Lindner, Joost HenningJoost HenningLindner2026-02-042026-02-042025Technische Universität Hamburg (2025)https://hdl.handle.net/11420/61076The increasing share of decentralized producers and flexible consumers complicates the maintenance of grid stability and raises the requirements for the control and monitoring of distribution networks. As part of a research project, a coordination function (COF) is being developed that performs day-ahead scheduling of flexible household loads to proactively prevent grid congestion. Since the schedule data of many households are incomplete, uncertainties arise in grid assessment and flexibility planning. This work addresses this issue and extends the existing concept by developing a forecasting method that provides the COF with a complete and reliable data basis. To this end, a data-driven approach is developed that applies the HDBSCAN clustering algorithm to categorize households with similar technical characteristics and consumption behavior, thereby forming representative load profiles. Missing schedules can thus be plausibly reconstructed. Subsequently, a gradient boosting regression refines the cluster-based forecasts by incorporating additional influencing factors such as weather, price, and grid data. The results demonstrate that the combination of clustering and regression analysis provides an effective approach for forecasting household schedules. This highly improves the data foundation for the COF and enables robust load forecasting even under limited data availability. The cluster forecast provides a high level of predictive accuracy at the aggregated level, while the regression analysis further improves accuracy at the household level. Individual households with highly specific consumption patterns can only be represented to a limited extent, which may affect the efficient use of flexibility in the grid but has only a minor impact on the overall reliability of the forecast. Future work should integrate real smart meter data to validate and improve the model quality and examine the use of deep learning-based methods to capture complex temporal patterns and short-term consumption changes even more accurately.dehttps://creativecommons.org/licenses/by-nc/4.0/Coordination functionforecasting household schedulesHDBSCAN clusteringgradient boosting regressionload forecastgrid flexibilityTechnology::621: Applied Physics::621.3: Electrical Engineering, Electronic EngineeringComputer Science, Information and General Works::006: Special computer methods::006.3: Artificial IntelligenceMaster Thesishttps://doi.org/10.15480/882.1656910.15480/882.16569Becker, ChristianChristianBeckerFischer, KathrinKathrinFischerNußbaum, FinnFinnNußbaumOther