REMBGO for Evaluation-Limited Optimization: CEC2022 Benchmarking and EAF-STATCOM Generalization

TL;DR

REMBGO is a refinement of EMBGO that integrates population preservation, scale-normalized exploration, and recovery mechanisms for evaluation-limited optimization. It is evaluated on CEC2022 and applied to EAF-STATCOM controller tuning. REMBGO's benefit depends on the evaluation budget.

Why It Matters

REMBGO's resource-aware approach is crucial for expensive engineering simulations with limited evaluation budgets. It enables efficient optimization without requiring excessive function evaluations. This contributes to practical applications where computational resources are restricted.

Key Findings

• REMBGO outperforms EMBGO under restricted evaluation budgets
• No single mechanism is responsible for REMBGO's performance
• REMBGO's benefit is budget-dependent and not universally superior