A method for suppressing mixed magnetotelluric noise in shallow water

Abstract

The present invention discloses a method for suppressing mixed magnetotelluric noise in shallow waters, comprising the following steps: reading a noisy marine magnetic field time series; setting phase space reconstruction parameters; constructing a phase space vector to obtain a noisy data matrix; pre-estimating the noise to obtain prior information about the noise; constructing the original noisy data matrix after noise whitening; performing singular value decomposition to obtain the eigenvectors and eigenvalue matrix of the noise-whitened data covariance matrix; selecting low-order principal components after noise-adjusted principal component transformation to reconstruct a denoised data matrix; restoring the denoised data matrix to a time series; performing Fourier transform to obtain amplitude spectra of marine magnetic field components, merging these to obtain a four-component amplitude spectrum of marine magnetic field components; detecting the frequency points of pulses in the four-component amplitude spectrum; setting all amplitudes corresponding to the pulse frequencies in the four-component amplitude spectrum to 0, and performing inverse Fourier transform to obtain a processed four-component time series. The present invention can better process marine magnetotelluric signals and improve data quality.

Description

Technical Field

[0001] The present invention relates to the technical field of electromagnetic signal processing, and in particular to a method for suppressing mixed magnetotelluric noise in shallow water areas. Background Art

[0002] Marine magnetotelluric sounding (MT) is a geophysical exploration technique that uses natural electromagnetic fields to study the electrical structure of the seafloor. The basic theory, data preprocessing, and interpretation of marine MT are similar to those of terrestrial MT. Natural MT signals collected on land are susceptible to human-induced noise, environmental noise, and instrument noise. However, in the ocean, high-frequency electromagnetic waves attenuate rapidly in the seawater layer, making the impact of human-induced noise on marine MT signals less pronounced. Therefore, the noise in marine MT data is primarily due to environmental noise and instrument noise. Environmental noise in marine MT data is primarily related to seawater motion (such as wave-induced magnetic noise); instrument noise is primarily associated with self-contained seafloor instruments and manifests as dense pulse noise in the frequency domain. Both types of noise alias in the time and frequency domains, forming a high-energy, wide-bandwidth mixed noise. This degrades the quality of magnetotelluric data and severely impacts the reliability of subsequent inversion and interpretation results.

[0003] In response to the above problems, CN111856590A proposes a method for suppressing ocean wave magnetic interference in marine magnetotelluric detection. The method uses an adaptive filtering algorithm to perform iterative calculations based on the correlation between the effective signal in the electric field input signal E and the magnetic field reference signal H, adaptively calculates the filtering parameter C, and reconstructs the magnetotelluric signal after suppressing the interference. However, the noise suppression effect of this application depends on the data quality of the reference signal, and only processes the environmental noise of the ocean wave induced magnetic interference, lacking a means to suppress instrument noise. However, the two types of noise are mixed together, and the characteristics they represent are more complex than those of a single noise. The existing noise suppression methods cannot effectively suppress mixed signals with complex characteristics, that is, the denoising effect in suppressing mixed noise is difficult to guarantee. Therefore, it is necessary to develop a noise suppression method for this complex noise. Summary of the Invention

[0004] The present invention provides a hybrid noise suppression method in marine magnetotelluric detection that does not rely on a reference signal, so as to better process marine magnetotelluric signals and improve data quality.

[0005] The present invention solves the technical problem by adopting the following technical solutions:

[0006] A method for suppressing mixed magnetotelluric noise in shallow water areas comprises the following steps:

[0007] Step S1, read the noisy ocean MT time series;

[0008] Step S2, setting phase space reconstruction parameters: time delay and embedding dimension; constructing phase space vectors through different time delays of the time series to obtain a noisy data matrix;

[0009] Step S3, pre-estimating the noise to obtain prior information of the noise;

[0010] Step S4, noise whitening, normalizes the noise level in each sample and removes the correlation of the noise;

[0011] Step S5, constructing the original data matrix after noise whitening;

[0012] Step S6, obtaining the eigenvectors and eigenvalue matrix of the data covariance matrix after noise whitening based on singular value decomposition;

