A CRP gather fidelity high signal-to-noise ratio processing method that meets the requirements of pre-stack waveform inversion
By combining error analysis and Shearlet transform with spatial weighted median filtering, the problem of low signal-to-noise ratio in CRP gathers was solved, and high signal-to-noise ratio CRP gather processing was achieved, meeting the requirements of pre-stack waveform inversion and improving well accuracy and seismic prediction accuracy.
Patent Information
- Application Number
- CN202211306118.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-25
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2042-10-25
AI Technical Summary
In pre-stack waveform inversion, existing technologies suffer from non-first-order reflections such as random noise, coherent noise, multiple waves, and converted waves in CRP gathers, resulting in low signal-to-noise ratios that fail to meet the requirements of high-resolution inversion. Furthermore, conventional processing methods compromise the fidelity of the gathers.
We employ techniques such as random noise suppression based on error analysis, Shearlet transform, and spatially weighted median filtering to improve the signal-to-noise ratio of CRP gathers in stages, ensuring that the relative amplitude relationship and geological information are not compromised during processing. We use Shearlet transform to distinguish between effective reflections and noise, and spatially weighted median filtering to suppress residual time differences.
It significantly improves the signal-to-noise ratio of CRP gathers, meets the requirements of pre-stack waveform inversion, improves well accuracy and seismic prediction accuracy, and is suitable for PWI pre-stack waveform inversion technology.
Smart Images

Figure CN117930338B_ABST
Abstract
Description
[0001] Technical Field: This invention belongs to the field of seismic data processing, and more specifically, this invention is a CRP gather fidelity and high signal-to-noise ratio processing method that meets the requirements of pre-stack waveform inversion.
[0002] Background: This invention is limited to providing high signal-to-noise ratio pre-stack CRP gathers for Prestack Waveform Inversion (PWI) technology. The basic assumption of PWI is that all information in the input data comes from real underground geological information, and the P-wave and S-wave velocities, densities, and other elastic parameters of the research object are inverted through changes in the waveform of the pre-stack gather. However, actual CRP gathers often contain residual random noise, coherent noise, multiples, migration noise, converted waves, and other non-primary reflections. Due to the special and novel nature of PWI technology, it has not yet been industrially applied in oil and gas field exploration in China, therefore, there are currently no dedicated processing methods for PWI technology in China.
[0003] First, commonly used gather optimization methods in the industry often aim for high-resolution output. Common techniques include deconvolution, inverse Q-filtering, and spectral brightening. These techniques alter the distribution of dominant frequency band energy within the CRP gather. The improved resolution often results in a shift of the dominant frequency to higher frequencies, leading to a loss of low-frequency information. However, low-frequency information is crucial for pre-stack inversion techniques like PWI. Therefore, conventional high-resolution processing techniques often suffer from poor fidelity. This invention targets PWI pre-stack waveform inversion technology, which demands high fidelity from pre-stack gathers. Therefore, this invention does not employ any of the aforementioned resolution-enhancing techniques, thus solving the problem of reduced CRP gather fidelity caused by conventional gather optimization.
[0004] Secondly, it is generally believed in the industry that when random noise and high-frequency signals are difficult to distinguish, random noise at the high-frequency end should not be suppressed rashly. The result of this approach is a low signal-to-noise ratio (SNR) for CRP (Continuous Resonance Process) gathers. This invention serves the needs of pre-stack waveform inversion in PWI (Programmable Wavelength Inversion). The input gather may not require signals that may exist at the high-frequency end, but it must be a CRP gather with a very high SNR. Therefore, it employs denoising techniques such as random noise suppression based on error analysis and Shearlet transform to significantly improve the SNR of the CRP gather. The SNR of the gather processed by this invention is higher than that of conventional gather optimization methods.
[0005] Pre-stack waveform inversion is a high-resolution inversion technique that has emerged in recent years. Theoretically, it can invert various reservoir parameters up to one-tenth of the wavelength. However, since pre-stack waveform inversion assumes that there is no noise in the input data, all signals are treated as seismic waveform changes caused only by underground geological information. Therefore, it requires the input CRP gather to have the highest possible signal-to-noise ratio. Based on this requirement, we invented this method, which can improve the signal-to-noise ratio of CRP gathers step by step.
[0006] Summary of the Invention: In view of the problems existing in the background technology, a high signal-to-noise ratio (SNR) processing method for CRP gathers that meets the requirements of pre-stack waveform inversion is invented. This method significantly improves the SNR of CRP gathers used for pre-stack waveform inversion. All processing of pre-stack CRP gathers is based on improving the SNR while preserving fidelity and amplitude. The CRP gathers processed by this method can meet the requirements of pre-stack waveform inversion.
[0007] To achieve the above objectives, a CRP gather fidelity high signal-to-noise ratio processing method that meets the requirements of pre-stack waveform inversion includes the following steps:
[0008] Step 1: Selection of offset method and optimization of CRP gathers based on pre-stack waveform inversion requirements;
[0009] Step 2: Perform time-frequency analysis on the CRP gathers that have passed the quality control in Step 1 and are divided into equal offsets. Select the target layer as the spectrum analysis window according to geological requirements, select the two sets of CRP gathers with the largest bandwidth, and select the set of data with better continuity of the phase axis of the target layer as the CRP gather for the next step of improving the signal-to-noise ratio optimization.
[0010] Step 3: Perform fidelity random noise suppression on the CRP gather obtained in Step 2;
[0011] Step 4: Perform Shearlet transformation on the CRP gathers obtained in Step 3 after local residual time difference correction and flattening of the in-phase axis without destroying the relative amplitude relationship, assuming that the seismic data of the CRP gathers contains effective reflections and noise;
[0012] Step 5: Perform spatial weighted median filtering on the noise-suppressed CRP gather obtained in Step 4 to further improve the signal-to-noise ratio of the CRP gather, in order to meet the requirement of high signal-to-noise ratio for gathers in pre-stack waveform inversion.
