Uncertainty quantification for modeling night-time ventilation in Stanford’s Y2E2 building

Abstract

Natural ventilation can significantly reduce the energy consumption of modern buildings, but robust design is challenging because of the many uncertainties involved. The present work aims to predict the 4-hour nightflush in Stanford’s Y2E2 building during 4 different nights. We employ an uncertainty quantification (UQ) framework that combines an integral model and a computational fluid-dynamics (CFD) model and propagate the uncertainty in the inputs using a polynomial chaos expansion method. The integral model solves a one-dimensional equation for the thermal mass temperature and an equation for the volume-averaged air temperature inside the building. The CFD instead solves for the three-dimensional flow and temperature field in the building, hence enabling direct computation of the heat transfer and discharge coefficients. We first performed a UQ study of the integral model and found that the measured air temperature is inside the 95% confidence interval of the mean prediction ∼80% of the times. Subsequently we employed CFD to improve the accuracy of the probability distributions for the heat transfer and discharge coefficients. In the final step we updated the integral model with these new probability distributions and showed that the standard deviation of the indoor temperature could be reduced up to ∼40%. These results demonstrate that the proposed framework can effectively quantify and reduce uncertainty in models used for the design of natural ventilation systems.