Answers are grounded in this paper and cite their sources.
Building parameter characterization is essential for energy performance assessment and retrofit planning, but detailed audits and calibrated simulations are difficult to scale across large building portfolios. This study proposes an inverse surrogate modelling framework to infer unknown building and operational parameters, including internal gains, after-hours plug-load operation, ventilation rate, after-hours ventilation, heating and cooling setback temperature offset, solar heat gain coefficient (SHGC) and effective envelope performance (U eff ), from the change-point model (CPM) features extracted from heating and cooling loads. Physics-based simulations were used to generate datasets for four building archetypes across four Canadian weather files, with three weather files used for model development and one reserved as an entirely unseen climate for generalization testing. For each simulation case, occupied and unoccupied heating and cooling CPMs were fitted, and their balance temperatures, slopes, intercepts, and fit-quality metrics were used as inputs to archetype-specific machine learning models. The trained models achieved strong predictive accuracy on the test set, with R 2 values generally ranging from 0.82 to 0.98 across the seven targets and four archetypes. Internal gains, U eff and temperature offset were among the most accurately predicted parameters, with R 2 values of 0.89–0.98 and CV(RMSE) values mostly below 0.18. When evaluated on the entirely unseen weather file, the models maintained promising generalization performance, with most outputs achieving R 2 values above 0.85. Overall, the proposed framework offers a scalable pathway for rapid virtual audits and portfolio-level building characterization.
No takes yet. Share an insight, caveat, or question.
Markarian et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: