Diagnosis of PV Array Faults Using RUSBoost

Solar photovoltaic (SPV) has become an inalienable part of the power system due to its numerous advantages over conventional energy sources. However, SPVs can be subjected to various kinds of faults which can degrade the overall performance of the system. Machine learning (ML) techniques may be useful to identify faults occurring in photovoltaic (PV) systems. In this paper, an ML technique called random under-sampling boosting (RUSBoost) has been applied to detect the different types of faults occurring on the DC side of the PV system. A test system of 4.8 kW(p) has been designed in MATLAB/Simulink environment for data acquisition of different operating conditions. Commonly used performance parameters have been used as features for the ML model. Thereafter, RUSBoost has been trained using features acquired from the test system. The work also investigates the optimum number of features required for fast and accurate detection of PV array faults. It has been found that, training the model with the current ratio, voltage ratio, power ratio, and array efficiency gives the best result with 99.6% training accuracy and 2.78 s of training time. The performance of RUSBoost is further compared to popular AdaBoost and bagged tree ensemble classifier algorithm to establish the efficacy of the applied ML technique.