Torsten Schwan, Sebastian Schmitt, Andrea Castellani, "Calibration of HVAC system models with monitoring data - Digital Twin meets measurement data", ESI FORUM IN DEUTSCHLAND , 2019.

Modern heat, ventilation and air-conditioning (HVAC) systems for buildings requires engineers to use increasingly more complex physical models to evaluate building performance in early design stages as well as during modernization and reconstruction phases. Those models often provide accurate results regarding total annual heat, cold and power consumption. However, achieving very accurate high temporal resolution results and evaluation of smart building control strategies in distributed energy systems is hardly possible. For this, such models would require the specification of a significant number of parameters and environmental conditions, which are mostly not available or at least very hard to collect. ESI ITI's SimulationX therefore offers the Green City library which provides suitable HVAC system models with interfaces to external control software. Furthermore, these models only require a reduced set of input parameters which can be easily obtained by simple manufacturer datasheets. Green City models can thus be used to evaluate building control with corresponding energy system reaction. However, there are often still significant differences between detailed simulation results and measurement data (e.g. switching frequencies) if a model has only been developed with manufacturer data and usual planning knowledge. To use such models for higher-order control strategies (i.e. model-in-the-loop) or system diagnostic purposes (i.e. automatic error detection), the model accuracy still has to be significantly increased. This paper shows an approach of a semi-supervised, partly-automated, data-driven calibration process of a complex HVAC system Digital Twin model in SimulationX with high-resolution monitoring data. It is furthermore, discussed how machine learning methods can be integrated and used to further automatiize such processes to reduce manual engineering effort.