Intelligent Detection of DER Islanding Events in Microgrids Using Deep Learning

This study presents a novel approach for Islanding Detection Techniques (IDT) in a Distributed Energy Resource (DER) system using a Convolution Neural Network-Long Short-Term Memory (CNN-LSTM) deep learning model. The DER system consists of a Photovoltaic (PV) module, a diesel generator set, and a Battery Energy Storage System (BESS) interconnected with the grid in the RSCAD-FX environment. The dataset comprises three operational scenarios: normal operation, Line-to-Ground (LG) fault, and islanding conditions. The model is trained using voltage and current signals measured at the Point of Common Coupling (PCC). The model integrates CNN for feature extraction and LSTM networks for temporal pattern recognition, achieving a high classification accuracy of 92.26%, which is a significant improvement over traditional techniques. Confusion matrix analysis revealed a strong performance of the proposed method. Islanding events were detected with an impressive 100% recall, while some normal operation cases were misclassified as islanding (1352 time samples). LG faults were classified with 99% precision and 95% recall, ensuring minimal false detection. The overall F1-score was 0.92. Receiver Operating Characteristic (ROC) curve analysis further validated the robustness of the method, showing high Area Under the Curve (AUC) values across all classes. The findings demonstrate that deep learning-based IDTs in DER systems are feasible and accurate, contributing to grid reliability and protection.