A method for identifying shale lamination types based on electrical imaging logging

Through the preprocessing and clustering algorithm of the electro-imaging logging data, combined with stratification clustering and sinusoidal interface fitting, the problem of difficult formation interface recognition in the electro-imaging logging images in the prior art is solved, and efficient and accurate traces are achieved Layer group type recognition and thickness calculation improve the robustness and efficiency of recognition.

CN118470364BActive Publication Date: 2024-11-08CHONGQING UNIVERSITY OF SCIENCE AND TECHNOLOGY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410537891.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-30
Publication Date
2024-11-08
Estimated Expiration
2044-04-30

AI Technical Summary

Technical Problem

In the prior art, the formation interface on the electrical imaging logging image, especially the stratigraphic surface, is displayed by the slight changes in the grayscale in the layer. The distance between adjacent interfaces is very close and the edges are weak, edge detection, image segmentation, and edges. Tracking and matching image processing methods are difficult to accurately extract the trajectories of these interfaces. The Hough transformation method is time-consuming and labor-intensive, and it is even more difficult to identify the situation where low-quality image and image data are damaged. It is through cores, microsheets and X-divergence. Such techniques to divide types are time-consuming and laborious, and can only describe the core segment.

Method used

The shale layer group type recognition method based on electrical imaging logging is adopted. The electro-imaging plate image data is generated by loading and decomposing the electrical imaging data, and the image center cluster is initialized using the K-means++ algorithm to perform the Euro-type distance calculation and Clustering, combining the circular connection area marking algorithm and the stratification clustering algorithm, extract the stratification pixel points and perform sine stratification interface fitting, and divide the stratification layer based on the electric imaging log clustering response characteristic value range of different stratification groups. group, calculate developmental density parameters.

Benefits of technology

It realizes accurate identification and rapid extraction of stratigraphic boundaries, efficiently and accurately identify sinusoidal curves, avoiding the absence of strands between two sinusoidal curves, and adopts the efficient equivalent thickness method to directly avoid repeated strand calculations when calculating density. , improves the robustness and efficiency of identification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118470364B_ABST
    Figure CN118470364B_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of application of electrical imaging logging data, and in particular to a shale laminae group type identification method based on electrical imaging logging, which comprises the following steps: loading and decoding electrical imaging data, generating electrical imaging plate image data after preprocessing; initializing an image center cluster by using a K-means++ algorithm, iterating a new cluster center according to a mean after calculating a Euclidean distance, and completing clustering after reaching a threshold or a number limit; binarizing N cluster values ​​after clustering to obtain a head-to-tail connected matrix that complies with the wellbore imaging principle; then performing a cyclic connected region labeling algorithm, extracting pixel points of laminae by using a bedding clustering algorithm, and then performing sinusoidal bedding interface fitting; dividing shale gas reservoir laminae based on the characteristic value range of electrical imaging logging clustering responses of different laminae groups, and using an improved Levenberg-Marquardt algorithm to directly, quickly and accurately extract boundaries, and efficiently, accurately and robustly identify quasi-sine curves, and finally completing shale laminae group type identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of application of electrical imaging logging data, and in particular to a method for identifying shale lamination group types based on electrical imaging logging. Background Art

[0002] At present, lamination not only records the paleoclimate and paleowater properties during shale deposition, but its development also restricts the quality of shale reservoirs and the effect of shale hydraulic fracturing. It is another important evaluation index of shale reservoirs after total organic carbon content (TOC), effective porosity, brittle minerals and gas content.

[0003] In the existing technology, image processing methods such as edge detection, image segmentation, edge tracking and matching are used to identify stratum boundaries, and Hough transform or improved Hough transform or even human-computer interaction is used to extract sinusoidal layer interfaces. At the same time, core, microscopic thin sections and X-diffraction techniques are used to classify types. The ultra-high resolution of electrical imaging logging makes it possible to intuitively observe and outline reservoir characteristics such as fractures, pores, and bedding from the image.

[0004] However, in the aforementioned prior art, the formation interfaces on the electrical imaging logging images, especially the bedding planes, are displayed through subtle changes in grayscale within the layers. The distance between adjacent interfaces is very close and the edges are weak. Image processing methods such as edge detection, image segmentation, edge tracking and matching are difficult to accurately extract the trajectories of these interfaces. The Hough transform method is time-consuming and labor-intensive, and it is even more difficult to identify low-quality images and situations where the image data is partially destroyed. At the same time, the method of classifying types through core, microscopic thin section and X-diffraction techniques is time-consuming and labor-intensive, and can only describe the coring section. Summary of the invention

[0005] The purpose of the present invention is to provide a method for identifying shale laminae group types based on electrical imaging logging, which solves the problem in the prior art that the formation interfaces on the electrical imaging logging images, especially the bedding planes, are displayed by subtle changes in grayscale within the layers, the distance between adjacent interfaces is very close, the edges are weak, and image processing methods such as edge detection, image segmentation, edge tracking and matching are difficult to accurately extract the trajectories of these interfaces. The Hough transform method is time-consuming and labor-intensive, and it is even more difficult to identify low-quality images and situations where the image data is partially destroyed. At the same time, the methods for classifying types through cores, microscopic thin sections and X-diffraction techniques are time-consuming and labor-intensive, and can only describe the coring sections.