[0013] The specific steps of step 1 are as follows: First, determine whether the offsets of the quality control CRP gather are equally divided. Record the offset value of the first CRP gather as offset1 (in meters), the offset value of the second CRP gather as offset2 (in meters), and the offset value of the last CRP gather as offset1. nThe unit is meters (m). Starting from the second track, subtract the offset of the adjacent track from the current track's offset distance to obtain the first offset difference Δ1 = offset2 - offset1, where Δ1 is in meters. Then, subtract the second track's offset distance from the third track's offset distance to obtain Δ2 = offset3 - offset2, where Δ2 is in meters, until the first offset difference Δ1 is obtained. n-1 =offset n -offset n-1 , where Δ n-1 The unit is m, and the quality control range is from Δ1 to Δ. n-1 If they are completely identical, then the quality control is passed, and this set of CRP gathers enters the next round of optimization. If the quality control Δ1 to Δ n-1 If they are not completely identical, the quality control will fail, the current CRP collection will be eliminated, and it cannot enter the next round of selection.
[0014] The specific implementation of step 2 is as follows: If both pre-stack time migration and pre-stack depth migration CRP gathers exist, the pre-stack depth migration gathers need to be converted from the depth domain to the time domain using the migration layer velocity model to obtain the time domain CRP gathers obtained by the pre-stack depth migration method, and then the target layer spectrum analysis and optimization are performed.
[0015] By performing Fourier transform-based data regularization on quality-controlled CRP gathers, the problem of low shallow coverage times can be improved, and the signal-to-noise ratio of CRP gathers can also be increased. Before performing FFT on CRP gathers, a cut-off process must be performed to remove the first arrival wave and the precoherent energy below the first arrival wave. The spectrum of the irregular data is obtained by applying non-uniform Fourier transform, as shown in the following equation:
[0016]
[0017] In equation (1), k is the frequency domain component, wavenumber, and the unit is units;
[0018] x s This represents the location of a spatially non-uniform sampling point, without units.
[0019] f(x s ) corresponds to x s The signal value at that location is a unitless scalar.
[0020] f(k) is the spectrum of the signal, which has no unit.
[0021] N is the number of non-uniform sample points, in units of one.
[0022] ΔD is the maximum space, measured in meters (m).
[0023] Δx s It is the spatial sampling interval, measured in meters (m).
[0024] In the data reconstruction process, the sparsest components with the highest energy in the frequency spectrum are first extracted. The seismic data is then transformed back to the time domain using IFFT. Next, the time-domain data after subtracting the highest energy component from the frequency domain is subjected to FFT again, and the highest energy component in the existing data is extracted again. This process is repeated iteratively, extracting each sparse frequency component according to its energy strength. By using a priori anti-spurious-frequency high-cutoff parameters, spurious frequencies can be prevented during inverse Fourier transform data reconstruction. The specific method is as follows:
[0025]
[0026]
[0027] In equations (2) and (3):
[0028] f k (x s () is the time-domain seismic data after IFFT, i.e., inverse Fourier transform;
[0029] k is the frequency domain component, wave number, and the unit is one;
[0030] x s This represents the location of a spatially non-uniform sampling point, without units.
[0031] f(x s ) corresponds to x s The signal value at that location is a unitless scalar.
[0032] f(k) is the spectrum of the signal, which has no unit.
[0033] It is the seismic data that is returned to the time domain by subtracting the strongest energy component from the input FFT data in the nth round and then passing it back through IFFT. After multiple rounds of iteration, the high signal-to-noise ratio CRP gather data of the target location can be reconstructed.
[0034] The specific method of step 3 is as follows: The 3D seismic data is transformed in both the X and Y directions using a 2D Z-transform to suppress random noise. This method has lower computational complexity and higher efficiency, achieving the same effect as the 3D transformation. First, the seismic data volume is subjected to a forward Fourier transform to obtain:
[0035]
[0036] In equation (4), D(x,y,ω) is the positive Fourier transform of the three-dimensional time-domain seismic data d(x,y,t);
[0037] x and y are spatial coordinates, in meters (m).
[0038] t is the vertical two-way reflection time, i.e., travel time, and its unit is ms;
[0039] e is the base of Euler's number, which is the natural logarithm, and has no unit.
[0040] i is the sequence number of the signal, which is a positive integer, i = 1, 2, ..., n;
[0041] ω is the frequency, and the unit is Hz;
[0042] Design a filter prediction operator f(x,y), assuming there are n sets of in-phase shafts with different velocities within this window. i (t), where the time difference between each in-phase axis in the x and y directions is Δt. x (i) and Δt y (i), where i = 1, 2, ..., n, then the two-dimensional seismic data f(x, y) of a specific frequency component can be expressed in the form of a two-dimensional Z-transform:
[0043]
[0044] In equation (5), F(Z1,Z2) is the Z-transform result of the signal f(x,y);
[0045] R(f) is the spectral domain representation of the in-phase axis R(t);
[0046] Expanding the right side of equation (9), we get:
[0047] E(Z1,Z2)=F(Z1,Z2)P(Z1,Z2) (6)
[0048] In equation (6), E(Z1,Z2) is the Z-transform result of the prediction error;
[0049] P(Z1,Z2) is the prediction error filter;
[0050] F(Z1,Z2) is the input signal;
[0051] Equation (10) applies the signal F(Z1,Z2) as a prediction error filter to obtain the result E(Z1,Z2);
[0052] The prediction error filter P(Z1,Z2) is actually unknown. Assuming that the prediction error is minimized after the 3D seismic data is predicted by the rectangular prediction operator, we construct an objective function and set its partial derivative with respect to the prediction operator to 0, resulting in the matrix equation:
[0053] R H ·P=R (7)
[0054] In equation (7), R H The Hermite matrix for multichannel autocorrelation of seismic data in the fx domain;
[0055] R is the multichannel autocorrelation matrix of the fx-domain seismic data;
[0056] P is the component matrix of the prediction operator;
[0057] The data processed by equation (7) is then subjected to inverse Fourier transform to obtain the pre-stack CRP gather in the tx domain, i.e., the time-space domain random noise suppressed.
[0058] The specific steps of step 4 are as follows: the CRP gathers obtained in step 2 after being cut off and suppressed for random noise are stacked, and the random noise suppression process in step S2 is repeated twice after stacking to obtain the stacked model data volume V. stkmodel Each CRP gather is matched with a single seismic trace at the same location, and the position of each amplitude is adjusted in the offset direction so that each signal of each phase axis in the CRP gather is calibrated to match the V signal from top to bottom. stkmodel At the same position, after all the traces of the current phase axis have been corrected and leveled without destroying the relative amplitude relationship, they are fixed in place. When correcting the residual time difference of each trace of the next phase axis, the position of the previous phase axis must remain unchanged. This process is repeated to level all the phase axes.
