Paper Conference

Proceedings of ASim Conference 2014: 2nd Asia conference of IBPSA-China, Japan, Korea


Deterministic vs. Stochastic Calibration of Energy Simulation Model for an Existing Building

Young-Jin Kim, Ki-Cheol Kim, Cheol-Soo Park

Abstract: Building performance simulation tools have been widely used for performance assessment, optimal design & control, fault detection/diagnosis, energy retrofit, etc. However, many studies (IBPSA 1987-2013) have reported that there still exists a significant performance gap between the reality and simulation output, partly caused by unknown simulation inputs. Therefore, model calibration has been widely used to estimate unknown inputs. However, calibration attempts in area of building simulation may fail due to the following reasons: [1] uncertainty in simulation inputs, [2] sensor errors, and [3] long sampling time in data measurement. This paper addresses the abovementioned issues for reliable simulation prediction. In this study, an existing office building was selected. This study presents two calibration approaches: deterministic vs. stochastic calibration. Deterministic calibration finds a set of unknown values which minimize the difference between the measured data and simulation outputs. Stochastic calibration is based on Bayesian approach and finds probability distribution of each unknown input. Since stochastic calibration is computation-demanding, a Gaussian Process Emulator (GPE) was introduced in this study as a surrogate model of the EnergyPlus model. It is shown that the stochastically calibrated model can predict better than the deterministically calibrated model and can reduce the variance of unknown inputs. In addition, the accumulated measured data with a sampling time as long as a day might be unsuitable for calibration work due to lack of ‘time-series trend’. This paper also pointed out the difference in calibration results when different sensor errors (-3%, 0%, +3%) exist and propose a future work to take it into account. * Corresponding author email: 586
Keywords: Model calibration, Bayesian theory, Gaussian Process Emulator, Monte Carlo simulation
Pages: 586 - 593