Photovoltaic DC Series Arc Fault Identification Method Based on Precise Noise Reduction Algorithm

Due to the influence of the external environment and the internal noise of the inverter, the noise harmonic injection may not be obvious in the initial stage of photovoltaic DC arc generation, resulting in the arc fault with strong concealment and difficult to detect. To solve this problem, an arc fault detection method based on improved empirical wavelet transform (IEWT) and improved singular value decomposition (ISVD) is proposed in this paper. Firstly, the improved empirical wavelet transform is used to realize the accurate segmentation of the noise spectrum, and then the construction dimension of the corresponding matrix of each component is reduced by improving the singular value decomposition, which not only achieves accurate noise reduction but also effectively reduces the amount of calculation of the algorithm. It makes the normal signal and the arc fault signal show obvious differentiation after noise reduction. In order to realize accurate detection of hidden arc, the appropriate transition interval is set according to each characteristic index, and the suspected arc fault signal is judged by the ratio of high and low frequency characteristic indexes. Finally, the improved entropy weight method (IEWM) is used to construct the fusion characteristic index to realize a fast and accurate diagnosis of arc fault. Through a series of comparative experiments, it can be proved that, compared with the commonly used AI method, the detection algorithm in this paper achieves 99.47% detection accuracy and improves the average calculation speed by 56.7%.