model predictive control, unscented Kalman filter, state and parameter estimation, Modelica
Model predictive control (MPC) for buildings is a promising approach to reduce the energy consumption of buildings while at the same time the thermal user comfort can be improved. The core of this control strategy consists of building models that can describe the thermal behavior of particular zones accurately. Grey-box models are frequently used modeling approaches for control-oriented models, however, these models often have limitations regarding their general applicability. Furthermore, the modeling and identification of models used in MPC still require significant effort and is one of the main obstacles for the actual practical implementation of building predictive control. This paper addresses these issues and presents a framework for the online state and parameter estimation of grey-box models. The results show that (1) this online simultaneous state and parameter estimation highly increases the multi-steps-ahead (up to 48 h) prediction performance, (2) this approach enables the models to adapt to changing environmental conditions and (3) it is possible to use only one pre-defined initial model to describe the thermal behavior of several different zones.