[0013] Step S7, performing noise adjustment principal component transformation;

[0014] Step S8, taking the first few rows of the matrix after noise adjustment principal component transformation to reconstruct the denoised data matrix;

[0015] Step S9, restoring the denoised data matrix to a time series based on the inverse transformation of the phase space reconstruction;

[0016] Step S10: using a bandpass filtering method to retain only the processing results within the frequency range affected by the wave-induced magnetic noise, and ultimately obtaining a magnetic field time series after suppressing the wave-induced magnetic noise;

[0017] Step S11: For frequency-domain dense impulse noise, first, perform a Fourier transform on the ocean MT four-component time series to obtain the ocean MT four-component amplitude spectrum; use a Hampel identifier to detect the pulse frequency points in the four-component amplitude spectrum; merge the four groups of pulse frequency points into one group; set all the amplitudes corresponding to the pulse frequency points in the four-component amplitude spectrum to 0; and perform an inverse Fourier transform to obtain the processed four-component time series.

[0018] Furthermore, in step S2, the noisy data matrix X It is expressed as follows:

[0019] ,

[0020] in, is a time series, is the length of the time series, is the time delay, is the embedding dimension.

[0021] Furthermore, in step S3, the method for pre-estimating the noise to obtain prior information of the noise is:

[0022] The original noisy data matrix is ​​filtered using a bandpass filter. X The noise of each time series segment corresponding to each row in the table is estimated, and a total of A time series segment corresponding to the frequency band affected by the wave-induced magnetic noise ; Finally, the estimated noise matrix is ​​obtained D :

[0023] ,

[0024] in, For the m The first time series segment elements.

[0025] Furthermore, in step S4, the noise whitening method is:

[0026] The noisy data matrix X and the estimated noise matrix D The mean of each row is subtracted from the mean of the corresponding row to obtain the decentralized original data matrix and the estimated noise matrix :

[0027] ,

[0028] ,

[0029] in, is the number of rows, is the number of columns,

[0030] Furthermore, in step S5, the original data matrix after noise whitening is As shown in the following formula:

[0031] ,

[0032] in and The noise covariance matrix is The eigenvector and eigenvalue matrix of ; 、 and satisfy .

[0033] Furthermore, in step S7, the noise adjustment principal component transformation is performed as follows:

[0034] ,

[0035] in, is the eigenvector of the data covariance matrix; the matrix Each row in Sort by signal-to-noise ratio in descending order.

[0036] Furthermore, in step S11, the pulse frequencies are detected and combined as follows:

[0037] Applying the Hample identifier in the frequency domain, the four-component amplitude spectrum , , , The frequency points of the pulses are detected and recorded as , , and ; Set four groups of pulse frequencies , , ,and Combined into a group, recorded as :

[0038] ,

[0039] in, Represents a union operation.

[0040] Furthermore, in step S11, the method for obtaining the processed four-component time series is:

[0041] The four-component amplitude spectrum , , , Medium pulse frequency The corresponding amplitudes are all set to 0:

[0042] ,

[0043] Inverse Fourier transform to obtain the processed four-component time series , , , .

[0044] Beneficial effects of the present invention:

[0045] The present invention mainly addresses the problem of high-energy, wide-bandwidth mixed noise in shallow-water magnetotelluric data, which is composed of seawater motion-induced magnetic noise and frequency-domain dense pulse noise, and causes serious distortion of magnetotelluric signals. A mixed noise suppression method in marine magnetotelluric detection that does not rely on reference signals is proposed. This method provides an effective technical solution for suppressing mixed noise in complex shallow-water environments and improving the quality of marine MT data. Compared with traditional marine magnetotelluric denoising methods, this method performs frequency division processing on noisy data based on the frequency domain distribution characteristics of mixed noise, combines phase space reconstruction in the time domain and noise-adjusted principal component analysis with the frequency domain Hampel identifier to achieve the suppression of mixed noise in shallow-water magnetotelluric data. The proposed method can effectively suppress mixed noise, correct the distortion of apparent resistivity and phase curves caused by noise, and significantly improve the quality of shallow-water MT data. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of the method of the present invention;

