Deep learning-based islanding strategy for DER microgrids using a hybrid ResNet-BiGRU model optimized by an improved white-faced capuchin algorithm

The integration of distributed energy resources (DERs) into modern power systems is steadily increasing to enhance system resilience. Traditional active and passive islanding detection (ID) techniques suffer from slow response time, large non-detection zones, and reduced accuracy under varying operating conditions. Furthermore, accurately distinguishing between islanding and fault events remains a complex task due to their comparable temporary features. To overcome these issues, this work proposes a novel ResNet-Bidirectional Gated Recurrent Unit-Self-Attention (RBG-SA) deep learning framework for robust islanding and fault event detection in distributed energy resource (DER) integrated power systems. The improved white-faced capuchin optimization algorithm (IWCOA) is utilized to adjust the weight parameters of the proposed model. The proposed RBG-SA model's efficacy in distinguishing islanding from fault conditions under various grid disturbances is demonstrated by evaluating it on accuracy, precision, recall, sensitivity, and specificity. The proposed method yields 98.2% accuracy, demonstrating the efficiency of the proposed system. The proposed model also reduces MSE and RMSE by 0.07% and 0.148%, respectively. Overall, the proposed method is effective at distinguishing between islanding and fault conditions under a variety of grid disruptions.