Data noise reduction method and device for multi-level magnetotelluric data and medium

Abstract

The embodiment of the invention discloses a data noise reduction method and device for multi-level magnetotelluric data and a medium, and relates to the technical field of noise reduction processing, and the method comprises the steps: collecting an original magnetotelluric data sequence, determining first noise reduction data, carrying out the variational mode decomposition of the first noise reduction data, and determining second noise reduction data; determining a first signal-to-noise ratio corresponding to the second noise reduction data, and when the first signal-to-noise ratio is smaller than a preset signal-to-noise ratio reference value, performing dynamic fusion on the residual signal to determine a fused residual signal; performing noise reduction processing on the fused residual signal, determining fused residual noise reduction data, performing residual recovery based on the fused residual noise reduction data, and determining current noise reduction data after residual recovery; when the second signal-to-noise ratio corresponding to the current noise reduction data is not smaller than the signal-to-noise ratio reference value, the current noise reduction data is output, the signal recovery requirement of weak effective information in the magnetotelluric data can be met, the signal fidelity is enhanced, and the deep noise processing and weak signal protection capabilities are improved.

Description

Technical Field

[0001] This specification relates to the technical field of noise reduction processing, and particularly to a data noise reduction method, device and medium for multi-level magnetotelluric data. Background Art

[0002] As an important means for deep geological structure exploration, the natural electromagnetic signals collected by magnetotelluric exploration have the characteristics of wide frequency band, weak energy, and being non-linear and non-stationary. Frequency domain noise reduction technology relies on the spectral separation characteristics of signals and noise. However, under low signal-to-noise ratio conditions, the effective spectrum is often completely submerged by noise, resulting in distortion of a large number of frequency points and high phase errors in the measured apparent resistivity-phase curve. Although time domain noise reduction methods can avoid the problem of spectral aliasing, they need to preset a noise empirical model (such as the fixed power frequency period assumption), have poor adaptability to non-periodic pulse interference (square wave, triangular wave mutation noise), and uniformly filtering abnormal waveforms will cause loss of effective signals, seriously weakening the deep geological response.

[0003] To break through the traditional limitations, the prior art (Patent CN 119046617A) proposes a combined noise reduction method for industrial pumps based on SVD and GWO-VMD. It initially reduces noise through multiresolution singular value decomposition (MSVD), combines the grey wolf algorithm (GWO) to optimize the parameters of variational mode decomposition (VMD), and finally selects modal components based on kurtosis-correlation coefficient. However, the above prior art still has obvious defects in the noise reduction effect for multi-level magnetotelluric (MT) data. There are significant differences between industrial pump signals and magnetotelluric signals. Industrial pump signals are periodic noises, while MT data are pulse and harmonic noises. It is necessary to protect weak signals and cope with ultra-wideband dynamic noises of 0.01 - 1000 Hz. The residual signal after noise reduction may contain effective signals. Especially for weak MT signals, directly discarding the residual signal may lead to loss of effective signals. And in the existing noise reduction process of multi-level magnetotelluric data, the noise reduction method is fixed, ignoring the use of MT signals and not considering the different impacts on MT signals under different geological analysis requirements.

[0004] Therefore, in the process of noise reduction for MT signals, the signal characteristics and geological analysis requirements of MT signals are ignored. Discarding the residual signal after noise reduction leads to loss of effective signals, and the noise reduction effect cannot meet the geological analysis requirements. Summary of the Invention

[0005] One or more embodiments of this specification provide a method, device, and medium for denoising multi-level magnetotelluric data to solve the following technical problems: In the process of denoising MT signals, the signal characteristics and geological analysis requirements of MT signals are ignored. During denoising, the residual signal is discarded, resulting in the loss of effective signals, and the denoising effect cannot meet the geological analysis requirements.

[0006] One or more embodiments of this specification adopt the following technical solutions: One or more embodiments of this specification provide a method for denoising multi-level magnetotelluric data. The method includes: collecting an original magnetotelluric data sequence, performing resolution singular value decomposition on the original magnetotelluric data sequence to determine first denoised data, performing variational mode decomposition on the first denoised data through a pre-determined set of optimization parameters to determine second denoised data; determining a first signal-to-noise ratio corresponding to the second denoised data, and when the first signal-to-noise ratio is less than a preset signal-to-noise ratio reference value, dynamically fusing the residual signal based on the first denoised data and the second denoised data to determine a fused residual signal; performing denoising processing on the fused residual signal to determine corresponding fused residual denoised data, and based on the fused residual denoised data, performing residual recovery to determine the current denoised data after residual recovery; when a second signal-to-noise ratio corresponding to the current denoised data is not less than the signal-to-noise ratio reference value, outputting the current denoised data.

[0007] One or more embodiments of this specification provide a device for denoising multi-level magnetotelluric data, including: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above method.

[0008] A non-volatile computer storage medium provided by one or more embodiments of this specification stores computer-executable instructions, and the computer-executable instructions are set to: execute the above method.

