeled approximate nondominated fronts.

Main Methods:

• Estimating nondominated front shapes by training a generalized simplex model.
• Developing an MOEA incorporating mating and environmental selection based on modeled fronts.
• Performance assessment via comparison with state-of-the-art algorithms on diverse benchmark problems.

Main Results:

• The proposed algorithm demonstrates consistent performance across various multiobjective optimization problems.
• Effective approximation of PFs with complicated geometries was achieved.
• Improved convergence and diversity in solution sets were observed.

Conclusions:

• The proposed generic front modeling method offers a robust approach for approximating complex Pareto fronts in MOEAs.
• The developed MOEA shows superior and consistent performance compared to existing methods.
• This work advances the field of evolutionary multiobjective optimization by addressing geometric complexities of PFs.