Coordination of Machine Learning and Current Modulation for DC Series Arc Fault Detection and Extinguishing for Module-Level PV Systems

A module-level power electronics can achieve the maximum power generation for each PV panel with overcoming the partial shading. However, it increases the possibility of DC series arc fault condition between the PV panel and converter. The current modulation can detect and extinguish the DC series arc fault condition using the characteristics of arc resistance. However, this method degrades the power generation according to inaccurate fault detection. This paper proposes the DC series arc fault detection and extinguishing with coordination of current modulation and machine learning method, which can improve the accuracy of fault detection through double screening. The proper machine learning algorithm can minimize the undesired trigger of current modulation, which improves the PV power generation at normal condition. The proposed method is possible to distinguish between arc fault and normal conditions regardless of impedance and external changes. The operational principle is analyzed to coordinate the current modulation and machine learning. The experimental results using 800-W prototype DC-DC converter can verify the arc fault detection and extinguishing