Longitudinal wave data denoising and waveform inversion imaging method for acoustic logging remote exploration

Abstract

The invention discloses a longitudinal wave data denoising and waveform inversion imaging method for acoustic logging remote exploration, and the method comprises the following steps: inputting original acoustic logging data, and carrying out the data conversion; inputting the converted data into a pre-trained network to obtain a signal mask and a noise mask; calculating a recovered time-frequency domain signal, converting the recovered time-frequency domain signal to a time domain, and carrying out denoising and frequency domain conversion; inputting initial longitudinal wave velocity data into the longitudinal wave velocity model, and constructing a discrete Helmholtz equation to obtain a forward modeling wave field and forward modeling prediction data; calculating a residual error between the forward modeling wave field data and the observation wave field data, and constructing a discrete dual Helmholtz equation to obtain a dual wave field; gradient information and iteration quantity are calculated respectively, and iteration is carried out on the longitudinal wave velocity model; and checking whether a convergence standard is reached. According to the method, reliable underground structure information can be provided for exploration and development of petroleum and natural gas, and powerful technical support can be provided for accurate and safe resource development.

Description

Technical Field

[0001] The invention relates to the technical field of acoustic well logging remote detection, and in particular to a longitudinal wave data denoising and waveform inversion imaging method for acoustic well logging remote detection. Background Art

[0002] In the field of acoustic logging remote detection, obtaining underground structure information is a crucial part of oil and gas exploration and development. When recording the wave field in the underground medium, traditional acoustic logging data is often affected by the noise from geological layer changes and the tool itself, resulting in reduced data quality and limiting the ability to accurately image.

[0003] Existing data processing methods have some challenges in denoising and waveform inversion. Traditional denoising methods may not be able to effectively distinguish between signals and noise, especially under complex geological conditions. At the same time, traditional waveform inversion methods may be constrained by problems such as insufficient resolution and model mismatch when restoring underground structures, limiting the accurate grasp of geological characteristics around the well.

[0004] Therefore, in order to improve the imaging accuracy and reliability of acoustic logging remote detection, a new type of P-wave data denoising and waveform inversion imaging method is urgently needed in this field. This method should be able to overcome the limitations of traditional methods, accurately extract underground structure information, and achieve high-resolution inversion of P-wave velocity, thereby providing a more reliable technical means for the precise exploration and safe development of oil and gas resources. Summary of the invention

[0005] The purpose of the present invention is to overcome the problem of lack of accurate velocity model around the well in the field of acoustic logging remote detection, and to provide a method for denoising P-wave data and waveform inversion imaging for acoustic logging remote detection. The method uses the observed P-wave field data, denoises it, and updates the P-wave velocity model around the well through continuous iteration, and finally obtains the best P-wave velocity model for interpreting the observed data. This innovation provides an important guarantee for achieving safe and efficient oil and gas development.

[0006] The present invention is implemented by adopting the following technical scheme: a method for denoising longitudinal wave data and waveform inversion imaging of acoustic well logging remote detection, comprising the following steps:

[0007] S1: Input the original acoustic logging data and perform data conversion;

[0008] S2: Input the converted data into the pre-trained network to obtain the signal mask and the noise mask;

[0009] S3: Calculate the restored time-frequency domain signal, convert it to the time domain and perform denoising and frequency domain conversion;

[0010] S4: inputting initial P-wave velocity data into the P-wave velocity model and constructing a discrete Helmholtz equation to obtain forward wave field and forward prediction data;

[0011] S5: Calculate the residual between the forward wavefield data and the observed wavefield data, and construct a discrete dual Helmholtz equation to obtain the dual wavefield;

[0012] S6: Calculate the gradient information and iteration amount respectively, and iterate the P-wave velocity model;

[0013] S7: Check whether the convergence standard is reached, if yes, then end, otherwise return to step S4.

[0014] Furthermore, step S1 includes the following sub-steps:

[0015] S11: Input the original acoustic logging longitudinal waveform data and related parameter files;

[0016] S12: Perform short-time Fourier transform on the input time series signal and convert it into the time-frequency domain to obtain a time-frequency diagram.

