Traveling Wave Front Identification In Mine Power Grid Based On LSTM

The stable operation of mine power grid is of great significance to ensure the safety of mine production and the personal safety of mine workers.With the expansion of the scale of the mine power grid,the original fault finding and processing methods of manual line patrol can not meet the production needs of mining enterprises.High-precision and high sampling frequency acquisition equipment is also promoted in the mine power grid.The fault location method based on traveling wave is also gradually applied to the mine power grid,and some application results have been achieved.However,there are many short lines in the mine power grid,and mine equipment is prone to produce complex high-frequency and high-frequency noise,which seriously affects the accuracy of traveling wave front identification.Therefore,the selection of traveling wave front identification method suitable for mine power grid is the key to improve the fault location of mine power grid,and also an important guarantee to improve the stable operation of mine power grid.First of all,this paper builds a cable line model in the mine power grid,and sets the ground fault point.Through triggering faults,the traveling wave signals under the mine power grid are collected,and the noise signals in the traveling wave signals are analyzed.Two methods are used respectively: autocorrelation function test and periodic function decomposition based on loess.The autocorrelation function test is used for qualitative analysis of the regular signals in the noise signals,and the decomposition method based on loess is used for full analysis of a group of signals The decomposition further verifies that the noise signal is not Gaussian white noise,but a periodic noise signal.In view of this regular noise signal,this paper proposes to use long and short-term memory network(LSTM)to extract the law of noise signal,and subtract the noise signal output by LSTM from the original signal to filter out the regular noise.When LSTM is trained to get the law of noise signal,the input signal cannot contain the information of traveling wave front,so this paper proposes a reverse search method based on lilliefos normal distribution to determine the time window of LSTM input signal.Which uses the change of distribution property of noise signal before and after the arrival of traveling wave front to determine the position of time window.Finally,the best super parameters are selected by LSTM to denoise the traveling wave signal,and the ADF stability test is carried out for the denoising waveform.The accurate arrival time of traveling wave front is determined by using the different properties of the stationarity before and after the arrival time of traveling wave front in the denoising waveform.In this paper,the traveling wave measuring device with sampling frequency of 1MHz is used to collect the traveling wave data of single-phase grounding fault in the mine power grid.The connection,setting,data signal upload and acceptance of the traveling wave measuring device are introduced,and the method is verified by the actual field data.At the same time,wavelet transform modulus maximum method,empirical mode decomposition modulus maximum method,difference and its threshold constraint method,high pass filtering method and low pass filtering method are used to calculate the arrival time of traveling wave front for comparison and result analysis.The results show that the method proposed in this paper is superior to the above methods,and the error of wave front identification is low.