On-Line Faults Detection and Classification in PV array Using Bayesian and K-Nearest Neighbor Classifier

Nowadays, Distributed Generation (DG), especially PV systems as a new source of power absorbed a high percentage of investment. Fault Detection and analysis in solar photovoltaic (PV) arrays are important issues to increase reliability, efficiency and safety in PV arrays. Due to PV’s non-linear characteristics, current-limiting nature, high fault impedances, low irradiance conditions, the PV grounding schemes, or inverter condition and protection system weakness, faults in PV arrays are not properly recognized. Therefore, to fill this protection gap, machine learning techniques has been proposed for fault detection based on PV array voltage and current measurements and irradiance and temperature in a grid connected 17.6 kw photovoltaic system.To determine the type and classification of the faults, choosing the best method of classification with high accuracy and finding suitable feature in commercial-scale photovoltaic arrays, are important issues that has not been done so far. The input data for using Bayesian and K-Nearest Neighbor Methods are the simulation results of different defined classes of the line to line and open circuit faults by various temperature and irradiance. The results have shown that using the suggested classification system is very successful in the detection and classification of faults in an array string.