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

概要

Metaheuristic comparisons are often reported at evaluation budgets far larger than an expensive engineering simulation can afford. This paper presents REMBGO, an evaluation-budget-aware refinement of EMBGO. Its contribution is not any single operator: it is the integration of population preservation, scale-normalized exploration and recovery mechanisms whose parent-relative benefit is concentrated under restricted function-evaluation budgets. REMBGO is evaluated on CEC2022 at a primary, application-aligned budget of 1040 function evaluations, with a surrounding budget ladder reported descriptively to characterise budget dependence rather than as a family of controlled hypotheses. A component-removal study finds no individual mechanism whose removal survives multiplicity correction, consistent with an integrated rather than a single-operator contribution. The algorithm is then applied to EAF--STATCOM controller tuning, and the resulting frozen gain vectors are transferred without retuning to twenty pre-specified unseen stochastic arc realizations ($210$ fixed-gain simulations, all feasible). The selected gain-vector main effect accounts for only $3.6$--$5.8\%$ of the crossed validation variation, against approximately $61.4\%$ for interaction plus residual, and the matched REMBGO--EMBGO paired differences are controller-dependent in sign and not statistically significant. REMBGO is therefore presented as a resource-aware refinement whose benefit must be stated together with the evaluation budget, not as a universally superior optimizer.

参考リンク

• Paper ↗

リンク先のライセンス・再現性・データ条件をご確認ください。自動分類や実装の目安は保証ではありません。

実装を始めるプロンプト

参考リンクを確認したうえで、次の内容をコピーして利用できます。

# 論文・記事
REMBGO for Evaluation-Limited Optimization: CEC2022 Benchmarking and EAF-STATCOM Generalization

# 概要
Metaheuristic comparisons are often reported at evaluation budgets far larger than an expensive engineering simulation can afford. This paper presents REMBGO, an evaluation-budget-aware refinement of…

# 原典
https://arxiv.org/abs/2610.13471

上記の手法を実装してください。実行前に依存関係とライセンスを確認してください。