Modal Detection Informed Classification Evaluation via Ensemble Networks for Expensive Constrained Multimodal...

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Expensive constrained multimodal optimization problems (ECMMOPs) involve complex simulations or experiments with multiple solutions and constraints.
• Limited function evaluations (FEs) pose significant challenges in accurately finding multiple optimal solutions while satisfying constraints.

Purpose of the Study:

• To develop an efficient algorithm for solving ECMMOPs.
• To address the challenges of multimodality, limited FEs, and complex constraints in optimization.

Main Methods:

• A surrogate-assisted self-clustering particle swarm optimization algorithm with modal detection informed classification evaluation (MDICE) was developed.
• Key components include a self-clustering update mechanism, a novel modal detection strategy, a modality-guided classification evaluation, and a surrogate-assisted feasibility search.

Main Results:

• MDICE demonstrated superior performance in identifying multiple optimal solutions and satisfying constraints.
• The algorithm efficiently utilizes limited function evaluations.
• Experimental results on 33 benchmark functions confirmed MDICE's effectiveness.

Conclusions:

• MDICE offers a robust and efficient solution for expensive constrained multimodal optimization problems.
• The proposed methods for modal detection and guided classification significantly improve optimization performance.
• MDICE outperforms existing state-of-the-art surrogate-assisted evolutionary algorithms.