GRID-ML

The main objective of the project is an automated process for robust and accurate fault detection and diagnosis in low and medium voltage grids using measurement data from multiple measurement devices distributed across the grid, reducing fault localization times compared to conventional approaches.

For this purpose, learning methods involving external knowledge sources must be explored for accurate and timely data-driven and robust localization of faults in medium and low voltage power systems based on a Digital Twin. Furthermore, robust approaches ensure reliable localization even with measurement data that is often missing or inaccurate in practice. For this purpose, the GRID-ML (GRID-Machine Learning)research project relies on data from heterogeneous measuring devices at different locations in the network for fault localization.

The overall objective of the project is the research of robust learning methods for the accurate and real-time data-driven localization of faults in medium and low-voltage power grids on the basis of a digital twin. The simulation of fault conditions in the digital twin forms the basis for generating a representative set of training and test data for learning procedures. In this context it has to be investigated which measurement and network topology data have to be available and which data or features can be used for localization. In general there are three sub-goals for this:

1. Quantify the robustness of existing methods under topology data errors, measurement errors, and data altered due to privacy procedures.
2. Ensure and improve the robustness against data errors
3. Transfer the methods and adapt them to real scenarios, with the goal of a trustworthy use of the developed methods at the distribution grid operator without loss of accuracy.

Federal Ministry of Education and Research
DLR Projektträger KMU Innovativ Nr 01|S23006B

01.07.2023 - 30.06.2026