[0047] Figure 2 It is a one-dimensional horizontal anisotropic geoelectric model diagram;

[0048] Figure 3 is the time series of pure MT signal, noisy MT data and denoised data;

[0049] Figure 4 The power spectrum density diagrams of pure MT signal, noisy MT data, and denoised data are shown;

[0050] Figure 5 These are the apparent resistivity and phase curves before and after mixed noise suppression. DETAILED DESCRIPTION

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0052] Reference Attachment Figure 1 The present invention provides a method for suppressing mixed magnetotelluric noise in shallow water, comprising the following steps:

[0053] Step S1: Read the noisy ocean MT time series , is the length of the time series;

[0054] Step S2, set phase space reconstruction parameters: time delay and embedding dimensions ; Constructed by different time delays of the time series dimensional phase space vector, and obtain the noisy data matrix X , as shown below:

[0055] .

[0056] Step S3: pre-estimate the noise to obtain the prior information of the noise. Considering the time-varying characteristics of the wave-induced magnetic noise and its energy concentration in a specific frequency range, the original noisy data matrix is ​​filtered by bandpass filtering. X The noise of each time series segment corresponding to each row in the table is estimated, and a total of m A time series segment corresponding to the frequency band affected by the wave-induced magnetic noise ; Finally, the estimated noise matrix is ​​obtained D :

[0057] ,

[0058] in, For the m The first time series segment elements.

[0059] Step S4, noise whitening, normalizes the noise level in each sample and removes the correlation of the noise, using the following method:

[0060] The noisy data matrix X and the estimated noise matrix D The mean of each row is subtracted from the mean of the corresponding row to obtain the decentralized original data matrix and the estimated noise matrix :

[0061] ,

[0062] ,

[0063] in, is the number of rows, is the number of columns.

[0064] Step S5: construct the original data matrix after noise whitening , as shown below:

[0065] ,

[0066] in and The noise covariance matrix is The eigenvector and eigenvalue matrix of ; 、 and satisfy .

[0067] Step S6, based on Get the data covariance matrix after noise whitening The eigenvector of and the eigenvalue matrix ;

[0068] Step S7: Perform noise adjustment principal component transformation in the following manner:

[0069] ,

[0070] in, is the eigenvector of the data covariance matrix; the matrix Each row in Arrange in descending order of signal-to-noise ratio, low-order Represents electromagnetic data with a high signal-to-noise ratio.

[0071] Step S8: Take the front of the matrix after noise adjustment principal component transformation q ( q < m ) rows to reconstruct the denoised data matrix ;

[0072] .

[0073] Step S9, restoring the denoised data matrix to a time series based on the inverse transformation of the phase space reconstruction;

[0074] Step S10: In order to avoid damaging signal components of other frequencies due to the wide processing frequency range, a bandpass filtering method is used to retain only the processing results within the frequency range affected by the wave-induced magnetic noise, and finally obtain the magnetic field time series after suppressing the wave-induced magnetic noise;

[0075] In step S11, the energy of the wave-induced magnetic noise is concentrated in the noise-affected frequency band. Therefore, a more accurate estimated noise can be obtained by bandpass filtering the noise-affected frequency band in step S3, and then the wave-induced magnetic noise can be suppressed based on steps S4 to S10. However, the frequency range affected by the frequency-domain dense pulse noise is large and the noise frequency is discrete, making it difficult to process based on steps S3 to S10. For the frequency-domain dense pulse noise, first, the ocean MT four-component time series is processed. , , , Perform Fourier transform to obtain the ocean MT four-component amplitude spectrum , , , in , and , They represent the magnetic field time series and amplitude spectrum after suppressing the wave-induced magnetic noise. In view of the fact that the frequency-domain dense impulse noise usually shows a typical and obvious high-value anomaly at a specific frequency point on the amplitude spectrum, the Hampel identifier is introduced in the frequency domain. , , , The frequency points of the pulses are detected and recorded as , , and . The four groups of pulse frequencies , , ,and Combined into a group, recorded as :

[0076] ,

[0077] in, Represents the union operation; the four-component amplitude spectrum , , , Medium pulse frequency The corresponding amplitudes are all set to 0:

[0078] ,

[0079] Inverse Fourier transform to obtain the processed four-component time series , , , .

