Probabilistic identification of inverse building model parameters

Probabilistic and nonlinear least squares parameter estimation methods are evaluated for inverse gray box model identification of a retail building. A detailed building energy simulation program is used to generate surrogate data for estimation of parameters. The most probable or optimal parameters from each method are compared through simulation of building zone temperature and thermal loads. The least squares method generally found solutions near probable regions of the posterior from the probabilistic approach, and simulation performance was very similar between best parameter sets. A brief overview of probabilistic estimation techniques is provided, along with potential improvements to the approach presented and brief discussion on its applicability for uncertainty quantification within the building science domain.