A Hierarchical Surrogate-Assisted Differential Evolution With Core Space Localization.

Area of Science:

• Optimization Algorithms
• Computational Intelligence
• Machine Learning

Background:

• Surrogate-assisted evolutionary algorithms (SAEAs) are crucial for expensive optimization problems (EOPs).
• Hierarchical SAEAs balance global exploration and local exploitation but struggle with high-dimensional EOPs (HEOPs) due to the curse of dimensionality.

Purpose of the Study:

• To propose a novel hierarchical surrogate-assisted differential evolution with core space localization (HSADE-CS) algorithm.
• To effectively address the challenges of high-dimensional expensive optimization problems (HEOPs).

Main Methods:

• Introduced a top-promising sampling strategy for global search to handle surrogate model uncertainties.
• Developed a core space localization (CSL) method to identify high-potential local search spaces.
• Implemented a fitness-independent adaptive parameter control for differential evolution (DE) using Minkowski distance to enhance local search.

Main Results:

• HSADE-CS demonstrated competitive performance on benchmark suites (e.g., CEC2014, CEC2017) with dimensions up to 500.
• The algorithm was successfully applied to a real-world problem: circular antenna array design optimization.
• Experimental results show HSADE-CS outperforms state-of-the-art SAEAs on HEOPs.

Conclusions:

• HSADE-CS effectively overcomes the curse of dimensionality in HEOPs.
• The proposed strategies significantly enhance the performance of surrogate-assisted local search.
• HSADE-CS offers a highly competitive and effective solution for complex optimization challenges.