Multisurrogate-Assisted Ant Colony Optimization for Expensive Optimization Problems With Continuous and Categorical...

Area of Science:

• Optimization algorithms
• Computational intelligence
• Machine learning

Background:

• Expensive optimization problems (EOPs) require efficient algorithms due to high computational costs.
• Existing surrogate-assisted evolutionary algorithms (SAEAs) primarily address continuous or combinatorial EOPs, leaving mixed-variable EOPs under-addressed.
• EOPs with continuous and categorical variables (EOPCCVs) present unique challenges for standard optimization techniques.

Purpose of the Study:

• To propose a novel algorithm, MiSACO, specifically designed for solving EOPCCVs.
• To enhance the efficiency of optimization processes for EOPCCVs by minimizing function evaluations.
• To develop a robust method that can handle the complexities of mixed-variable optimization problems.

Main Methods:

• Developed a multisurrogate-assisted ant colony optimization algorithm (MiSACO).
• Incorporated two core strategies: multisurrogate-assisted selection using Radial Basis Function (RBF) and Least-Squares Boosting Tree (LSBT), and surrogate-assisted local search.
• Employed RBF and LSBT for robust selection from offspring solutions and utilized RBF with sequential quadratic optimization to refine continuous variables.

Main Results:

• MiSACO demonstrated effectiveness in solving EOPCCVs across three sets of test problems and two real-world cases.
• The algorithm successfully handled mixed-variable optimization problems with a limited number of function evaluations.
• The proposed multisurrogate strategy proved robust in addressing different types of EOPCCVs.

Conclusions:

• MiSACO provides an effective solution for expensive optimization problems with continuous and categorical variables.
• The algorithm's dual strategies enhance optimization performance and robustness, particularly when function evaluations are costly.
• MiSACO represents a significant advancement in tackling complex mixed-variable optimization challenges.