[0006] To achieve the above-mentioned purpose, the present invention provides a shale laminae group type identification method based on electrical imaging logging, which loads and decodes electrical imaging data and generates electrical imaging plate image data after preprocessing;

[0007] Convert the data from floating point to unsigned integer, then set the blank strips to null values ​​on the dynamic graph, and finally equalize the full-well segment image by setting the window length and step length to obtain a dynamic grayscale image.

[0008] Initialize the image center cluster using the K-means++ algorithm, calculate the Euclidean distance and iterate the new cluster center according to the mean, and complete the clustering after reaching the threshold or the number limit;

[0009] After the clustered N values ​​are binarized, the matrix is ​​connected end to end to better conform to the principle of wellbore imaging;

[0010] Then, the circular connected region marking algorithm is performed, and the head-to-tail connected regions are merged to obtain the coordinate points of each closed laminae;

[0011] After extracting the pixel points of the laminae through the bedding clustering algorithm, the sinusoidal bedding interface fitting can be performed;

[0012] Based on the characteristic value range of electrical imaging logging clustering response of different laminae groups, the shale gas reservoir laminae are divided, and the corresponding laminae group development density parameters are calculated.

[0013] In the dynamic grayscale image, a data point is randomly selected from the data set as the initial center point. For the subsequent cluster centers, the shortest distance from each data point to the currently selected cluster center set is calculated, that is, the square of the distance to the nearest cluster center point. The calculation model is: D(x) 2 =min i ||xc i || 2 ;

[0014] Where D(x) 2 : represents the square of the distance from the data point x to the cluster center point, D represents the distance; min i : Indicates that the minimum value is selected after calculating the distance of all cluster center points, min i The i in is an index, representing the center point of the i-th cluster; ||xc i ||: represents the Euclidean distance from the data point x to the center point ci of the i-th cluster, || || represents the norm or distance of the vector;

[0015] Calculate the probability distribution, which represents the probability of selecting the next cluster center point. The probability is inversely proportional to the square of the distance to ensure that the data points farther away from the selected center point have a greater probability of being selected. The calculation model is: The next cluster center point is randomly selected from the data set using probability distribution, and the above process is repeated until K initial cluster centers are selected, completing the K-means++ initialization process;

[0016] Where, P(x): represents the probability of selecting data point x as the next cluster center, P represents the probability; D(x) 2 : represents the square of the distance from the data point x to the cluster center, which is mentioned earlier; ∑ x′ D(x′) 2 : represents the sum of the squares of the distances from all data points x′ to the cluster center, ∑ represents the sum;

[0017] For each data point x i , calculate the distance to all cluster centers, and select the nearest cluster center c j , which can be represented by the Euclidean distance metric, where the data point x i Assigned to cluster C j In the calculation model,

[0018] In the formula, C j : represents the data point x i is assigned to cluster j, C represents the cluster; x i : represents the i-th data point, x i is a data point; c j : represents the center point of cluster j, c j is the cluster center; ||x i -c j ||: represents data point x i To cluster center c j The Euclidean distance of For all cluster center points c k , where K is the total number of clusters;

[0019] For each cluster C j , recalculate its center point c j , that is, take the mean of all data points in the cluster, which will become the new cluster center point for the next round of iteration. The calculation model is:

[0020] In the formula, c j : represents the center point of cluster j, c j is the cluster center; C j : represents the set of all data points in cluster j, C j is a set of data points; |C j ∣: represents the number of data points in cluster j, ∣·∣ represents the size of the set; x i ∈C j ∑x i :For all data points x in cluster j i , sum their coordinates; 1 / |C j∣: Represents the inverse of the number of data points, used to calculate the mean of the data points.

[0021] Among them, the shale gas reservoir lamina is divided into organic-clay lamina group, siliceous-clay lamina group and calcareous-clay lamina group.

