Performance assessment of selective machine learning techniques for improved PV array fault diagnosis

Abstract

Solar photovoltaics (SPV) are susceptible to various kinds of faults which can diminish overall performance of the system. Proper fault diagnosis strategy needs to be developed to accurately identify the faults for smooth operation of the photovoltaic (PV) systems. Machine learning (ML) can be used to diagnose the faults in PV arrays. In this paper, three powerful machine learning algorithms i.e., categorical boosting (CatBoost), light gradient boosting method (LGBM), and extreme gradient boosting (XGBoost) have been selected for investigating their efficacy to diagnose different PV array faults. A PV system has been designed in MATLAB/Simulink environment using real time irradiance and temperature data acquired from grid connected PV System of National Institute of Technology Agartala. The constructed dataset is used to extract features including one new index to train these algorithms in Python 3.7. Promising results have been achieved using these algorithms as average detection and classification accuracy of 99.996% and 99.745% has been noted by implementing LGBM, followed by CatBoost, and XGBoost respectively. Moreover, these algorithms reduce the computational time significantly with LGBM leading the chart with training time of 0.053 and 0.375 s for fault detection and classification. These algorithms have been compared with random forest (RF) technique to exhibit their proficiency in fault diagnosis of PV arrays. • Assessment of CatBoost, LGBM and XGBoost algorithms for PV array fault diagnosis. • A new performance index based on array capture loss of PV array. • Selected methods detect various types of faults in PV array with high accuracy. • Applied techniques can detect and classify faults with very less training time. • Methods are scrutinized with various performance indicators to confirm efficacy.