A many-objective evolutionary algorithm based on three states for solving many-objective optimization problem.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Evolutionary Computation

Background:

• Many-objective optimization problems (MaOPs) involve simultaneously optimizing five or more objectives.
• Existing algorithms face challenges such as Pareto resistance and difficulties in maintaining population diversity.
• These challenges hinder the effective identification of high-quality solutions in MaOPs.

Purpose of the Study:

• To propose a novel many-objective evolutionary algorithm (MOEA/TS) designed to overcome limitations in current MaOP research.
• To enhance the performance of evolutionary algorithms in handling problems with a large number of objectives.
• To improve both solution quality and diversity in many-objective optimization.

Main Methods:

• Development of a feature extraction operator to guide the evolution using high-quality solution features.
• Introduction of the "individual importance degree" concept within Pareto front layers to differentiate solutions.
• Implementation of a repulsion field method to ensure population diversity and even distribution on the Pareto front.
• Design of a concurrent algorithm framework with three distinct, switchable states for specialized tasks.

Main Results:

• The proposed MOEA/TS algorithm demonstrated competitive performance compared to seven advanced many-objective optimization algorithms.
• Experimental results indicate improved ability to handle Pareto resistance and maintain population diversity.
• The feature extraction and individual importance degree effectively assisted in evolving higher-quality solutions.

Conclusions:

• MOEA/TS offers a promising approach for tackling many-objective optimization problems.
• The algorithm's novel components effectively address key challenges in the field.
• The concurrent framework and state-switching mechanism contribute to its enhanced competitiveness.