Decision Tree Based Method for Detecting Islanding using Pattern Recognition with HOG Features

Abstract

The ever-increasing demand for electricity drives the expansion of Distributed Generation (DG). Almost all DG sources are renewable in the long run. One of the major issues related with excessive penetration of DG sources is islanding. The islanding may endanger the clients and their equipment. This paper suggests a novel approach for detecting islanding based upon image categorisation using Decision Tree (DT). Images are utilised to generate the histogram of oriented gradient (HOG) features in order to distinguish between nonislanding and islanding events. The spectrogram images are obtained from the time-series signal which are derived from the point of common coupling. These images are used for extracting the features utilising histogram of oriented gradient, which is then used as a feature vector input for the DT classifier for training as well as testing. Rate of change of negative sequence voltage (ROCONSV) parameter is being used for deriving the spectrogram image. 5, 10, 15 and 20-fold cross-validations is being used for assessing the DT classifier’s effectiveness. The results from classification result demonstrates that detection of islanding detection using DT classifier and histogram of oriented gradient feature being utilised for image classification achieved excellent results with 100% accuracy and validation error of 0.13462.