A special geological body identification method based on RPCA background modeling

Abstract

The application belongs to the technical field of seismic signal processing and interpretation, and specifically discloses a special geological body identification method based on RPCA background modeling, which comprises the following steps: S1, a special geological body identification model is constructed based on an RPCA algorithm, and seismic data matrix is decomposed into a sparse sharp noise matrix and a low-rank structure information matrix; S2, the low-rank structure information matrix is calculated through an SVD decomposition algorithm, the sparse sharp noise matrix is calculated through a soft threshold algorithm, and the low-rank structure information matrix and the sparse sharp noise matrix are updated through iterative residuals until the residuals of the seismic data two-dimensional matrix, the sparse sharp noise matrix and the low-rank structure information matrix meet the set conditions; S3, the sparse sharp noise matrix is output, which is special geological body data, and the low-rank structure information matrix is output, which is a remaining stratum background. The application can identify special geological bodies, and improves the interpretation efficiency, accuracy and applicability of special geological bodies. The application is suitable for the identification of special geological bodies.

Description

Technical Field

[0001] This invention belongs to the field of seismic signal processing and interpretation technology, specifically a method for identifying special geological bodies based on RPCA background modeling. Background Technology

[0002] With increasing energy demand and the continuous development of oil and gas exploration, most of the world's large structural oil and gas reservoirs, such as anticline traps and fault traps, have been discovered. Geological exploration personnel are now shifting their focus to identifying hidden oil and gas reservoirs within special geological bodies such as fracture-cavity formations and river channels. Traditionally, the interpretation of these special geological bodies was largely done manually. However, with the deepening of oil and gas exploration and the increasing volume of seismic data, the efficiency of manual interpretation is no longer sufficient to meet production demands.

[0003] Currently, interpreters typically use multi-attribute volume coherence, seismic facies identification, wave dynamics characteristics, and lithology to assist in the interpretation of special geological bodies using existing technologies and software. However, manual interpretation is very inefficient, requiring a significant amount of time and effort. Identification is even more difficult and prone to large errors in low signal-to-noise ratio and complex geological structures, often necessitating manual correction after identification, which further reduces work efficiency. In recent years, some deep learning-based identification methods using forward modeling as labels have been developed and are beginning to be applied to seismic facies interpretation. However, because intelligent methods are still in their early stages and highly specific, current fracture-cavity identification methods cannot achieve accurate automatic interpretation results, and no specifically designed intelligent identification methods have been developed for volcanoes and river channels. Summary of the Invention

[0004] The purpose of this invention is to provide a special geological body identification method based on RPCA background modeling, so as to achieve accurate automatic identification and interpretation of special geological bodies such as fissures, volcanoes, and river channels.

[0005] To achieve the above objectives, the present invention employs the following technical methods:

[0006] A method for identifying special geological bodies based on RPCA background modeling includes the following steps performed sequentially:

[0007] S1. Construct a special geological body identification model based on the RPCA algorithm. Decompose the seismic data of the target work area into a two-dimensional seismic data matrix and input it into the special geological body identification model. The two-dimensional seismic data matrix is ​​decomposed into a sparse sharp noise matrix and a low-rank structure information matrix.

[0008] S2. Calculate the low-rank structure information matrix using the SVD decomposition algorithm; calculate the sparse sharp noise matrix using the soft thresholding algorithm; update the low-rank structure information matrix and the sparse sharp noise matrix by iterating the residuals between the two-dimensional seismic data matrix and the sparse sharp noise matrix and the low-rank structure information matrix until the residuals between the two-dimensional seismic data matrix and the sparse sharp noise matrix and the low-rank structure information matrix meet the set threshold.

[0009] S3. Output the sparse sharp noise matrix and the low-rank structure information matrix respectively. The output sparse sharp noise matrix is ​​the data of special geological bodies, and the low-rank structure information matrix is ​​the background of the remaining strata.