[0059] f(t) CRP =d(t) CRP +noise(t) (8)
[0060] In equation (8), f(t) CRP This is data from Daoji.
[0061] d(t) CRP This is a concentrated and effective seismic reflection;
[0062] noise(t) is random noise in the trace set;
[0063] For f(t) CRP After performing the Shearlet transformation, we get:
[0064] C(i,j,k)=SH[f(t) CRP ]= <f(t) CRP ,φ i,j,k > (9)
[0065] In equation (9), C(i,j,k) is the input data f(t). CRP Shearlet coefficients obtained through the Shearlet transformation are dimensionless, and SH[·] represents the inner product operator;
[0066] φ i,j,k The parent function of the Shearlet transform is the continuous shear wave, where i is the scale, j is the shearing amount, and k is the translation parameter, all in milliseconds.
[0067] Let the coefficient
[0068] n is the number of all Shearlet coefficients;
[0069] σ is the average value of the summation of each sample in the entire Shearlet matrix;
[0070] The criterion A for distinguishing effective reflections from random noise in the Shearlet domain is determined by the following formula (unitless):
[0071]
[0072] When |C(i,j,k)|≥A i When the reflection is valid, no processing is performed on C(i,j,k), and it has no unit.
[0073] When |C(i,j,k)|<A i When this condition is identified as noise, the C(i,j,k) value at that point is marked as 0, resulting in a new coefficient matrix C. denoise (i,j,k);
[0074] The reprocessed coefficient matrix is then subjected to an inverse Shearlet transform to obtain the noise-removed data:
[0075]
[0076] In equation (11), Seismic data representing effective reflections from CRP gathers;
[0077] C denoise (i,j,k) is the new coefficient matrix after removing the noisy Shearlet coefficients;
[0078] φ(i,j,k) is the generating function of the Shearlet transform, where i is the scale, j is the shearing amount, and k is the translation parameter, all in milliseconds (ms).
[0079] This step suppresses noise in the CRP gather.
[0080] The specific implementation of step 5 is as follows: weighted median filtering of three-dimensional data can better suppress residual time difference. The pre-stack CRP gather can directly push the two-dimensional prediction method of CRP gather to not only statistically analyze the continuity of the phase axis of a CRP gather in three-dimensional space, but also take into account whether the corresponding phase axes of this phase axis in several or a dozen adjacent CRP gathers still have the same spatial continuity. The residual time difference is suppressed by using spatial three-dimensional weighted median filtering.
[0081] The number of cells in the weighted median filter of the main survey line is: No._cells_inline=2INT(x / Δx)+1 (12)
[0082] The number of cells in the weighted median filter for the crossline is: No._cells_crossline=2INT(y / Δy)+1 (13)
[0083] In equations (12) and (13), INT indicates that the result calculated in parentheses is rounded down;
[0084] x and y represent the calculation ranges on the main survey line and the connecting survey line, respectively, in meters (m).
[0085] Δx and Δy are the top elements for the main survey line and connecting survey line directions, and the unit is meters (m).
[0086] The radius of the ellipse along the main survey line is: R inline =[INT(No._cells_inline / 2)+0.5]Δx (14)
[0087] The radius of the ellipse of the connecting survey line is: R crossline =[INT(No._cells_crossline / 2)+0.5]Δy (15)
[0088] The weights are:
[0089] In equation (16), the weight is dimensionless, x and y are the absolute distances from any point to the calculation point along the main survey line / connecting survey line, in meters; R inline and R crossline The radius of the ellipse at the calculation point along the main survey line / connecting survey line is in meters; Decay is the attenuation coefficient, the magnitude of which determines the weight of the weighted median filter.
[0090]
[0091] In equation (17), V K This represents the non-abnormal amplitude value within the aperture, and has no unit.
[0092] N represents the number of non-abnormal amplitude values within the aperture, expressed in units of ;
[0093] Weight is the weighted value used in the calculation and has no unit.
[0094] After spatially weighted median filtering, the signal-to-noise ratio of the pre-stack CRP gather is significantly improved.
[0095] The present invention has the following beneficial effects:
[0096] A high signal-to-noise ratio (SNR) processing method for CRP gathers that meets the requirements of pre-stack waveform inversion is based on the assumption that there is no noise in the input data. Regardless of the signal, it is treated as a seismic waveform change caused only by underground geological information. Therefore, the input CRP gathers are required to have the highest possible SNR. Based on this requirement, we invented this method to improve the SNR of CRP gathers step by step. It has achieved good results in the application of multiple actual seismic data blocks in Daqing Oilfield. The CRP gathers processed by this method can meet the requirements of pre-stack waveform inversion.
[0097] Attached image description: Figure 1 This is a flowchart of the operation of a CRP gather fidelity high signal-to-noise ratio processing method that meets the requirements of pre-stack waveform inversion, as disclosed in this invention.
[0098] Figure 2a The non-equal-division offset is a commercial software Omega Kirchhoff pre-stack time-off CRP gather;
[0099] Figure 2b It is a CRP gather that has been converted back to the time domain after pre-stack depth migration using the commercial software PG18;
[0100] Figure 2c It is a pre-stack time migration CRP gather using the Kirchhoff integral method with equal-division offsets;
[0101] Figure 2d It is a Q-offset CRP gather with equally divided offsets;
[0102] Figure 3 It is the CRP gather after the selected gathers have undergone data regularization processing based on Fourier transform;
[0103] Figure 4 It is a CRP gather after fidelity random noise suppression;
[0104] Figure 5 It is a CRP gather after the in-phase axis is flattened by correcting the local residual time difference of the in-phase axis without destroying the relative amplitude relationship;
[0105] Figure 6 It is a CRP gather after Shearlet transform to improve the signal-to-noise ratio of the effective reflected signal;
[0106] Figure 7 It is the final CRP gather after spatial weighted median filtering.
[0107] Detailed Description of Embodiments: The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.
