DC Series Arc Fault Detection and Localization Based on Photovoltaic Module‐Level Power Optimizer

ABSTRACT Under the scaling application of new energy power generation systems, electrical safety risks in photovoltaic power stations have become increasingly prominent. As a typical electrical hazard, DC‐side series arc faults can readily trigger severe incidents such as electrical fires due to arc high‐temperature effects if not timely addressed. To solve the challenges of detecting and precisely locating series arc faults, this paper proposes a novel arc fault detection and localization method for photovoltaic systems based on power optimizers. Firstly, a feature‐enhanced principal component analysis is applied to reduce the dimensionality of multidimensional features obtained from wavelet decomposition of current signals, effectively fusing arc fault characteristics. A sliding window adaptive threshold algorithm then dynamically updates feature thresholds in real time, significantly enhancing the detection system's robustness against varying photovoltaic operating conditions. Subsequently, attenuation patterns of current signal characteristics with transmission distance and frequency are theoretically derived, and a neural network‐based localization model is constructed using multioptimizer signal inputs. Finally, an experimental platform based on power optimizer was established to verify the feasibility of the feature‐enhanced PCA algorithm in real‐time detection and the precise fault localization ability of neural networks on the upper computer. This method achieved a fault detection accuracy of 96% and a positioning accuracy of 97.62%.