Robust DC Series Arc Fault Detection in Photovoltaic Systems via Multi-Feature Fusion and Heterogeneous Ensemble Learning

DC arc fault is one of the typical faults in photovoltaic (PV) systems, which can lead to fire accidents in severe case. Therefore, it is necessary to develop an effective arc fault detection technology. Aiming at the arc fault diagnosis with multiple noise disturbances, especially irradiance changes, this paper proposes a robust DC series arc fault (SAF) detection method in PV systems based on multi-feature fusion and heterogeneous ensemble learning. The collected current signals are segmented with sliding windows, and the DC component is removed. Subsequently, time-domain, frequency-domain, simplified Lempel-Ziv complexity (SLZC) and refined composite multiscale fuzzy entropy (RCMFE) features are extracted from the current subsequences. Besides, these segmented subsequences are decomposed by Variational Mode Decomposition (VMD) into several modes with different frequency values, the features of which in time and frequency domains are also obtained. Then Relief is adopted for feature selection to eliminate redundant features. Finally, combining the selected features, heterogeneous ensemble learning model is developed for realizing SAF detection. The experimental results show that the proposed method can not only effectively extract the features of arc fault, but also has high accuracy and robustness with an improvement of the detection accuracy by a range of 0.12% to 4.74%.