Machine Learning Enhanced Grey-Box Modeling for Building Thermal Modeling

In this work, two approaches for grey-box thermal system and time-varying heat disturbance modeling of buildings are proposed. They combine machine-learning-based modeling with grey-box modeling to form a hybrid approach. The first approach simultaneously estimates the system model parameters and estimator gain for a system model augmented with an integrating disturbance model. With the estimator gain and collected input-output data, the disturbance is estimated, and a predictive recurrent neural network disturbance model is fit using the estimates. The second approach simultaneously estimates the system model parameters and the predictive disturbance model parameters. Both approaches are applied to a building zone example to demonstrate and analyze them.