Data-driven analysis of complex systems (DRACOS)



Our research aims to solve complex problems of our time, focusing on sustainability and especially the energy transition and the power grid. Methodologically, the group “Data-driven analysis of complex systems (DRACOS)” combines exploratory data analysis, physical modeling and machine learning methods. One focus here is on the interpretability of the models: Instead of “black box” predictions, we develop transparent models.


Mitigation of climate change requires a fundamental transformation of our energy system. Power plants based on fossil fuels must be replaced by renewable power sources, such as wind and solar power. This energy transition (“Energiewende”) towards a sustainable energy system raises numerous complex challenges, as power generation becomes more uncertain, while simultaneously more operational data becomes available. Hence, data-driven approaches become feasible and even necessary to fully understand the energy systems of today and tomorrow across all scales.

Machine learning and artificial intelligence can handle these enormous amounts of data but need to do so in a transparent way. Obtaining classifications or forecasts without explanations limits their use severely. Hence, we aim to explain “black box” models using interpretability tools or develop outright interpretable models. For example, an algorithm should explain which external factors, such as the feed-in of photovoltaic systems, the current electricity price or the time of day, are relevant for its prediction of the power grid frequency or household consumption. This transparency then enables synergies from machine and human models: Where is the machine better than the human? What can we learn from this for our human models and thus make them better?


Images adapted – Icon made by Becris from


Our contributions are summarized in our vision and mission statements:

Vision: Enable a sustainable and collaborative future, where all energy generation is sustainable,  data informs decisions and models are understood by everyone.

Mission: We develop and apply state-of-the-art machine learning methods to understand complex energy systems from households to transmission systems. We emphasize the openness of data and interpretability of models to enable collaboration among scientists and industry. We engage the general public by sharing and explaining our results.   

Applications and partner

In our research, we utilize data from the Energy Lab 2.0 and frequency measurements of various power grids. We collaborate with colleagues at KIT, other universities and Helmholtz centers, e.g. within Helmholtz AI (Helmholtz Artificial Intelligence Cooperation Unit).

For further information, contact Benjamin Schäfer.