[0080] The present invention mainly addresses the problem of high-energy, wide-bandwidth mixed noise in shallow-water magnetotelluric data, which is composed of seawater motion-induced magnetic noise and frequency-domain dense pulse noise, and causes serious distortion of magnetotelluric signals. A mixed noise suppression method in marine magnetotelluric detection that does not rely on reference signals is proposed. This method provides an effective technical solution for suppressing mixed noise in complex shallow-water environments and improving the quality of marine MT data. Compared with traditional marine magnetotelluric denoising methods, this method performs frequency division processing on noisy data based on the frequency domain distribution characteristics of mixed noise, combines phase space reconstruction in the time domain and noise-adjusted principal component analysis with the frequency domain Hampel identifier to achieve the suppression of mixed noise in shallow-water magnetotelluric data. The proposed method can effectively suppress mixed noise, correct the distortion of apparent resistivity and phase curves caused by noise, and significantly improve the quality of shallow-water MT data.

[0081] refer to Figure 2 , is a one-dimensional horizontal anisotropic geoelectric model diagram. In order to verify the feasibility of suppressing the mixed noise in shallow water magnetotelluric data proposed in this invention, Figure 2 The electromagnetic field simulated by the horizontal anisotropic geoelectric model is used as the effective electromagnetic signal. The mixed noise is constructed by synthesizing the wave-induced magnetic field and the synthetic frequency-domain dense pulse noise and mixed into the effective signal to achieve the aliasing of the effective signal and the mixed noise. The three-dimensional random wave-induced magnetic field formula is used to simulate the wave-induced magnetic noise. The wave spectrum is set to PM-SWOP spectrum, the geomagnetic field intensity is 52000nT, the seawater layer thickness is 21m, the magnetic declination is 54°, the magnetic inclination is -8°, and the seawater resistivity is The simulation results are X and Y components of the wave-induced horizontal magnetic field time series, which match the duration of the MT signal. The X and Y components of the frequency-domain dense impulse noise time series are obtained by superimposing multiple cosine functions of different frequencies with random amplitudes and phases. The noisy MT magnetic field component is obtained by adding the simulated MT magnetic field signal, the wave-induced magnetic noise, and the frequency-domain impulse noise.

[0082] Figure 3 The time series of pure MT signal, noisy MT data and denoised data are shown. Figure 3It can be seen that after adding mixed noise to the pure MT signal, the obtained noisy MT data and the pure MT signal have significant deviations at all times, affecting the MT signal quality at all times. Using the mixed noise suppression method proposed in this invention to process the synthesized noisy MT data, the resulting denoised data is very close to the pure MT signal. After processing, the signal-to-noise ratio of the Hx component increased from -1.3196dB to 13.2021dB, the normalized root mean square error decreased from 18.7888% to 3.5303%, and the correlation coefficient between the denoised signal and the noise-free signal increased from 0.6562 to 0.9762; the signal-to-noise ratio of the Hy component increased from -0.4687dB to 14.7691dB, the normalized root mean square error decreased from 20.7953% to 3.5981%, and the correlation coefficient between the denoised signal and the noise-free signal increased from 0.6876 to 0.9821; the signal quality is significantly improved.

[0083] Figure 4 The power spectrum density plots of the pure MT signal, noisy MT data and denoised data are shown. Figure 4 As can be seen, after adding mixed noise to the pure MT signal, the power spectral density plot of the noisy MT data shows obvious strong energy abnormal bumps and dense pulses. Using the mixed noise suppression method proposed in this paper to process the synthesized noisy MT data, the resulting denoised data closely matches the power spectral density of the pure MT signal, and the strong energy abnormal bumps and dense pulses in the power spectral density plot are almost completely eliminated.

