energy model, Bayesian calibration
Calibration of building energy models is important to ensure accurate modeling of existing buildings. Typically this calibration is done manually by modeling experts, which can be both expensive and time consuming. Â Additionally, biases of the individual modelers will creep into the process. Â Many methods of automated calibration have been developed which reduce costs, time and biases, including optimization using genetic or swarming algorithms, machine learning methods, and Bayesian methods. Â Bayesian methods differ significantly from the other optimization and machine learning methods in that inputs are assumed to be uncertain and main goal is not to match the prediction to the measured data as closely as possible, but to reduce the uncertainty in the inputs in a manner consistent with the measured data. Â Bayesian methods are particularly useful when there are model inputs that have high sensitivity and high uncertainty and where there is limited measured data to use for calibration. In this paper, the basic concepts of Bayesian calibration are explained and a typical application and results are demonstrated.