Transforming Building Energy Management: Sparse, Interpretable, and Transparent Hybrid Machine Learning for Probabilistic Classification and Predictive Energy Modelling

• DOI: 10.3390/architecture5020024
• Publication Date: 2025-04-01T14:59:59Z
• Authors: Yiping Meng; Yiming Sun; Sergio Rodriguez; Binxia Xue
• Keywords: Model Predictive Control

The building sector, responsible for 40% of global energy consumption, faces increasing demands sustainability and efficiency. Accurate consumption forecasting is essential to optimise performance reduce environmental impact. This study introduces a hybrid machine learning framework grounded in Sparse, Interpretable, Transparent (SIT) modelling enhance management. Leveraging the REFIT Smart Home Dataset, integrates occupancy pattern analysis, appliance-level prediction, probabilistic uncertainty quantification. clusters occupancy-driven usage patterns using K-means Gaussian Mixture Models, identifying three distinct household profiles: high-energy frequent occupancy, moderate-energy variable low-energy irregular occupancy. A Random Forest classifier employed pinpoint key appliances influencing with drop-in accuracy analysis verifying their predictive power. Uncertainty quantifies classification confidence, revealing ambiguous periods linked appliance patterns. Additionally, time-series decomposition predictions are contextualised seasonal dynamics, enhancing interpretability. Comparative evaluations demonstrate framework’s superior transparency over traditional single models, including Support Vector Machines (SVM) XGBoost Matlab 2024b Python 3.10. By capturing behaviours accounting inherent uncertainties, this research provides actionable insights adaptive proposed SIT model can contribute sustainable resilient smart systems, paving way efficient management strategies.