Answers are grounded in this paper and cite their sources.
Building performance simulation (BPS) provides high-fidelity assessments of thermal, moisture, comfort, and HVAC behaviour, but computational costs limit large-scale optimisation and uncertainty analysis. Surrogate models reduce these costs, yet conventional validation may overlook physical consistency, robustness, uncertainty calibration, and decision preservation. This study reviews 79 peer-reviewed studies (2019–2026) and proposes a six-tier validation framework covering statistical accuracy, sensitivity preservation, physical consistency, out-of-distribution (OOD) robustness, uncertainty calibration, and downstream decision fidelity. The framework is applied to a bio-based residential building across six Moroccan climate zones using EnergyPlus–HAMT and three surrogates: XGBoost, a Physics-Informed Neural Network, and an FNO-style tabular neural operator. XGBoost achieves R2=0.997, while Tier 3 yields a five-seed mean HAMT-referenced MVR of 39.51%. Results show that low a-priori directional-discrepancy rates do not necessarily indicate agreement with local HAMT behaviour, demonstrating that predictive accuracy alone cannot guarantee reference fidelity or downstream decision preservation.
No takes yet. Share an insight, caveat, or question.
Moutaouakil et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: