Hybrid calibration methodology for building energy models coupling sensor data and stochastic modeling

Calibrated detailed energy models are often used to identify economic deep retrofit opportunities in existing buildings. But the uncertainty in building performance makes it technically unrealistic to reach a single “best fit” model without extensive sub-metering and domain experts on the project which is typically labor intensive. To address this problem, researchers have been adopting stochastic modeling as a more reliable approach to calibrate building energy models. A set of all plausible models is found, rather than a best fit model, which accounts for uncertainties in existing building properties and conditions. In addition, sensor data collected from within a building can be used to identify key operational characteristics such as setpoint temperatures, carbon-dioxide levels, light levels, and temperature setbacks. This paper presents a hybrid calibration methodology for building energy models using a combination of short-term wireless sensor data, 15-min interval smart meter data and stochastic modeling. The hybrid approach provides a means to calibrate the operational variables and physical variables separately, reducing potential bias and errors and to reach a set of plausible model solutions. A case study is presented to demonstrate the strength of the calibration methodology.