Handling Constrained Multiobjective Optimization Problems via Bidirectional Coevolution.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Evolutionary Computation

Background:

• Constrained multiobjective optimization problems (CMOPs) present challenges due to conflicting objectives and constraints.
• Optimal solutions for CMOPs often reside on the boundaries of the feasible region.
• Existing methods may struggle to efficiently search these boundaries from both feasible and infeasible sides.

Purpose of the Study:

• To propose a novel constrained multiobjective evolutionary algorithm (BiCo) that addresses the challenges of CMOPs.
• To enhance the search capability by exploring the feasible region boundary from both feasible and infeasible directions.
• To improve the convergence and diversity of solutions for CMOPs.

Main Methods:

• BiCo employs a bidirectional coevolutionary approach with two populations: a main population and an archive population.
• The main population is updated using a modified NSGA-II with constraint-domination to move towards the feasible region and Pareto front.
• The archive population is updated using nondominated sorting and angle-based selection to approach the Pareto front from the infeasible side, ensuring diversity.

Main Results:

• BiCo demonstrated competitive performance against eight state-of-the-art constrained multiobjective evolutionary optimizers.
• Comprehensive experiments on benchmark and real-world CMOPs validated the algorithm's effectiveness.
• The bidirectional approach allows BiCo to effectively approach the Pareto front from complementary directions.

Conclusions:

• BiCo offers a robust and effective strategy for solving constrained multiobjective optimization problems.
• The bidirectional coevolutionary framework enhances the ability to locate Pareto-optimal solutions on constraint boundaries.
• BiCo shows significant potential for practical applications requiring efficient CMOP solutions.