In the low-voltage distribution network, the tripping criterion of the residual current devices is usually a fixed threshold. The incorrect detection of leakage current signal may lead to tripping delay or mis-trip of the devices in electric shock accidents, which may be caused by the equipment vibration, harmonics, transient disturbances, etc. Hence, this study proposes an artificial intelligence method for quickly identifying electric shock faults to improve the function of the device. Firstly, wavelet entropy removes noise from the electric shock signal. Then, the autoencoder is applied to extract the total leakage current waveform as the feature information. Finally, the back-propagation neural network classifies the electric shock types. The simulation and experiment platforms are established to obtain experimental samples and validate the reliability of the proposed method. Compared to other alternative methods, the proposed method shows outstanding identification ability, which demonstrates its applicability in reality.

Autoencoder, back-propagation, electric shock faults identification, low-voltage distribution network,wavelet entropy denoising