Building energy model calibration: A detailed case study using sub-hourly measured data

Abstract

This paper details a manual calibration of a 132 zones, highly monitored 5 434 m2 office building located in France. The calibration is conducted over one year of simulation at a five minute time-step. The model is mostly fed with input files containing measured data for each simulation time-step. First simulation results show that the model cannot be considered as being calibrated considering usual guidelines based on statistical indicators calculated on monthly energy consumption. Instead of going through a classical tuning process to match measured and simulated monthly energy consumption data, the choice was made to focus on explaining those discrepancies using available building information and monitoring data. A first dynamical heating and cooling powers analysis has been made, resulting in good correlation with measured values and explaining most of the discrepancies previously noticed, followed by a dynamical temperature analysis performed at the zone scale, and completed by a global statistical analysis at the building scale for the 132 zones of the building. Results showed very good agreement between measured and simulated temperature values with 90% of data points ranging between −1.72 °C and 2.02 °C of error. This demonstrate that a model that at first glance does not perform well considering particular statistical indicators can show an accurate dynamical behavior and a good performance when other variables and performance criteria are considered. This is followed by a discussion of the possible model improvements along with the limitations of statistical analysis compared to dynamic physical analysis for calibration performance assessment. The paper concludes with several proposals to improve the guidelines for calibration procedures involving the use of detailed monitoring data.