[0010] As a limitation: in step S1, the seismic data of the target work area is decomposed into a two-dimensional seismic data matrix along the longitudinal or transverse slices of the data volume. The special geological body identification model is as follows:

[0011]

[0012] In the formula, ||A|| * λ is the kernel norm of matrix A, ||E||1 is the 1 norm of matrix E, E is the sparse sharp noise matrix, A is the low-rank structural information matrix, D is the two-dimensional matrix of seismic data of the target work area; λ is the weight of the sparse sharp noise related to the modeling intensity parameters. Where m is the number of columns in the seismic data matrix of the target work area, n is the number of rows in the seismic data matrix of the target work area, and a is the modeling intensity parameter, 0 < a < 1.

[0013] As a further limitation: In step S2, when the computer hardware performance cannot handle the amount of data in the two-dimensional matrix of seismic data of the target work area, before calculating the low-rank structure information matrix and the sparse sharp noise matrix, the two-dimensional matrix of seismic data of the target work area is first divided into multiple matrix blocks along the vertical direction, with overlapping parts between adjacent matrix blocks. The calculation is performed on each matrix block of the seismic data of the target work area, and each matrix block is decomposed into a sparse sharp noise matrix and a low-rank structure information matrix; when the computer hardware performance can handle the amount of data in the two-dimensional matrix of seismic data of the target work area, the two-dimensional matrix of seismic data of the target work area is directly decomposed into a sparse sharp noise matrix and a low-rank structure information matrix.

[0014] As a further clarification: Step S3 specifically involves the following steps: If the two-dimensional seismic data matrix of the target work area was segmented in step S2, then the calculation results of the seismic data of each matrix block obtained in step S2 are superimposed. The weight of the non-overlapping part of each block is 1, the weight of the superimposed part of the left block of the adjacent block from left to right is 1-g / G, and the weight of the superimposed part of the right block of the adjacent block from left to right is g / G, where G is the number of horizontal sampling points in the overlapping part of the adjacent blocks, and 0≤g≤G. The sparse sharp noise matrix calculated for each block is multiplied by its corresponding weight and then added to output the sparse sharp noise matrix E. The low-rank structure information matrix calculated for each block is multiplied by its corresponding weight and then added to output the low-rank structure information matrix A. If the two-dimensional seismic data matrix of the target work area was not segmented in step S2, then the calculated sparse sharp noise matrix is ​​the sparse sharp noise matrix E, and the calculated low-rank structure information matrix is ​​the low-rank structure information matrix A. The output sparse sharp noise matrix E is the special geological body data, and the low-rank structure information matrix A is the remaining stratigraphic background.

[0015] As another limitation: in step S2, the low-rank structure information matrix is ​​calculated using the SVD decomposition algorithm, specifically as follows:

[0016] If m ≤ n, then define matrix X. m*n ,

[0017] X m*n =D m*n -E m*n (1)

[0018] Let matrix X m*n Multiplying it by its own transpose yields a symmetric matrix C. m*m ,

[0019] X m*n *X m*n T =C m*m (2)

[0020] For C m*m Perform finite eigenvalue decomposition to obtain the first k principal eigenvectors U. m*k and principal eigenvalues ​​D k ,

[0021]

[0022] For the principal eigenvalue D k Taking the square root of each point in the matrix, we get S. k ,

[0023]

[0024] X m*n Transpose multiplied by Um*k , thus obtaining V′ n*k ,

[0025] V′ n*k =X m*n T *U m*k (5)

[0026] For V′ n*k Divide all n values ​​in each row by S k The number at the corresponding position in the middle is obtained.

[0027]

[0028] V′ n*k After all rows are calculated according to formula (6), V is obtained. n*k U m*k Multiply by S k Multiply by V n*k Transpose to get X m*n Low-rank information matrix

[0029]

[0030] That is, A m*n ,

[0031]

[0032] In the above formula, D k It is C m*m The main eigenvalues, D k yes All eigenvalues, U m*k It is C m*m The main feature vector, U m*k yes All feature vectors, U m*k yes The left singular vector, S k yes The singular value, V n*k T yes The right singular vector, V′ n*k Each row contains n values. S represents k Zhongyu The number at the corresponding position; k = 0.3 × min(m, n), representing the first 30% of the eigenvalues ​​after decomposition;

[0033] If m > n, then define matrix X. n*m ,

[0034] T

[0035] X n*m =(D m*n -E m*n (9)

