Data-driven understanding of energy systems

My goal is to understand complex energy systems from a data-driven point of view, i.e. without assuming too much prior knowledge. 

I pay special attention to making any forecast or analysis transparent and explainable, using and developing interpretable machine learning tools. Thereby, I wish to identify drivers and risks for energy systems and help to design future systems, e.g. involving a high share of wind generation, electric cars, heat pumps etc.

Recent highlights

With uncertain generation and demand patterns, we require also probabilistic approaches to power systems. We showcased how boosting and neural networks transparently quantify system stability in a tabular data set (ACM e-energy Nikoltchovska 2025 et al).

As power systems become more volatile, we require treatment of highly-resolved time series. Hence we worked on forecasting of highly-resolved data (ACM e-energy 2024 Pütz et al) and on how we can identify complex power grid dynamics purely from data (ACM e-energy 2024 Wen et al)
These results were also presented at ICLR and NeurIPS workshops by Climate Change AI.

Constructing new transmission lines in power grids may induce Braess' Paradox, i.e. reduce system stability. Our approach, identifies Braessian lines and may guide grid extension plans. Nature Communications 2022  (I also published a German public science article in Spektrum)

Contact

If any of this sounds interesting to you, feel free to get in touch