[0017] Furthermore, step S2 is specifically as follows: inputting the time-frequency graph into a pre-trained deep convolutional denoising neural network DNN(*), and outputting two masks, namely a noise mask and a signal mask.

[0018] Furthermore, step S3 includes the following sub-steps:

[0019] S31: multiplying the signal mask and the time-frequency domain data to obtain a restored time-frequency domain signal;

[0020] S32: converting the restored time-frequency domain signal into the time domain;

[0021] S33: Convert the denoised time domain data into frequency domain data.

[0022] Further, step S31 is specifically as follows: Hadamard product is performed on the noise mask and the signal mask with the time-frequency diagram to obtain the restored time-frequency domain signal Y noise and Y signal :

[0023] Y noise =Y⊙M noise ;

[0024] Y signal =Y⊙M signal ;

[0025] Among them, ⊙ represents the Hadamard product, that is, the multiplication of corresponding elements.

[0026] Furthermore, step S4 includes the following sub-steps:

[0027] S41: Obtaining an initial prediction model for the wellbore peripheral longitudinal wave velocity based on previous seismic exploration data;

[0028] S42: Constructing the Helmholtz equation:

[0029] (Δ+w 2 m)u=q;

[0030] And obtain the forward wave field:

[0031] q=A -1 u;

[0032] Among them, q is the sub-wave function, u is the forward wave field, w is the frequency, m is the square of the slowness of longitudinal wave propagation, Δ is the spatial Laplace operator, A is the discrete sum Helmholtz matrix, and after obtaining the forward wave field, a projection operator P(*) is used to calculate the forward simulation data

[0033]

[0034] Furthermore, step S5 includes the following sub-steps:

[0035] S51: Calculate the residual between the forward modeled wave field data and the observed wave field data;

[0036] S52: The calculated residual is used as a dual source, a discrete dual Helmholtz equation is constructed, and the equation is solved to obtain a dual wave field.

[0037] Furthermore, step S52 is specifically as follows: constructing a dual matrix A of the discrete Helmholtz matrix A T , and use δd(w) as the source term of the dual wave field to calculate the dual wave field v:

[0038] v=A -T δd(w).

[0039] Furthermore, step S6 includes the following sub-steps:

[0040] S61: Perform cross-correlation operation on the calculated forward wave field and dual wave field to calculate the gradient information required for the iteration of the P-wave velocity model:

[0041]

[0042] Among them, diag(*) is the operation of converting a vector into a diagonal matrix, and conj(*) is the operation of taking the conjugate of a complex vector;

[0043] S62: combining the calculated gradient information of the P-wave velocity model with the total variation information of the P-wave velocity model to calculate the model iteration amount of the P-wave velocity;

[0044] S63: Iterating the longitudinal wave velocity model using the calculated model iteration amount of the longitudinal wave velocity.

[0045] Furthermore, step S62 is specifically as follows:

[0046] Calculate the total variation gradient of the current model:

[0047]

[0048] Add g1 and βg2 to get the new gradient vector:

[0049] g=g1+βg2;

[0050] Among them, β is the penalty factor set in advance, and then combined with the set step size γ, the model iteration update amount is calculated:

[0051] δm=-γg.

[0052] The beneficial effects of the present invention are:

[0053] The present invention utilizes the P-wave field information observed in acoustic logging and adopts a waveform inversion method combined with total variation information to iteratively update and invert the P-wave velocity model within 10-100 meters around the well. This innovation provides a solid velocity model support for achieving high-precision acoustic logging long-range detection imaging.

[0054] The present invention utilizes a deep learning method to denoise the acquired wave field data, effectively improving the availability of the data and providing solid data support for achieving high-precision acoustic logging remote detection imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0056] Figure 1 It is a flow chart of the present invention;

[0057] Figure 2 It is a schematic diagram of the correlation data in the time-frequency domain;

[0058] Figure 3 It is a schematic diagram of input longitudinal wave field data;

[0059] Figure 4 It is a schematic diagram of the longitudinal wave field data after denoising;

[0060] Figure 5 It is a schematic diagram of longitudinal wave field data in the frequency domain;

[0061] Figure 6 is a schematic diagram of the input initial P-wave velocity model;

