基于注意力机制的光伏系统故障电弧检测与定位研究

The series arc fault caused by insulation damage and other reasons may threaten the stable operation of photovoltaic systems seriously. Meanwhile, the ability of arc fault detection may be affected by the impedance network in photovoltaic systems, and the reliability of time-frequency detection methods is reduced. To address the challenge of detecting and locating arc fault caused by impedance networks, this paper establishes a DC arc fault experimental platform with photovoltaic impedance network module. The arc fault experiments with different current levels, loads, and simulated lengths of lines are conducted. The current signal is analyzed through fast Fourier transform (FFT) spectrum and wavelet transform. The amplitude ratio and lift ratio indicators are constructed to evaluate the difference of arc fault feature before and after the photovoltaic impedance network. The weakening effect of photovoltaic impedance network on arc fault feature is analyzed. The three-point symmetric differential energy operator (DEO3S) is applied to the wavelet reconstruction signal to enhance the feature of each frequency band within 100 kHz that the arc fault detection is effectively improved. The arc fault detection and location algorithm based on Long Short-term Memory network with Attention mechanism is proposed according to the law of attenuation of feature with the line grows. The location of series arc fault within 0~80 m is achieved with a maximum error of no more than 4m. It may provide crucial theoretical and technical references for developing arc fault detection modules in photovoltaic systems.