[0022] Among them, in the circular connected region marking, the calculation model is: Where K is the number of assumed clusters, and blank strips are ignored; C1, C2, …, C K is the point set obtained after clustering; C i is the point set of the i-th class; the points corresponding to these K classes are respectively formed into K binary images, denoted as B1, B2,…, B K , B K The first column of is appended to the end of the matrix;

[0023] Label each binary image with eight connected regions to obtain a label image set L1, L2, ..., L K ;

[0024] Calculate L K The unique value u in the last column is used to find the ordinate set corresponding to the element in the last column whose value is equal to u. The abscissa set X is the coordinate of the data in the first column, that is, all are 0. The calculation model is: Y = argwhere(L K:,-1 =u);

[0025] Where, Y: represents the found vertical coordinate set, which is the index set; L K :, -1 : represents the last column of matrix L, L is a matrix, K: represents taking all rows, -1 represents taking the data of the last column; argwhere(·): represents finding the index set of elements that meet the conditions;

[0026] Calculate L K The unique value u of the point X,Y in 2, L K The amplitude of all points in which the value is equal to u2 is u, and the calculation model is: L k [L k = =u2] = u;

[0027] In the formula, u2: represents the unique value of the point X, Y, u2 is a unique value; L k [L k = = u2]: represents the matrix L k Find all elements whose values ​​are equal to u2; u: represents the value to be updated;

[0028] Removing the last column implements the Circular Connected Component Labeling algorithm.

[0029] Among them, after the pixel points of the laminae are extracted by the bedding clustering algorithm, the sinusoidal bedding interface fitting can be performed, which includes the following steps:

[0030] The three parameters of amplitude, phase and depth are randomly initialized. In this process, the calculation model is

[0031] In the formula, y: represents the display value of the formation on the electrical imaging; A: represents the amplitude of the formation display, that is, the magnitude of the formation amplitude; x: represents the position coordinates along the cylindrical well wall; φ: represents the phase, that is, the phase offset of the sine function; h0: represents the base height of the formation;

[0032] After the three parameters predicted by the random forest are input, they are input into the sine fitting algorithm as initialization parameters. After the sine fitting algorithm fits three new parameters, they are added to the parameter pool. After continuous iteration, the random forest parameter initialization can be realized.

[0033] Vectorize the data to increase the computation speed.

[0034] Then calculate the residual, Jacobian matrix and incremental direction in turn, and decide whether to continue fitting or output the results based on the conditions. Finally, append the output results to the parameter pool.

[0035] The process of vectorizing the data to increase the computing speed includes the following sub-steps:

[0036] Assuming there are n data points, the calculation model is: (x i ,y i ) i=1,2,…,n ;

[0037] In the formula, n: represents the number of data points, n is a positive integer; x i : represents the horizontal coordinate of the i-th data point, i ranges from 1 to n; y i : represents the ordinate of the i-th data point, i ranges from 1 to n; i: represents the index of the data point, i is an integer ranging from 1 to n;

[0038] The goal is to fit a nonlinear function, and the calculation model is:

[0039]

[0040] In the formula, f(x; φ, A, h): represents a nonlinear function, where x is the independent variable, φ, A, and h are parameters; φ: represents the phase parameter, which affects the phase offset of the sine function; A: represents the amplitude parameter, which affects the amplitude of the sine function; h: represents the reference height parameter;

[0041] The data points are expressed as vectors, and the calculation model is: X = [x1, x2, ..., x n ] T ,Y=[y1,y2,…,y n ] T ;

[0042] Where, X: represents the horizontal coordinate vector of the data point, which is a column vector containing the horizontal coordinates x1, x2, ..., xn of all data points, where n is the number of data points; Y: represents the vertical coordinate vector of the data point, which is also a column vector containing the vertical coordinates y1, y2, ..., yn of all data points, where n is the number of data points;

[0043] The goal is to calculate the residual vector R, where each element is the difference between the actual observation and the model prediction. The calculation model is:

[0044] Where, R: represents the residual vector, which is a column vector containing the residual of each data point, that is, the difference between the actual observed value and the model predicted value; Y: represents the ordinate vector of the data point, that is, the actual observed value; f(x; φ, A, h): represents the model predicted value, which is the nonlinear function value calculated based on the given parameters φ, A, h and the horizontal coordinate x of the data point;

[0045] By vectorizing the residual computation, we are able to efficiently compute the residuals for the entire dataset and use these residuals for parameter updates and fit optimization in subsequent steps.

[0046] Among them, the Jacobian matrix is ​​the partial derivative matrix of the model function with respect to the parameter vector. Each of its elements represents the partial derivative of the model function with respect to a certain parameter. The calculation model is:

[0047]

[0048] Wherein, ΔA: represents a small increment of amplitude A, which can be a small positive number and is used to calculate partial derivatives; Δf(x; φ, A, h): represents the change of model function f(x; φ, A, h), that is, the change of model function value when amplitude A increases by ΔA; f(x; φ, A+ΔA, h): represents the new model function value calculated after amplitude A increases by ΔA; f(x; φ, A, h): represents the original model function value, that is, the value of amplitude A;

[0049] The model for calculating the partial derivative of amplitude A is:

[0050]

