Dobschinski, JanJanDobschinskiJost, DominikDominikJostGirón Cruz, Pedro JoséPedro JoséGirón CruzGoldmaier, Ann-KatrinAnn-KatrinGoldmaierHollermann, Dinah ElenaDinah ElenaHollermannRodriguez Santiago, JuanJuanRodriguez Santiago2024-11-262025-05-132025-06-172024-11-262024-10https://publica.fraunhofer.de/handle/publica/47932010.1049/icp.2024.3837To be able to efficiently provide flexibility from renewable energy (RE) driven heat pump systems optimized forecast systems are indispensable. Forecasts of weather-dependent generation (mainly wind and solar) but also forecasts of heat pump operation are needed to create cross-sectoral operational plannings. Wind and solar forecasting systems are well researched and established in the energy industry. In contrast, there is very little knowledge in the development of forecasts for an optimised heat pump operation with focus on a provision of flexible power. The ‘FlexLearn’ project has set itself the task of solving this problem. Regarding the electric grid integration of heat pumps and the utilization of flexibility potentials from the perspective of an electricity grid operator the focus of this paper lies in the simulation and forecast of the electric power consumption of single heat pumps but also of larger heat pump portfolios. The primary objective of the study is to test methods for predicting the electricity consumption of heat pumps in a 15-minute time resolution with forecast horizons of a few hours. The particular challenge here lies in mapping the on and off times of the heat pump, which lead to large gradients in electricity consumption that are generally difficult to predict. The paper starts with an overview about the temporal behaviour of single and aggregated heat pump. Based on this analysis, challenges and opportunities with regard to the development of reliable and qualitative forecast methods are formulated. To simulate and forecast the electric power consumption of heat pump systems two approaches have been tested. One approach is based on a fully comprehensive building simulation to model the various thermal flows. The second approach is purely data-driven. As benchmark a simple persistence model is used. Based on this, a multi-linear regression model was used to combine information on the recent characteristics of the electricity consumption with measurements/forecasts of the outside temperature. The paper ends with an outlook on possibilities for determining flexibility potentials of heat pump systems on the basis of real-world SCADA data.enForecastingheat pumpMLjournal article