[0108] A high signal-to-noise ratio (SNR) CRP gather processing method that meets the requirements of pre-stack waveform inversion includes the following steps:
[0109] Step 1: Selection of offset method and optimization of CRP gathers based on pre-stack waveform inversion requirements;
[0110] Step 2: Perform time-frequency analysis on the CRP gathers that have passed the quality control in Step 1 and are divided into equal offsets. Select the target layer as the spectrum analysis window according to geological requirements, select the two sets of CRP gathers with the largest bandwidth, and select the set of data with better continuity of the phase axis of the target layer as the CRP gather for the next step of improving the signal-to-noise ratio optimization.
[0111] Step 3: Perform fidelity random noise suppression on the CRP gather obtained in Step 2;
[0112] Step 4: Perform Shearlet transformation on the CRP gathers obtained in Step 3 after local residual time difference correction and flattening of the in-phase axis without destroying the relative amplitude relationship, assuming that the seismic data of the CRP gathers contains effective reflections and noise;
[0113] Step 5: Perform spatial weighted median filtering on the noise-suppressed CRP gather obtained in Step 4 to further improve the signal-to-noise ratio of the CRP gather, in order to meet the requirement of high signal-to-noise ratio for gathers in pre-stack waveform inversion.
[0114] The specific steps of step 1 are as follows: First, determine whether the offsets of the quality control CRP gather are equally divided. Record the offset value of the first CRP gather as offset1 (in meters), the offset value of the second CRP gather as offset2 (in meters), and the offset value of the last CRP gather as offset1. n The unit is meters (m). Starting from the second track, subtract the offset of the adjacent track from the current track's offset distance to obtain the first offset difference Δ1 = offset2 - offset1, where Δ1 is in meters. Then, subtract the second track's offset distance from the third track's offset distance to obtain Δ2 = offset3 - offset2, where Δ2 is in meters, until the first offset difference Δ1 is obtained. n-1 =offset n -offset n-1 , where Δ n-1 The unit is m, and the quality control range is from Δ1 to Δ. n-1 If they are completely identical, then the quality control is passed, and this set of CRP gathers enters the next round of optimization. If the quality control Δ1 to Δ n-1 If they are not completely identical, the quality control will fail, the current CRP collection will be eliminated, and it cannot enter the next round of selection.
[0115] The specific implementation of step 2 is as follows: If both pre-stack time migration and pre-stack depth migration CRP gathers exist, the pre-stack depth migration gathers need to be converted from the depth domain to the time domain using the migration layer velocity model to obtain the time domain CRP gathers obtained by the pre-stack depth migration method, and then the target layer spectrum analysis and optimization are performed.
[0116] By performing Fourier transform-based data regularization on quality-controlled CRP gathers, the problem of low shallow coverage times can be improved, and the signal-to-noise ratio of CRP gathers can also be increased. Before performing FFT on CRP gathers, a cut-off process must be performed to remove the first arrival wave and the precoherent energy below the first arrival wave. The spectrum of the irregular data is obtained by applying non-uniform Fourier transform, as shown in the following equation:
[0117]
[0118] In equation (1), k is the frequency domain component, wavenumber, and the unit is units;
[0119] x s This represents the location of a spatially non-uniform sampling point, without units.
[0120] f(x s ) corresponds to x s The signal value at that location is a unitless scalar.
[0121] f(k) is the spectrum of the signal, which has no unit.
[0122] N is the number of non-uniform sample points, in units of one.
[0123] ΔD is the maximum space, measured in meters (m).
[0124] Δx s It is the spatial sampling interval, measured in meters (m).
[0125] In the data reconstruction process, the sparsest components with the highest energy in the frequency spectrum are first extracted. The seismic data is then transformed back to the time domain using IFFT. Next, the time-domain data after subtracting the highest energy component from the frequency domain is subjected to FFT again, and the highest energy component in the existing data is extracted again. This process is repeated iteratively, extracting each sparse frequency component according to its energy strength. By using a priori anti-spurious-frequency high-cutoff parameters, spurious frequencies can be prevented during inverse Fourier transform data reconstruction. The specific method is as follows:
[0126]
[0127]
[0128] In equations (2) and (3):
[0129] fk (x s () is the time-domain seismic data after IFFT, i.e., inverse Fourier transform;
[0130] k is the frequency domain component, wave number, and the unit is one;
[0131] x s This represents the location of a spatially non-uniform sampling point, without units.
[0132] f(x s ) corresponds to x s The signal value at that location is a unitless scalar.
[0133] f(k) is the spectrum of the signal, which has no unit.
[0134] It is the seismic data that is returned to the time domain by subtracting the strongest energy component from the input FFT data in the nth round and then passing it back through IFFT. After multiple rounds of iteration, the high signal-to-noise ratio CRP gather data of the target location can be reconstructed.
[0135] The specific method of step 3 is as follows: The 3D seismic data is transformed in both the X and Y directions using a 2D Z-transform to suppress random noise. This method has lower computational cost and higher efficiency, achieving the same effect as the 3D transformation. First, the seismic data volume is subjected to a forward Fourier transform to obtain:
[0136]
[0137] In equation (4), D(x,y,ω) is the positive Fourier transform of the three-dimensional time-domain seismic data d(x,y,t);
[0138] x and y are spatial coordinates, in meters (m).
[0139] t is the vertical two-way reflection time, i.e., travel time, and its unit is ms;
[0140] e is the base of Euler's number, which is the natural logarithm, and has no unit.
[0141] i is the sequence number of the signal, which is a positive integer, i = 1, 2, ..., n;
[0142] ω is the frequency, and the unit is Hz;
[0143] Design a filter prediction operator f(x,y), assuming there are n sets of in-phase shafts with different velocities within this window. i (t), where the time difference between each in-phase axis in the x and y directions is Δt. x (i) and Δt y (i), where i = 1, 2, ..., n, then the two-dimensional seismic data f(x, y) of a specific frequency component can be expressed in the form of a two-dimensional Z-transform:
[0144]
[0145] In equation (5), F(Z1,Z2) is the Z-transform result of the signal f(x,y);
[0146] R(f) is the spectral domain representation of the in-phase axis R(t);
[0147] Expanding the right side of equation (9), we get:
[0148] E(Z1,Z2)=F(Z1,Z2)P(Z1,Z2) (6)
[0149] In equation (6), E(Z1,Z2) is the Z-transform result of the prediction error;
[0150] P(Z1,Z2) is the prediction error filter;
[0151] F(Z1,Z2) is the input signal;
[0152] Equation (10) applies the signal F(Z1,Z2) as a prediction error filter to obtain the result E(Z1,Z2);
[0153] The prediction error filter P(Z1,Z2) is actually unknown. Assuming that the prediction error is minimized after the 3D seismic data is predicted by the rectangular prediction operator, we construct an objective function and set its partial derivative with respect to the prediction operator to 0, resulting in the matrix equation:
[0154] R H ·P=R (7)
[0155] In equation (7), R H The Hermite matrix for multichannel autocorrelation of seismic data in the fx domain;
[0156] R is the multichannel autocorrelation matrix of the fx-domain seismic data;
[0157] P is the component matrix of the prediction operator;
[0158] The data processed by equation (7) is then subjected to inverse Fourier transform to obtain the pre-stack CRP gather in the tx domain, i.e., the time-space domain random noise suppressed.