[0051] In the formula, Represents the partial derivative of amplitude A, that is, the rate of change of function f(x; φ, A, h) with respect to amplitude A; △f(x; φ, A, h) / △A: represents the ratio between the change of model function f(x; φ, A, h) and the change of amplitude A;

[0052] The calculation model of the partial derivative of the model function with respect to the phase difference is:

[0053]

[0054] In the formula It represents the partial derivative of the phase difference φ, that is, the rate of change of the function f(x; φ, A, h) with respect to the phase difference φ; A: represents the amplitude, the parameter in the cosine function; cos(x+φ): represents the cosine value of x+φ, where x is the independent variable;

[0055] The calculation model for constructing the Jacobian matrix is:

[0056] By calculating the Jacobian matrix, it can be used in each iteration of the Levenberg-Marquardt algorithm to update parameters and optimize the fit;

[0057] Where, J: represents the Jacobian matrix, which is a row vector containing the partial derivatives of the model function f(x; φ, A, h) with respect to the parameters A, φ, h; It represents the partial derivative of the model function f(x; φ, A, h) with respect to the amplitude A, that is, the rate of change of the amplitude; represents the partial derivative of the model function f(x; φ, A, h) with respect to the phase difference φ, that is, the rate of change of the phase difference; It represents the partial derivative of the model function f(x;φ,A,h) with respect to the reference height h, that is, the rate of change of the reference height.

[0058] The calculation model for calculating the incremental direction is: Δp = (J T *J+λ*diag(J T *J)) -1 *J T *R, where diag(J T *J) is used to adjust the effect of parameter changes on the incremental direction. λ is the adjustment factor used to control the size of the step size. If the fitting result becomes better after the parameter update, λ is reduced; if the fitting result becomes worse, λ is increased;

[0059] Wherein, Δp: represents the change in the parameter, which is a column vector indicating how to adjust the parameter to minimize the residual under the current parameter value; λ: represents the regularization parameter, which is used to control the strength of regularization; J T : represents the transpose of the Jacobian matrix J; JT *J: represents the product of the Jacobian matrix J transpose and itself; diag(J T *J): means converting the product of the Jacobian matrix J transpose and itself into a diagonal matrix, that is, only retaining the elements on the main diagonal; J T *R: represents the product of the transposed Jacobian matrix J and the residual vector R; (J T *J+λ*diag(J T *J)) -1 :Represents the matrix (J T *J+λ*diag(J T *J))'s inverse matrix.

[0060] Among them, after updating the parameter estimate, the sum of squares of the new residual vector is calculated and compared with the previous value. If it is smaller and reaches the preset convergence threshold, the algorithm stops iterating and the parameter estimate is considered to have converged; if it is smaller but does not reach the convergence threshold, the iteration continues, the parameter value is updated and the residual is calculated again. If it is larger, the step size of the parameter update is adjusted according to the current parameter estimate and the residual vector, and the iteration continues. The calculation model is:

[0061] In the formula, Represents the sum of squares of the residual vector of the previous iteration; Yf(x; p+Δp) 2 : represents the sum of squares of the new residual vector calculated after updating the parameter value; threshold: represents the preset convergence threshold, which is used to determine whether the change in the residual is small enough. If it is less than the threshold, the parameter estimation is considered to have converged; condition5: represents the stopping condition of the iteration. If If the sum of the squares of the residual vector minus the new residual vector is greater than the threshold, the iteration continues; otherwise, the iteration stops.

[0062] The beneficial effects of the shale lamination group type identification method based on electrical imaging logging of the present invention are as follows: by adopting the bedding clustering algorithm, the stratum boundary is accurately identified, and the boundary can be directly, quickly and accurately extracted; then, by using the improved Levenberg-Marquardt algorithm for lamination identification, the sinusoidal curve can be identified efficiently, accurately and robustly; finally, the bedding clustering algorithm can effectively avoid the situation where there is no lamination between two sinusoidal curves; the thickness calculation also adopts the efficient equivalent thickness calculation method; when calculating the density, the situation of repeated calculation of the lamination can be directly avoided according to the center depth of the lamination. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0064] Figure 1 It is a comparison diagram of the original image and the clustering algorithm of the present invention.

[0065] Figure 2 It is a comparison diagram between the eight-connected region labeling algorithm and the circularly connected region labeling algorithm of the present invention.

[0066] Figure 3 It is a comparison diagram of the original image and the bedding clustering algorithm of the present invention.

[0067] Figure 4 It is the wellbore and formation trajectory diagram and the development diagram of the present invention.

[0068] Figure 5 It is a parameter initialization flow chart of the present invention.

[0069] Figure 6 It is the original image and the final effect image of the parameter pool data of the present invention.

[0070] Figure 7 This is a typical characteristic diagram of the organic matter-clay lamina group of Well Z9 of the present invention.

[0071] Figure 8 This is a typical characteristic diagram of the siliceous-clay lamina group of the Z10 well of the present invention.

