Photovoltaic Systems Status Monitoring Utilizing Machine Learning Technique

This article proposes a Data-based condition monitoring technique (CM) to observe the operational status of a Photovoltaic (PV) system. Without any need for more sensors than traditional measurements installed in the DC Combiner Box which are one Voltage sensor for the array and one Current sensor for each string, Kernel-based Gaussian process regression (GPR) is capable of nonlinear approximation, So it is utilized to empirically model the DC power of PV generator and compute residuals which are used for detecting faults. The residuals go via an exponentially weighted moving averages filter. Kernel density estimation (KDE) is employed in this study to detect faults by examining the univariate residuals. The four failure situations that have received the most attention in this study are the PV string open circuit fault, partial shading, shorted PV modules, and module deterioration or degradation. Using information from a 5, 310Wp gridconnected solar power system simulated on Matlab Simulink. Through this study, six kernel functions were tested for the GPR model where two of them were customized, These functions are Quadratic, Matern, Exponential, Squared Exponential, Dot product Plus White kernel, and Matern kernel Plus quadratic kernel, by performing advanced preprocessing, and comparing results and performance metrics we found that MATERN and Customized Matern Kernelbased GPR models perform better than other kernel-based GPR to monitor PV systems.