A Kriging-Assisted Evolutionary Algorithm With Dual Perspectives and Dual Indicators for Expensive Robust...

Area of Science:

• Optimization Algorithms
• Computational Science
• Machine Learning

Background:

• Expensive robust multiobjective optimization problems (ExRMOPs) require balancing solution optimality and robustness.
• Existing evolutionary algorithms often focus on either average or worst-case performance, neglecting their complementarity.

Purpose of the Study:

• To propose a novel Kriging-assisted evolutionary algorithm (KPI) that addresses the limitations of existing methods for ExRMOPs.
• To develop a method that adaptively balances optimality and robustness by considering both average and worst-case perspectives.

Main Methods:

• Development of a dual-perspective aggregation function (DPAF) that weighs average and worst-case performance based on population stability.
• Introduction of a dual-indicator candidate selection strategy using robust optimality and diversity indicators.
• Utilizing Kriging surrogate models for efficient handling of expensive function evaluations.

Main Results:

• The proposed KPI algorithm demonstrates superior performance in solving ExRMOPs compared to existing approaches.
• Extensive experiments on benchmark test suites and a real-world application validate the effectiveness of KPI.
• The dual-perspective approach and indicator-based selection strategy contribute to improved robust optimization.

Conclusions:

• KPI offers an effective solution for ExRMOPs by adaptively balancing optimality and robustness.
• The proposed DPAF and candidate selection strategy are key innovations for robust evolutionary optimization.
• The findings suggest a promising direction for future research in expensive multiobjective optimization.