[0072] Fig. 9 This is a typical characteristic diagram of the calcareous-clay lamina group of the Z9 well of the present invention.

[0073] Fig.10 This is the cluster density diagram of the three types of laminae groups in the X3 well of the present invention.

[0074] Fig.11 This is the comprehensive result diagram of the identification of the X9 well stratum group type of the present invention.

[0075] Fig.12 It is the overall technical roadmap of the present invention. DETAILED DESCRIPTION

[0076] Embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0077] See also Figures 1 to 12 The present invention provides a method for identifying shale lamination group types based on electrical imaging logging, comprising the following steps:

[0078] Load and decode the electrical imaging data, and generate electrical imaging plate image data after preprocessing;

[0079] Convert the data from floating point to unsigned integer, then set the blank strips to null values ​​on the dynamic graph, and finally equalize the full-well segment image by setting the window length and step length to obtain a dynamic grayscale image.

[0080] Initialize the image center cluster using the K-means++ algorithm, calculate the Euclidean distance and iterate the new cluster center according to the mean, and complete the clustering after reaching the threshold or the number limit;

[0081] After the clustered N values ​​are binarized, the matrix is ​​connected end to end to better conform to the principle of wellbore imaging;

[0082] Then, the circular connected region marking algorithm is performed, and the head-to-tail connected regions are merged to obtain the coordinate points of each closed laminae;

[0083] After extracting the pixel points of the laminae through the bedding clustering algorithm, the sinusoidal bedding interface fitting can be performed;

[0084] Based on the characteristic value range of electrical imaging logging clustering response of different laminae groups, the shale gas reservoir laminae are divided, and the corresponding laminae group development density parameters are calculated.

[0085] Furthermore, in the dynamic grayscale image, a data point is randomly selected from the data set as the initial center point. For the subsequent cluster centers, the shortest distance from each data point to the currently selected cluster center set is calculated, that is, the square of the distance to the nearest cluster center point. The calculation model is: D(x) 2 =min i ||xc i || 2 ;

[0086] Where D(x) 2 : represents the square of the distance from the data point x to the cluster center point, D represents the distance; min i : Indicates that the minimum value is selected after calculating the distance of all cluster center points, min i The i in is an index, representing the center point of the i-th cluster; ||xc i ||: represents the Euclidean distance from the data point x to the center point ci of the i-th cluster, || || represents the norm or distance of the vector;

[0087] Calculate the probability distribution, which represents the probability of selecting the next cluster center point. The probability is inversely proportional to the square of the distance to ensure that the data points farther away from the selected center point have a greater probability of being selected. The calculation model is: The next cluster center point is randomly selected from the data set using probability distribution, and the above process is repeated until K initial cluster centers are selected, completing the K-means++ initialization process;

[0088] Where, P(x): represents the probability of selecting data point x as the next cluster center, P represents the probability; D(x) 2 : represents the square of the distance from the data point x to the cluster center, which is mentioned earlier; ∑ x′ D(x′) 2 : represents the sum of the squares of the distances from all data points x′ to the cluster center, ∑ represents the sum;

[0089] For each data point x i , calculate the distance to all cluster centers, and select the nearest cluster center c j , which can be represented by the Euclidean distance metric, where the data point x i Assigned to cluster C j In the calculation model,

[0090] In the formula, C j : represents the data point x i is assigned to cluster j, C represents the cluster; x i : represents the i-th data point, x i is a data point; c j : represents the center point of cluster j, c j is the cluster center; ||x i -c j ||: represents data point x i To cluster center c j The Euclidean distance of For all cluster center points c k , where K is the total number of clusters;

[0091] For each cluster C j , recalculate its center point c j , that is, take the mean of all data points in the cluster, which will become the new cluster center point for the next round of iteration. The calculation model is:

[0092] In the formula, c j : represents the center point of cluster j, c j is the cluster center; C j : represents the set of all data points in cluster j, C j is a set of data points; |C j ∣: represents the number of data points in cluster j, ∣·∣ represents the size of the set; x i ∈C j ∑x i :For all data points x in cluster j i , sum their coordinates; 1 / |C j∣: Represents the inverse of the number of data points, used to calculate the mean of the data points.

[0093] Furthermore, the shale gas reservoir laminae are divided into organic-clay laminae, siliceous-clay laminae and calcareous-clay laminae.