[0036] Let matrix X n*m Multiplying it by its own transpose yields a symmetric matrix C. n*n ,

[0037] X n*m *X n*m T =C n*n (10)

[0038] For C n*n Perform finite eigenvalue decomposition to obtain the first k principal eigenvectors U. n*k and principal eigenvalues ​​D k ,

[0039]

[0040] For the principal eigenvalue D k Taking the square root of each point in the matrix, we get S. k ,

[0041]

[0042] X n*m Transpose multiplied by U n*k , thus obtaining V′ m*k ,

[0043] V′ m*k =X n*m T *U n*k (13)

[0044] For V′ m*k Divide all m values ​​in each row by S k The number at the corresponding position in the middle is obtained.

[0045]

[0046] V′ m*k After all rows of formula (14) are calculated, V is obtained. m*k U n*k Multiply by S k Multiply by V m*k Transpose to get X n*m Low-rank information matrix

[0047]

[0048] The transpose of A is A m*n ,

[0049]

[0050] In the above formula, D k It is C n*n The main eigenvalues, D k yes All eigenvalues, U n*k It is C n*n The main feature vector, U n*k yes All feature vectors, U n*k yes The left singular vector, S k yes The singular value, V m*k T yes The right singular vector, V′ m*k Each row contains m values. S represents k Zhongyu The number at the corresponding position; k = 0.3 × min(m, n), representing the first 30% of the eigenvalues ​​after decomposition;

[0051] The sparse, sharp noise matrix E is calculated using a soft thresholding algorithm. m*n Specifically:

[0052] (When P1 < 0, P1 = 0; when P1 > 0, retain the original value of P1)

[0053] (When P2>0, P2=0; when P2<0, retain the original value of P2)

[0054] E m*n =P1+P2

[0055] Among them, S Y The sum of Y is the weighted sum of Y from each iteration, where Y is the residual, and Y = D. m*n -A m*n -E m*n ; S0 is D m*n The maximum component of the singular value decomposition, where x is the number of iterations;

[0056] The residuals between the two-dimensional seismic data matrix and the sparse sharp noise matrix and the low-rank structure information matrix are iteratively updated until the residual Y between the two-dimensional seismic data matrix and the sparse sharp noise matrix and the low-rank structure information matrix satisfies the set conditions.

[0057] As a further constraint: the residual Y of the seismic data two-dimensional matrix, the sparse sharp noise matrix, and the low-rank structure information matrix in step S2 satisfies the following condition: in, S d For matrix D m*n The standard deviation of each point is obtained from the statistics.

[0058] As another limitation, the method further includes step S4, which specifically involves recombining the two-dimensional special geological body data and the two-dimensional remaining stratigraphic background obtained in step S3 into a three-dimensional special geological body and a three-dimensional remaining stratigraphic background, respectively.

[0059] The beneficial effects achieved by this invention, due to the adoption of the above-described solution, compared with the prior art, are as follows:

[0060] This invention provides a special geological body identification method based on RPCA background modeling. It decomposes the seismic data of the target work area into a two-dimensional seismic data matrix by slicing the data volume longitudinally or laterally. This method can identify volumetric or surface attributes of special geological bodies such as fractures, volcanoes, and river channels. During longitudinal slicing, it can analyze the data independently in two directions, combine the results from both directions, or analyze along any specified line direction. By adjusting the modeling intensity parameters, it can be applied to actual seismic data from different work areas and with varying data characteristics. By using SVD decomposition of some eigenvalues ​​and adding some simple matrix operations, the size of subsequent computational matrices and the amount of computation are reduced, thereby reducing hardware resource consumption and significantly improving computational efficiency. Compared with existing deep learning methods, this method improves the interpretation efficiency, accuracy, and applicability of special geological bodies. It is applicable to complex geological conditions and low seismic data quality, providing reliable data for subsequent geological modeling and reservoir prediction.

[0061] This invention is applicable to the identification of special geological bodies. Attached Figure Description

[0062] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0063] Figure 1 This is a flowchart illustrating a special geological body identification method based on RPCA background modeling, according to an embodiment of the present invention.

