Decomposition-informed deep learning for wind-power forecasting: A CEEMDAN→VMD hybrid with feature extraction and per-component deeplearners
分解情報を活用した深層学習による風力発電予測：CEEMDAN→VMDハイブリッドと特徴抽出・成分別深層学習器の組み合わせ (AI 翻訳)
NOUNANGNONHOU CT, DIDAVI KBA, AZA-GNANDJI MR
🤖 gxceed AI 要約
日本語
本論文は、CEEMDANとVMDの二重分解およびGRU-Attentionを用いたハイブリッド深層学習モデル（B6）を提案し、7つの風力発電所のデータで評価した。1時間、4時間、24時間先予測において、従来手法と比較してMAEとRMSEを25-35%削減し、100MWグリッドの予備力要件を26-30%低減することを示した。
English
This paper proposes a hybrid deep learning model (B6) combining CEEMDAN-VMD dual decomposition with a GRU-attention backbone for wind power forecasting. Evaluated on seven wind farms, it reduces MAE and RMSE by 25-35% over baselines for 1h, 4h, and 24h horizons, translating to 26-30% spinning reserve reduction for a 100 MW grid.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
日本では風力発電の導入拡大が進む一方、系統安定性が課題である。本手法は予測精度向上により出力変動対策の効率化に寄与し、FIT後の市場統合や需給調整市場での評価向上に貢献する可能性がある。
In the global GX context
Globally, this work advances the state-of-the-art in short-term wind forecasting by integrating multi-scale decomposition with deep learning. It offers a scalable solution for grid operators and renewable asset managers to reduce reserve requirements and curtailment, supporting higher renewable penetration.
👥 読者別の含意
🔬研究者:Novel hybrid decomposition-deep learning framework (CEEMDAN-VMD-GRU-Attention) with strong empirical validation across multiple sites and horizons.
🏢実務担当者:Operational accuracy gains (25-35% error reduction) directly reduce spinning reserve costs and curtailment; can be integrated into existing EMS systems.
🏛政策担当者:Demonstrates that advanced forecasting can lower grid integration costs for wind power, supporting policy targets for renewable energy share and grid stability.
📄 Abstract（原文）
<title>Abstract</title> <p>Accurate wind power forecasting remains a critical challenge for grid stability and renewable energy integration, particularly in emerging power systems with limited flexibility. This study proposes a hybrid decomposition–learning framework that combines the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Variational Mode Decomposition (VMD) with a deep gated recurrent network (B6-CEEMDAN-VMD). The dual decomposition enables the extraction of multi-scale temporal features, while the GRU–attention backbone captures nonlinear dependencies and temporal dynamics across horizons. The proposed model was evaluated on data from seven wind farms using 1-hour, 4-hour, and 24-hour forecast horizons and benchmarked against persistence, linear regression, ARIMA, gradient boosting (GBM), and standalone deep models. Results demonstrate that B6 outperforms all baselines, reducing mean absolute error (MAE) and root mean squared error (RMSE) by 25--35% compared to GBM and by more than 40% compared to persistence forecasts. The model also exhibits strong spatial robustness, maintaining low forecasting errors across geographically diverse sites. Operationally, these accuracy gains translate into a 26--30% reduction in spinning reserve requirements for a 100 MW grid, contributing to more efficient dispatch, reduced curtailment, and lower backup fuel costs. Overall, the CEEMDAN–VMD–Deep hybridization provides a scalable, physics-consistent, and data-efficient solution for improving short-term and day-ahead wind forecasting in emerging renewable energy systems</p>
🔗 Provenance — このレコードを発見したソース
- Research Square https://doi.org/10.21203/rs.3.rs-9856569/v1first seen 2026-06-12 04:29:20
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。