[0159] The specific steps of step 4 are as follows: the CRP gathers obtained in step 2 after being cut off and suppressed for random noise are stacked, and the random noise suppression process in step S2 is repeated twice after stacking to obtain the stacked model data volume V. stkmodel Each CRP gather is matched with a single seismic trace at the same location, and the position of each amplitude is adjusted in the offset direction so that each signal of each phase axis in the CRP gather is calibrated to match the V signal from top to bottom. stkmodelAt the same position, after all the traces of the current phase axis have been corrected and leveled without destroying the relative amplitude relationship, they are fixed in place. When correcting the residual time difference of each trace of the next phase axis, the position of the previous phase axis must remain unchanged. This process is repeated to level all the phase axes.
[0160] f(t) CRP =d(t) CRP +noise(t) (8)
[0161] In equation (8), f(t) CRP This is data from Daoji.
[0162] d(t) CRP This is a concentrated and effective seismic reflection;
[0163] noise(t) is random noise in the trace set;
[0164] For f(t) CRP After performing the Shearlet transformation, we get:
[0165] C(i,j,k)=SH[f(t) CRP ]= <f(t) CRP ,φ i,j,k > (9)
[0166] In equation (9), C(i,j,k) is the input data f(t). CRP Shearlet coefficients obtained through the Shearlet transformation are dimensionless, and SH[·] represents the inner product operator;
[0167] φ i,j,k The parent function of the Shearlet transform is the continuous shear wave, where i is the scale, j is the shearing amount, and k is the translation parameter, all in milliseconds.
[0168] Let the coefficient
[0169] n is the number of all Shearlet coefficients;
[0170] σ is the average value of the summation of each sample in the entire Shearlet matrix;
[0171] The criterion A for distinguishing effective reflections from random noise in the Shearlet domain is determined by the following formula (unitless):
[0172]
[0173] When |C(i,j,k)|≥A i When the reflection is valid, no processing is performed on C(i,j,k), and it has no unit.
[0174] When |C(i,j,k)|<A i When this condition is identified as noise, the C(i,j,k) value at that point is marked as 0, resulting in a new coefficient matrix C. denoise (i,j,k);
[0175] The reprocessed coefficient matrix is then subjected to an inverse Shearlet transform to obtain the noise-removed data:
[0176]
[0177] In equation (11), Seismic data representing effective reflections from CRP gathers;
[0178] C denoise (i,j,k) is the new coefficient matrix after removing the noisy Shearlet coefficients;
[0179] φ(i,j,k) is the generating function of the Shearlet transform, where i is the scale, j is the shearing amount, and k is the translation parameter, all in milliseconds (ms).
[0180] This step suppresses noise in the CRP gather.
[0181] The specific implementation of step 5 is as follows: weighted median filtering of three-dimensional data can better suppress residual time difference. The pre-stack CRP gather can directly push the two-dimensional prediction method of CRP gather to not only statistically analyze the continuity of the phase axis of a CRP gather in three-dimensional space, but also take into account whether the corresponding phase axes of this phase axis in several or a dozen adjacent CRP gathers still have the same spatial continuity. The residual time difference is suppressed by using spatial three-dimensional weighted median filtering.
[0182] The number of cells in the weighted median filter of the main survey line is: No._cells_inline=2INT(x / Δx)+1 (12)
[0183] The number of cells in the weighted median filter for the crossline is: No._cells_crossline=2INT(y / Δy)+1 (13)
[0184] In equations (12) and (13), INT indicates that the result calculated in parentheses is rounded down;
[0185] x and y represent the calculation ranges on the main survey line and the connecting survey line, respectively, in meters (m).
[0186] Δx and Δy are the top elements for the main survey line and connecting survey line directions, and the unit is meters (m).
[0187] The radius of the ellipse along the main survey line is: R inline=[INT(No._cells_inline / 2)+0.5]Δx (14)
[0188] The radius of the ellipse of the connecting survey line is: R crossline =[INT(No._cells_crossline / 2)+0.5]Δy (15)
[0189] The weights are:
[0190] In equation (16), the weight is dimensionless, x and y are the absolute distances from any point to the calculation point along the main survey line / connecting survey line, in meters; R inline and R crossline The radius of the ellipse at the calculation point along the main survey line / connecting survey line is in meters; Decay is the attenuation coefficient, the magnitude of which determines the weight of the weighted median filter.
[0191]
[0192] In equation (17), V K This represents the non-abnormal amplitude value within the aperture, and has no unit.
[0193] N represents the number of non-abnormal amplitude values within the aperture, expressed in units of ;
[0194] Weight is the weighted value used in the calculation and has no unit.
[0195] After spatially weighted median filtering, the signal-to-noise ratio of the pre-stack CRP gather is significantly improved. Example:
[0196] The seismic data processed in this embodiment is from the target area of 50km in the Gulong Shale Oilfield X test area of Daqing Oilfield. 2 The 3D seismic data, collected in 2016 by Daqing Geophysical Exploration Company in the Bayanchagan area west of Changyuan, has a 10m×10m area. The observation system type is a central shot with receivers on both sides, with a maximum offset greater than 3200m and 192 seismic scans. A single geophone with a 10Hz main frequency was used for single-point reception, and the excitation method was a single-point explosive source with a charge weight of 4-6kg. Both the shot distance and receiver distance were 160m. The target layer T2 has an aspect ratio greater than 0.8, making it a typical high-density, wide-azimuth acquisition and observation system. In shale oil exploration, problems such as thin gold box structures and inaccurate target layer structures exist, necessitating improvements in seismic prediction accuracy and well-to-well consistency.
