This study uses several artificial intelligence approaches to detect and estimate electrical faults in photovoltaic (PV) farms. The fault detection approaches of random forest, logistic regression, naive Bayes, AdaBoost, and CN2 rule induction were selected from a total of 12 techniques because they produced better decisions for fault detection. The proposed techniques were designed using distributed PV current measurements, plant current, plant voltage, and power. Temperature, radiation, and fault resistance were treated randomly. The proposed classification model was created using the Orange platform. A classification tree was visualized, consisting of seven nodes and four leaves, with a depth of four levels and edge width relative to parents. Thirty fault features attributes, four of them major, supported fault detection through the selected algorithms. The different fault types occurring in a PV farm were considered, including string fault, string-to-ground fault, and string-to-string fault. The selected classifiers were evaluated, and their performance was compared with respect to the important decision factors of precision, recall, classification accuracy, F-measure, specificity, and area under the receiver-operating curve. Using Simulink/MATLAB, a grid-connected 250-kW PV farm was implemented, including the converters control. Results confirmed that AdaBoost achieved the best performance.

AdaBoost algorithm; fault detection; logistic regression; Orange data mining; photovoltaic farm