Fracture detection method and device

Abstract

The invention discloses a fracture detection method and device, and relates to the technical field of oil and gas reservoir exploration, and the key points of the technical scheme are that an image classification problem is converted into an image segmentation problem through carrying out deep learning trunk fracture recognition based on deep learning image segmentation; a U-Net-based deep learning network is used to realize high-precision identification of fractures; on the basis of structure tensor and disorder degree detection, development characteristics of minor faults and cracks around the fault are engraved; and based on the conventional coherence and amplitude curvature, identifying a trunk fracture and a part of minor faults. According to the fracture detection method and device suitable for the complex strike-slip fracture development area, fracture identification of the complex strike-slip fracture development area is rapidly and accurately achieved, and effective technical support is provided for deployment of fracture control type reservoir body well drilling.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas reservoir exploration, and more specifically, to a fracture detection method and device. Background Art

[0002] Accurately identifying the fracture system is an important part of seismic data interpretation, and the interpretation efficiency and quality directly affect the progress of oil and gas exploration and development work. In recent years, "fault-karst body" oil and gas reservoirs related to deep and large strike-slip fault zones have been continuously discovered in the Tarim Basin, and many high-yield oil and gas wells have emerged, showing the huge potential of this unconventional oil and gas reservoir. The fault-karst body oil and gas reservoir in the Tarim Basin is a series of special oil and gas reservoirs distributed along the strike-slip fault zone formed after the late oil and gas charging, after the large strike-slip fault fracture zone that has been active for a long time (from the Cambrian to the Silurian) is modified by deep acid solution to form a fracture-cavity type reservoir. The Ordovician, the target layer of the oil and gas reservoir, generally has a burial depth exceeding 7000m, and has the characteristics of deep burial, strong fracture control, diverse reservoir types, and complex reservoir control factors. Affected by the low resolution of seismic data, strong heterogeneity of carbonate reservoirs, and complex distribution laws of fractures and cavities, the fracture zone in this area is characterized by difficult boundary determination and unclear internal characteristics, and it is difficult to identify and describe the fault-controlled reservoir body.

[0003] For the above reasons, the present invention provides a fracture detection method and device applicable to areas with complex strike-slip fractures, so as to quickly and accurately identify fractures in areas with complex strike-slip fractures and provide effective guidance for actual production. Summary of the Invention

[0004] To achieve the above object, the present invention provides the following technical solution: A fracture detection method, including the following steps:

[0005] S1. According to the seismic data of the research area, perform deep learning based on a convolutional neural network to construct a three-dimensional fracture probability volume.

[0006] S2. Based on the three-dimensional amplitude data volume of the research area, construct a structure tensor matrix, and calculate to obtain a three-dimensional structure tensor attribute volume.

[0007] S3. Based on the structure tensor matrix, obtain the corresponding structure tensor eigenvalues λ1, λ2, λ3 at each location, where λ1≥λ2≥λ3>0.

[0008] Construct a three-dimensional clutter attribute volume, and the clutter S calculation formula is as follows:

[0009]

[0010] S4. Perform an inverse transformation on the three-dimensional clutter attribute volume to obtain an inverse three-dimensional clutter attribute volume, and the inverse transformation formula is as follows:

[0011] newS = max(S) - (S * (max(S) - min(S)))

[0012] In the formula, newS is the reverse disorder degree, max(S) is the maximum value of the disorder degree, min(S) is the minimum value of the disorder degree, and S is the disorder degree.

[0013] S5. Generate a three-dimensional coherence attribute volume and a three-dimensional amplitude curvature attribute volume based on the seismic data in the research area.

[0014] S6. Normalize the three-dimensional fracture probability volume, the three-dimensional tensor thinning attribute volume, the reverse three-dimensional disorder degree attribute volume, the three-dimensional coherence attribute volume, and the three-dimensional amplitude curvature attribute volume and map them to 0 - 1. Add the normalized three-dimensional fracture probability volume, the three-dimensional tensor thinning attribute volume, the reverse three-dimensional disorder degree attribute volume, the three-dimensional coherence attribute volume, and the three-dimensional amplitude curvature attribute volume to construct a three-dimensional fusion attribute volume. The greater the fusion attribute value, the higher the probability of the occurrence of faults, fractures, and reservoirs.

