Abstract
Arc fault detection in photovoltaic systems is crucial, since it may cause incidents like fires and explosions. So far, most existing methods rely on an arc's local features and do not characterize arc faults globally, which may lead to detection failure in noisy environments. In this paper, a fundamentally different method is proposed that relies on an arc's global features instead of local ones. The core idea of the method is that the physical mechanisms of the arc fault signals and the normal signals are so different that they are thought to be generated by two independent sources. Based on this insight, independent component analysis (ICA) is introduced to decompose the photovoltaic system's DC currents. By using ICA, the DC current signals with an arc fault can be decomposed into two independent signals, while the normal signals without an arc fault cannot be decomposed into two such independent signals. This indicates that arc faults can be detected by using the concept of "independence". Then, the dynamic time warping algorithm was used to determine the independence level of the ICA outputs so as to realize end-to-end arc fault detection. Experimental results showed that our method has better performance than traditional methods in terms of detection accuracy and robustness against environmental disturbances.