Robust adaptive model predictive building climate control

Abstract

Model Predictive Control (MPC) is a promising algorithm for building climate control. However, the construction of an accurate building model is very challenging since identification experiments on buildings are, due to time and usage constraints, difficult to conduct and the construction data are often only known approximately. This paper proposes the use of a recently developed robust, adaptive MPC as a possible solution for handling this model uncertainty. Starting from an initial set of possible models, the algorithm relies on recursive set membership identification to shrink this set in each step of the closed-loop operation. The MPC enforces comfort constraints for all the models in the current set and hence achieves robust constraint satisfaction even during the adaptation phase. The proposed algorithm is compared to a standard, non-robust, non-adaptive MPC and to a non-robust, adaptive MPC in simulations using the EnergyPlus building simulation software.