[0009] The above at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects: Through the technical solution of the embodiments of this specification, the original magnetotelluric data sequence corresponding to the target geological analysis area in the target geological analysis requirement information is collected, ensuring the regional matching between the signal source and the analysis requirement; The MSVD algorithm is introduced to perform multi-level decomposition on the non-stationary signal MT signal for preliminary noise reduction. Compared with the single-level linear decomposition SVD, it can perform multi-scale and high-resolution processing on the low-frequency noise in the MT signal, effectively retaining the target signal while suppressing the noise; Aiming at problems such as mode mixing generated during the decomposition of residual pulse signals, the Thomson function is introduced on the basis of the VMD algorithm to perform adaptive soft threshold attenuation on the components exceeding the threshold, avoiding hard truncation distortion, and being more effective in suppressing the synthesis of pulse signals than the traditional VMD algorithm; The high convergence characteristic of the GWO algorithm is used to optimize the key parameters in the VMD. During the optimization process, the eigenvector library of electromagnetic interference in the target geological analysis area is introduced as noise prior information. By constructing a composite fitness function that combines envelope entropy, decomposition results, and noise feature matching degree, target-guided adjustment of the VMD parameters is realized, ensuring the matching between the noise prior knowledge and the target geological analysis area, not only greatly reducing the possibility of mode mixing, but also improving the calculation efficiency by optimizing the number of modes and significantly shortening the calculation time; The dynamic residual feedback method and weighted fusion mechanism are used to further process the effective signals in the residuals, which can meet the signal recovery requirements of weak effective information in magnetotelluric data, enhance the signal fidelity, and improve the ability of deep noise processing and weak signal protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings described below are only some embodiments recorded in this specification. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings: Figure 1 It is a schematic flowchart of a data noise reduction method for multi-level magnetotelluric data provided by the embodiments of this specification; Figure 2 It is a schematic diagram of evaluation parameters after processing noise signals by different methods provided by the embodiments of this specification; Figure 3 It is a schematic diagram of the denoising effect of a measured magnetotelluric data provided by the embodiments of this specification; Figure 4 It is a schematic diagram of the structure of a data noise reduction device for multi-level magnetotelluric data provided by the embodiments of this specification. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the scope of protection of this specification.

[0012] The embodiments of this specification provide a method for data denoising of multi-level magnetotelluric data. It should be noted that the execution subject in the embodiments of this specification can be a server or any device with data processing capabilities. Figure 1 The following is a schematic flowchart of a method for data denoising of multi-level magnetotelluric data provided by the embodiments of this specification, as Figure 1 shown, mainly including the following steps: Step S101, determine the target geological analysis requirement information, collect the original magnetotelluric data sequence corresponding to the target geological analysis area in the target geological analysis requirement information, perform resolution singular value decomposition on the original magnetotelluric data sequence to determine the first denoised data, and perform variational mode decomposition on the first denoised data through a pre-determined optimization parameter set to determine the second denoised data.

[0013] Among them, the optimization parameter set is determined based on an electromagnetic interference feature library corresponding to the pre-constructed target geological analysis area.

[0014] In an embodiment of this specification, to determine the target geological analysis requirement information, the geological analysis type and the target geological analysis area are included in the target geological analysis requirement information. Based on the target geological analysis area, the corresponding original magnetotelluric data sequence is collected. The original MT data sequence can be obtained through a v5-2000 type MT sounding instrument and preprocessed. It should be noted that the magnetotelluric data, multi-level magnetotelluric data, and MT data in the embodiments of this specification all represent the same type of data. The original MT data sequence here is the target denoised data that needs to be denoised. According to the characteristics of MT data, in order to improve the denoising effect, the embodiments of this specification preprocess the original MT data sequence after obtaining the original MT data sequence. In an embodiment of this specification, the preprocessing includes mean removal and segmentation. The purpose of mean removal is to eliminate the DC component of the signal, make the data evenly distributed around zero, and facilitate further processing. Segmentation can reduce the influence of noise and improve the stability of subsequent spectrum analysis and impedance tensor estimation. Both play an important role in improving the signal quality, reducing the influence of outliers, and improving the effect of subsequent denoising methods.

[0015] In one embodiment of the present specification, the data after preprocessing is subjected to MSVD decomposition, and the detailed components D of each layer are k superimposed to obtain MSVD reconstructed data. The MSVD reconstruction steps are as follows during this process: Step (a): Assume there is a 1-D data X(t) = , where N is the length of the 1-D data. In the embodiment of the present specification, N represents the total number of sampling points of the original magnetotelluric data sequence, that is, the length of the original magnetotelluric data sequence. Construct the multi-dimensional Hankel matrix of X(t): (1) Step (b): Perform singular value decomposition (SVD) processing on to obtain the singular value matrix .

[0016] Step (c): Select the main singular values according to the magnitudes of the singular values, perform reconstruction on the matrix using the inverse SVD operation, and restore the reconstructed matrix to the approximate component and the fine component .

[0017] Step (d): Return to Step (a) to construct the matrix using the approximate component, and stop the loop when the decomposition level reaches the optimum.

[0018] According to the above steps, perform MSVD decomposition on the preprocessed MT data sequence that needs noise reduction processing, and superimpose the detailed components of each layer to obtain the MSVD reconstructed data, that is, the first noise-reduced data.

[0019] Perform variational mode decomposition on the first noise-reduced data through a pre-determined set of optimization parameters to determine the second noise-reduced data, specifically including: optimizing the parameters of the variational mode decomposition through the grey wolf optimization algorithm and the electromagnetic interference feature library corresponding to the pre-constructed target geological analysis area to determine the set of optimization parameters, where the set of optimization parameters includes the number of mode decompositions and the quadratic penalty factor; performing variational mode decomposition on the first noise-reduced data using the set of optimization parameters to obtain the number of mode components equal to the number of mode decompositions; calculating the instantaneous frequency threshold for each mode component, and determining the second noise-reduced data through the Thomson function and the instantaneous frequency threshold.

