Detection of Arc Faults in PV Systems Using Compressed Sensing

As photovoltaic systems grow in size, there has been an increasing desire to automate the detection of arc faults. Any automated system for arc detection must be as fast and accurate as possible: delayed detection of electrical arcs can lead to fire and considerable system damage, while false positives that cause preventative system shutdown are associated with a significant financial cost. In this article, we present a novel approach to detect arcs in dc microgrids via their high-frequency (HF) spectral pattern using ideas from compressed sensing. The acquisition and analysis of HF signals using analog-to-digital converter technology typically requires costly hardware and is not feasible for on-site installation at power plants. However, sparsifying the signal by filtering everything but a narrow HF band enables the use of a modulated wideband converter to sample the arc signature at sub-Nyquist frequencies. We then calculate a characteristic band power within the selected spectrum slice over time and show that it can be used to reliably detect arc events via simple thresholding. We have evaluated our methods on both simulated and experimentally generated arc signals. Finally, we perform statistical analysis of power distributions using linear discriminant analysis in order to identify the frequency range best suited for arc detection.