Bayesian parameter estimation of resistor-capacitor models for building thermal dynamics

Accurate and computationally efficient building energy models are critical to the development of online or pseudo-online control strategies and other building management activities. However, such models need to overcome the large uncertainty involved with continuously changing occupant activities and building status. The present study uses unscented Kalman filtering (UKF) in the model parameter estimation for simple yet accurate resistor-capacitor (RC) models to develop reliable building energy models. The estimation procedure, mathematical operations, and other estimation enhancing techniques are presented in detail. Synthetic and measured data were used to validate and evaluate the methodology. The obtained model shows better performance when compared with a model that was calibrated using genetic algorithms in a previous study. This remarkable model performance shows that UKF can enable timely online model update and improve the model predictability.