[0020] In one embodiment of this specification, the Grey Wolf Optimizer (GWO) algorithm is used to jointly optimize the number of modal decompositions k and the quadratic penalty factor α of the VMD algorithm, and a feature vector library of electromagnetic interference (including power frequency harmonics, square waves, triangular waves, etc.) in this target geological analysis area is added as noise prior information during the optimization process. Using the selected k and α, the first denoised data after MSVD denoising is decomposed by VMD into the optimal number of modal components, the instantaneous frequency threshold of each modal component is calculated, and the Thomson function is used to perform weighted processing on the part of each modal component that exceeds the threshold, and the modal components that do not exceed the threshold are retained. The optimized modal components are added together to reconstruct the main sequence of the MT data, that is, the second denoised data is determined.

[0021] Through the grey wolf optimization algorithm and the pre-constructed electromagnetic interference feature library corresponding to this target geological analysis area, the parameters of this variational mode decomposition are optimized to determine the optimized parameter set, specifically including: collecting magnetotelluric noise samples of various noise types in this target geological analysis area to extract the electromagnetic interference feature vectors corresponding to each noise type, and constructing an electromagnetic interference feature library based on the electromagnetic interference feature vectors corresponding to each noise type; determining the noise matching degree evaluation index corresponding to this decomposition modal component according to the decomposition modal component after this resolution singular value decomposition and multiple electromagnetic interference feature vectors; using this noise matching degree evaluation index and the preset envelope entropy function to determine the fitness function, and using this grey wolf optimization algorithm to optimize the parameters of this variational mode decomposition to determine the optimized number of modal decompositions and the quadratic penalty factor.

[0022] In one embodiment of this specification, the steps of the VMD algorithm are as follows: First, the overall constrained variational model of VMD is expressed as follows: (2) Among them, is the signal to be analyzed, is the modal function, is the center frequency of each order of mode, The unit impulse function is used for Hilbert transform, represents the partial derivative with respect to time. Introduce the quadratic penalty factor and the Lagrange multiplier , and convert the constrained variational problem into an unconstrained variational problem. The model is expressed as follows: (3) In the formula, is used to ensure the reconstruction accuracy of the signal, Make the constraint conditions more stringent. Finally, VMD adaptively decomposes the signal to be analyzed into narrowband modal components . During the research on VMD decomposition, it is found that the accuracy of VMD decomposition is mainly affected by the number of modal decompositions k and the quadratic penalty factor α. These two parameters must be determined in advance in practical applications. Therefore, the embodiments of this specification adopt the GWO algorithm to optimize the key parameters of VMD. And in the optimization process, the eigenvector library of typical electromagnetic interference (including power frequency harmonics, square waves, triangular waves, etc.) is introduced as noise prior information. In the fitness function based on envelope entropy, the matching degree evaluation index of the current decomposition result and the predefined noise eigenvector is superimposed to realize the target-guided adjustment of VMD parameters.

[0023] The GWO algorithm is used to optimize the key parameters of VMD. The mathematical models of the distance D between the gray wolf individual and the prey and the position update X(t + 1) of the gray wolf are as follows:

[0024]

[0025]

[0026]

[0027]

[0028] Among them, t is the current iteration number, , X are the position vectors of the prey and the gray wolf respectively; A and C are coefficients; is the convergence factor, linearly decreasing from 2 to 0 as the iteration progresses; r1 and r2 are random vectors generated in [0, 1], and M is the maximum number of iterations. The wolf pack chases the prey under the leadership of the alpha wolf, and its hunting model is as follows:

[0029]

[0030]

[0031] In the formula, α, β, δ respectively represent three alpha wolves, and ω is regarded as the eliminated wolf, , , are the direction vectors between the prey and the three alpha wolves respectively. Z represents the distance between the prey and the wolf pack, , , are the distances between the alpha wolves α, β, δ and the prey. , , respectively represent the positions of three leading wolves. Vectors A and C are coefficient vectors. Z(t + 1) represents the distance between the prey and the wolf pack after t iterations.

[0032] Magnetotelluric signals are often mixed with various noises such as power frequency harmonics, transient pulses, and square wave oscillations. Especially for different geological analysis regions, there may be multiple types of noises in their environments. The traditional noise reduction method of fixed threshold filtering cannot adapt to the noise characteristics of magnetotelluric signals. The time-frequency characteristics of different noises are significantly different. For example, power frequency noise is narrowband concentrated, and pulse noise is broadband burst. A single filtering model cannot adaptively distinguish and process them; directly discarding specific singular value components may accidentally damage weak useful signals, such as the low-frequency components of geological structure responses. By collecting typical noise samples (such as substation power frequency interference, lightning pulses, and switching power supply oscillations) in the target geological analysis region, extracting their electromagnetic interference feature vectors to form a prior knowledge base, that is, the regional electromagnetic interference feature library corresponding to the target geological analysis region. Comparing each modal component after SVD decomposition with the feature library and calculating the noise matching degree index can accurately identify the noise attributes of the components, avoiding the simple discarding or retaining of components by traditional methods, and instead realizing the intelligent weighted fusion of noise type perception in the analysis region.