[0094] Furthermore, in the circularly connected region marking, the calculation model is: Where K is the number of assumed clusters, and blank strips are ignored; C1, C2, ..., C K is the point set obtained after clustering; C i is the point set of the i-th class; the points corresponding to these K classes are respectively formed into K binary images, denoted as B1, B2, ..., B K , B K The first column of is appended to the end of the matrix;

[0095] Label each binary image with eight connected regions to obtain a label image set L1, L2, ..., L K ;

[0096] Calculate L K The unique value u in the last column is used to find the ordinate set corresponding to the element in the last column whose value is equal to u. The abscissa set X is the coordinate of the data in the first column, that is, all are 0. The calculation model is: Y = argwhere(L K:,-1 =u);

[0097] Where, Y: represents the found vertical coordinate set, which is the index set; L K :, -1 : represents the last column of matrix L, L is a matrix, K: represents taking all rows, -1 represents taking the data of the last column; argwhere(·): represents finding the index set of elements that meet the conditions;

[0098] Calculate L K The unique value u2 of the X,Y point value in L K The amplitude of all points in which the value is equal to u2 is u, and the calculation model is: L k [L k = =u2] = u;

[0099] In the formula, u2: represents the unique value of the point X, Y, u2 is a unique value; L k [L k = = u2]: represents the matrix L k Find all elements whose values ​​are equal to u2; u: represents the value to be updated;

[0100] Removing the last column implements the Circular Connected Component Labeling algorithm.

[0101] Furthermore, after extracting the pixel points of the laminae through the bedding clustering algorithm, the sinusoidal bedding interface fitting can be performed, which includes the following steps:

[0102] The three parameters of amplitude, phase and depth are randomly initialized. In this process, the calculation model is

[0103] In the formula, y: represents the display value of the formation on the electrical imaging; A: represents the amplitude of the formation display, that is, the magnitude of the formation amplitude; x: represents the position coordinates along the cylindrical well wall; φ: represents the phase, that is, the phase offset of the sine function; h0: represents the base height of the formation;

[0104] After the three parameters predicted by the random forest are input, they are input into the sine fitting algorithm as initialization parameters. After the sine fitting algorithm fits three new parameters, they are added to the parameter pool. After continuous iteration, the random forest parameter initialization can be realized.

[0105] Vectorize the data to increase the computation speed.

[0106] Then calculate the residual, Jacobian matrix and incremental direction in turn, and decide whether to continue fitting or output the results based on the conditions. Finally, append the output results to the parameter pool.

[0107] Furthermore, in vectorizing the data to increase the computing speed, the following sub-steps are included:

[0108] Assuming there are n data points, the calculation model is: (x i ,y i ) i=1,2,…,n ;

[0109] In the formula, n: represents the number of data points, n is a positive integer; x i : represents the horizontal coordinate of the i-th data point, i ranges from 1 to n; y i : represents the ordinate of the i-th data point, i ranges from 1 to n; i: represents the index of the data point, i is an integer ranging from 1 to n;

[0110] The goal is to fit a nonlinear function, and the calculation model is:

[0111]

[0112] In the formula, f(x; φ, A, h): represents a nonlinear function, where x is the independent variable, φ, A, and h are parameters; φ: represents the phase parameter, which affects the phase offset of the sine function; A: represents the amplitude parameter, which affects the amplitude of the sine function; h: represents the reference height parameter;

[0113] The data points are expressed as vectors, and the calculation model is: X = [x1, x2, ..., x n ] T ,Y=[y1,y2,…,y n ] T ;

[0114] Where X: represents the horizontal coordinate vector of the data point, which is a column vector containing the horizontal coordinates x1, x2,…, x n , where n is the number of data points; Y: represents the ordinate vector of the data points, which is also a column vector containing the ordinates of all data points y1, y2,…, y n , again n is the number of data points;

[0115] The goal is to calculate the residual vector R, where each element is the difference between the actual observation and the model prediction. The calculation model is:

[0116] Where, R: represents the residual vector, which is a column vector containing the residual of each data point, that is, the difference between the actual observed value and the model predicted value; Y: represents the ordinate vector of the data point, that is, the actual observed value; f(x; φ, A, h): represents the model predicted value, which is the nonlinear function value calculated based on the given parameters φ, A, h and the horizontal coordinate x of the data point;

[0117] By vectorizing the residual computation, we are able to efficiently compute the residuals for the entire dataset and use these residuals for parameter updates and fit optimization in subsequent steps.

[0118] Furthermore, the Jacobian matrix is ​​the partial derivative matrix of the model function with respect to the parameter vector. Each element of the matrix represents the partial derivative of the model function with respect to a certain parameter. The calculation model is:

[0119]

[0120] Wherein, ΔA: represents a small increment of amplitude A, which can be a small positive number and is used to calculate partial derivatives; Δf(x; φ, A, h): represents the change of model function f(x; φ, A, h), that is, the change of model function value when amplitude A increases by ΔA; f(x; φ, A+ΔA, h): represents the new model function value calculated after amplitude A increases by ΔA; f(x; φ, A, h): represents the original model function value, that is, the value of amplitude A;

[0121] The model for calculating the partial derivative of amplitude A is:

[0122]

