Paper Conference

Proceedings of ASim Conference 2016: 3rd Asia conference of IBPSA-China, Japan, Korea


Uncertainty Quantification of Building Performance Simulation using Gaussian Process Emulator and Polynomial Chaos Expansion

Y. Kim

Abstract: Uncertainty Quantification (UQ) based on Monte Carlo Sampling (MCS) methods has been widely used for decision making problems. UQ is useful to address stochastic nature by quantifying risks of predicted outputs. However, for successful implementation of UQ, it takes significant modeling efforts and computation time. For handling the aforementioned issue, this study introduces two meta-models (Gaussian Process Emulator [GPE] and Polynomial Chaos Expansion [PCE]) which can be regarded as a surrogate model of a dynamic whole-building simulation model. In this study, the GPE and PCE are compared in terms of prediction capability and model flexibility under the different number of training data and inputs. In the paper, it is discussed whether two meta-models would be able to produce high performance qualities with acceptable computation time.
Keywords: Uncertainty Quantification, Monte Carlo Sampling, Gaussian Process, Polynomial Chaos, Building simulation