Data-driven model predictive control for buildings with glass façade and thermally activated building structure

• A DMPC is compared to a 2-point controller for heating, cooling and shading. • The real-data-based simulation environment emulates operation of an office building. • The optimiser uses a matrix form of the Mixed-Integer Linear Programming algorithm. • Constraint functions ensure comfort; the objective function minimizes energy/cost. • A comprehensive parametric variation shows the DMPC superiority in all cases. This study presents the development and testing of a Data-driven Model Predictive Control (DMPC) strategy for optimizing energy efficiency in buildings with glass façades, shading systems, and Thermally Activated Building Structures (TABS). The DMPC approach utilizes a grey-box state space model with data-driven parameter identification. The optimization algorithm is Mixed-Integer Linear Programming (MILP). It allows accounting for thermal comfort through constraints, while the objective function minimizes energy or costs in a reduced solution space. A detailed simulation environment has been developed for the investigations and validated through long-term experiments on a real building. Based on full-year simulations, the DMPC is compared against standard rule-based controllers, varying parameters such as shading control, prediction horizon, and optimization frequency. Results show that the DMPC consistently outperforms traditional control methods, achieving lower energy consumption and costs while maintaining comfortable room temperatures. Across the simulated variants, the DMPC reduces total energy demand –and similarly the associated costs– for the investigated zones in a range of 37% to 47%, compared to the best variant of the rule-based controller with automatic shading control. The robustness of the DMPC algorithm across various settings demonstrates its potential for practical implementation in building energy management systems, particularly in structures with significant solar gains and thermal mass.