[0033] Specifically, according to the geographical analysis information of the target geological analysis region, collection points are arranged in strong interference regions within the region. For example, high-precision MT collection stations are arranged in strong interference regions such as around substations and industrial areas to continuously collect pure noise segments. For example, the number is greater than 10,000 groups, covering core noise types such as power frequency harmonics corresponding to 50 / 60 Hz and its multiples, square wave pulses generated by the action of switching equipment, and triangular wave pulses corresponding to inverter disturbances. Each sample needs to be marked with the noise type, intensity level (dBμV / m), and time segment mark. The anti-aliasing sampling technology is used with a sampling rate ≥1 kHz to ensure complete capture of pulse details, and the noise events are intercepted and separated through a sliding time window. Here, the window length of the sliding time window can be 0.5 s, and the overlap rate is 30%.

[0034] Extract the features of the collected magnetotelluric noise samples to obtain the electromagnetic interference feature vectors corresponding to each noise type. In one example, perform an FFT transform (4096 points) on each noise segment, calculate the proportion of the energy in the 0-100 Hz frequency band, calculate the total energy in the 0-100 Hz frequency band, and calculate the low-frequency energy in the 0-50 Hz frequency band. The proportion of the frequency band energy is the ratio of the total energy to the low-frequency energy, and the proportion of the frequency band energy is used to quantify the degree of high-frequency noise pollution. For example, the proportion of the power frequency harmonic frequency band energy is approximately 1.2, and the proportion of the impulse noise frequency band energy is greater than 2.5. In addition, perform a kurtosis analysis on the noise samples in the time domain to extract the kurtosis features. Generally, the square wave pulse has a sharp and steep change, and the kurtosis is greater than 15. The triangular wave pulse has a gentle transition, and the kurtosis is approximately 3. In addition, perform a zero-crossing rate detection on each noise sample to determine the number of times the signal crosses the zero level per unit time. According to the above three features, extract the features of multiple samples of each type of noise to construct a feature triple corresponding to each noise type, and determine the electromagnetic interference feature vector. It should be noted that the median of each feature can be taken as the template value, and in the above manner, an electromagnetic interference feature library is constructed. According to the decomposed mode components after the resolution singular value decomposition and the multiple electromagnetic interference feature vectors in the electromagnetic interference feature library, determine the noise matching degree evaluation index corresponding to the decomposed mode components. In this process, perform the same processing as the construction of the feature library for each decomposed mode component, segment with the same window length, and synchronously calculate the triple features. Calculate the Euclidean distance between the decomposed mode component and each electromagnetic interference feature vector in the electromagnetic interference feature library, and take the minimum Euclidean distance as the noise matching degree evaluation index corresponding to this decomposed mode component.

[0035] Through the above technical solution, traditional methods rely on manually setting thresholds or fixed-band filtering, and it is difficult to deal with variable-frequency harmonics and composite noises. Through the feature vector matching of the noises in the target geological analysis area, the automatic identification and separation of noise types are realized, effectively improving the detection rate of power frequency noises; the feature library supports online update. For example, if new high-speed rail interference samples are added, the noise type recognition ability can be dynamically expanded.

[0036] With this noise matching degree evaluation index and the preset envelope entropy function, determine the fitness function, specifically including: generate a dynamic weight coefficient according to the proportional relationship between the current envelope entropy value and the preset maximum reference entropy value, and adjust the noise matching degree evaluation index based on this dynamic weight coefficient to determine the environmental noise reference item; based on this envelope entropy function, determine the signal purity item, and through this signal purity item and this environmental noise reference item, determine this fitness function.

[0037] In an embodiment of this specification, use the improved envelope entropy function and the noise matching degree evaluation index as the fitness function. The mathematical model of the standard envelope entropy is: (4) is the envelope entropy; is the normalized envelope energy of the modal component (IMF) obtained by the j-th variational mode decomposition; is the envelope energy of the IMF, obtained through the Hilbert transform. Equation (4) extracts the time-domain envelope signal of the modal component through the Hilbert transform, calculates the normalized probability of its energy distribution, and then obtains the Shannon entropy value characterizing the randomness of the signal.

[0038] Construct a two-dimensional function that includes signal purity evaluation and noise separation evaluation, dynamically balance the guiding role of the two evaluation indicators for parameter optimization, and the overall form of the fitness function is as follows: (5) where, is the signal purity term determined based on the envelope entropy function, is the environmental noise reference term determined by adjusting the noise matching degree evaluation index based on the dynamic weight coefficient. Specifically, { u i ( t )} is the set of IMF components; F IMF is the eigenvector of all current IMFs; F noise,j represents the eigenvector of the j-th noise template. Among them, Dist(F IMF , F noise,j ) = ∥F IMF - F noise,j ∥2 is the noise matching degree evaluation index, which is used to measure the similarity between the characteristics of the IMF in the decomposition result and the noise template, that is, the minimum distance between the eigenvector of the decomposition result and the eigenvector of the template in the noise library.

[0039] According to the proportional relationship between the current envelope entropy value and the preset maximum reference entropy value , generate a dynamic weight coefficient, establish a negative correlation adjustment mechanism between the envelope entropy value and the weight coefficient, so that the fitness function automatically switches the optimization focus in different noise environments. The calculation method of the dynamic weight coefficient is as follows: (6) where, λ is the dynamic weight coefficient, λ0 is the initial value of the maximum weight factor, set to 0.5, represents the maximum reference value of the envelope entropy.

