Intelligent Islanding Classification with Optimal k-Nearest Neighbors Technique for Single Phase Grid Integrated PV System

To meet the power demands, grid integration of distributed generation systems (DGs) is fast growing in modern power systems. Islanding scenarios are taken into consideration as a concern for the stable functioning of utilities in the grid-integrated (GI) functioning of these systems. In this work, a method for classifying islanding scenarios and quickly identifying the grid anomalies is developed. Data sets that correspond to different grid anomalies are generated in the classification process and trained the data sets by using a machine learning algorithm. To understand this research, a 3.5 kW transformerless single-phase grid-integrated photovoltaic system (GIPV) is simulated and analyzed in Typhoon hardware-in-loop (HIL) software to obtain various data sets for grid anomalies. By using the k-nearest neighbor (kNN) classifier, the data sets are trained for islanding classification mechanism with the efficiency of 98.4%. The kNN classifier has capability to perform classification effectively at a rate of about 26000 observations per second.