Calculating true energy savings: a validated model approach

Deep renovation of public buildings is a central strategy for reducing energy use, yet the gap between predicted and realized savings remains a barrier to the financial viability of energy performance contracts. A fundamental difficulty is that savings cannot be directly measured and must instead be derived from validated energy models. This paper proposes a structured framework in which a physics-based model and a data-driven model are each assigned to a distinct contractual function and demonstrates the consequences of applying each model outside its intended role. A dynamic hourly simulation following HRN EN ISO 52016-1 is applied to a renovated ten-story academic building to quantify guaranteed renovation savings with Monte Carlo uncertainty propagation. A feedforward neural network trained on multi-season data from a second monitored building predicts postrenovation energy use and assesses setpoint sensitivity. To enable direct comparison, the neural network is also applied to the renovated building using only the available single-season dataset. The physics-based model reproduces measured heating energy within approximately 10% over a full heating season and confirms renovation savings of 48.9% with a 90% confidence interval of +/- 5%. The neural network approaches established calibration thresholds on the multi-season building and predicts 5-6% savings per degree of setpoint reduction but fails to generalize from a single partial heating season. The results show that physics-based models are essential for establishing pre-renovation baselines, while data-driven models are best suited for operational monitoring once multi-season data are available.