ABSTRACT Accurate, time‐resolved heat load profiles are essential for realistic district heating simulations, enabling optimized network operation and long‐term transformation planning. In this work, we systematically evaluate modeling approaches for generating such profiles, comparing data‐driven methods (e.g., regression and machine learning) with physics‐based techniques (e.g., differential equation model). Using a large dataset of Danish residential buildings, we present context‐aware, specific heat load profiles based on a gradient boosting approach, achieving substantially higher accuracy than conventional standard load profiles from the gas sector. Feature importance analysis confirms that the key impacts on the model align with findings from the pattern analysis, reinforcing the physical plausibility of the predictions. The resulting profiles are adaptable to varying building stock compositions and were partially validated through simulation and calibration in a real district heating network.
Heidrich et al. (Sun,) studied this question.