Prediction of heat pump demand profiles with few Dependencies: A combination of statistical and physical modelling

• High accuracy 24 h prediction of heat pump profile. • Large validation set (time and number) for heat pumps. • Low dependence on (external) data. The transition to renewable energy sources, along with the growing adoption of heat pumps used for space heating, presents new challenges for the electric grid and particular for peak load management. A crucial part of exploiting the temporal flexibility of heat pumps is the accurate short-term prediction of their electric demands. This paper introduces a novel predictive model for heat pump electricity consumption profiles by combining a simple physical model and statistical features from limited training periods. The presented approach is validated using real-world data from 70 heat pumps from various sources. In comparison to a baseline model, we achieve a 20% reduction in NRMSE. Two model refinements result in only modest enhancements. Our key conclusion is that in the setting of limited training data, the simple model presented is capable of yielding accurate predictions, its further improvement, however, is not trivial.