[0123] In the formula, Represents the partial derivative of amplitude A, that is, the rate of change of function f(x; φ, A, h) with respect to amplitude A; △f(x; φ, A, h) / △A: represents the ratio between the change of model function f(x; φ, A, h) and the change of amplitude A;

[0124] The calculation model of the partial derivative of the model function with respect to the phase difference is:

[0125]

[0126] In the formula, It represents the partial derivative of the phase difference φ, that is, the rate of change of the function f(x; φ, A, h) with respect to the phase difference φ; A: represents the amplitude, the parameter in the cosine function; cos(x+φ): represents the cosine value of x+φ, where x is the independent variable;

[0127] The calculation model for constructing the Jacobian matrix is:

[0128] By calculating the Jacobian matrix, it can be used in each iteration of the Levenberg-Marquardt algorithm to update parameters and optimize the fit;

[0129] Where, J: represents the Jacobian matrix, which is a row vector containing the partial derivatives of the model function f(x; φ, A, h) with respect to the parameters A, φ, h; It represents the partial derivative of the model function f(x; φ, A, h) with respect to the amplitude A, that is, the rate of change of the amplitude; represents the partial derivative of the model function f(x; φ, A, h) with respect to the phase difference φ, that is, the rate of change of the phase difference; It represents the partial derivative of the model function f(x;φ,A,h) with respect to the reference height h, that is, the rate of change of the reference height.

[0130] Furthermore, the calculation model for calculating the incremental direction is: Δp=(J T *J+λ*diag(J T *J)) -1 *J T *R, where diag(J T *J) is used to adjust the effect of parameter changes on the incremental direction. λ is the adjustment factor used to control the size of the step size. If the fitting result becomes better after the parameter update, λ is reduced; if the fitting result becomes worse, λ is increased;

[0131] Wherein, Δp: represents the change in the parameter, which is a column vector indicating how to adjust the parameter to minimize the residual under the current parameter value; λ: represents the regularization parameter, which is used to control the strength of regularization; J T : represents the transpose of the Jacobian matrix J; JT *J: represents the product of the Jacobian matrix J transpose and itself; diag(J T *J): means converting the product of the Jacobian matrix J transpose and itself into a diagonal matrix, that is, only retaining the elements on the main diagonal; J T *R: represents the product of the transposed Jacobian matrix J and the residual vector R; (J T *J+λ*diag(J T *J))-1: represents the matrix (J T *J+λ*diag(J T *J))'s inverse matrix.

[0132] Furthermore, after updating the parameter estimate, the sum of squares of the new residual vector is calculated and compared with the previous value. If it is smaller and reaches the preset convergence threshold, the algorithm stops iterating and the parameter estimate is considered to have converged. If it is smaller but does not reach the convergence threshold, the iteration continues, the parameter value is updated and the residual is calculated again. If it is larger, the step size of the parameter update is adjusted according to the current parameter estimate and the residual vector, and the iteration continues. The calculation model is:

[0133] In the formula, Represents the sum of squares of the residual vector of the previous iteration; Yf(x; p+Δp) 2 : represents the sum of squares of the new residual vector calculated after updating the parameter value; threshold: represents the preset convergence threshold, which is used to determine whether the change in the residual is small enough. If it is less than the threshold, the parameter estimation is considered to have converged; condition5: represents the stopping condition of the iteration. If If the sum of the squares of the residual vector minus the new residual vector is greater than the threshold, the iteration continues; otherwise, the iteration stops.

[0134] In this embodiment, a bedding clustering algorithm is proposed by adopting reverse thinking. First, the whole of different laminae groups is separated. Then, the boundary is directly, quickly and accurately extracted according to the entity, which is used for the next step of sinusoidal curve detection of the laminae group boundary. The traditional sinusoidal curve identification of the interface from the imaging image generally adopts Hough transform and least squares method. Although Hough transform can identify multiple curves at a time, it is slow, low in precision, and it is very easy to confuse the upper and lower layers; the least squares method is fast and accurate, but it is difficult to identify the boundaries of irregular laminae groups. Combining the advantages and disadvantages of the two algorithms, the improved Levenberg-Marquardt algorithm for laminae identification in this application can identify sinusoidal curves efficiently, accurately and robustly. Due to the irregularity of the stratum interface, the boundary pixel points are blurred, which seriously affects the type classification in the case of thin laminae groups; at the same time, there may be no laminae between the layer interfaces, which may lead to errors in calculating the laminae thickness by the height difference between two sinusoidal curves. The bedding clustering algorithm of the present application can effectively avoid the situation where there is no lamination between two sinusoidal curves. The thickness calculation also adopts an efficient equivalent thickness calculation method. When calculating the density, the repeated calculation of laminations can be directly avoided according to the center depth of the lamination.

