Hybrid Deep Learning Enhanced Angstrom Framework for Global Horizontal Irradiance Prediction
グローバル水平面日射量予測のためのハイブリッド深層学習強化Angstromフレームワーク (AI 翻訳)
ARVIND K, Gangadharappa M
🤖 gxceed AI 要約
日本語
本論文は、太陽光発電の効率向上に重要な全球水平面日射量（GHI）の予測に対して、経験的なAngstrom式とリカレントニューラルネットワーク（LSTM, GRU, BiLSTM）を統合したハイブリッドモデルを提案。インドの2地点の気象データを用いた実験では、Angstrom-BiLSTMモデルが最高精度（R²=0.9893）を達成。SHAP分析により、Angstrom放射や気象特徴量の重要性を確認した。
English
This paper proposes a hybrid deep learning framework that integrates the empirical Angstrom equation with recurrent neural networks (LSTM, GRU, BiLSTM) for accurate Global Horizontal Irradiance (GHI) prediction, critical for photovoltaic energy systems. Experiments on meteorological data from two Indian sites show the Angstrom-BiLSTM model achieves the best performance (R²=0.9893), with SHAP analysis highlighting key features.
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
日本では太陽光発電の出力予測が系統安定化に不可欠であり、本手法は日本の気象データにも応用可能。特に、変動性の高い日射量予測の改善により、再生可能エネルギーの導入拡大と電力システムの効率化に貢献する可能性がある。
In the global GX context
Accurate solar irradiance forecasting is crucial for grid integration of solar PV globally. The hybrid empirical-deep learning approach demonstrated here can be adapted to different climates, supporting the expansion of renewable energy and grid stability, which are key to the global energy transition.
👥 読者別の含意
🔬研究者:The hybrid Angstrom-deep learning method offers a new direction for solar forecasting research, potentially improving generalization across diverse climates.
🏢実務担当者:Solar farm operators and grid managers can apply this model to improve day-ahead power output predictions, reducing imbalance costs.
🏛政策担当者:Policymakers supporting renewable energy should note that advanced forecasting technologies like this can enhance grid reliability and facilitate higher solar penetration.
📄 Abstract（原文）
<title>Abstract</title> <p>Accurate estimation of Global Horizontal Irradiance (GHI) is critical in making photovoltaic energy production, renewable energy systems, and smart grids more efficient. Nonetheless, the highly non-linear and time-changing nature of atmospheric conditions poses challenges to GHI estimation via standard prediction methods. This paper presents an innovative approach towards GHI estimation based on the integration of the empirical Angstrom equation with sophisticated recurrent neural networks. In total, six predictive models, including LSTM, GRU, BiLSTM, Angstrom-LSTM, Angstrom-GRU, and Angstrom-BiLSTM, are designed and examined based on meteorological data collected from two locations in India: Panaji and Udaipur. This approach combines preprocessing, temporal learning, extraction of features based on the Angstrom radiation equation, and explainability of artificial intelligence via SHAP analysis. The experimental findings show superior performance of the hybrid Angstrom-based models compared to conventional deep learning structures. Out of all the models proposed, the Angstrom-BiLSTM model was found to have yielded the most accurate predictions with RMSE = 28.7960, MAE = 12.7803, R² = 0.9893, and Willmott 0.9973, reflecting the high level of agreement between the real and predicted irradiance values. It was also proved through the use of the SHAP value analysis that the Angstrom radiation and weather-related features played an important role in improving the accuracy of predictions. The results showed that by combining the empirical information about solar radiation with the bi-directional deep learning algorithms, it is possible to obtain a more reliable forecast.</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-9937780/v1first seen 2026-06-10 04:32:39 · last seen 2026-06-16 04:30:22
🔔 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。