AI-Augmented Energy Management in Smart Buildings: A Hybrid Approach for Sustainable Operation

Buildings account for nearly one-third of global energy consumption, driving the need for sustainable energy practices and smarter energy management. This paper proposes a methodological framework to forecast photovoltaic (PV) generation and household energy consumption at high temporal resolution, enabling residents to optimize appliance usage and reduce dependence on grid input. The framework integrates remotely sensed meteorological data, historical appliance-level consumption records, and machine learning models (Gradient Boosting, LightGBM, XGBoost, and Random Forest) to provide short- and medium-term forecasts. Data from a smart household over five months were used to train and evaluate the models, with forecasts generated for 48-hour and 144-hour horizons. Hyperparameter tuning was performed using both Grid Search and Optuna. Results show that LightGBM and Gradient Boosting models outperformed others in forecast accuracy, with Optuna-tuned models offering marginal improvements. The findings highlight the feasibility of combining AI and satellite-derived data to enhance household-level energy efficiency, identify high-impact appliances, and support cleaner production practices through informed energy management. This scalable framework can aid the transition toward smarter, more sustainable residential energy systems.