A Multi-feature Fusion-Based Method to Recognize Series DC Arc Fault in Photovoltaic System

In allusion to such defects in current photovoltaic (abbr. PV) generation systems as less fault characteristic quantity of series DC arc and the difficulty to identify and locate them, a multi-feature fusion-based method to identify the series DC arc fault in PV generation system was proposed. Firstly, an experimental platform was built to collect current signals under normal operation of PV system and series DC arc fault occurred in PV system and the noise reduction was performed by wavelet transform. Secondly, for the post-noise reduction signals the change rates of the average of current and the characteristic quantity of cycle maximal difference in time-domain were extracted, and the energy and energy ratio of each frequency band in frequency domain were extracted, and by use of ensemble empirical mode decomposition (abbr. EEMD) the intrinsic mode function (abbr. IMF) of each order signal was obtained to calculate the cosine similarity between each order IMF fault signal and normal signal, and the energy entropy feature of IMF of the arc with lower similarity was extracted. Thirdly, the feature space of fault arc was built by the multi-dimensional feature vector that was constituted by the features in time domain, frequency domain and the feature of energy entropy, and the boundary parameters of fault space were determined by experiments, and then the characteristic criteria in time-domain, frequency-domain and that in energy entropy were obtained, and according to multi-dimensional characteristic criteria the DC arc fault detection could be implemented. Finally, both feasibility and accuracy of the proposed method are verified by the analysis on the results of experiments.