[0062] Figure 7 is the schematic diagram of the calculated forward wave field;

[0063] Figure 8 is a schematic diagram of the calculated dual wave field;

[0064] Fig. 9 is a schematic diagram of the calculated gradient;

[0065] Fig.10 is a schematic diagram of the calculated model update amount;

[0066] Fig.11 This is a schematic diagram of the final inversion results. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0068] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0069] Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0070] See also Figures 1 to 11 A method for denoising longitudinal wave data and waveform inversion imaging of acoustic logging remote detection comprises the following steps:

[0071] Step 1: Input the original acoustic logging longitudinal waveform data and related parameter files;

[0072] Step 2: Convert the input time domain data into time-frequency domain data;

[0073] Step 3: Input the time-frequency domain data into the pre-trained network to obtain the signal mask and noise mask;

[0074] Step 4: Multiply the signal mask and the time-frequency domain data to obtain the restored time-frequency domain signal;

[0075] Step 5: Convert the recovered time-frequency domain signal to the time domain;

[0076] Step 6: Convert the denoised time domain data into frequency domain data;

[0077] Step 7: Input initial P-wave velocity data;

[0078] Step 8: Based on the input longitudinal wave velocity model, the discrete Helmholtz equation is constructed and solved to obtain the forward wave field and forward prediction data;

[0079] Step nine: Calculate the residual between the forward modeled wave field data and the observed wave field data;

[0080] Step 10: Use the calculated residual as the dual source, construct the discrete dual Helmholtz equation, and solve it to obtain the dual wave field;

[0081] Step 11: Perform cross-correlation operation on the calculated forward wave field and dual wave field to calculate the gradient information required for the iteration of the P-wave velocity model;

[0082] Step 12: combining the calculated gradient information of the P-wave velocity model with the total variation information of the P-wave velocity model to calculate the model iteration amount of the P-wave velocity;

[0083] Step 13: Iterate the P-wave velocity model using the calculated P-wave velocity model iteration amount;

[0084] Step 14: Check whether the convergence standard is met. If yes, end; otherwise, return to step 8.

[0085] The specific process is as follows:

[0086] Step 1: Input original data: input original acoustic logging vertical waveform data and related parameter files. Figure 3 The input longitudinal wave waveform data is displayed.

[0087] Step 2: Convert the input time domain data into time-frequency domain data, perform short-time Fourier transform on the input time series signal d(t) and convert it into the time-frequency domain to obtain the time-frequency diagram Y:

[0088] Y = SIFT(d(t));

[0089] Figure 2 (i) in FIG. 1 shows the time-frequency diagram Y. FIG.

[0090] Step 3: Input the time-frequency domain data into the pre-trained network to obtain the signal mask and the noise mask, input Y into a pre-trained deep convolutional denoising neural network DNN(*), and output two masks, namely the noise mask M and noise and signal mask Msignal :

[0091] M noise ,M signal = DNN(d(t));

[0092] Among them, M noise and M signal are matrices of the same size as Y, and their values ​​are all 0 or 1.

[0093] Step 4: Multiply the signal mask with the time-frequency domain data to obtain the restored time-frequency domain signal. noise and M signal Do Hadamard product with Y respectively to get Y noise and Y signal :

[0094] Y noise =Y⊙Mn oise ;

[0095] Y signal =Y⊙M signal ;

[0096] Where ⊙ represents the Hadamard product, that is, the corresponding elements are multiplied. Figure 2 (ii) and (iii) show Y signal and Y noise .

[0097] Step 5: Convert the recovered time-frequency domain signal to the time domain and signal Perform inverse short-time Fourier transform to obtain denoised data

[0098] Figure 4 The data after denoising is shown.

[0099] Step 6: Sampling the frequency domain data and denoising the time domain data Perform Fourier transform FFT(*):

[0100] So as to obtain frequency domain data Then, among all the frequency data, select the data within the frequency range required for inversion imaging. Figure 5 The frequency domain data after Fourier transformation are shown.

[0101] Step seven, obtaining an initial prediction model of the wellbore peripheral longitudinal wave velocity based on the previous seismic exploration data, and obtaining an initial prediction model of the wellbore peripheral longitudinal wave velocity based on the previous seismic exploration data.

