Islanding Classification with Optimized k-Nearest Neighbors for Three Phase Grid Connected Photovoltaic System

The grid penetration of distributed generation systems is rapidly increasing to meet the energy demands. In the grid connected operation of these systems, the islanding scenarios are considered as a critical threat for stable operation of the utility. This paper aims at developing an islanding classification technique to efficiently detect the grid abnormalities and classify the islanding scenario. The process involves generation of data sets corresponding to various grid abnormalities, and training them with a machine learning classifier. To realize this development, a 10 kW three-phase grid connected photovoltaic (PV) system is simulated and compiled in Typhoon hardware-in-loop (HIL) environment to get different data sets for grid abnormality conditions. Further, the data is trained with the k-nearest neighbor (kNN) classified to develop the islanding classification mechanism. The efficiency of trained classifier is 94.2 %, and has the capability for efficient classification at a speed of approximately 17000 observations per second.