Influence of data acquisition on the Bayesian calibration of urban building energy models

Abstract

Urban building energy models become increasingly important to estimate the dynamic heat demand of building groups, when planning energy saving measures. Archetype solutions can help to reduce the necessary data to model buildings in the urban context. Nevertheless, additional data and Bayesian calibration can be used to reduce the uncertainty in estimated parameter values. We model and calibrate three office buildings in Germany with different information levels using one year of hourly measurements. Starting with an archetype building energy model (BEM), the facade area and orientation, the window share, the physical properties, and the heating times are specified gradually before calibrating. The calibrated models are evaluated in CV(RMSE) and R2 and put in the context of ASHRAE Guideline 14 [1]. The calibration results with more specified parameters in the initial BEM are absorbing less error, because of the eliminated uncertainty before the calibration. Furthermore, the calibration of the most advanced BEM is performed with different periods of real measurement data. The calibrated BEMs indicate a similar data fit for calibrations with three months of measurement data from quarter 1, 2, and 4 of the year and with semi-annual measurement data.