Influences of energy data on Bayesian calibration of building energy model

Every building has different (and fuzzy) characteristics and contains complex sub-systems that affect each other. Therefore, significant uncertainties exist when modeling an entire building as a system. Calibration is necessary and able to reduce many sources of these uncertainties. Bayesian calibration is one of the automatic calibration methods that has been utilized in various applications. However, few researches were found that investigated the influences of quality and quantity of measured data used for the calibration. Moreover, Bayesian calibration requires considerable computing cost due to the inherent iteration attribute. This paper proposes the use of informative data to produce more accurate Bayesian calibration with reduced computing time. The measured energy data are classified by statistical classification methods. Using different energy measurement data, the study compares and analyzes the calibration outcomes with three criteria: input parameter estimation accuracy, energy use prediction accuracy, and overall computing time. The results show that the calculation time and the accuracy of the calibration are distinct for different selections of the data for calibration. Proper data should be used in comprehensive consideration of purpose, computing time and accuracy of calibration. Using informative data for calibration is able to keep similar accuracy but with 44% reduction in computing time compared to the use of all data.