This page contains automatically translated content.

M. Sc. Dominik Köhler

Reinforcement Learning for Cognitive Energy Systems (RL4CES)

Email: dkoehler[at]uni-kassel[dot]de

Publications

[ 2025 ]

2025[ to top ]

PPTopoGym: Towards an RL Environment for Topology Actions on Power Grids.Köhler, Dominik; Hassouna, Mohamed; Degtyar, Dmitry; Krauß, Jonas; Brendlicger, Kurt; Scholz, Christoph. In Workshop on Machine Learning for Sustainable Power Systems (ML4SPS 2025), ECML PKDD. Springer, 2025.
- [ Abstract ]
- [ BibTeX ]
- [ EndNote ]
Power grids are facing growing challenges, arising from the increasing share of renewable energy sources and highly variable loads, in- troducing significant variability into both power generation and demand. This variability can create transmission line bottlenecks, which in turn lead to overheating. Transmission system operators address this prob- lem by rerouting the power flow via topological control, which is more cost-effective than redispatch measures. While reinforcement learning has shown promise in optimizing such topology changes, existing reinforce- ment learning environments exhibit notable differences from real-world operations. To support research and bridge the gap to real-world ap- plications, we present PPTopoGym—a reinforcement learning environ- ment for evaluating the impact of topology actions on the power flow in pandapower-based grids. It allows for simulations with variable load and generation profiles, and supports direct computation of key grid security metrics, i.e., N−0 and N−1 security constraints for each time step. PP- TopoGym addresses key limitations of existing environments by enabling realistic grid modeling and the integration of real-world data. The code is publicly available.
@inproceedings{kohler2025pptopogym, abstract = {Power grids are facing growing challenges, arising from the increasing share of renewable energy sources and highly variable loads, in- troducing significant variability into both power generation and demand. This variability can create transmission line bottlenecks, which in turn lead to overheating. Transmission system operators address this prob- lem by rerouting the power flow via topological control, which is more cost-effective than redispatch measures. While reinforcement learning has shown promise in optimizing such topology changes, existing reinforce- ment learning environments exhibit notable differences from real-world operations. To support research and bridge the gap to real-world ap- plications, we present PPTopoGym—a reinforcement learning environ- ment for evaluating the impact of topology actions on the power flow in pandapower-based grids. It allows for simulations with variable load and generation profiles, and supports direct computation of key grid security metrics, i.e., N−0 and N−1 security constraints for each time step. PP- TopoGym addresses key limitations of existing environments by enabling realistic grid modeling and the integration of real-world data. The code is publicly available.}, author = {Köhler, Dominik and Hassouna, Mohamed and Degtyar, Dmitry and Krauß, Jonas and Brendlicger, Kurt and Scholz, Christoph}, booktitle = {Workshop on Machine Learning for Sustainable Power Systems (ML4SPS 2025), ECML PKDD}, keywords = {itegpub}, note = {(accepted)}, publisher = {Springer}, title = {PPTopoGym: Towards an RL Environment for Topology Actions on Power Grids}, year = 2025 }
%0 Conference Paper %1 kohler2025pptopogym %A Köhler, Dominik %A Hassouna, Mohamed %A Degtyar, Dmitry %A Krauß, Jonas %A Brendlicger, Kurt %A Scholz, Christoph %B Workshop on Machine Learning for Sustainable Power Systems (ML4SPS 2025), ECML PKDD %D 2025 %I Springer %T PPTopoGym: Towards an RL Environment for Topology Actions on Power Grids %X Power grids are facing growing challenges, arising from the increasing share of renewable energy sources and highly variable loads, in- troducing significant variability into both power generation and demand. This variability can create transmission line bottlenecks, which in turn lead to overheating. Transmission system operators address this prob- lem by rerouting the power flow via topological control, which is more cost-effective than redispatch measures. While reinforcement learning has shown promise in optimizing such topology changes, existing reinforce- ment learning environments exhibit notable differences from real-world operations. To support research and bridge the gap to real-world ap- plications, we present PPTopoGym—a reinforcement learning environ- ment for evaluating the impact of topology actions on the power flow in pandapower-based grids. It allows for simulations with variable load and generation profiles, and supports direct computation of key grid security metrics, i.e., N−0 and N−1 security constraints for each time step. PP- TopoGym addresses key limitations of existing environments by enabling realistic grid modeling and the integration of real-world data. The code is publicly available.
Flow-Attentional Graph Neural Networks.Plettenberg, Pascal; Köhler, Dominik; Sick, Bernhard; Thomas, Josephine M. In Transactions on Machine Learning Research. 2025.
- [ Abstract ]
- [ BibTeX ]
- [ EndNote ]
- [ URL ]
Graph Neural Networks (GNNs) have become essential for learning from graph-structured data. However, existing GNNs do not consider the conservation law inherent in graphs associated with a flow of physical resources, such as electrical current in power grids or traffic in transportation networks, which can lead to reduced model performance. To address this, we propose flow attention, which adapts existing graph attention mechanisms to satisfy Kirchhoff’s first law. Furthermore, we discuss how this modification influences the expressivity and identify sets of non-isomorphic graphs that can be discriminated by flow attention but not by standard attention. Through extensive experiments on two flow graph datasets (electronic circuits and power grids), we demonstrate that flow attention enhances the performance of attention-based GNNs on both graph-level classification and regression tasks.
@article{plettenberg2025flow, abstract = {Graph Neural Networks (GNNs) have become essential for learning from graph-structured data. However, existing GNNs do not consider the conservation law inherent in graphs associated with a flow of physical resources, such as electrical current in power grids or traffic in transportation networks, which can lead to reduced model performance. To address this, we propose flow attention, which adapts existing graph attention mechanisms to satisfy Kirchhoff’s first law. Furthermore, we discuss how this modification influences the expressivity and identify sets of non-isomorphic graphs that can be discriminated by flow attention but not by standard attention. Through extensive experiments on two flow graph datasets (electronic circuits and power grids), we demonstrate that flow attention enhances the performance of attention-based GNNs on both graph-level classification and regression tasks.}, author = {Plettenberg, Pascal and Köhler, Dominik and Sick, Bernhard and Thomas, Josephine M.}, journal = {Transactions on Machine Learning Research}, keywords = {itegpub}, title = {Flow-Attentional Graph Neural Networks}, year = 2025 }
%0 Journal Article %1 plettenberg2025flow %A Plettenberg, Pascal %A Köhler, Dominik %A Sick, Bernhard %A Thomas, Josephine M. %D 2025 %J Transactions on Machine Learning Research %T Flow-Attentional Graph Neural Networks %U https://openreview.net/forum?id=tOzg7UxTPD %X Graph Neural Networks (GNNs) have become essential for learning from graph-structured data. However, existing GNNs do not consider the conservation law inherent in graphs associated with a flow of physical resources, such as electrical current in power grids or traffic in transportation networks, which can lead to reduced model performance. To address this, we propose flow attention, which adapts existing graph attention mechanisms to satisfy Kirchhoff’s first law. Furthermore, we discuss how this modification influences the expressivity and identify sets of non-isomorphic graphs that can be discriminated by flow attention but not by standard attention. Through extensive experiments on two flow graph datasets (electronic circuits and power grids), we demonstrate that flow attention enhances the performance of attention-based GNNs on both graph-level classification and regression tasks.