[0064] Figure 2 The effect of layer flattening on the results in this embodiment of the invention;

[0065] Figure 3 This is a schematic diagram of the block-based weighted superposition method according to an embodiment of the present invention;

[0066] Figure 4 This is a schematic diagram illustrating the effect of modeling intensity parameters on the contrast intensity between special geological bodies and the background of remaining strata in a seismic data slice according to an embodiment of the present invention.

[0067] Figure 5 This is a schematic diagram of a partial seismic data slice input for a work area and a partial special geological body output for a certain work area according to an embodiment of the present invention;

[0068] Figure 6 This is a schematic diagram of a partial seismic data slice input for a work area and a partial special geological body output for a certain work area according to an embodiment of the present invention. Detailed Implementation

[0069] The present invention will be further described below with reference to the embodiments. However, those skilled in the art should understand that the present invention is not limited to the following embodiments. Any improvements and equivalent changes made based on the specific embodiments of the present invention are within the scope of protection of the claims of the present invention.

[0070] Example: A method for identifying special geological bodies based on RPCA background modeling

[0071] A special geological body identification method based on RPCA background modeling, such as Figure 1 As shown, the steps are performed sequentially:

[0072] S1. Construct a special geological body identification model based on the RPCA algorithm, decomposing the seismic data of the target work area into a two-dimensional matrix D along the longitudinal or transverse slices of the data volume; before decomposing the seismic data of the target work area along the longitudinal slices of the data volume, layer-by-layer flattening processing can be performed to cope with complex geological conditions, such as... Figure 2 As shown; the two-dimensional matrix D of seismic data of the target work area is input into the special geological body identification model, and the two-dimensional matrix D is decomposed into a sparse sharp noise matrix E and a low-rank structural information matrix A;

[0073] The special geological body identification model is as follows:

[0074]

[0075] In the formula, ||A|| * λ is the nuclear norm of matrix A, ||E||1 is the 1 norm of matrix E, E is the sparse sharp noise matrix, A is the low-rank structural information matrix, D is the seismic data matrix of the target work area; λ is the weight of the sparse sharp noise related to the modeling intensity parameters. Where m is the number of columns in the seismic data matrix of the target work area, n is the number of rows in the seismic data matrix of the target work area, and a is the modeling intensity parameter, 0 < a < 1;

[0076] S2. When the computer hardware performance cannot handle the data volume of the two-dimensional seismic data matrix of the target work area, the two-dimensional seismic data matrix of the target work area is divided into blocks along the vertical direction, and the block division method is as follows: Figure 3 As shown, the data is divided into multiple matrix blocks. The portion between adjacent solid lines constitutes one matrix block, and the portion between adjacent dashed lines constitutes another matrix block. There is a 50% overlap between adjacent matrix blocks. Calculations are performed on each matrix block of the seismic data for the target work area. Figure 3 The values ​​represented by the dashed and solid lines in the upper part are related to... Figure 3 The weights corresponding to the lower matrix blocks are decomposed into a sparse sharp noise matrix and a low-rank structure information matrix for each matrix block. When the computer hardware performance can handle the amount of data in the two-dimensional matrix of seismic data of the target work area, the two-dimensional matrix of seismic data of the target work area is directly decomposed into a sparse sharp noise matrix and a low-rank structure information matrix.

[0077] The low-rank structure information matrix is ​​calculated using the SVD decomposition algorithm, specifically as follows:

[0078] If m ≤ n, then define matrix X. m*n ,

[0079] X m*n =D m*n -E m*n (1)

[0080] Let matrix X m*n Multiplying it by its own transpose yields a symmetric matrix C. m*m ,

[0081] X m*n *X m*n T =C m*m (2)

[0082] For C m*m Perform finite eigenvalue decomposition to obtain the first k principal eigenvectors U. m*k and principal eigenvalues ​​D k ,

[0083]

[0084] For the principal eigenvalue D k Taking the square root of each point in the matrix, we get S. k ,

[0085]

[0086] X m*nTranspose multiplied by U m*k , thus obtaining V′ n*k ,

[0087] V′ n*k =X m*n T *U m*k (5)

[0088] For V′ n*k Divide all n values ​​in each row by S k The number at the corresponding position in the middle is obtained.

