Adaptive Weighted Strategy Based Integrated Surrogate Models for Multiobjective Evolutionary Algorithm.

Area of Science:

• Computational intelligence
• Optimization algorithms
• Machine learning

Background:

• Integrated models struggle with multimodal and noisy problems due to lacking uncertainty evaluation.
• Radial basis function models excel with small, noisy datasets but are dimension-insensitive.
• Gaussian process regression offers prediction fitness and crucial uncertainty evaluation.

Purpose of the Study:

• To propose an adaptive weighted strategy-based integrated surrogate model.
• To address noisy multiobjective evolutionary problems effectively.
• To enhance convergence and diversity in population-based algorithms.

Main Methods:

• An indicator-based multiobjective evolutionary framework was employed.
• A weighted combination of radial basis function and Gaussian process regression was introduced.
• A U-learning sampling scheme was adopted for performance improvement and evaluation.

Main Results:

• The proposed algorithm demonstrated effectiveness on 12 benchmark test problems.
• Validation was performed on hybrid optimization problems, including sample construction and parameter determination.
• Experimental results confirmed the feasibility and effectiveness of the novel method.

Conclusions:

• The adaptive weighted integrated surrogate model successfully tackles noisy multiobjective evolutionary problems.
• The integration of radial basis function and Gaussian process regression with U-learning enhances optimization performance.
• The proposed method offers a robust solution for complex optimization challenges in various applications.