[0040] Calculate the instantaneous frequency threshold for each such modal component, specifically including: determining the median value of all sampling points of the modal component, calculating the absolute deviation of each sampling point value in the modal component relative to the median value, and determining the median dispersion index of the absolute deviation; calculating the signal kurtosis distribution form parameter of the modal component, and determining the form adjustment coefficient based on the ratio of a preset adjustment constant to the signal kurtosis distribution form parameter; determining the basic threshold according to the ratio of the median dispersion index and the form adjustment coefficient, and performing environmental compensation on the basic threshold through the total number of sampling points obtained in advance to determine the instantaneous frequency threshold.

[0041] Through the Thomson function and the instantaneous frequency threshold, the Thomson function is a function used in the electromagnetic field domain and is used to process outliers exceeding the threshold in the modal component. Use the Thomson function Perform weighted processing on the part of each modal component that exceeds the threshold, retain the modal components that do not exceed the threshold, add up the optimized modal components, and reconstruct the main sequence of MT data, that is, determine the second denoised data. The specific implementation process of performing weighted processing on the part of each modal component that exceeds the threshold is as follows: Frequency threshold And the Thomson function Is: (7) (8) Among them, N represents the total number of sampling points of the original magnetotelluric data sequence, Is the value of the jth sampling point in the ith IMF taken in the entire time domain, Is for The sequence takes the median, that is, the median value of the jth sampling point in the ith IMF, Is the absolute median deviation of the ith IMF, that is, the absolute deviation, which can detect protruding pulse noise. Is the adaptive coefficient of the ith IMF component, that is, the form adjustment coefficient, , Is the kurtosis of each IMF component, that is, the signal kurtosis distribution form parameter, and c is the adjustment coefficient set to 0.6745.

[0042] Taking As the threshold, the process of removing pulse noise from the magnetotelluric data is as follows: (9) Formula (9) represents the processing method for each sampling point j of each IMF component (that is, the ith modal component). For each sampling point j of each IMF component, first judge whether the absolute value of the sampling point is greater than the threshold , that is If it is greater than the threshold value, the first rule is executed ; if it is not greater than the threshold value (i.e., less than or equal to), the second rule is executed . is the instantaneous frequency threshold calculated for the i-th modal component. This threshold is used to distinguish normal signal points from abnormal noise points, usually the critical value of impulse noise. When a sampling point is considered a noise point (i.e., exceeding the threshold), the Thomson function is used to weight (attenuate) this point. is a value between 0 and 1, such that the part exceeding the threshold is exponentially attenuated, so as to achieve the purpose of suppressing noise. For points that do not exceed the threshold, they are considered valid signals, so the original values are retained unchanged. It should be noted that this formula is actually a soft threshold processing process. Compared with the hard threshold processing method of directly setting the points exceeding the threshold to zero, the soft threshold can more smoothly attenuate noise by multiplying a factor less than 1, avoiding signal distortion.

[0043] Step S102, determine the first signal-to-noise ratio corresponding to the second noise reduction data and the signal-to-noise ratio reference value corresponding to the target geological analysis requirement. When the first signal-to-noise ratio is less than the preset signal-to-noise ratio reference value, perform dynamic fusion on the residual signal according to the first noise reduction data and the second noise reduction data to determine the fused residual signal, so as to facilitate the analysis of weak signals in the residual signal and avoid discarding weak signals.

[0044] In an embodiment of this specification, to determine the first signal-to-noise ratio corresponding to the second noise reduction data, the signal-to-noise ratio SNR formula is: (10) In the formula, and are respectively the original effective signal and the synthesized time series after denoising, that is, the amplitude of the data sequence corresponding to the second noise reduction data, and n is the sampling point. When the first signal-to-noise ratio is less than the preset signal-to-noise ratio reference value, it indicates that the current denoising effect does not meet the requirements.

[0045] In the field of magnetotelluric exploration, the useful signal power being at least 3 times (4.7 dB) the noise power is a key critical value to ensure that the apparent resistivity-phase curve is basically available. However, to obtain reliable data that can be used for fine inversion and interpretation, a higher signal-to-noise ratio is usually required, and the preset signal-to-noise ratio reference value needs to be specifically determined according to different working conditions and industry guidelines. It should be noted that the signal-to-noise ratio reference value in the embodiments of this specification can be directly set to 8 db according to experience. It can also be based on the signal-to-noise ratio reference value corresponding to the target geological analysis requirement and the geological analysis requirement. When the first signal-to-noise ratio is not less than the preset signal-to-noise ratio reference value, the second noise reduction data is used as the final denoised output signal.

[0046] When determining the reference SNR value corresponding to the target geological analysis requirement, the following method can be adopted: Combine the parameters of the target geological body within the target geological analysis area. Here, the parameters of the target geological body include the resistivity and burial depth of the target geological body, and set the reference geological SNR value corresponding to the geological conditions according to empirical data. For example, through testing, count the SNRs corresponding to the analysis requirements under different geological conditions during geological analysis, obtain the reference geological SNR values corresponding to different geological conditions through statistical analysis, and construct a mapping relationship table, where the geological conditions can be distinguished by geological classification.

[0047] On the basis of determining the reference geological SNR value corresponding to the geological conditions, consider the task type of the target geological analysis task and correct the SNR. It should be noted that the task types here include tectonic boundary positioning type, lithology identification type, and fluid detection type. Preset the type SNR correction amount corresponding to each task type. For example, the correction amount for the tectonic boundary positioning type is 0 dB, the correction amount for the lithology identification type is 1.5 ± 0.3 dB, and the correction amount for the fluid detection type is 2.5 ± 0.5 dB. In the above way, it is ensured that the obtained reference SNR value matches the target geological analysis requirement.

