MACHINE LEARNING BASED FAULT DIAGNOSIS SCHEME FOR GRID-CONNECTED PV SYSTEM

Abstract

Photovoltaic (PV) systems have recieved a lot of attention in recent decades due to their accessibility and advancements in PV technology. The protection of PV systems from faults such as String to String (SS), String to Ground (SG), Open circuit (OC), and partial shading are the key challenges to the realization of cost-effective and environmentally friendly PV systems. Such unusual circumstances reduce the maximum available PV power. Partial shading and breakdowns in a PV array must therefore be noticed quickly for enhanced system efficiency and reliability. The significant fault current in PV systems can be detected using the existing safety devices in PV systems, such as fuses and residual current detectors. The flowing fault current being of low order is not significant enough for current protection devices to detect if the solar and/or fault mismatch is modest and the fault resistance is high. As a result, under cloudy and low irradiance conditions, the traditional protection devices fail to identify problems, resulting in reliability concerns and photovoltaic fire threats. In this context, a fault diagnosis scheme for PV systems is presented in this paper, which includes feature extraction using the Discrete wavelet transform, and classification of various defects on the PV system using Decision tree.