A Novel Approach for Detecting Islanding Using Machine Learning Technique

The consistent demand for electricity is driving the expansion of Distributed Generation (DG). Almost all DG sources are renewable in the long term. The excessive penetration of DG sources is one of the primary issues associated with islanding. The clients and their equipment may be at risk as a result of islanding. This study suggests a Machine Learning (ML) classifier-based approach for the detection of islanding that is predicated on the categorisation of images. In order to differentiate between non-islanding and islanding events, the Histogram of Orientated Gradient (HOG) characteristics is generated using images. The spectrogram images are produced by the time-series signal that is derived from the point of common coupling. The HOG technique is employed to extract the features from these images, and the resulting feature vector is utilised as an input for the ML classifier for both training and testing purposes. The Rate of Change of Voltage (ROCOV) parameter is employed to generate the spectrogram's image. We are employing 5-fold cross-validation to assess the efficacy of the ML classifier. The classification results indicate that the use of an ML classifier and the HOG feature provides excellent results and minimal validation error when used to classify images for islanding detection.