[0197] We adopted the internationally advanced PWI pre-stack waveform inversion technology. The input gather was the CRP gather processed by the conventional processing software Omega. However, the PWI results still could not meet the requirements for fine earthquake prediction and accurate modeling. The analysis revealed that there was strong interference such as coherent noise and random noise in the input pre-stack gather, and the overall signal-to-noise ratio of the CRP gather was too low. Therefore, it is necessary to reprocess the CRP gather using the method of this invention.
[0198] Step 1: Migration Method Selection and CRP Gather Optimization Based on Pre-Stack Waveform Inversion Requirements: Through extensive migration method processing experiments, we collected pre-stack time migration CRP gathers from the commercial software Omega Kirchhoff with non-equally divided offsets, such as... Figure 2a As shown, the CRP gathers converted back to the time domain after pre-stack depth migration using the commercial software PG18 are as follows: Figure 2b As shown, the pre-stack time migration CRP gather using the Kirchhoff integral method with equally divided offsets is as follows: Figure 2c As shown, the Q-offset CRP gather with equally divided offsets is as follows: Figure 2d As shown, according to this method, based on the principle of equally divided offsets, the signal-to-noise ratio cannot be too low, thus eliminating [this method]. Figure 2a Figure 2d The two sets of Taoist texts that remain will proceed to the next step.
[0199] Step 2: Perform time-frequency analysis on the CRP gathers that have passed the quality control in Step 1, dividing the offsets equally. Select the target layer as the spectral analysis window according to geological requirements, and choose the two CRP gathers with the largest bandwidth. Select the set with better continuity of the target layer's phase axis as the CRP gather for the next step of signal-to-noise ratio (SNR) improvement optimization. Through time-frequency analysis, the CRP gathers from the pre-stack depth migration back to the time domain have more low frequencies, but the low-frequency SNR is not high, which will affect the application effect of this invention. The pre-stack time migration CRP gathers with equally divided offsets are the set with better continuity of the phase axis, and are ultimately the preferred choice. Figure 2c Moving on to the next step, in subsequent processing, due to the problem that the near-track energy is abnormally weak due to the objective influence of the seismic acquisition and observation system on the non-equally divided offset, this influence must be taken into account, and the participation of the nearest trace data should be consciously avoided in subsequent inversion.
[0200] Step 3: Perform fidelity random noise suppression on the CRP gather obtained in Step 2: Figure 3 The CRP gather for the Guogu shale oil level X well point shows that the 1.82s to 1.9s interval is the shale oil-rich interval of the Qingshankou Formation, which is the target layer for this invention. The strongest energy phase axis at 1.92s is T2, a standard layer in the Songliao Basin. The typical seismic response characteristics of the upper Qingshankou Formation are three peaks and three troughs. The shallowest peak is the phase axis at 1.82s, and the deepest trough is the thick trough phase axis from 1.87s to 1.9s. However, from... Figure 3It is evident that T2 alone clearly exhibits a large number of short-period offset arcs, low-frequency coherent noise from top to bottom, and a significant amount of random noise in its upper part. These factors disrupt the continuity of the three distinct wave peak phase axes that should exist in the upper part of the Qingshankou Formation of T2, failing to meet the requirements of PWI pre-stack waveform inversion technology. Therefore, random noise suppression processing based on prediction error analysis is necessary. Figure 3 The CRP gather is used as input data. After processing using Equations 1 to 7, the CRP gather data undergoes an inverse Fourier transform to obtain the pre-stack CRP gather after suppressing random noise in the tx domain. See [link to relevant documentation]. Figure 4 Then... Figure 4 The residual time difference of the gather is eliminated without disrupting the relative amplitude relationship. Specifically, the CRP gather with suppressed random noise is used. Figure 4 After stacking, the random noise suppression process in step S2 is repeated twice to obtain the stacked model data volume Vstkmodel. Each CRP gather is matched with the only seismic trace at the same location. The position of each amplitude is adjusted in the offset direction so that each trace signal of each phase axis in the CRP gather is calibrated to the same position in Vstkmodel from top to bottom. After the local residual time difference correction of the current phase axis is completed without destroying the relative amplitude relationship, it is fixed. When correcting the residual time difference of each trace of the next phase axis, the position of the previous phase axis must remain unchanged. This process is repeated to complete the flattening of all phase axes. For example, with Figure 4 The gather uses a time window of 1.4s-1.95s. After forming the stacked model traces, comparison begins with the trace with the closest offset. If there is no remaining time difference, no up or down time shift is performed. If there is a 2ms time difference between the 7th trace (T2) and the model trace, then the entire 7th trace is shifted by 2ms to make its T2 position completely consistent with the model trace. This process is repeated for all traces to correct for remaining time differences, resulting in... Figure 5 CRP collection.
[0201] Step 4: Perform Shearlet transformation on the CRP gather obtained in Step 3 after local residual time difference correction and flattening of the in-phase axis without destroying the relative amplitude relationship, to suppress... Figure 5 CRP channel concentration noise, non-first-reflection signal: Figure 5 Using the CRP gather as input data, and through Shearlet transform and judgment and inverse transform according to Equations 8 to 11, a high SNR pre-stack gather with suppressed non-first-time reflection information is obtained. The processed CRP gather is shown below. Figure 6 Through comparative analysis, it can be clearly seen that the continuity of the single peak phase axis of the gather from 1.82s to 1.87s is significantly enhanced after the offset distance of 900m.
[0202] Step 5: The noise-suppressed CRP gather obtained in Step 4 is subjected to spatially weighted median filtering to further improve the signal-to-noise ratio (SNR) of the CRP gather, thus meeting the high SNR requirement for pre-stack waveform inversion. Figure 6 Using the CRP gather as input data, data-driven adaptive spatial weighted median filtering is performed through formulas 11 to 17. This step effectively suppresses noise within the in-phase axis at near offsets (i.e., offsets less than 900m from 1.82s to 1.9s in this example), restoring the original shape and relative amplitude relationship of the in-phase axis at near offsets, thus obtaining... Figure 7 The final processing result will Figure 7 The CRP gathers were used again as input data for PWI pre-stack waveform inversion, yielding inversion results for P-wave velocity, density, and other elastic parameters consistent with the well logging information. The well-to-well agreement rate was higher than 80%, verifying the correctness and reliability of the results presented in this invention. This invention also addresses the issue that the signal-to-noise ratio and overall quality of CRP gathers obtained through conventional commercial software processing methods, which were insufficient for PWI pre-stack waveform inversion (as shown in the four examples in Figure 2), by processing them into... Figure 7 This high signal-to-noise ratio, high-quality CRP gather that meets the requirements of PWI pre-stack waveform inversion verifies the high fidelity and high quality of the processing results of this invention. Each step of this invention can be implemented through programming or existing mainstream commercial software, and it has broad prospects for promotion and application. It will provide strong technical support for accurate seismic prediction of thin interlayer reservoirs, i.e. meter-level reservoirs, in fields such as shale oil and tight oil.
