Robust Multiobjective Evolutionary Algorithm Based on Surrogate-Assisted Robust Distance Metric.

Area of Science:

• Computational intelligence
• Optimization algorithms
• Evolutionary computation

Background:

• Traditional robust multiobjective evolutionary algorithms (RMOEAs) demand extensive sampling, leading to high computational costs in real-world scenarios.
• The need for efficient methods to achieve robust optimal solutions in computationally intensive applications is critical.

Purpose of the Study:

• To develop a surrogate-assisted robust multiobjective evolutionary algorithm (RMOEA-SA) that significantly reduces function evaluations.
• To introduce a novel robust distance metric (RDM) integrated with a surrogate model for enhanced robustness measurement.
• To improve the trade-off between solution robustness and optimality by augmenting the objective space.

Main Methods:

• Implementation of a radial basis function (RBF) surrogate model to approximate fitness values, reducing the number of direct function evaluations.
• Development of a robust distance metric (RDM) that utilizes the RBF surrogate model to quantify solution robustness.
• Augmentation of the objective space by incorporating the RDM value as an additional objective for selection.

Main Results:

• The RBF surrogate model effectively approximates fitness values, substantially decreasing computational load during robust optimization.
• The RDM, assisted by the RBF surrogate, accurately measures solution robustness.
• Selection in the augmented objective space successfully balances robustness and optimality.

Conclusions:

• The proposed RMOEA-SA demonstrates superior feasibility and effectiveness compared to existing algorithms.
• The RBF surrogate model and RDM integration offer a computationally efficient approach to robust multiobjective optimization.
• The method shows promise for complex real-world applications requiring robust optimal solutions.