Single channel electric shock signals blind source separation algorithm based on local mean decomposition

Abstract: Residual current protection device (RCD) is a kind of protection device in low voltage system and has been widely applied in preventing of grid leakage and protecting of peoples life and property safety. At present, the action threshold of residual current protection device is 30 mA, but there is no direct relation between the action setting value and the electric shock current passing through the electric shock body. When heavy load is put into operation or weather changes or electric shock occurs, residual current protection device often occurs misoperation or rejection. Therefore, extracting electric shock current from the residual current and setting a new protection action criterion are of great significance for improving the operational reliability of residual current protection device. Because electric shock accident is unpredictable, it is difficult to extract the electric shock current from the residual current of low voltage power network exactly. A method of extracting the electric shock current by combining the local mean decomposition (LMD) with blind signal separation is proposed. In general, the observed signals used for blind source separation are multi-channel signals. When electric shock occurs, the residual current signals contain the electric shock current signals, normal leakage current signals and noise signals. The residual current signal is a single channel signal, and at least one virtual channel needs to be constructed. Therefore, using the local mean decomposition method, the residual current signal is adaptively decomposed into the sum of several product functions(PF), and each product function is equal to the product of an amplitude modulated signal and a frequency modulated signal. Computing the similarity coefficient between each component and the original signal, the modal components with the largest similarity coefficient and greater than 0.8 are used as virtual channels for blind source separation. Two channels of blind source separation are constructed by combining virtual channels with residual current signals, and the problem of single channel blind source signal separation was solved. Then FastICA algorithm was used to extracte the electric shock currents from the residual current signals. The results shown that LMD method has less decomposition component, shorter calculation time compared with empirical mode decomposition (EMD) method, and can avoid the disadvantage of endpoint effect in the decomposing process of EMD. When electric shock accident occur in single-phase circuit, the average correlation coefficients between the original electric shock current and the electric shock current extracted by LMD-FastICA and EMD-FastICA are 0.937 4 and 0.925 3 respectively, the average relative errors are 0.096 2 and 0.109 8 respectively. The relative error ranges of peak factor of decomposition signal by EMD-FastICA and LMD-FastICA is from 0.012 to 0.155 and from 0.001 to 0.103 respectively. When electric shock accident occur in three-phase circuit, the average correlation coefficients between the original electric shock current and the electric shock current extracted by LMD-FastICA and EMD-FastICA are 0.962 4 and 0.948 9 respectively, and the average relative errors are 0.056 4 and 0.081 55 respectively. The calculation time of LMD-FastICA(0.032 s) is shorter than that of EMD-FastICA(0.129 s). The research results lay a theoretical foundation for development of new residual current protection device based on the action of electric shock current.