[0015] The present invention is further configured such that in S3, each component of the structure tensor matrix is filtered. Smooth filtering is performed in the direction parallel to the fault plane, and sharpening filtering is performed in the direction perpendicular to the fault plane. The eigenvalues λ1, λ2, and λ3 are obtained from the structure tensor matrix after filtering.

[0016] The present invention is further configured such that in S3, the Laplacian of Gaussian operator is used for sharpening filtering. The LoG formula of the Laplacian of Gaussian operator is as follows:

[0017]

[0018] In the formula, ε 1 、ε 2 、ε 3 are respectively the three-dimensional seismic body line, the trace direction, and the two-way travel time direction, and σ 1 、σ 2 、σ 3 are respectively the standard deviations in the directions of ε 1 、ε 2 、ε 3 , and σ 1 、σ 2 、σ 3 are given values and satisfy

[0019] The present invention is further configured such that the result after filtering with the Laplacian of Gaussian operator in S3 is used as the input, and after more than three iterations, the sharpening filtering result is output.

[0020] The present invention is further configured such that: The specific steps of S2 are: constructing a structure tensor matrix based on the three-dimensional amplitude data volume of the research area. For each point, calculate the structure tensor values in different directions, and take the maximum structure tensor value as the structure tensor value of this point, and finally form a three-dimensional structure tensor attribute volume.

[0021] The present invention is further configured such that: Before constructing the structure tensor matrix, perform smoothing filtering on the three-dimensional amplitude data volume of the research area.

[0022] The present invention is further configured such that: Perform standardization processing on the three-dimensional tensor thinning attribute volume and the reverse three-dimensional clutter degree attribute volume, add the standardized three-dimensional tensor thinning attribute volume and the reverse three-dimensional clutter degree attribute volume, and construct a three-dimensional skeletonization attribute volume. Given a threshold value A, use the skeletonization attribute value greater than A as the judgment criterion for small faults and fractures around the fault.

[0023] The present invention is further configured such that: Given a threshold value a1 based on the three-dimensional tensor thinning attribute volume, given a threshold value a2 based on the reverse three-dimensional clutter degree attribute volume, and the sum of the standardized a1 and a2 is A.

[0024] The present invention is further configured such that: Perform standardization processing on the three-dimensional coherence attribute volume and the three-dimensional amplitude curvature attribute volume, add the standardized three-dimensional coherence attribute volume and the three-dimensional amplitude curvature attribute volume, and construct a fracture spatial distribution attribute volume. Given a threshold value B, use the fracture spatial distribution attribute value greater than B as the judgment criterion for main fractures and small faults.

[0025] The present invention is further configured such that: Given a threshold value b1 based on the three-dimensional coherence attribute volume, given a threshold value b2 based on the three-dimensional amplitude curvature attribute volume, and the sum of the standardized b1 and b2 is B.

[0026] The present invention is further configured such that: Calculate the derivative of the three-dimensional structure tensor attribute volume along the direction perpendicular to the fault, and interpolate the position where the derivative is 0, which is the position of the fault ridge.

[0027] The present invention is further configured such that: Calculate the derivative of the three-dimensional fracture probability volume along the direction perpendicular to the fault, and interpolate the position where the derivative is 0, which is the position of the fault ridge.

[0028] The present invention is further configured such that: Based on the seismic data of the research area, establish a three-dimensional seismic model of the research area. Use a given 3D selection box to pick up the fracture development information in the three-dimensional seismic model as the original sample. Generate multiple different fracture development information through random model simulation according to the original sample as the synthetic samples. Use the original sample as the training set, and the synthetic samples as the training set and the test set, perform deep learning based on the U-Net convolutional neural network, obtain the probability distribution of fractures existing in each part of the three-dimensional seismic model, and generate a three-dimensional fracture probability volume.

