Islanding Detection in PV-Grid System by Random Forest and Spectrogram Based LBP

Islanding is a significant obstacle along with excessive penetration of DG sources. The islanding could cause harm to the customers and their equipment's. The IEEE 1547 DG interconnection regulations stipulate that the DG must be disconnected and the islanding must be observed within two seconds. Basically, this paper presents a novel method for detecting islanding based upon image pattern recognition using random forest (RF) classification technique is proposed. Images are utilized to derive the local binary pattern (LBP) characteristics so as to identify non-islanding as well as islanding events. Hence, the time series signal is obtained and transformed into spectrogram images at the point of common connection. These images are utilized to extract the features using histogram of oriented gradient, which is then utilized as a feature vector input for the RF classifier for training as well as testing. It uses rate of change of negative sequence voltage parameter to derive spectrogram image. The RF classifier's effectiveness is assessed with 5-fold cross-validations. The results from classification demonstrates that detection of islanding using RF classifier and LBP feature being used for image pattern recognition achieved excellent results with 98.25% accuracy and detection time of 182 milliseconds.