Abstract: Arc faults have always been a concern for PV systems.The existing techniques that rely on maximal difference in the time domain,or a frequency analysis in the frequency domain,often result in false positives or false negatives because the signal is the only part to be paid attention.The noise caused by the switching frequency and the harmonics of the inverters in photovoltaic system often overlaps the arc signal partially,which makes fault detecting more difficult.This paper presented a method of arc fault analysis and diagnosis using the wavelet packet decomposition to extract the frequency domain feature vector and the statistical method to extract the time domain feature vector.The fault is diagnosed by the position of the system feature plane,which is composed by the feature vector in time and frequency.The characteristics of arc signal was analyzed and the model of arc was established.The method of extracting the feature vector was researched,and the system feature plane was built with the certain key parameters.The results of simulations and experiments show that the proposed method with the low false-positive rate can realize the diagnosis effectively without the interference of the inverter switching frequency and be more applicative than the traditional ones.

Key words: DC arc, fault detection, wavelet theory, characteristic plane