Abstract
This study proposes a new fault diagnosis algorithm for a photovoltaic (PV) solar system implemented on the bases of diode distribution and two classifiers based on ensemble bagged trees and neural pattern recognition techniques. The proposed approach is designed by adding two diodes connected to the upper and lower terminals of each PV string. The two diodes are connected to prevent the reverse direction of faulted string currents, thereby preventing high injecting fault currents. Given that the string fault current is prevented owing to diode interaction, this dynamic interaction provides new states during the fault periods considered to detect string faults. Owing to the distributed diodes, system voltage is not reduced during fault disturbances and voltage is not considered an input to the proposed detection techniques. Monitoring currents at the upper and lower ends of each string is required as input for the proposed approaches to detect string fault types. The proposed approaches are used to detect string fault types, such as cell-to-string negative terminal, cell-to-cell in the same string, and string-to-string faults. The proposed model is applied for 400 kW, and the PV system consists of 4 arrays interconnected with 1200 V AC grid built-in MATLAB/Simulink. Experimental results validate the newly suggested approaches. (c) 2020 Elsevier Ltd. All rights reserved.