Abstract
The effectiveness of model predictive control in residential buildings depends on computationally efficient and accurate component models, such as thermal energy storage, to optimize heating and cooling systems. Existing high-fidelity models, using computational fluid dynamics and multi-nodal approaches, are often too computationally intensive for real-time model predictive control applications, while simpler lumped models fail to accurately capture thermal stratification and its impact on the heating system performance. To address these limitations, an innovative reduced-order model is introduced. It simulates thermocline dynamics like formation and movement while only employing two thermal nodes. The model was validated experimentally across diverse operational scenarios, including charging, discharging, and sequential cycles. It was found to be two orders of magnitude faster for the specified simulation timestep and boundary conditions, while consistently achieving superior accuracy compared to finite difference white box models, reducing RMSE and MAE by up to 46% and 44% respectively. As such, the model is a highly effective tool for real-time model-based control applications, enabling improved energy management and energy efficiency in buildings.