Research On Magnetotelluric Signal-noise Identification And Separation Based On Variational Mode Decomposition And Fractal

With the rapid development of the national economy and heavy industry production,a variety of human-induced electromagnetic interferences have become increasingly serious,and the development of magnetotelluric(MT)sounding in ore-concentration areas is facing severe challenges.For this reason,how to precisely extract weak MT signals from strong interference has become a research hotspot.In this paper,the variational mode decomposition and fractal are applied to the signal-noise identification and separation of the MT field.The main research are as follows:(1)The basic principles of VMD and matching pursuit(MP)algorithm are studied.The processing effects of VMD and empirical mode decomposition(EMD)and inherent time-scale decomposition(ITD)are compared and analyzed.We discussed the selection of modal number K and the denoisiong performance of the strong interference,which by observing center frequency,similarity(NCC)and signal-to-noise ratio(SNR).The matching pursuit is used to secondary signal-noise separation on VMD reconstructed signal.(2)The basic principle of detrended fluctuation analysis(DFA)algorithm is studied.By computing the scale exponent of the original signal,the value of the modal number K is obtained adaptively,and the scale exponent of each modal is estimated.We selected the scaleexponent of modal greater than or equal to 0.75 to reconstruct the useful MT signal.(3)Four kinds of characteristic parameters of fractal box dimension,Higuchi fractal dimension,fuzzy entropy and approximate entropy of MT signal are analyzed,quantitatively distinguished the useful MT signals and strong interference.The multiple characteristic parameters and fuzzy C-means clustering are combined to perform signal-noise identification,and the identified as a strong interference signal is eliminated pertinently by using wavelet soft threshold denoising.(4)The basic principle of multifractal spectrum is studied.The non-uniformity and irregularity characteristic parameters of MT data are analyzed.Then,two characteristic parameters are input into support vector machine(SVM)training,and a signal-noise classification model is generated.We only use MP technique to suppress noise which identified as signal with interference.The results of simulation experiment,EMTF numerical simulation,Qinghai test site and measured data processing show that the proposed method deliberately eliminates large-scale strong interference in the MT time domain sequence.Compared with the remote reference method,the VMD-based overall processing method,the wavelet soft threshold overall processing method and matching pursuit overall processing method.The obtained apparent resistivity-phase curve of measured sites is smootherand more stable,which greatly improves the quality of MT data in low frequency band.