DC series arc fault detection method using LDA-CatBoost for PV system under shaded condition

Abstract Detecting DC series arc faults (SAFs) in photovoltaic (PV) systems is challenging due to shaded conditions. A detection method for SAFs in PV systems focusing on shaded conditions is proposed in this paper. By analyzing voltage–current characteristics under different shaded scenarios and the interaction of arc and PV sources to ignite arc faults, the DC SAFs are classified into three types: strong, weak, and intermittent arcs, and then the DC arc features under various shaded conditions are investigated from the current signals. Then, a linear discriminant analysis (LDA)-CatBoost method is proposed to detect and identify SAFs using the features from extracting both time- and frequency-domain current signals. Experimental results demonstrate that the detection and identification accuracy is 98% under various shaded conditions. It significantly improves SAF detection applicability compared with other methods.