Physics-informed neural networks as virtual sensors for domestic heat pumps
To ensure efficient operation of heat pump systems, comprehensive monitoring of the operational status is needed. However, due to cost constraints in the domestic setting, key parameters such as refrigerant mass flow rate are not monitored. This has led to growing interested in virtual sensors which use machine learning along with commonly available measurements to predict these missing parameters. To enhance the accuracy as well as interpretability of machine learning models, research has shifted to physics-informed machine learning. In this paper a short description of physics-informed machine learning is provided, followed by a case study detailing a physics-informed virtual sensor for the refrigerant mass flow rate of a domestic heat pump. The proposed model achieves a prediction accuracy of 93% with greater interpretability than that of a standard machine learning model.