Handling Constrained Many-Objective Optimization Problems via Problem Transformation.

Area of Science:

• Optimization
• Evolutionary Computation
• Multi-objective Optimization

Background:

• Constrained many-objective optimization problems (CMaOPs) require balancing objective optimization and constraint satisfaction.
• Existing feasibility-driven algorithms often prioritize constraints, potentially sacrificing population diversity and leading to suboptimal solutions in high-dimensional spaces.

Purpose of the Study:

• To propose a problem transformation technique to convert CMaOPs into dynamic CMaOPs (DCMaOPs).
• To develop a tailored algorithm, DCNSGA-III, that effectively handles both constraints and multiple objectives simultaneously.
• To improve the ability of algorithms to navigate large, discrete infeasible regions in the search space.

Main Methods:

• A problem transformation technique is introduced to create DCMaOPs from CMaOPs.
• The reference-point-based NSGA-III algorithm is adapted to the DCMaOP framework, resulting in DCNSGA-III.
• A mating selection mechanism and an environmental selection operator are designed to generate and select high-quality ε-feasible solutions.

Main Results:

• DCNSGA-III was evaluated on benchmark CMaOPs with 3, 5, 8, 10, and 15 objectives.
• The algorithm was compared against six state-of-the-art constrained many-objective evolutionary algorithms (CMaOEAs).
• Experimental results demonstrated the high competitiveness of the proposed DCNSGA-III algorithm.

Conclusions:

• The proposed problem transformation and the DCNSGA-III algorithm offer a competitive approach for solving CMaOPs.
• The method effectively balances objective optimization and constraint satisfaction, particularly in challenging high-dimensional scenarios.
• DCNSGA-III shows promise in overcoming limitations of existing feasibility-driven CMaOEAs.