Microscopic Fluctuations in Power-Grid Frequency Recordings at the Subsecond Scale
Schäfer, B.; Rydin Gorjão, L.; Yalcin, G. C.; Förstner, E.; Jumar, R.; Maass, H.; Kühnapfel, U.; Hagenmeyer, V.
2023. (H. Sayama, Ed.) Complexity, 2023, Art.-Nr.: 2657039. doi:10.1155/2023/2657039
Spatial heterogeneity of air pollution statistics in Europe
He, H.; Schäfer, B.; Beck, C.
2022. Scientific Reports, 12 (1), Artikel-Nr.: 12215. doi:10.1038/s41598-022-16109-2
Machine learning approach towards explaining water quality dynamics in an urbanised river
Schäfer, B.; Beck, C.; Rhys, H.; Soteriou, H.; Jennings, P.; Beechey, A.; Heppell, C. M.
2022. Scientific Reports, 12 (1), Art.-Nr.: 12346. doi:10.1038/s41598-022-16342-9
Secondary control activation analysed and predicted with explainable AI
Kruse, J.; Schäfer, B.; Witthaut, D.
2022. Electric Power Systems Research, 212, Art.-Nr.: 108489. doi:10.1016/j.epsr.2022.108489
Predicting the power grid frequency of European islands
Lund Onsaker, T.; Nygård, H. S.; Gomila, D.; Colet, P.; Mikut, R.; Jumar, R.; Kühnapfel, U.; Maass, H.; Hagenmeyer, V.; Witthaut, D.; Schäfer, B.
2022, September 27. doi:10.48550/arXiv.2209.15414
Transformer Neural Networks for Building Load Forecasting
Hertel, M.; Ott, S.; Neumann, O.; Schäfer, B.; Mikut, R.; Hagenmeyer, V.
2022. 36. Conference on Neural Information Processing Systems (NeurIPS 2022), Online, November 28–December 10, 2023
Initial analysis of the impact of the Ukrainian power grid synchronization with Continental Europe
Böttcher, P. C.; Rydin Gorjão, L.; Beck, C.; Jumar, R.; Maass, H.; Hagenmeyer, V.; Witthaut, D.; Schäfer, B.
2022. Energy Advances, 2 (1), 91–97. doi:10.1039/D2YA00150K
Evaluation of Transformer Architectures for Electrical Load Time-Series Forecasting
Hertel, M.; Ott, S.; Schäfer, B.; Mikut, R.; Hagenmeyer, V.; Neumann, O.
2022. Proceedings - 32. Workshop Computational Intelligence: Berlin, 1. - 2. Dezember 2022. Hrsg.: H. Schulte, F. Hoffmann; R. Mikut, 93–110, KIT Scientific Publishing
Understanding Braess’ Paradox in power grids
Schäfer, B.; Pesch, T.; Manik, D.; Gollenstede, J.; Lin, G.; Beck, H.-P.; Witthaut, D.; Timme, M.
2022. doi:10.48550/arXiv.2209.13278
Understanding Braess’ Paradox in power grids
Schäfer, B.; Pesch, T.; Manik, D.; Gollenstede, J.; Lin, G.; Beck, H.-P.; Witthaut, D.; Timme, M.
2022. Nature Communications, 13 (1), Art.-Nr.: 5396. doi:10.1038/s41467-022-32917-6
Inferring Topology of Networks With Hidden Dynamic Variables
Schmidt, R.; Haehne, H.; Hillmann, L.; Casadiego, J.; Witthaut, D.; Schafer, B.; Timme, M.
2022. IEEE Access, 10, 76682–76692. doi:10.1109/ACCESS.2022.3191665
Boost short-term load forecasts with synthetic data from transferred latent space information
Heidrich, B.; Mannsperger, L.; Turowski, M.; Phipps, K.; Schäfer, B.; Mikut, R.; Hagenmeyer, V.
2022. DACH+ Conference on Energy Informatics German Federal Ministry for Economic Affairs and Energy
Boost short-term load forecasts with synthetic data from transferred latent space information
Heidrich, B.; Mannsperger, L.; Turowski, M.; Phipps, K.; Schäfer, B.; Mikut, R.; Hagenmeyer, V.
2022. Energy Informatics, 5 (S1), Article no: 20. doi:10.1186/s42162-022-00214-7
Data-driven load profiles and the dynamics of residential electricity consumption
Anvari, M.; Proedrou, E.; Schäfer, B.; Beck, C.; Kantz, H.; Timme, M.
2022. Nature Communications, 13 (1), Art.-Nr.: 4593. doi:10.1038/s41467-022-31942-9
Validation Methods for Energy Time Series Scenarios From Deep Generative Models
Cramer, E.; Gorjao, L. R.; Mitsos, A.; Schafer, B.; Witthaut, D.; Dahmen, M.
2022. IEEE Access, 10, 8194–8207. doi:10.1109/ACCESS.2022.3141875
Secondary frequency control stabilising voltage dynamics
Tchuisseu, E. B. T.; Dongmo, E.-D.; Procházka, P.; Woafo, P.; Colet, P.; Schäfer, B.
2022. European Journal of Applied Mathematics, 1–17. doi:10.1017/S095679252100036X
Phase and Amplitude Synchronization in Power-Grid Frequency Fluctuations in the Nordic Grid
Rydin Gorjao, L.; Vanfretti, L.; Witthaut, D.; Beck, C.; Schäfer, B.
2022. IEEE Access, 10, 18065–18073. doi:10.1109/ACCESS.2022.3150338
Open database analysis of scaling and spatio-temporal properties of power grid frequencies
Rydin Gorjão, L.; Jumar, R.; Maass, H.; Hagenmeyer, V.; Yalcin, G. C.; Kruse, J.; Timme, M.; Beck, C.; Witthaut, D.; Schäfer, B.
2020. Nature Communications, 11 (1), Art.-Nr. 6362. doi:10.1038/s41467-020-19732-7
An ensemble-averaged, cell density-based digital model of zebrafish embryo development derived from light-sheet microscopy data with single-cell resolution
Kobitski, A. Y.; Otte, J. C.; Takamiya, M.; Schäfer, B.; Mertes, J.; Stegmaier, J.; Rastegar, S.; Rindone, F.; Hartmann, V.; Stotzka, R.; Garcia, A.; Wezel, J. van; Mikut, R.; Strähle, U.; Nienhaus, G. U.
2015. Scientific Reports, 5 (8601), 1–10. doi:10.1038/srep08601
realTox: Real-time imaging of toxicant impact in whole organisms at single cell resolution
Otte, J. C.; Kobitskiy, A.; Stegmaier, J.; Takamiya, M.; Hartmann, V.; Schäfer, B.; Garcia, A.; Wezel, J. van; Stotzka, R.; Nienhaus, G. U.; Mikut, R.; Strähle, U.
2013. Jahrestagung der Deutschen Gesellschaft für Experimentelle und Klinische Pharmakologie und Toxikologie (DGPT), Halle, 5.-7. März 2013