EKF based self-adaptive thermal model for a passive house

Abstract Model predictive control (MPC) allows the integration of weather forecasts and of the expected building thermal behavior into the energy management system of buildings. The MPC algorithm requires an accurate but also computationally efficient mathematical model of the building thermal behavior. In this paper, several lumped-parameter thermal models of a passive house with an integrated photovoltaic system are compared to evaluate the model complexity needed to capture the basic thermal behavior of the entire building. In order to reduce implementation costs, the state and parameters of the finally chosen 1R1C model are estimated online with an extended Kalman filter (EKF). In addition, this self-adaptive thermal model provides online estimations of the unmeasured heat flows caused by the inhabitants. The results show that the EKF yields a robust convergence of the parameters after approximately three weeks and that the adapted model is able to generate a prediction of the heat demand for several days. The predicted reference room temperature shows average deviations of less than 1 °C for two-day predictions and of less than 3 °C for four-day predictions. Therefore, the proposed self-adaptive thermal building model is well suited to be used in a MPC environment.