1713. Leveraging similarities in Grid Topologies for Learning in Power Systems
Invited abstract in session WD-49: Advances in Conic Optimization and Applications, stream Conic and polynomial optimization.
Wednesday, 14:30-16:00Room: Parkinson B10
Authors (first author is the speaker)
| 1. | Tim Graulich
|
| DTU Management, Technical University of Denmark | |
| 2. | Marie Münster
|
| DTU Management Engineering, Technical University of Denmark | |
| 3. | Bissan Ghaddar
|
| Ivey Business School | |
| 4. | Rasmus Bramstoft
|
| Technical University of Denmark |
Abstract
The optimal power flow (OPF) problem, particularly the alternating current formulation (AC-OPF), is fundamental to the operation of electricity grids. However, it presents a significant computational challenge due to its non-linear and non-convex nature. In recent years, machine learning (ML) has gained attention as a promising tool for enhancing the efficiency of solving the AC-OPF problem. While many studies focus on improving the predictive performance of these ML models, a frequently remaining caveat is the fixed topology requirement for these techniques. Thus, whenever the grid topology changes, new training samples need to be generated, and the models retrained. An expensive process, especially for large systems.
This study aims to investigate the potential of identifying similarities between different grid topologies, especially common generator characteristics, to improve the adaptive performance of ML methods. This research seeks to reduce the issue of the fixed topology requirement, remaining in many proposed ML solutions and ultimately enhance the decision-making process for power grid operations.
Keywords
- OR in Energy
- Machine Learning
Status: accepted
Back to the list of papers