A hybrid approach for uncertainty quantification in dynamic building energy systems modeling

Accurate estimation of building energy system performance is critical for its optimal design and operation. However, uncertainties in building energy modeling (BEM) can cause significant discrepancies between predictions and real-world performance. To address these issues, this paper proposes a hybrid modeling framework integrating EnergyPlus with Quantile Regression Random Forests (QRRF) method to predict the cooling load of radiant cooling panels.

The best variant achieved CV(RMSE) of 16.70% for 1-hour and 28.34% for 5-minute predictions, outperforming both physics-based and data-driven models. Moreover, by incorporating simulated indoor dynamics instead of relying on measurements, the hybrid model maintains coverage probability while more accurately estimating lower-tail loads. These findings underscore the effectiveness of uncertainty-aware hybrid modeling over a building’s life-cycle.