[0102] Step 8: Calculate the forward wave field and forward prediction data, and construct the following Helmholtz equation:

[0103] (Δ+w 2 m)u=q;

[0104] Discrete form: Au = q;

[0105] And solve to obtain the forward wave field: q = A -1 u;

[0106] Among them, q is the sub-wave function, u is the forward wave field, w is the frequency, m is the square of the slowness of longitudinal wave propagation, Δ is the spatial Laplace operator, A is the discrete sum Helmholtz matrix, and after obtaining the forward wave field, a projection operator P(*) is used to calculate the forward simulation data

[0107] Figure 7 Display the calculated forward wavefield.

[0108] Step nine, the step of calculating the residual includes: calculating the residual between the forward wave field data and the observed wave field data:

[0109]

[0110] Step 10: Calculate the dual wave field and construct the dual matrix A of the discrete Helmholtz matrix A T , and use δd(w) as the source term of the dual wave field to calculate the dual wave field v:

[0111] v=A -T δd(w); Figure 8 The calculated dual wavefield is shown.

[0112] Step 11: Calculate the gradient information required for the iteration of the P-wave velocity model: Calculate the differential of the discrete Helmholtz matrix A with respect to the square of the slowness m of the P-wave propagation

[0113]

[0114] And the forward wave field and the dual wave field are cross-correlated to calculate the gradient information required for the iteration of the P-wave velocity model:

[0115] Among them, diag(*) is an operation that converts a vector into a diagonal matrix, and conj(*) is an operation that takes the conjugate of a complex vector. Fig. 9 The calculated gradients are shown.

[0116] Step 12, calculating the model iteration amount of the longitudinal wave velocity includes: calculating the gradient of the total variation of the current model:

[0117] Add g1 and βg2 to get the new gradient vector:

[0118] g=g1+βg2;

[0119] Where β is a pre-set penalty factor; and then combined with the set step size γ, the model iteration update amount is calculated: δm = -γg; Fig.10 The calculated model updates are shown.

[0120] Step 13: Update the model and calculate the model after iteration:

[0121]

[0122] Step 14: determine whether the convergence condition is met and check whether the convergence standard is reached. If yes, then end; otherwise, return to step 8; Fig.11 The final inversion results are shown.

[0123] Based on the above embodiments, the present invention has at least the following technical effects:

[0124] The present invention uses the P-wave field information observed in acoustic logging and adopts a waveform inversion method combined with total variation information to iteratively update and invert the P-wave velocity model within 10-100 meters around the well. This innovation provides a solid velocity model support for high-precision acoustic logging remote detection imaging. The present invention uses a deep learning method to denoise the acquired wave field data, effectively improving the availability of the data and providing solid data support for high-precision acoustic logging remote detection imaging.

[0125] In general, the present invention can not only provide reliable underground structure information for oil and gas exploration and development, but also provide strong technical support for accurate and safe resource development. In the oil and gas industry, the application of the present invention is expected to promote exploration and development work towards a more efficient and sustainable direction.

[0126] For the aforementioned embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily required by the present application.

[0127] The above embodiments describe the basic principles and main features of the present invention and the advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the changes and modifications made by those skilled in the art shall be within the scope of protection of the appended claims of the present invention without departing from the spirit and scope of the present invention.

Claims

1. A method for denoising longitudinal wave data and waveform inversion imaging of acoustic logging remote detection, characterized in that: The steps include: S1: Input the original acoustic logging data and perform data conversion; S2: Input the converted data into the pre-trained network to obtain the signal mask and the noise mask; S3: Calculate the restored time-frequency domain signal, convert it to the time domain and perform denoising and frequency domain conversion; S4: inputting initial P-wave velocity data into the P-wave velocity model and constructing a discrete Helmholtz equation to obtain forward wave field and forward prediction data; S5: Calculate the residual between the forward wavefield data and the observed wavefield data, and construct a discrete dual Helmholtz equation to obtain the dual wavefield; S6: Calculate the gradient information and iteration amount respectively, and iterate the P-wave velocity model; S7: Check whether the convergence standard is reached, if yes, then end, otherwise return to step S4.

