Series DC Arc Fault Detection in Photovoltaic System Based on Multi-feature Fusion and SVM

The photovoltaic (PV) system working environment, which contains a large number of interferences, is easy to produce series direct current (DC) arc. Aiming at the problems of poor anti-interference ability of time domain detection and difficulty in feature extraction, an arc fault detection method based on Hilbert spectrum analysis (HSA), singular value decomposition (SVD) and support vector machine (SVM) is proposed. By high-pass filtering and ensemble empirical mode decomposition (EEMD), the influence of different loads and noises on feature extraction is overcome. The intrinsic mode functions (IMFs) are used to perform HSA and SVD to obtain the maximum value of the Hilbert marginal spectrum and singular values, which add the maximum current change difference as the multi-feature fusion. The particle swarm optimization (PSO) is used to optimize the penalty coefficient c and the Gaussian kernel parameter g of SVM based on the constructed feature vector. It has been verified by experiments that it can achieve a better series DC arc fault detection effect.