建物エネルギー予測と再エネ考慮型系統需要推定のための説明可能機械学習とルールベース意思決定支援
Explainable Machine Learning and Rule-Based Decision Support for Building Energy Prediction and Renewable-Aware Grid Demand Estimation (原題)
Yasin Ghad Yi, Muhammad Saqib S, SAMKARI HS, Shahzad T, F. Allehyani M, Saeed M, Tehseen M
🤖 gxceed AI 要約
日本語
IoTと説明可能AI（XAI）を統合し、建物エネルギー管理の運用支援フレームワークを提案。合成ベンチマークデータでRidge回帰がMAPE5.45%・R²0.63を達成し、LIME・SHAPで温度とHVAC使用が主要因と確認。DiCEによる感度分析とルールベース意思決定モジュールで、再エネ考慮型の系統需要推定と運用アクションを提示する。
English
This study integrates IoT and explainable AI (XAI) for building energy management, combining Ridge regression forecasting (MAPE 5.45%, R² 0.63) with LIME/SHAP interpretation and DiCE counterfactual analysis. A rule-based decision module translates predictions into operational actions and renewable-aware net grid demand estimates. Presented as an interpretable prototype on synthetic benchmark data, not a validated control system.
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
建物の省エネ・再エネ運用は日本のGX推進（省エネ法・ZEH・需給調整市場）と直結する。説明可能AIによる運用判断は、企業の脱炭素投資の説明責任やESG開示における定量的根拠として活用しうる。
In the global GX context
Building-level energy optimization and renewable-aware demand estimation feed directly into corporate Scope 2 accounting and grid decarbonization. Explainable, rule-based decision support aligns with growing disclosure demands for verifiable, auditable emissions-reduction measures under ISSB/CSRD.
👥 読者別の含意
🔬研究者:XAIとルールベース運用を組み合わせた建物エネルギー管理の再現可能なプロトタイプ設計を学べる。
🏢実務担当者:建物の省エネ運用判断を説明可能な形で支援し、Scope 2削減策の根拠づくりに応用できる。
🏛政策担当者:需給調整や省エネ政策において、説明可能なAI運用支援の可能性と限界を把握する材料になる。
📄 Abstract（原文）
This study presents an integrated IoT and Explainable Artificial Intelligence (XAI) framework for building energy management, addressing the gap between passive load forecasting and interpretable, rule-based operation support. Using a publicly available synthetic benchmark dataset of ten sensor features — Temperature, Humidity, SquareFootage, Occupancy, HVAC (Heating, Ventilation, and Air Conditioning) usage, Lighting usage, Renewable Energy, Holiday status, Hour, and Month — a Ridge Regression model is trained and evaluated, achieving a Mean Absolute Percentage Error (MAPE) of 5.45% and an R² score of 0.6292 on a held-out 40% test set, with a five-fold cross-validated R² of 0.60 ± 0.05. Among fourteen candidate regressors, the regularized linear model gives the best accuracy on this dataset while remaining fully transparent. Model behaviour is interpreted directly from the Ridge coefficients — identifying Temperature and HVAC usage as the dominant drivers — and this ranking is corroborated by two established XAI methods, LIME and SHAP. A directional sensitivity (what-if) analysis based on Diverse Counterfactual Explanations (DiCE) explores how feature changes shift the predicted consumption. A rule-based Decision Module (RDDM) translates sensor readings and predictions into four operational outputs: Status Alerts, IoT Action Plans, System Suggestions, and a renewable-aware Net Grid Demand estimate. Over 1,000 records the RDDM shows moderate positive correlations between predicted consumption and triggered interventions (Pearson r = 0.411, Spearman ρ = 0.384, p < 0.001); its reported 100% rule-consistency is an internal-consistency property that holds by construction and is not a measure of operational accuracy. The framework is presented as an interpretable, reproducible prototype on a synthetic benchmark rather than a validated control system
🔗 Provenance — このレコードを発見したソース
- Research Square https://doi.org/10.22541/authorea.15009479/v1first seen 2026-09-30 04:35:22 · last seen 2026-10-02 04:21:59
🔔 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。