2. The method for denoising longitudinal wave data and waveform inversion imaging of acoustic logging remote detection according to claim 1, characterized in that: Step S1 includes the following sub-steps: S11: Input the original acoustic logging longitudinal waveform data and related parameter files; S12: Perform short-time Fourier transform on the input time series signal and convert it into the time-frequency domain to obtain a time-frequency diagram.

3. A method for denoising longitudinal wave data and waveform inversion imaging of acoustic logging remote detection as claimed in claim 2, characterized in that: Step S2 specifically includes: inputting the time-frequency graph into a pre-trained deep convolutional denoising neural network DNN(*), and outputting two masks, namely a noise mask and a signal mask.

4. A method for denoising longitudinal wave data and waveform inversion imaging of acoustic logging remote detection as claimed in claim 3, characterized in that: Step S3 includes the following sub-steps: S31: multiplying the signal mask and the time-frequency domain data to obtain a restored time-frequency domain signal; S32: converting the restored time-frequency domain signal into the time domain; S33: Convert the denoised time domain data into frequency domain data.

5. The method for denoising longitudinal wave data and waveform inversion imaging of acoustic logging remote detection as claimed in claim 4, characterized in that: Step S31 is as follows: Hadamard product the noise mask and the signal mask with the time-frequency diagram to obtain the restored time-frequency domain signal Y noise and Y signal : AND noise =Y⊙M noise ; AND signal =Y⊙M signal ; Among them, ⊙ represents the Hadamard product, that is, the multiplication of corresponding elements.

6. A method for denoising longitudinal wave data and waveform inversion imaging of acoustic logging remote detection as claimed in claim 5, characterized in that: Step S4 includes the following sub-steps: S41: Obtaining an initial prediction model for the wellbore peripheral longitudinal wave velocity based on previous seismic exploration data; S42: Constructing the Helmholtz equation: (Δ+w 2 m)u=q; And obtain the forward wave field: q=A -1 u; Among them, q is the sub-wave function, u is the forward wave field, w is the frequency, m is the square of the slowness of longitudinal wave propagation, Δ is the spatial Laplace operator, A is the discrete sum Helmholtz matrix, and after obtaining the forward wave field, a projection operator P(*) is used to calculate the forward simulation data 7. A method for denoising longitudinal wave data and waveform inversion imaging of acoustic logging remote detection as claimed in claim 6, characterized in that: Step S5 includes the following sub-steps: S51: Calculate the residual between the forward modeled wave field data and the observed wave field data; S52: The calculated residual is used as a dual source, a discrete dual Helmholtz equation is constructed, and the equation is solved to obtain a dual wave field.

8. A method for denoising longitudinal wave data and waveform inversion imaging of acoustic logging remote detection as claimed in claim 7, characterized in that: Step S52 is specifically: constructing the dual matrix A of the discrete Helmholtz matrix A T , and use δd(w) as the source term of the dual wave field to calculate the dual wave field v: v=A -T δd(w)。 9. A method for denoising longitudinal wave data and waveform inversion imaging of acoustic well logging remote detection as claimed in claim 8, characterized in that: Step S6 includes the following sub-steps: S61: Perform cross-correlation operation on the calculated forward wave field and dual wave field to calculate the gradient information required for the iteration of the P-wave velocity model: Among them, diag(*) is the operation of converting a vector into a diagonal matrix, and conj(*) is the operation of taking the conjugate of a complex vector; S62: combining the calculated gradient information of the P-wave velocity model with the total variation information of the P-wave velocity model to calculate the model iteration amount of the P-wave velocity; S63: Iterating the longitudinal wave velocity model using the calculated model iteration amount of the longitudinal wave velocity.

10. A method for denoising longitudinal wave data and waveform inversion imaging of acoustic logging remote detection as claimed in claim 9, characterized in that: Step S62 is specifically as follows: Calculate the total variation gradient of the current model: Add g1 and βg2 to get the new gradient vector: g=g1+βg2; Among them, β is the penalty factor set in advance, and then combined with the set step size γ, the model iteration update amount is calculated: δm=-γg.

Images