Claims
1. A CRP gather fidelity high signal-to-noise ratio processing method that meets the requirements of pre-stack waveform inversion, characterized in that: Includes the following steps: Step 1: Selection of offset method and optimization of CRP gathers based on pre-stack waveform inversion requirements; Step 2: Perform time-frequency analysis on the CRP gathers that have passed the quality control in Step 1 and are divided into equal offsets. Select the target layer as the spectrum analysis window according to geological requirements, select the two sets of CRP gathers with the largest bandwidth, and select the set of data with better continuity of the phase axis of the target layer as the CRP gather for the next step of improving the signal-to-noise ratio optimization. Step 3: Perform fidelity random noise suppression on the CRP gather obtained in Step 2; Step 4: Perform Shearlet transformation on the CRP gathers obtained in Step 3 after local residual time difference correction and flattening of the in-phase axis without destroying the relative amplitude relationship, assuming that the seismic data of the CRP gathers contains effective reflections and noise; Step 5: Perform spatial weighted median filtering on the noise-suppressed CRP gather obtained in Step 4 to further improve the signal-to-noise ratio of the CRP gather, in order to meet the requirement of high signal-to-noise ratio for gathers in pre-stack waveform inversion.
2. The CRP gather fidelity high signal-to-noise ratio processing method for meeting the requirements of pre-stack waveform inversion as described in claim 1, characterized in that: The specific steps of step 1 are as follows: First, determine whether the offsets of the quality control CRP gather are equally divided. Record the offset value of the first CRP gather as offset1 (in meters), the offset value of the second CRP gather as offset2 (in meters), and the offset value of the last CRP gather as offset1. n The unit is meters (m). Starting from the second track, subtract the offset of the adjacent track from the current track's offset distance to obtain the first offset difference Δ1 = offset2 - offset1, where Δ1 is in meters. Then, subtract the second track's offset distance from the third track's offset distance to obtain Δ2 = offset3 - offset2, where Δ2 is in meters, until the first offset difference Δ1 is obtained. n-1 =offset n -offset n-1 , where Δ n-1 The unit is m, and the quality control range is from Δ1 to Δ. n-1 If they are completely identical, then the quality control is passed, and this set of CRP gathers enters the next round of optimization. If the quality control Δ1 to Δ n-1 If they are not completely identical, the quality control will fail, the current CRP collection will be eliminated, and it cannot enter the next round of selection.
3. The CRP gather fidelity high signal-to-noise ratio processing method for meeting the requirements of pre-stack waveform inversion as described in claim 1, characterized in that: The specific implementation of step 2 is as follows: If both pre-stack time migration and pre-stack depth migration CRP gathers exist, the pre-stack depth migration gathers need to be converted from the depth domain to the time domain using the migration layer velocity model to obtain the time domain CRP gathers obtained by the pre-stack depth migration method, and then the target layer spectrum analysis and optimization are performed. By performing Fourier transform-based data regularization on quality-controlled CRP gathers, the problem of low shallow coverage times can be improved, and the signal-to-noise ratio of CRP gathers can also be increased. Before performing FFT on CRP gathers, a cut-off process must be performed to remove the first arrival wave and the precoherent energy below the first arrival wave. The spectrum of the irregular data is obtained by applying non-uniform Fourier transform, as shown in the following equation: In equation (1), k is the frequency domain component, wavenumber, and the unit is units; x s This represents the location of a spatially non-uniform sampling point, without units. f(x s ) corresponds to x s The signal value at that location is a unitless scalar. f(k) is the spectrum of the signal, which has no unit. N is the number of non-uniform sample points, in units of one. ΔD is the maximum space, measured in meters (m). Δx s It is the spatial sampling interval, measured in meters (m). In the data reconstruction process, the sparsest components with the highest energy in the frequency spectrum are first extracted. The seismic data is then transformed back to the time domain using IFFT. Next, the time-domain data after subtracting the highest energy component from the frequency domain is subjected to FFT again, and the highest energy component in the existing data is extracted again. This process is repeated iteratively, extracting each sparse frequency component according to its energy strength. By using a priori anti-spurious-frequency high-cutoff parameters, spurious frequencies can be prevented during inverse Fourier transform data reconstruction. The specific method is as follows: In equations (2) and (3): f k (x s () is the time-domain seismic data after IFFT, i.e., inverse Fourier transform; k is the frequency domain component, wave number, and the unit is one; x s This represents the location of a spatially non-uniform sampling point, without units. f(x s ) corresponds to x s The signal value at that location is a unitless scalar. f(k) is the spectrum of the signal, which has no unit. It is the seismic data that is returned to the time domain by subtracting the strongest energy component from the input FFT data in the nth round and then passing it back through IFFT. After multiple rounds of iteration, the high signal-to-noise ratio CRP gather data of the target location can be reconstructed.