[0089]

[0090] V′ n*k After all rows are calculated according to formula (6), V is obtained. n*k U m*k Multiply by S k Multiply by V n*k Transpose to get X m*n Low-rank information matrix

[0091]

[0092] That is, A m*n ,

[0093]

[0094] In the above formula, D k It is C m*m The main eigenvalues, D k yes All eigenvalues, U m*k It is C m*m The main feature vector, U m*k yes All feature vectors, U m*k yes The left singular vector, S k yes The singular value, V n*k T yes The right singular vector, V′ n*k Each row contains n values. S represents k Zhongyu The number at the corresponding position; k = 0.3 × min(m, n), representing the first 30% of the eigenvalues ​​after decomposition;

[0095] If m > n, then define matrix X. n*m ,

[0096] T

[0097] X n*m =(D m*n -E m*n (9)

[0098] Let matrix X n*m Multiplying it by its own transpose yields a symmetric matrix C. n*n ,

[0099] X n*m *X n*m T =C n*n (10)

[0100] For C n*n Perform finite eigenvalue decomposition to obtain the first k principal eigenvectors U. n*k and principal eigenvalues ​​D k ,

[0101]

[0102] For the principal eigenvalue D k Taking the square root of each point in the matrix, we get S. k ,

[0103]

[0104] X n*m Transpose multiplied by U n*k , thus obtaining V′ m*k ,

[0105] V′ m*k =X n*m T *U n*k (13)

[0106] For V′ m*k Divide all m values ​​in each row by S k The number at the corresponding position in the middle is obtained.

[0107]

[0108] V′ m*k After all rows of formula (14) are calculated, V is obtained. m*k U n*k Multiply by S k Multiply by V m*k Transpose to get X n*m Low-rank information matrix

[0109]

[0110] The transpose of A is A m*n ,

[0111]

[0112] In the above formula, D k It is C n*n The main eigenvalues, D k yes All eigenvalues, U n*k It is C n*n The main feature vector, U n*k yes All feature vectors, U n*k yes The left singular vector, S k yes The singular value, V m*k T yes The right singular vector, V′ m*k Each row contains m values. S represents k Zhongyu The number at the corresponding position; k = 0.3 × min(m, n), representing the first 30% of the eigenvalues ​​after decomposition;

[0113] The sparse, sharp noise matrix E is calculated using a soft thresholding algorithm. m*n Specifically:

[0114] (When P1 < 0, P1 = 0; when P1 > 0, retain the original value of P1)

[0115] (When P2>0, P2=0; when P2<0, retain the original value of P2)

[0116] E m*n =P1+P2

[0117] Among them, S Y The sum of Y is the weighted sum of Y from each iteration, where Y is the residual, and Y = D. m*n -A m*n -E m*n ; S0 is D m*n The maximum component of the singular value decomposition, where x is the number of iterations;

[0118] The low-rank structure information matrix and the sparse sharp noise matrix are updated by iteratively calculating the residuals between the two-dimensional seismic data matrix, the sparse sharp noise matrix, and the low-rank structure information matrix, until the residuals between the two-dimensional seismic data matrix and the sparse sharp noise matrix and the low-rank structure information matrix satisfy the following conditions: in, S d For matrix D m*n The standard deviation of each point is obtained statistically.

[0119] S3. If the two-dimensional matrix of seismic data for the target work area was divided into blocks in step S2, then the calculation results of the seismic data for each matrix block obtained in step S2 are superimposed. The weight of the non-overlapping part of each block is 1. The weight of the superimposed part of the left block of the adjacent block from left to right is 1-g / G, and the weight of the superimposed part of the right block of the adjacent block from left to right is g / G, where G is the number of horizontal sampling points in the overlapping part of the adjacent blocks, 0≤g≤G; the sparse sharp noise matrix E calculated for each block m*n Multiply by their respective weights and then sum them to output a sparse, sharp noise matrix E, and the low-rank structure information matrix A calculated for each block. m*n Multiply by their respective weights and then sum them to output a low-rank structural information matrix A; if the two-dimensional matrix of seismic data for the target work area was not partitioned in step S2, the calculated sparse and sharp noise matrix E m*n That is, the sparse, sharp noise matrix E, and the calculated low-rank structure information matrix A. m*n This is the low-rank structure information matrix A; the output sparse sharp noise matrix E is the special geological body data, and the low-rank structure information matrix A is the remaining stratigraphic background.

