Research
Data-driven understanding
The aim of my research is to obtain a better understanding of complex systems, in particular complex energy systems, such as the power grid, and thereby support the energy transition.
In my group “Data-driven analysis of complex systems (DRACOS)”, we combine exploratory data analysis, physical modeling and machine learning methods. One focus is on the interpretability of the models: Instead of “black box” predictions, transparent models are being developed, e.g interpretable machine learning or explainable artificial intelligence (XAI).
For example, when forecasting the dynamics of the power grid frequency or the consumption of a household, the 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. Such transparency then enables synergies from machine and human models: Are the results obtained by the machine consistent with human models? Where is the machine performing better than the human? What can we learn from the machine to improve upon human-made models?
Forecasting energy time series
Energy systems are mostly characterized by time series, ranging from changing prices, dynamic demand trajectories to volatile renewable generation. Hence, developing tayolor-made forecasting tools is critical for current and future energy systems. I focus in particular on highly-resolved time series, going down to e.g. 1-second resolution, as I am convinved that these time scales will become more important in the future.
Read more at ACM e-Energy 2024 Pütz et al
Reinforcement learning for complex energy systems
With economic development, the complexity of infrastructure has increased drastically and so has the need for more control options. Complementing classical control options, such as model-repdictive control, Reinforcement Learning (RL) has emerged as a promising solution. I focus in my research how RL can be used in energy systems and how its actions can be explained to users and experts.
Read more at ExEn 2024
Understanding energy systems with eXplainable AI (XAI)
Numerous factors influence the dynamics of energy systems. Complementing human-made models, I pursue the idea of machine-made models, e.g. to predict the power grid stability. However, I pay special attention to the transparency of these models, i.e. the goal is to obtain eXplainable artificial intelligence (XAI), either with post-hoc explanations or with inherently interpretable models.
Read more about XAI applied to power grids in Patterns, ACM e-energy 2023 or Energy and AI
Combining physical insights and machine learning
Power systems are experiencing rapid stochastic fluctuations and slower effects of scheduling. Human-designed models and machine- and thereby data-based methods are two complementary approaches to describe these systems. In my research, I integrate domain knowledge into the design of artificial intelligence systems, e.g. training a neural network combined with a stochastic model.
Read more in PRX Energy 2023
Modelling of driven and stochastic energy systems
Power systems are intrinsically driven by deterministic changes, e.g. the load increases in the morning and then decreases during the late evening. In addition, short-term stochastic changes introduce fluctuations, e.g. when demand fluctuates on a short time scale or (volatile renewable) generation is not constant. I wish to quantify, understand and emulate such driven and stochastic systems. Aside from stochastic modelling, we employ also symbolic regression.
Read more in IEEE Access or IEEE PowerTech 2023 or ACM e-Energy 2024 Wen et al
Information for potential applicants
I am always open to applications. Please include why you want to work with me and also consider applying for external funding as my group's funding is limited.
