Self-Adaptive AdamW-Guided Optimization: A Learning-Driven Metaheuristic for Solving Complex Real-World Engineering...

Area of Science:

• Numerical Optimization
• Metaheuristic Algorithms
• Machine Learning Optimization

Background:

• Continuous optimization problems are increasingly complex, particularly in black-box and strongly coupled environments.
• Existing methods may struggle with efficiency and convergence in these challenging scenarios.

Purpose of the Study:

• To introduce a novel adaptive gradient-guided metaheuristic, Self-Adaptive AdamW-Guided Optimization (SAWG).
• To enhance optimization performance in complex continuous and black-box environments without explicit gradient information.

Main Methods:

• SAWG constructs population-based pseudo-gradients to guide optimization.
• It integrates AdamW mechanisms: adaptive moment estimation, step-size regulation, and weight decay.
• A stagnation-aware adaptive control strategy balances exploration and exploitation, preventing premature convergence.

Main Results:

• SAWG demonstrated excellent optimization performance across CEC2017 and CEC2020 benchmark suites and engineering problems.
• Comparative analysis against nine other optimizers showed SAWG's strong adaptability and competitiveness.
• Statistical analysis confirmed SAWG's effectiveness on various numerical optimization tasks.

Conclusions:

• SAWG is a high-performance optimizer suitable for complex numerical optimization tasks.
• It offers a novel and effective approach, particularly in challenging optimization landscapes.
• The method shows significant potential for advancing continuous optimization techniques.