[0120] S4. Re-synthesize the two-dimensional special geological body data and the two-dimensional remaining stratum background obtained in step S3 into a three-dimensional special geological body and a three-dimensional remaining stratum background, respectively.

[0121] The special geological body identification method based on RPCA background modeling in this embodiment is applied to a certain work area for special geological body identification. First, seismic data slices of the work area are input, and the modeling intensity parameter 'a' is continuously adjusted until the contrast intensity between the special geological body and the remaining stratigraphic background in the seismic data slices of the work area meets the interpreter's requirements. The effect of adjusting the modeling intensity parameter 'a' on the contrast intensity between the special geological body and the remaining stratigraphic background is as follows: Figure 4 As shown, the data for special geological bodies and the remaining stratigraphic background are then output according to the above steps. Figure 5 The upper part shows a slice of seismic data from the input work area, and the lower part shows a portion of the output special geological bodies, specifically fracture-cavity formations. Figure 6The left half shows a slice of seismic data from the input work area, and the right half shows a portion of the output special geological bodies, including volcanoes. Finally, the calculated two-dimensional special geological body data and the two-dimensional remaining stratigraphic background are recombined into three-dimensional special geological bodies and three-dimensional remaining stratigraphic background, respectively. The special geological bodies are then displayed in three dimensions using GeoEast software to perform quality control and obtain interpretation results.

Claims

1. A method for identifying special geological bodies based on RPCA background modeling, characterized in that, include: S1. Construct a special geological body identification model based on the RPCA algorithm. Decompose the seismic data of the target work area into a two-dimensional seismic data matrix and input it into the special geological body identification model. The two-dimensional seismic data matrix is ​​decomposed into a sparse sharp noise matrix and a low-rank structure information matrix. In step S1, the seismic data of the target work area is decomposed into a two-dimensional seismic data matrix along the longitudinal or transverse slices of the data volume. The special geological body identification model is as follows: In the formula, ||A|| * Let ||E||1 be the kernel norm of matrix A, ||E||1 be the 1 norm of matrix E, E be a sparse sharp noise matrix, A be a low-rank structure information matrix, and D be a two-dimensional matrix of seismic data for the target work area. λ is the weight of the sparse sharp noise matrix related to the modeling intensity parameter; Where m is the number of columns in the two-dimensional matrix of seismic data of the target work area, n is the number of rows in the two-dimensional matrix of seismic data of the target work area, and a is the modeling intensity parameter, 0 < a < 1; S2. Calculate the low-rank structure information matrix using the SVD decomposition algorithm; calculate the sparse sharp noise matrix using the soft thresholding algorithm; update the low-rank structure information matrix and the sparse sharp noise matrix by iterating the residuals between the two-dimensional seismic data matrix and the sparse sharp noise matrix and the low-rank structure information matrix until the residuals between the two-dimensional seismic data matrix and the sparse sharp noise matrix and the low-rank structure information matrix meet the set conditions. In the SVD decomposition algorithm, only the first 30% of the feature values ​​are decomposed; Calculate the sparse sharp noise matrix using the soft thresholding algorithm. Specifically: in, The sum of Y is the weighted sum of Y from each iteration, where Y is the residual. ; , for The maximum component of the singular value decomposition, where x is the number of iterations; The conditions set are as follows: ,in, ,S d For matrix The standard deviation of each point in the statistical analysis is obtained; In step S2, when the computer hardware performance cannot handle the amount of data in the two-dimensional matrix of seismic data of the target work area, before calculating the low-rank structure information matrix and the sparse sharp noise matrix, the two-dimensional matrix of seismic data of the target work area is first divided into multiple matrix blocks along the vertical direction. There is an overlap between adjacent matrix blocks. The calculation is performed on each matrix block of the seismic data of the target work area, and each matrix block is decomposed into a sparse sharp noise matrix and a low-rank structure information matrix. S3. Output the sparse sharp noise matrix and the low-rank structure information matrix respectively. The output sparse sharp noise matrix is ​​the data of special geological bodies, and the low-rank structure information matrix is ​​the background of the remaining strata.

