A Coevolutionary Algorithm Based on Dominance and Decomposition for Constrained Multiobjective Optimization.

Area of Science:

• Optimization
• Evolutionary Computation
• Computational Intelligence

Background:

• Constrained multiobjective optimization problems (CMOPs) are challenging research areas.
• Existing constrained multiobjective evolutionary algorithms (CMOEAs) often use dominance-based or decomposition-based frameworks independently.
• These frameworks possess complementary strengths for different problem types, suggesting potential benefits from integration.

Purpose of the Study:

• To propose a novel coevolutionary algorithm that synergistically integrates dominance-based and decomposition-based frameworks for CMOPs.
• To leverage the respective advantages of each framework for improved search performance.
• To address the limitations of isolated framework usage in existing CMOEAs.

Main Methods:

• A coevolutionary algorithm is developed with two coevolved populations, each utilizing a different framework.
• The dominance-based population optimizes a dynamic problem with a tolerance-based selection strategy for diversity.
• The decomposition-based population adapts its focus from unconstrained to constrained Pareto fronts using stage identification, objective switching, and relevance-based selection.

Main Results:

• The proposed algorithm demonstrates superior performance compared to 11 state-of-the-art CMOEAs.
• Performance advantages were validated across four benchmark suites and five real-world CMOPs.
• Information sharing between populations during evolution enhances mutual reinforcement.

Conclusions:

• Integrating dominance-based and decomposition-based frameworks in a coevolutionary manner offers significant performance improvements for CMOPs.
• The proposed algorithm effectively leverages complementary strengths, overcoming limitations of isolated approaches.
• This integrated strategy provides a promising direction for advancing CMOEA research.