[0048] Since the magnetotelluric effective signal (such as the response of deep fault zones) is usually a weak signal in the microvolt level and is easily masked by noise; the initial noise reduction may over-suppress the useful components overlapping with the noise frequency band, such as the 0.1 - 1 Hz low-frequency band signal. When the first SNR is less than the preset reference SNR value, it indicates that the residual noise energy in the current noise reduction result has exceeded the effective signal, so the subsequent residual recovery process is executed.

[0049] Residual recovery is used to solve the technical problem of the intertwining of stubborn noise and deep weak signals in magnetotelluric exploration. When the SNR output by the noise reduction channel does not reach the preset threshold, it means that the conventional noise reduction process is difficult to balance the contradiction between noise suppression and signal fidelity. Over-suppressing noise will damage the microvolt-level geological response hidden in the residuals, while retaining the residuals will cause the noise to continuously contaminate the effective signal. Through an innovative dynamic fusion mechanism, the two types of residual signals left by the initial noise reduction are transformed into a resource library that can be purified twice, realizing the precise separation of noise and weak signals. The direct abandonment of residuals by traditional methods actually permanently discards some effective signals, especially resulting in the irreversible loss of key information such as the fracture zone characteristics of deep formation responses and thin coal seam interfaces.

[0050] Based on the first noise reduction data and the second noise reduction data, perform dynamic fusion on the residual signal to determine the fused residual signal, specifically including: determining the first residual data corresponding to the first noise reduction data and the second residual data corresponding to the second noise reduction data; respectively calculating the first time-domain energy data of the first residual data and the second time-domain energy data of the second residual data, and generating a dynamic weight parameter combination based on the first time-domain energy data and the second time-domain energy data; performing weighted fusion on the first residual data and the second residual data through the dynamic weight parameter combination to determine the fused residual signal.

[0051] In an embodiment of the present specification, the first residual data corresponding to the first noise reduction data is the mixed component remaining after the preliminary noise reduction by MSVD, and the first noise reduction data is the MSVD reconstructed signal , through the formula , calculate the first residual data, where is the first residual data, is the original sampling sequence, that is, the original magnetotelluric data sequence. The second noise reduction data is the reconstructed signal after the secondary noise reduction by VMD , and the second residual data corresponding to the second noise reduction data is the high-frequency pulses and oscillatory noises remaining after the secondary noise reduction. Through the formula , calculate the second residual data .

[0052] Generate a dynamic weight parameter combination based on the first time-domain energy data and the second time-domain energy data, specifically including: determining the total energy of the first time-domain energy data and the second time-domain energy data, and generating the first weight coefficient corresponding to the first residual data with the ratio of the first time-domain energy data to the total energy; determining the second weight coefficient corresponding to the second residual data according to the first weight coefficient and the preset normalization constraint condition.

[0053] In an embodiment of the present specification, the fused residual signal The formula is: (11) Among them, , are weights set based on the residual signal energy, and the preset normalization constraint condition is to satisfy .

[0054] (12) Among them, is the residual between the original data and the data after MSVD noise reduction (i.e., the residual of the first noise reduction), is the residual between the reconstructed signal data after MSVD-VMD joint noise reduction and the MSVD reconstructed signal (i.e., the residual of the second noise reduction). The two residuals are weighted and fused through formula (11) to obtain the fused residual signal . Calculate the first time-domain energy data of the first residual data and the second time-domain energy data of the second residual data respectively. The signal energy here is defined as the sum of the squares of the amplitudes of each sampling point, which represents the square of the two-norm of the signal, i.e., the energy. When determining the first weight coefficient in the dynamic fusion weight combination, take the square value of the two-norm of the first residual data and divide it by the sum of the square value of the two-norm of this residual signal and the square value of the two-norm of the second residual data. Through the dynamic weight parameter combination, the first residual data and the second residual data are weighted and fused to determine the fused residual signal. When the proportion of the MSVD residual energy is high, the low-frequency noise processing is strengthened. When the proportion of the MSVD-VMD residual energy is high, the high-frequency noise suppression is strengthened.

[0055] Through the above residual recycling method, the cyclic regeneration of signal components is realized. The low-frequency weak signals hidden in the MSVD residuals and the high-frequency pulses remaining in the VMD residuals are dynamically weighted and fused. The low-frequency weak signals hidden in the SVD residuals are weighted and strengthened because they carry the deep geological response. The high-frequency pulses remaining in the VMD residuals are gradually attenuated due to their pure noise attributes, enabling the recovery of deep weak signals and significantly repairing the distortion of the apparent resistivity curve.

[0056] Step S103: Denoise the fused residual signal to determine the corresponding fused residual denoised data, and perform residual recycling based on the fused residual denoised data to determine the current denoised data after residual recycling.

[0057] In an embodiment of this specification, use the MSVD-VMD main noise reduction channel to denoise the fused residual signal after fusion. After obtaining the denoised fused residual denoised data, perform residual recycling on the fused residual denoised data and merge it with the main signal into the reconstructed MT data sequence. The main signal here is the reconstructed signal after MSVD-VMD secondary noise reduction, i.e., the second denoised data, to determine the current denoised data after residual recycling. It should be noted that in the subsequent iterative loop process, use the second denoised data as the original MT data sequence for noise reduction and residual recycling. The recovered residual signal is merged with the reconstructed signal after MSVD-VMD secondary noise reduction in the current loop to determine the denoised data after residual recycling.