[0029] The present invention is further configured such that: in S1, 3Dwidget is used to pick up the fracture development information in the three-dimensional seismic model.

[0030] The present invention also provides a fracture detection device, including a storage medium and a processor, and a computer program is stored on the storage medium. The processor is used to implement the above-mentioned fracture detection method when executing the computer program.

[0031] In summary, compared with the prior art, the present invention has the following beneficial effects: by carrying out deep learning backbone fracture recognition based on deep learning image segmentation, the present invention converts the image classification problem into an image segmentation problem, and uses a deep learning network based on U-Net to achieve high-precision fracture recognition; based on structure tensor and clutter detection, the development characteristics of small faults and fractures around the carved fault are characterized; based on conventional coherence and amplitude curvature, the main fractures and some small faults are identified. The present invention provides a fracture detection method and device applicable to complex strike-slip fracture development areas, so as to quickly and accurately realize fracture recognition in complex strike-slip fracture development areas, and provide effective technical support for the deployment of drilling for fault-controlled reservoirs. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic flow chart of the fracture detection method in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] The technical solutions of the present invention will be clearly described below in conjunction with the description of the drawings. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the invention.

[0034] Embodiment

[0035] As Figure 1 shown, a fracture detection method provided in a preferred embodiment of the present invention includes the following steps:

[0036] S1. Based on the seismic data of the research area, a three-dimensional seismic model of the research area is established. Use a given 3D selection box to pick up the fracture development information in the three-dimensional seismic model as the original sample. Multiple different fracture development information is generated by random model simulation according to the original sample as the synthetic sample. The original sample is used as the training set, and the synthetic sample is used as the training set and the test set to perform deep learning based on the U-Net convolutional neural network to obtain the probability distribution of fractures existing in each part of the three-dimensional seismic model, and generate a three-dimensional fracture probability volume. In this embodiment, 3Dwidget is used to pick up the fracture development information in the three-dimensional seismic model

[0037] Calculate the derivative of the three-dimensional fracture probability volume along the direction perpendicular to the fault, and interpolate the positions where the derivative is 0, which are the positions of the fault ridges. For each image sample, the direction perpendicular to the fault plane can be calculated from the scanned fault dip angle image to maximize the accuracy of fault identification.

[0038] S2. Based on the three-dimensional amplitude data volume of the study area, construct a structure tensor matrix: Taking a certain point as the center, create a matrix O in any direction, calculate the covariance matrix R of O, obtain the gradient vector g by differentiating R along the x, y, and z directions respectively, calculate the covariance matrix T of g, and T is the structure tensor; Rotate the eigenvectors of T along the given direction v to obtain T', and calculate the eigenvalues λ1, λ2, λ3 of T', where λ1≥λ2≥λ3>0; In this embodiment, the direction v is the direction perpendicular to the fault plane. For each point, calculate the structure tensor values in different directions and take the maximum structure tensor value as the structure tensor value of this point, and finally form a three-dimensional structure tensor attribute volume to identify the fractures around the fault. Calculate the derivative of the three-dimensional structure tensor attribute volume along the direction perpendicular to the fault, and interpolate the positions where the derivative is 0, which are the positions of the fault ridges. For each image sample, the direction perpendicular to the fault plane can be calculated from the scanned fault dip angle image to maximize the accuracy of fault identification.

[0039] To construct the structure tensor matrix, it is necessary to perform image enhancement processing on the three-dimensional amplitude data volume of the study area, that is, set the corresponding time window and trace number range, and perform smoothing filtering on the three-dimensional amplitude data volume of the study area.

[0040] S3. Obtain the corresponding structure tensor eigenvalues λ1, λ2, λ3 at each location based on the structure tensor matrix, where λ1≥λ2≥λ3>0.

