A deep learning–based CNN approach for accurate islanding detection in microgrids using droop-controlled signal behavior
This paper proposes an approach that leverages a Convolutional Neural Network (CNN) as a classifier for islanding detection. The proposed method is distinguished by monitoring the individual response of each DG in the system, making it suitable for multi-DG systems. The method monitors the behavior of droop-controlled signals within distributed generation (DG) controllers, analyzing their deviations from expected droop characteristics under various operating conditions. These responses are formatted into two-dimensional images representing droop behavior and fed into the CNN for classification. The approach demonstrates high accuracy of 98.68% in distinguishing between intentional islanding, unintentional islanding, and other normal disturbances. The performance of the model is capped by an outstanding Area Under the Curve (AUC) value of 0.99. Simulation results validate the robustness of the method in accurately detecting islanding events in microgrids.
Authors
- Elhossiny Ibrahim (ORCID: https://orcid.org/0000-0001-5421-2170)
- Mahmoud M. Elgamasy (ORCID: https://orcid.org/0000-0003-3023-082X)
Institutions
- Majmaah University (SA)
- Arab East Colleges (SA)
- Menoufia University (EG)
Publication Details
- Journal
- Scientific Reports
- Published
- 2026-11-19
- DOI
- https://doi.org/10.1038/s41598-026-81296-1
- Primary Topic
- Islanding Detection in Power Systems
- Type
- article
- Field-Weighted Citation Impact
- 0.00