[0058] Step S104: When the second signal-to-noise ratio corresponding to the current denoised data is not less than the signal-to-noise ratio reference value, output the current denoised data to perform geological analysis on the target geological analysis area based on the current denoised data.

[0059] In one embodiment of this specification, the signal-to-noise ratio is calculated for the current noise-reduced data after residual recovery. When the second signal-to-noise ratio corresponding to the current noise-reduced data is not less than the signal-to-noise ratio reference value, the current noise-reduced data is output and the noise reduction process ends. When the second signal-to-noise ratio corresponding to the current noise-reduced data is less than the signal-to-noise ratio reference value, the current noise-reduced data is output and the noise reduction process ends.

[0060] The method further includes: after determining the current noise-reduced data after residual recovery, performing a noise reduction count operation to determine the current noise reduction iteration count; when the current noise reduction iteration count reaches a preset iteration count threshold, outputting the current noise-reduced data.

[0061] In one embodiment of this specification, after determining the current noise-reduced data after residual recovery, a noise reduction count operation is performed to determine the current noise reduction iteration count. That is to say, the noise reduction process after residual recovery is the first noise reduction. When the current noise reduction iteration count reaches a preset iteration count threshold, the current noise-reduced data is output to avoid excessive iteration. Based on the current noise-reduced data, subsequent geological analysis is performed on the target geological analysis area to ensure the availability of the signal.

[0062] To further analyze the noise reduction performance of the embodiments of this specification, different evaluation parameters are used to deeply analyze its stability and effectiveness. Figure 2 It is a schematic diagram of the evaluation parameters after different methods provided by the embodiments of this specification process noise signals. Through Figure 3 The noise reduction ability of the embodiments of this specification is further verified. To simulate the influence of noise on the effective signal at different signal-to-noise ratio levels, different intensities of noise are introduced into the simulated signal. Figure 3 It is a schematic diagram of the noise reduction effect of a measured magnetotelluric data provided by the embodiments of this specification. As Figure 3 shown, similar and continuous noise interference can be observed in the Ex and Ey channels, while there is complex noise mixed with pulse signals in the Hx and Hy magnetic channels. The amplitudes of these interference signals far exceed the effective MT signal, resulting in a low signal-to-noise ratio of the MT data. Reconstruction is performed using the MSVD-VMD algorithm with dynamic residual feedback, and the noise reduction results are as shown in (e)- Figure 3 in (h) Figure 3 shown. The noise amplitude has decreased, the square wave noise in the electric channel has been effectively suppressed, and the effective signal has been retained. The waveform in the magnetic channel is more stable and continuous, without abnormal spikes, and is basically close to the time domain characteristics of the natural magnetotelluric signal. The experimental results prove the effectiveness and superiority of the embodiments of this specification in noise data processing.

[0063] Through the technical solution of the embodiments of this specification, the original magnetotelluric data sequence corresponding to the target geological analysis area in the target geological analysis requirement information is collected, ensuring the regional matching between the signal source and the analysis requirement; the MSVD algorithm is introduced to perform multi-level decomposition on the non-stationary signal MT signal for preliminary noise reduction. Compared with the SVD of single-level linear decomposition, it can perform multi-scale and high-resolution processing on the low-frequency noise in the MT signal, and can effectively retain the target signal while suppressing the noise; for problems such as mode mixing generated during the decomposition of the residual pulse signal, the Thomson function is introduced on the basis of the VMD algorithm to perform adaptive soft threshold attenuation on the components exceeding the threshold, avoiding hard truncation distortion, and being more effective in suppressing the synthesis of pulse signals than the traditional VMD algorithm; the high convergence characteristic of the GWO algorithm is used to optimize the key parameters in the VMD. During the optimization process, the eigenvector library of electromagnetic interference in the target geological analysis area is introduced as noise prior information. By constructing a composite fitness function that integrates envelope entropy, decomposition result, and noise feature matching degree, the target-guided adjustment of the VMD parameters is realized, ensuring the matching between the noise prior knowledge and the target geological analysis area, not only greatly reducing the possibility of mode mixing, but also improving the calculation efficiency by optimizing the number of modes and significantly shortening the calculation time; the dynamic residual feedback method and the weighted fusion mechanism are used to further process the effective signals in the residual, which can meet the signal recovery requirements of the weak effective information in the magnetotelluric data, enhance the signal fidelity, and improve the ability of deep noise processing and weak signal protection.

[0064] The embodiments of this specification also provide a data noise reduction device for multi-level magnetotelluric data, as Figure 4 shown. The device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above method.

[0065] The embodiments of this specification also provide a non-volatile computer storage medium storing computer-executable instructions, and the computer-executable instructions are set to: execute the above method.

[0066] The embodiments in this specification are all described in a progressive manner. The same or similar parts between the embodiments can be referred to each other, and the key points of each embodiment are the differences from other embodiments. In particular, for the embodiments of the device, equipment, and non-volatile computer storage medium, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments.

[0067] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, various changes and modifications can be made to one or more embodiments of this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included within the scope of the claims of this specification.