[0041] Construct a three-dimensional clutter attribute volume based on the amplitude gradient vector to carve the development characteristics of small faults and fractures around the fault. The clutter S calculation formula is as follows:

[0042]

[0043] S4. Perform an inverse transformation on the three-dimensional clutter attribute volume to obtain the inverse three-dimensional clutter attribute volume. The inverse transformation formula is as follows:

[0044] newS = max(S) - (S * (max(S) - min(S)))

[0045] In the formula, newS is the inverse clutter, max(S) is the maximum clutter, min(S) is the minimum clutter, and S is the clutter. Determine the azimuth and dip angle corresponding to the maximum inverse clutter as the azimuth and dip angle of the fault plane to infer the stress direction and magnitude on the object or structure when the fracture occurs, so as to determine the fracture mechanism.

[0046] Standardize the 3D tensor thinning attribute volume and the reverse 3D clutter degree attribute volume, convert the data to the same scale, add the standardized 3D tensor thinning attribute volume and the reverse 3D clutter degree attribute volume to construct a 3D skeletonization attribute volume. Given a threshold value A, use the condition that the skeletonization attribute value is greater than A as the criterion for judging small faults and fractures around the fault.

[0047] Combined with the actual well drilling data and geological exploration data, based on the 3D tensor thinning attribute volume combined with the actual well drilling data and geological exploration data, a threshold value a1 is given, and based on the reverse 3D clutter degree attribute volume, a threshold value a2 is given. The sum of the standardized a1 and a2 is A.

[0048] S5. Generate a 3D coherence attribute volume and a 3D amplitude curvature attribute volume based on the seismic data in the study area.

[0049] The 3D coherence attribute volume is obtained by dividing the study area into multiple 3D grids. Each 3D grid contains multiple seismic traces. Calculate the coherence attribute values of adjacent seismic traces in each 3D grid, and use the maximum or minimum coherence attribute value in each 3D grid as the coherence attribute value of the 3D grid, and output the 3D coherence attribute volume of the study area.

[0050] The 3D amplitude curvature attribute volume is obtained by taking the second-order lateral derivative of the seismic data amplitude. First, calculate the first-order derivatives in the main survey line direction and the tie survey line direction using the seismic amplitude or energy. The obtained energy gradient attribute itself can reflect abnormal geological bodies, which is generally called the amplitude energy gradient. Then, take the second-order derivative of it to obtain the amplitude surface, and finally calculate each amplitude curvature attribute according to the fitting of this surface. In principle, the amplitude energy gradient is the manifestation of the edge of the geological body, so the spatial distribution of the geological body cannot be obtained through the threshold value. While the amplitude curvature converts the amplitude energy gradient into an attribute reflecting the envelope of the geological body, and the spatial distribution of the geological body can be obtained through the threshold value.

[0051] Standardize the 3D coherence attribute volume and the 3D amplitude curvature attribute volume, convert the data to the same scale, add the standardized 3D coherence attribute volume and the 3D amplitude curvature attribute volume to construct a fracture spatial distribution attribute volume. Given a threshold value B, use the condition that the fracture spatial distribution attribute value is greater than B as the criterion for judging major fractures and small faults.

[0052] Combined with the actual well drilling data and geological exploration data, based on the 3D coherence attribute volume, a threshold value b1 is given, and based on the 3D amplitude curvature attribute volume, a threshold value b2 is given. The sum of the standardized b1 and b2 is B.

[0053] S6. Normalize the three-dimensional fracture probability volume, three-dimensional tensor thinning attribute volume, reverse three-dimensional clutter degree attribute volume, three-dimensional coherence attribute volume, and three-dimensional amplitude curvature attribute volume and map them to 0 - 1. Add the normalized three-dimensional fracture probability volume, three-dimensional tensor thinning attribute volume, reverse three-dimensional clutter degree attribute volume, three-dimensional coherence attribute volume, and three-dimensional amplitude curvature attribute volume to construct a three-dimensional fusion attribute volume. The higher the fusion attribute value, the higher the probability of the occurrence of faults, fractures, and reservoirs.

[0054] Specifically, in S3, each component of the structure tensor matrix is filtered. Smooth filtering is performed in the direction parallel to the fault plane, and sharpening filtering is performed in the direction perpendicular to the fault plane. The eigenvalues λ1, λ2, and λ3 are obtained from the structure tensor matrix after filtering.

