Automated reduced-order model generation of radiant heating terminals

Reduced-order building models offer a promising alternative to traditional white-box approaches, providing accurate thermodynamic representations of buildings with minimal input data. However, existing reduced-order models primarily focus on indoor temperature fluctuations, limiting their utility for controlling real-world hydronic heating systems, which rely on supply water temperature and flow rate. This study extends conventional reduced-order models by explicitly representing heat emitters with generic, simplified models. Model parameters are estimated via a global optimization routine. The models were trained and tested on data from 10 real-world residential buildings with monitored hydronic heating systems featuring either radiators or radiant floors. Results indicate that dynamic models (featuring thermal capacitances) are required to accurately capture heating system behaviour, achieving normalized root mean squared errors (NRMSE) between 3-10% in heat transfer predictions. Radiant floors exhibit higher estimated heat capacity than radiators, reflecting their increased thermal mass. Training models on systems using advanced controls, such as model predictive control (MPC), decreases model accuracies, particularly for radiant floors. These findings enhance the applicability of reduced-order models for real-world hydronic heating control, bridging the gap between simplified building energy models and practical implementation.