A Surrogate-Assisted Differential Evolution Algorithm for High-Dimensional Expensive Optimization Problems.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Surrogate-assisted evolutionary algorithms (SAEAs) commonly employ Radial Basis Function (RBF) and Kriging models.
• High-dimensional and computationally expensive problems pose significant challenges for traditional optimization methods.

Purpose of the Study:

• To propose a novel Global and Local Surrogate-Assisted Differential Evolution algorithm (GL-SADE).
• To enhance convergence speed and global search capability for high-dimensional expensive optimization problems.

Main Methods:

• GL-SADE utilizes a global RBF model for trend estimation and a local Kriging model for focused search.
• The local Kriging model prioritizes points with high predicted fitness and uncertainty to prevent premature convergence.
• A reward search strategy is incorporated to exploit promising regions identified by the local Kriging model.

Main Results:

• Experiments on benchmark functions (30-200 dimensions) demonstrate GL-SADE's effectiveness and efficiency.
• Comparative analysis against four state-of-the-art algorithms validates GL-SADE's superior performance.
• Application to an airfoil optimization problem confirms its practical utility.

Conclusions:

• GL-SADE offers a robust approach for tackling high-dimensional, expensive optimization problems.
• The hybrid global-local surrogate strategy effectively balances exploration and exploitation.
• The algorithm shows significant potential for real-world engineering design optimization.