[0084] Robust impedance estimation is performed on the synthetic MT data before and after suppressing mixed noise, and the apparent resistivity and phase are calculated as follows: Figure 5 As shown. Figure 5 It can be seen that for the synthetic MT data after adding mixed noise, its apparent resistivity and phase curves are distorted in the range of >0.1Hz, significantly deviating from the true values. The apparent resistivity and phase curves obtained by processing the mixed noise suppression method proposed in this invention are smoother and more continuous, and are numerically closer to the true values. These processing results demonstrate that the algorithm proposed in this invention can effectively suppress mixed noise in shallow water magnetotelluric data and obtain more reliable apparent resistivity and phase curves by estimating impedance. This also shows that the algorithm proposed in this invention is effective.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for suppressing mixed magnetotelluric noise in shallow water, characterized in that: The steps include: Step S1, read the noisy ocean MT time series; Step S2, setting phase space reconstruction parameters: time delay and embedding dimension; constructing phase space vectors through different time delays of the time series to obtain a noisy data matrix; Step S3: pre-estimate the noise to obtain prior information of the noise. The method is: The original noisy data matrix is ​​filtered using a bandpass filter. X The noise of each time series segment corresponding to each row in the table is estimated, and a total of A time series segment corresponding to the frequency band affected by the wave-induced magnetic noise ; Finally, the estimated noise matrix is ​​obtained : ， in, For the The first time series segment elements; Step S4, noise whitening, normalizes the noise level in each sample and removes the correlation of the noise; the decentralization method is: The noisy data matrix X and the estimated noise matrix The mean of each row is subtracted from the mean of the corresponding row to obtain the decentralized original data matrix and the estimated noise matrix : ， in, is the number of rows, is the number of columns; Step S5, constructing the original noisy data matrix after noise whitening; Step S6, obtaining the eigenvectors and eigenvalue matrix of the data covariance matrix after noise whitening based on singular value decomposition; Step S7, performing noise adjustment principal component transformation; Step S8, taking the first few rows of the matrix after noise adjustment principal component transformation to reconstruct the denoised data matrix; Step S9, restoring the denoised data matrix to a time series based on the inverse transformation of the phase space reconstruction; Step S10: using a bandpass filtering method to retain only the processing results within the frequency range affected by the wave-induced magnetic noise, and ultimately obtaining a magnetic field time series after suppressing the wave-induced magnetic noise; Step S11: For frequency-domain dense impulse noise, first, perform a Fourier transform on the ocean MT four-component time series to obtain the ocean MT four-component amplitude spectrum; detect the pulse frequency points in the four-component amplitude spectrum using a Hampel identifier; merge the four groups of pulse frequencies into one group; set all the amplitudes corresponding to the pulse frequencies in the four-component amplitude spectrum to 0; and perform an inverse Fourier transform to obtain the processed four-component time series. The method of detecting and merging pulse frequencies is as follows: Applying the Hample identifier in the frequency domain, the four-component amplitude spectrum , , , The frequency points of the pulses are detected and recorded as , , and ; Set four groups of pulse frequencies , , and Combined into a group, recorded as : ， in, Represents a union operation; The way to obtain the processed four-component time series is: The four-component amplitude spectrum , , , Medium pulse frequency The corresponding amplitudes are all set to 0: ， Inverse Fourier transform to obtain the processed four-component time series , , , .

2. The shallow water area magnetotelluric hybrid noise suppression method according to claim 1 is characterized in that: In step S2, the noisy data matrix X It is expressed as follows: ， in, is a time series, is the length of the time series, is the time delay, is the embedding dimension.

3. The shallow water area magnetotelluric hybrid noise suppression method according to claim 2, characterized in that: In step S5, the original data matrix after noise whitening is As shown in the following formula: ， in and The noise covariance matrix is The eigenvector and eigenvalue matrix of ; 、 and satisfy .

4. The shallow water area magnetotelluric hybrid noise suppression method according to claim 3 is characterized in that: In step S7, the noise adjustment principal component transformation is performed as follows: ， in, is the eigenvector of the data covariance matrix; the matrix Z Each row in Sort by signal-to-noise ratio in descending order.

Images