2. The method for identifying special geological bodies based on RPCA background modeling according to claim 1, characterized in that, In step S2, when the computer hardware performance can handle the amount of data in the two-dimensional matrix of seismic data of the target work area, the two-dimensional matrix of seismic data of the target work area is directly decomposed into a sparse sharp noise matrix and a low-rank structure information matrix.

3. The method for identifying special geological bodies based on RPCA background modeling according to claim 2, characterized in that, Step S3 is as follows: If the two-dimensional matrix of seismic data for the target work area was segmented in step S2, the calculation results of the seismic data for each matrix block obtained in step S2 are superimposed. The weight of the non-overlapping part of each block is 1. The weight of the superimposed part of the left block of the adjacent block from left to right is 1-g / G. The weight of the superimposed part of the right block of the adjacent block from left to right is g / G. G is the number of sampling points in the horizontal direction of the overlapping part of the adjacent blocks, 0≤g≤G. The sparse sharp noise matrix calculated for each block is multiplied by its corresponding weight and then added to output the sparse sharp noise matrix E. The low-rank structure information matrix calculated for each block is multiplied by its corresponding weight and then added to output the low-rank structure information matrix A. If the two-dimensional matrix of seismic data for the target work area was not segmented in step S2, the calculated sparse sharp noise matrix is ​​the sparse sharp noise matrix E, and the calculated low-rank structure information matrix is ​​the low-rank structure information matrix A. The output sparse sharp noise matrix E is the special geological body data, and the low-rank structure information matrix A is the remaining stratigraphic background.

4. A method for identifying special geological bodies based on RPCA background modeling according to any one of claims 1-3, characterized in that, In step S2, the low-rank structure information matrix is ​​calculated using the SVD decomposition algorithm, specifically as follows: If m ≤ n, then define a matrix. , （1） matrix Multiplying it by its own transpose yields a symmetric matrix. , right Perform finite eigenvalue decomposition to obtain the first k principal eigenvectors. and principal eigenvalues , （3） For the main eigenvalues Taking the square root of each point in the matrix, we get... , （4） Transpose ,get , （5） right Divide all n values ​​in each row The number at the corresponding position in the middle is obtained. , After all rows are calculated according to formula (6), we get , take take Transpose, and you get Low-rank information matrix , Right now , In equations (1) to (8), yes The main eigenvalues, yes All eigenvalues, yes The main feature vectors, yes All feature vectors, yes The left singular vector, yes singular values, yes The right singular vector, express Each row contains n values. express Zhongyu The number at the corresponding position; , represents the first 30% of eigenvalues ​​after decomposition; If m > n, then define the matrix. , matrix Multiplying it by its own transpose yields a symmetric matrix. , right Perform finite eigenvalue decomposition to obtain the first k principal eigenvectors. and principal eigenvalues , （11） For the main eigenvalues Taking the square root of each point in the matrix, we get... , （12） Transpose ,get , （13） right Divide all m values ​​in each row The number at the corresponding position in the middle is obtained. , After all rows of formula (14) are calculated, we get , take take Transpose, and you get Low-rank information matrix , The transpose of is , （16） In equations (9) to (16), yes The main eigenvalues, yes All eigenvalues, yes The main feature vectors, yes All feature vectors, yes The left singular vector, yes singular values, yes The right singular vector, express Each row contains m values. express Zhongyu The number at the corresponding position; , represents the first 30% of eigenvalues ​​after decomposition.

5. A method for identifying special geological bodies based on RPCA background modeling according to any one of claims 1-3, characterized in that, The method also includes step S4, which specifically involves recombining the two-dimensional special geological body data and the two-dimensional remaining stratigraphic background obtained in step S3 into a three-dimensional special geological body and a three-dimensional remaining stratigraphic background, respectively.

Images