Smarter Models for Smarter Buildings

Buildings are getting smarter. From automated heating systems to solar panels feeding energy back into the grid, the way we manage energy in buildings is changing fast. At the heart of this transformation is a deceptively simple idea: store heat when it's cheap or abundant, and use it when it's needed most. Thermal energy storage (TES) tanks do exactly this, but getting the most out of them requires software smart enough to plan ahead and react in real time.

Advanced control systems, known as Model Predictive Control (MPC), need to run thousands of simulations in the blink of an eye to figure out the best heating strategy at any given moment. To do that, they rely on mathematical models of the storage tank, and those models need to be both accurate and extremely fast. Until now, the two qualities have been hard to achieve at the same time.

A new scientific publication by Eurac Research, titled “Modelling thermocline dynamics in thermal energy storages: A computationally efficient reduced-order model approach” addresses this challenge by introducing an innovative reduced-order model (ROM) capable of accurately capturing thermocline dynamics while remaining suitable for real-time control applications.

The study focuses on accurately modelling thermal stratification, the key physical phenomenon governing TES performance. While high-fidelity computational fluid dynamics (CFD) and multi-nodal models can describe these processes in detail, they are often too computationally demanding for MPC. Conversely, simplified lumped models lack the ability to reproduce stratification effects with sufficient accuracy. The proposed ROM bridges this gap through a grey-box modelling approach that combines physical insight with minimal calibration effort.

"Our objective was to develop a model that preserves the essential physics of thermocline evolution while remaining fast enough for real-time optimization," explains Tim Diller, co-author of the study. "The result is a formulation that is two orders of magnitude faster than established models, while actually improving prediction accuracy."

The numbers speak for themselves. In a nine-hour test simulating a full cycle of charging and discharging driven by a heat pump, the new model completed its simulation in just 0.015 seconds, while cutting temperature prediction errors by up to 46% compared to conventional approaches. That kind of speed and accuracy opens the door to much more responsive and efficient building control.

The model was tested extensively in the Energy Exchange Laboratory at Eurac Research, across a range of realistic scenarios. It consistently matched real-world measurements closely, outperforming established reference models in every test. Beyond the performance gains, one of the model's greatest strengths is that it remains physically transparent. It doesn't just produce numbers, it reflects what is genuinely happening inside the tank. This makes it easier to trust, easier to explain to engineers and building managers, and easier to connect to the broader systems that control heating and cooling in a building.

This work represents a meaningful step forward in making building energy systems smarter and more flexible. By making thermal storage easier to model and control in real time, it helps unlock the full potential of heat pumps and renewable energy bringing us a little closer to buildings that are not just efficient, but truly intelligent.