2016 ASHRAE Winter Conference

Simplified simulation models can be used for determination of effectiveness of energy-conservation measures in design stage as well as assessing performance optimization measures in retro-commissioning process. This presentation will deal with the general approaches that are followed to calibrate simplified simulations – auto and manual. The presentation will also show some examples on how the calibrated simulation can be used in practice.

Bayesian calibration is one of many methods that can be used for calibrating building energy models.   In Bayesian calibration, modelers start from an assumption that parameters to be calibrated are uncertain through assignment of probability distribution functions. The Bayesian calibration algorithm adjusts the parameter probability distribution functions by comparison of model predictions to measured data.  The posterior parameter distributions parameters are statistically more consistent with the measured data than the prior distributions with the probability distributions representing the confidence of the values of the input parameters given both the model of the buildings and the observed values from the building.

This paper summarizes a method for testing model calibration procedures. We call this a pure test method because it tests only the calibration procedure and not the correctness of the associated simulation program. This is accomplished by using the simulation program (used with a calibration method) to generate synthetic/surrogate building energy use data for a pre and post retrofit test case.  Thus the “correct” inputs are precisely known and the calibration technique can be tested for closure on the correct pre-retrofit model input values and the post retrofit energy savings along with the usual “goodness of fit” criterion.