Constrained Kalman Filter for Identification of Semiphysical Building Thermal Models

Model-based energy management of buildings through a model predictive control framework proved to be the promising solution for improving the energy efficiency of the building sector. The keystone for further improvements and real implementation is to have reliable and accurate mathematical models that are simple enough to be used in real-time control. In the work presented, a general identification framework for acquiring a control-oriented semiphysical thermal model of a building based on a modified constrained unscented Kalman filter algorithm is proposed. Experiments performed on a living-lab skyscraper building show that the adopted algorithm applied to short-term operation data outperforms the standard Kalman filter forms, both in convergence rate and numerical stability. The proposed method thus paves the way to the fast deployment of model-based energy management strategies in buildings.