AI-enabled energy baselines for verified building decarbonization

Accurately quantifying energy savings in retrofitted and operational buildings remains challenging due to dynamic occupancy, weather variability, and changing operational conditions. This study proposes an AI-enabled Energy Conservation Calculation (ECC) framework to establish dynamic energy baselines and verify energy and carbon savings under real-world operation in Singapore hotel. High-resolution operational data are integrated with a hybrid LSTM-XGBoost architecture to capture temporal patterns and nonlinear interactions in building energy use, while ECC logic translates model outputs into auditable and policy-compliant savings metrics aligned with the Green Mark certification system. The framework was deployed across multiple commercial, residential, and mixed-use buildings over a three-year period. Results show consistently strong predictive performance, with root mean square error generally below 5% across heterogeneous building types. Verified outcomes indicate cumulative emissions reductions of 3221 tCO₂e and energy use intensity improvements exceeding 60% in selected retrofitted cases. Beyond performance evaluation, the framework supports closed-loop operational optimization and produces audit-ready outputs suitable for certification and sustainability-linked finance. These results show that combining dynamic AI-based baselining with standardized energy accounting enables reliable verification of decarbonization outcomes in operational buildings.