[0055] Specifically, in S3, Gaussian filtering is used for smooth filtering to smooth the short-wavelength artifacts of the image. The Gaussian filtering formula is as follows:

[0056]

[0057] In the formula, x is the component to be filtered, μ is the mean of all X, σ is the variance of all x, and f(x) is the output result after filtering.

[0058] Specifically, in S3, the Laplacian of Gaussian (LoG) operator is used for sharpening filtering. The LoG formula of the Laplacian of Gaussian operator is as follows:

[0059]

[0060] In the formula, ε 1 、ε 2 、ε 3 are the three-dimensional seismic body line, trace direction, and two-way travel time direction respectively. σ 1 、σ 2 、σ 3 are the standard deviations in the directions of ε 1 、ε 2 、ε 3 respectively. σ 1 、σ 2 、σ 3 are given values and satisfy

[0061] Specifically, in S3, the result after filtering with the Laplacian of Gaussian operator is used as the input. After more than three iterations, the sharpened filtering result is output.

[0062] This embodiment also provides a fracture detection device, including a storage medium and a processor. A computer program is stored on the storage medium. The processor is used to implement the above fracture detection method when executing the computer program.

[0063] In summary, in this embodiment, by carrying out deep learning main fault fracture recognition based on deep learning image segmentation, the image classification problem is converted into an image segmentation problem, and a deep learning network based on U-Net is used to achieve high-precision fracture recognition; based on structure tensor and clutter detection, the development characteristics of small faults and fractures around the carved fault are detected; based on conventional coherence and amplitude curvature, the main faults and some small faults are identified. This embodiment provides a fault detection method and device applicable to complex strike-slip fault development areas to quickly and accurately identify faults in complex strike-slip fault development areas, providing effective technical support for the deployment of drilling for fault-controlled reservoirs.

[0064] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A fracture detection method, characterized in that: It includes the following steps: S1. Based on the seismic data in the study area, perform deep learning based on a convolutional neural network to construct a three-dimensional fracture probability volume; S2. Based on the three-dimensional amplitude data volume in the study area, construct a structure tensor matrix, and calculate to obtain a three-dimensional structure tensor attribute volume; S3. Obtain the corresponding structure tensor eigenvalues λ at each location based on the structure tensor matrix 1 , λ 2 , λ 3 , λ 1 ≥ λ 2 ≥ λ 3 > 0; Construct a three-dimensional clutter attribute volume, and the calculation formula for clutter S is as follows: S4. Perform an inverse transformation on the three-dimensional clutter attribute volume to obtain an inverse three-dimensional clutter attribute volume. The inverse transformation formula is as follows: newS = max(S) - (S * (max(S) - min(S))) In the formula, newS is the inverse clutter, max(S) is the maximum clutter value, min(S) is the minimum clutter value, and S is the clutter; S5. Based on the seismic data in the study area, generate a three-dimensional coherence attribute volume and a three-dimensional amplitude curvature attribute volume; S6. Normalize the three-dimensional fracture probability volume, three-dimensional tensor thinning attribute volume, inverse three-dimensional clutter attribute volume, three-dimensional coherence attribute volume, and three-dimensional amplitude curvature attribute volume and map them to 0-1. Add the normalized three-dimensional fracture probability volume, three-dimensional tensor thinning attribute volume, inverse three-dimensional clutter attribute volume, three-dimensional coherence attribute volume, and three-dimensional amplitude curvature attribute volume to construct a three-dimensional fusion attribute volume; the greater the fusion attribute value, the higher the probability of the occurrence of faults, fractures, and reservoirs.

2. A fracture detection method according to claim 1, characterized in that: In S3, each component of the structure tensor matrix is filtered. Smooth filtering is performed in the direction parallel to the cross-section, and sharpening filtering is performed in the direction perpendicular to the cross-section; the eigenvalues λ 1 , λ 2 , λ 3 are obtained from the structure tensor matrix after filtering.