Claims

1. A data denoising method for multi-level magnetotelluric data, characterized in that The method includes: Determine the target geological analysis requirement information, collect the original magnetotelluric data sequence corresponding to the target geological analysis area in the target geological analysis requirement information, perform resolution singular value decomposition on the original magnetotelluric data sequence to determine the first noise-reduced data, and perform variational mode decomposition on the first noise-reduced data through a pre-determined set of optimization parameters to determine the second noise-reduced data, where the set of optimization parameters is determined based on an electromagnetic interference feature library corresponding to the pre-constructed target geological analysis area; Determine the first signal-to-noise ratio corresponding to the second noise-reduced data and the signal-to-noise ratio reference value corresponding to the target geological analysis requirement. When the first signal-to-noise ratio is less than the preset signal-to-noise ratio reference value, perform dynamic fusion on the residual signal according to the first noise-reduced data and the second noise-reduced data to determine the fused residual signal, so as to facilitate the analysis of weak signals in the residual signal and avoid discarding weak signals; Perform noise reduction processing on the fused residual signal to determine the corresponding fused residual noise-reduced data, and perform residual recovery based on the fused residual noise-reduced data to determine the current noise-reduced data after residual recovery; When the second signal-to-noise ratio corresponding to the current noise-reduced data is not less than the signal-to-noise ratio reference value, output the current noise-reduced data to perform geological analysis on the target geological analysis area based on the current noise-reduced data.

2. The data noise reduction method for multi-level magnetotelluric data according to claim 1, wherein Performing variational mode decomposition on the first noise-reduced data through a pre-determined set of optimization parameters to determine the second noise-reduced data specifically includes: Optimize the parameters of the variational mode decomposition through the grey wolf optimization algorithm and an electromagnetic interference feature library corresponding to the pre-constructed target geological analysis area to determine a set of optimization parameters, where the set of optimization parameters includes the number of mode decompositions and the quadratic penalty factor; Perform variational mode decomposition on the first noise-reduced data using the set of optimization parameters to obtain the number of mode components of the mode decomposition; Calculate the instantaneous frequency threshold for each mode component, and determine the second noise-reduced data through the Thomson function and the instantaneous frequency threshold.

3. A data noise reduction method for multi-level magnetotelluric data according to claim 2, characterized in that, Optimize the parameters of the variational mode decomposition through the grey wolf optimization algorithm and an electromagnetic interference feature library corresponding to the pre-constructed target geological analysis area to determine a set of optimization parameters specifically including: Collect magnetotelluric noise samples of various noise types in the target geological analysis area to extract the electromagnetic interference feature vectors corresponding to each noise type, and construct an electromagnetic interference feature library based on the electromagnetic interference feature vectors corresponding to each noise type; Determine the noise matching degree evaluation index corresponding to the decomposition mode component according to the decomposition mode component after the resolution singular value decomposition and multiple electromagnetic interference feature vectors to quantify the environmental noise interference in the target geological analysis area; Determine the fitness function with the noise matching degree evaluation index and a preset envelope entropy function, and optimize the parameters of the variational mode decomposition through the grey wolf optimization algorithm to determine the optimized number of mode decompositions and the quadratic penalty factor.

4. A data noise reduction method for multi-level magnetotelluric data according to claim 3, characterized in that, Determine a fitness function based on the noise matching degree evaluation index and a preset envelope entropy function, specifically including: Generate a dynamic weight coefficient according to the proportional relationship between the current envelope entropy value and the preset maximum reference entropy value, and adjust the noise matching degree evaluation index based on the dynamic weight coefficient to determine an environmental noise reference term; Determine a signal purity term based on the envelope entropy function, and determine the fitness function through the signal purity term and the environmental noise reference term.

5. A data noise reduction method for multi-level magnetotelluric data according to claim 2, characterized in that, Calculate an instantaneous frequency threshold for each of the modal components, specifically including: Determine the median value of all sampling points of the modal component, calculate the absolute deviation of each sampling point value in the modal component relative to the median value, and determine the median dispersion index of the absolute deviation; Calculate the signal kurtosis distribution form parameter of the modal component, and determine a form adjustment coefficient based on the ratio of a preset adjustment constant to the signal kurtosis distribution form parameter; Determine a basic threshold according to the ratio of the median dispersion index to the form adjustment coefficient, and perform environmental compensation on the basic threshold through the total number of sampling points obtained in advance to determine the instantaneous frequency threshold.

6. A data noise reduction method for multi-level magnetotelluric data according to claim 1, characterized in that Perform dynamic fusion on the residual signal according to the first noise reduction data and the second noise reduction data to determine a fused residual signal, specifically including: Determine a first residual data corresponding to the first noise reduction data and a second residual data corresponding to the second noise reduction data; Calculate a first time-domain energy data of the first residual data and a second time-domain energy data of the second residual data respectively, and generate a dynamic weight parameter combination based on the first time-domain energy data and the second time-domain energy data; Perform weighted fusion on the first residual data and the second residual data through the dynamic weight parameter combination to determine the fused residual signal.

7. A data noise reduction method for multi-level magnetotelluric data according to claim 6, characterized in that, Generate a dynamic weight parameter combination based on the first time-domain energy data and the second time-domain energy data, specifically including: Determine the total energy of the first time-domain energy data and the second time-domain energy data, and generate a first weight coefficient corresponding to the first residual data based on the ratio of the first time-domain energy data to the total energy; Determine a second weight coefficient corresponding to the second residual data according to the first weight coefficient and a preset normalization constraint condition.

8. A data noise reduction method for multi-level magnetotelluric data according to claim 1, characterized in that, The method further includes: After determining the current noise reduction data after residual recovery, perform a noise reduction times counting operation to determine the current noise reduction iteration times; When the current noise reduction iteration times reach a preset iteration times threshold, output the current noise reduction data.

9. A data noise reduction device for multi-level magnetotelluric data, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1-8.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are set to: execute the method according to any one of claims 1-8.

Images