Convolutional Neural Network-Based Intelligent Protection Strategy for Microgrids

Microgrids experience significantly different fault currents in different operating scenarios, which make microgrid protection challenging. Existing intelligent protection schemes rely on the extraction of appropriate fault features using statistical parameters. The selection of these features is difficult in a microgrid because of its various operating scenarios. This study develops a convolutional neural network-based intelligent fault protection strategy (CNNBIPS) for microgrids that inherently integrates the feature extraction and classification process. The proposed strategy is directly applicable to three-phase (TP) current signals; thus, it does not require any separate feature extractor. In the proposed CNNBIPS, TP current signals sampled by the protective relays are used as an input to three different CNNs. The CNNs apply convolution and pooling operations to extract the features from the input signals. Then, fully connected layers of the CNNs employ the features to develop fault-type, phase, and location information. To analyse the efficacy of the proposed design, we execute exhaustive simulations on a standard microgrid test system. The results confirm the effectiveness of the proposed strategy in terms of detection accuracy, security, and dependability. Moreover, comparisons with previous methods show that the proposed approach outperforms the existing microgrid protection schemes.