3. A fracture detection method according to claim 2, characterized in that: S3 uses a Gaussian-Laplacian operator for sharpening filtering. The LoG formula of the Gaussian-Laplacian operator is as follows: where ε 1 , ε 2 , ε 3 are the line, trace direction and two-way travel time direction of the 3D seismic volume respectively, σ 1 , σ 2 , σ 3 are the standard deviations in the directions of ε 1 , ε 2 , ε 3 respectively, and σ 1 , σ 2 , σ 3 are given values and satisfy 4. A fracture detection method according to claim 3, characterized in that: The result after filtering with the Gaussian-Laplacian operator in S3 is used as the input. After more than three iterations, the sharpened filtering result is output.

5. A fracture detection method according to claim 1, characterized in that: The specific steps of S2 are: based on the three-dimensional amplitude data volume in the study area, construct a structure tensor matrix; for each point, calculate its structure tensor values in different directions, and take the maximum structure tensor value as the structure tensor value of this point, and finally form a three-dimensional structure tensor attribute volume.

6. A fracture detection method according to any one of claims 1-5, characterized in that: Perform smoothing filtering on the three-dimensional amplitude data volume in the study area before constructing the structure tensor matrix.

7. A fracture detection method according to claim 1, characterized in that: Perform standardization processing on the three-dimensional tensor thinning attribute volume and the inverse three-dimensional clutter attribute volume. Add the standardized three-dimensional tensor thinning attribute volume and the inverse three-dimensional clutter attribute volume to construct a three-dimensional skeletonization attribute volume; given a threshold value A, use the skeletonization attribute value greater than A as the judgment criterion for small faults and fractures around the fault.

8. A fracture detection method according to claim 7, characterized in that: Based on the three-dimensional tensor thinning attribute volume, a threshold value a1 is given, and based on the inverse three-dimensional clutter attribute volume, a threshold value a2 is given. The sum of the standardized a1 and a2 is A.

9. A fracture detection method according to claim 1, characterized in that: The three-dimensional coherence attribute volume and the three-dimensional amplitude curvature attribute volume are standardized, and the standardized three-dimensional coherence attribute volume and the three-dimensional amplitude curvature attribute volume are added to construct a fracture spatial distribution attribute volume; a threshold value B is given, and the fracture spatial distribution attribute value greater than B is used as the judgment criterion for main fractures and small faults.

10. A fracture detection method according to claim 9, characterized in that: Based on the three-dimensional coherence attribute volume, a threshold value b1 is given, and based on the three-dimensional amplitude curvature attribute volume, a threshold value b2 is given. The sum of the standardized b1 and b2 is B.

11. A fracture detection method according to claim 1, characterized in that: Calculate the derivative of the three-dimensional structure tensor attribute volume along the direction perpendicular to the fault, and interpolate the position where the derivative is 0, which is the position of the fault ridge.

12. A fracture detection method according to claim 1, characterized in that: Calculate the derivative of the three-dimensional fracture probability volume along the direction perpendicular to the fault, and interpolate the position where the derivative is 0, which is the position of the fault ridge.

13. A fracture detection method according to claim 1, characterized in that: Based on the seismic data in the study area, establish a three-dimensional seismic model of the study area; use a given 3D selection box to pick up the fracture development information in the three-dimensional seismic model as the original sample; generate multiple different fracture development information through random model simulation according to the original sample as the synthetic sample; use the original sample as the training set and the synthetic sample as the training set and the test set to perform deep learning based on the U-Net convolutional neural network to obtain the probability distribution of fractures existing in each part of the three-dimensional seismic model, and generate a three-dimensional fracture probability volume.

14. A fracture detection method according to claim 13, characterized in that: In S1, a 3D widget is used to pick up the fracture development information in the three-dimensional seismic model.

15. A fracture detection device, characterized in that: It includes a storage medium and a processor, and a computer program is stored on the storage medium; the processor is used to implement the fracture detection method according to any one of claims 1-14 when executing the computer program.

Images