4. The CRP gather fidelity high signal-to-noise ratio processing method that meets the requirements of pre-stack waveform inversion as described in claim 1, characterized in that: The specific method of step 3 is as follows: The 3D seismic data is transformed in both the X and Y directions using a 2D Z-transform to suppress random noise. This method has lower computational cost and higher efficiency, achieving the same effect as the 3D transformation. First, the seismic data volume is subjected to a forward Fourier transform to obtain: In equation (4), D(x,y,ω) is the positive Fourier transform of the three-dimensional time-domain seismic data d(x,y,t); x and y are spatial coordinates, in meters (m). t is the vertical two-way reflection time, i.e., travel time, and its unit is ms; e is the base of Euler's number, which is the natural logarithm, and has no unit. i is the sequence number of the signal, which is a positive integer, i = 1, 2, ..., n; ω is the frequency, and the unit is Hz; Design a filter prediction operator f(x,y), assuming there are n sets of in-phase shafts with different velocities within this window. i (t), where the time difference between each in-phase axis in the x and y directions is Δt. x (i) and Δt y (i), where i = 1, 2, ..., n, then the two-dimensional seismic data f(x, y) of a specific frequency component can be expressed in the form of a two-dimensional Z-transform: In equation (5), F(Z1,Z2) is the Z-transform result of the signal f(x,y); R(f) is the spectral domain representation of the in-phase axis R(t); Expanding the right side of equation (9), we get: E(Z1,Z2)=F(Z1,Z2)P(Z1,Z2) (6) In equation (6), E(Z1,Z2) is the Z-transform result of the prediction error; P(Z1,Z2) is the prediction error filter; F(Z1,Z2) is the input signal; Equation (10) applies the signal F(Z1,Z2) as a prediction error filter to obtain the result E(Z1,Z2); The prediction error filter P(Z1,Z2) is actually unknown. Assuming that the prediction error is minimized after the 3D seismic data is predicted by the rectangular prediction operator, we construct an objective function and set its partial derivative with respect to the prediction operator to 0, resulting in the matrix equation: R H ·P=R (7) In equation (7), R H The Hermite matrix for multichannel autocorrelation of seismic data in the fx domain; R is the multichannel autocorrelation matrix of the fx-domain seismic data; P is the component matrix of the prediction operator; The data processed by equation (7) is then subjected to inverse Fourier transform to obtain the pre-stack CRP gather in the tx domain, i.e., the time-space domain random noise suppressed.
5. The CRP gather fidelity high signal-to-noise ratio processing method that meets the requirements of pre-stack waveform inversion as described in claim 1, characterized in that: The specific steps of step 4 are as follows: the CRP gathers obtained in step 2 after being cut off and suppressed for random noise are stacked, and the random noise suppression process in step S2 is repeated twice after stacking to obtain the stacked model data volume V. stkmodel Each CRP gather is matched with a single seismic trace at the same location, and the position of each amplitude is adjusted in the offset direction so that each signal of each phase axis in the CRP gather is calibrated to match the V signal from top to bottom. stkmodel At the same position, after all the traces of the current phase axis have been corrected and leveled without destroying the relative amplitude relationship, they are fixed in place. When correcting the residual time difference of each trace of the next phase axis, the position of the previous phase axis must remain unchanged. This process is repeated to level all the phase axes. f(t) CRP =d(t) CRP +noise(t) (8) In equation (8), f(t) CRP This is data from Daoji. d(t) CRP This is a concentrated and effective seismic reflection; noise(t) is random noise in the trace set; For f(t) CRP After performing the Shearlet transformation, we get: C(i,j,k)=SH[f(t) CRP ]=<f(t) CRP ,φ i,j,k > (9) In equation (9), C(i,j,k) is the input data f(t). CRP Shearlet coefficients obtained through the Shearlet transformation are dimensionless, and SH[·] represents the inner product operator; φ i,j,k The parent function of the Shearlet transform is the continuous shear wave, where i is the scale, j is the shearing amount, and k is the translation parameter, all in milliseconds. Let the coefficient n is the number of all Shearlet coefficients; σ is the average value of the summation of each sample in the entire Shearlet matrix; The criterion A for distinguishing effective reflections from random noise in the Shearlet domain is determined by the following formula (unitless): When |C(i,j,k)|≥A i When the reflection is valid, no processing is performed on C(i,j,k), and it has no unit. When |C(i,j,k)|<A i When this is identified as noise, the C(i,j,k) value at that point is marked as 0, resulting in a new coefficient matrix C. denoise (i,j,k); The reprocessed coefficient matrix is then subjected to an inverse Shearlet transform to obtain the noise-removed data: In equation (11), Seismic data representing effective reflections from CRP gathers; C denoise (i,j,k) is the new coefficient matrix after removing the noisy Shearlet coefficients; φ(i,j,k) is the generating function of the Shearlet transform, where i is the scale, j is the shearing amount, and k is the translation parameter, all in milliseconds (ms). This step suppresses noise in the CRP gather.
6. The CRP gather fidelity high signal-to-noise ratio processing method for meeting the requirements of pre-stack waveform inversion as described in claim 1, characterized in that: The specific implementation of step 5 is as follows: weighted median filtering of three-dimensional data can better suppress residual time difference. The pre-stack CRP gather can directly push the two-dimensional prediction method of CRP gather to not only statistically analyze the continuity of the phase axis of a CRP gather in three-dimensional space, but also take into account whether the corresponding phase axes of this phase axis in several or a dozen adjacent CRP gathers still have the same spatial continuity. The residual time difference is suppressed by using spatial three-dimensional weighted median filtering. The number of cells in the weighted median filter of the main survey line is: No._cells_inline=2INT(x / Δx)+1 (12) The number of cells in the weighted median filter for the crossline is: No._cells_crossline=2INT(y / Δy)+1 (13) In equations (12) and (13), INT indicates that the result calculated in parentheses is rounded down; x and y represent the calculation ranges on the main survey line and the connecting survey line, respectively, in meters (m). Δx and Δy are the top elements for the main survey line and connecting survey line directions, and the unit is meters (m). The radius of the ellipse along the main survey line is: R inline =[INT(No._cells_inline / 2)+0.5]Δx (14) The radius of the ellipse of the connecting survey line is: R crossline =[INT(No._cells_crossline / 2)+0.5]Δy (15) The weights are: In equation (16), the weight is dimensionless, x and y are the absolute distances from any point to the calculation point along the main survey line / connecting survey line, in meters; R inline and R crossline The radius of the ellipse at the calculation point along the main survey line / connecting survey line is in meters; Decay is the attenuation coefficient, the magnitude of which determines the weight of the weighted median filter. Output amplitude at each weighted median filter calculation point In equation (17), V K This represents the non-abnormal amplitude value within the aperture, and has no unit. N represents the number of non-abnormal amplitude values within the aperture, expressed in units of ; Weight is the weighted value used in the calculation and has no unit. After spatially weighted median filtering, the signal-to-noise ratio of the pre-stack CRP gather is significantly improved.