[0135] When extracting laminae coordinate points based on the bedding clustering algorithm: first, the electrical imaging data is loaded and decoded, and the electrical imaging plate image data is generated after preprocessing; then the image center cluster is initialized with the K-means++ algorithm, and the new cluster center is iterated according to the mean after calculating the Euclidean distance, and clustering is completed after reaching the threshold or the number limit; then the N values ​​after clustering are binarized respectively, and the matrix is ​​connected end to end to better conform to the principle of wellbore imaging, and then the circular connected region marking algorithm is performed, and the end-to-end connected regions are merged to obtain the coordinate points of each closed laminae.

[0136] When loading and cleaning electrical imaging data: the acquired electrical imaging raw data is in dlis format, which needs to be loaded and decoded. The raw data is preprocessed to generate electrical imaging plate image data. At this time, the raw plate data is extremely large for long well sections. Taking the FMI full-bore dynamic map as an example, the 500-meter well section imaging has 70.92 million data points, and each data point is still floating point type; therefore, data cleaning is required. First, convert the data from floating point type to unsigned integer type, and then set the blank strips to null values ​​on the dynamic map. Finally, set the window length and step length to equalize the full-well section image to obtain a dynamic grayscale image.

[0137] Then, dynamic grayscale image clustering is performed: the initial center selection mechanism of K-means++ is introduced into Mini-Batch K-means, and the convergence speed and clustering effect of the algorithm are improved by selecting more representative initial centers. At the same time, the Mini-Batch learning method is adopted to reduce the computational overhead of each iteration, making the algorithm more suitable for processing large-scale data; after the clustering algorithm (assuming K = 6, ignoring blank strips), the amount of data is reduced from N*256 to N*6, but the impact on the lamination can be ignored, refer to Figure 8 As shown, the integrity of the data is preserved while the amount of data is greatly reduced.

[0138] Marking of cyclic connected regions: Since underground electrical imaging logging actually measures cylindrical shapes, there are plate intervals, and plate data may be lost. Therefore, a cyclic connected region marking algorithm is designed to effectively solve the problem of images being connected at the beginning and end, thereby avoiding the failure of lamination recognition caused by the loss of intermediate plate data; the conventional eight-connected region marking algorithm ignores the connectivity of the image when applied to wellbore imaging, and cannot identify the actually connected areas that are disconnected in the middle. Fig. 9 As shown in the figure, the cyclic connected region labeling algorithm connects the image end to end to realize the connected region labeling of the cylindrical image; finally, the K types of labels are merged to obtain a complete connected region labeling map. Each color represents a separate layer group.

[0139] In the improved Levenberg-Marquardt fitting of laminae interface: Figure 4 As shown in the wellbore expansion image, the bedding plane cutting the wellbore appears as a single-cycle sine curve on the image, so it is necessary to perform a sine fitting on the bedding plane;

[0140] After extracting the pixel points of the laminae through the bedding clustering algorithm, the sinusoidal bedding interface fitting can be performed. First, the three parameters of amplitude, phase and depth are randomly initialized, and then the data is vectorized to increase the calculation speed. Then the residual, Jacobian matrix and incremental direction are calculated in turn. According to the conditions, it is determined whether to continue fitting or output the results, and finally the output results are attached to the parameter pool. After the parameter pool data is input into the random forest, the initial parameter prediction will be performed. As the calculation iteration proceeds, the initialization parameter prediction becomes more and more accurate, and the calculation speed becomes faster and faster.

[0141] In random forest parameter initialization: Compared with traditional initialization methods, such as random or empirical selection, not only will the amount of calculation be greatly increased, but also the accuracy will decrease when the number of iterations is constant. Therefore, this application uses random forest in machine learning to initialize parameters.

[0142] In vectorized residual calculation: Instead of calculating the residual for each data point one by one, the data points are represented as vectors and vectorized operations (such as matrix multiplication and element-wise operations) are used to calculate the residual for the entire data set. This can take advantage of the optimized implementation of matrix operations and improve computational efficiency.

[0143] In the analytical-numerical synthesis method for calculating the Jacobian matrix: use analytical methods or numerical methods to calculate partial derivatives. Numerical methods are suitable for general nonlinear functions, which are easy to implement but have limited calculation accuracy; analytical methods are suitable for simple and differentiable functions, and can provide accurate derivative calculation results, but the derivation process may be cumbersome. Fitting a sinusoidal curve with a fixed frequency of 1, considering that the accuracy of the phase in electrical imaging is greater than the amplitude, a combination of analytical and numerical methods is proposed for calculation.

[0144] Then, actual drawing verification is performed: after completing the sine matching, the data in the parameter pool is redrawn into the original graph, such as Fig. 9 As shown, since more attention is paid to the boundaries between laminae, the upper or lower boundaries of the laminae are usually extracted first.

[0145] Then the laminae group imaging logging response characteristics analysis is carried out: the sampling rate of electrical imaging is 0.00254 meters, but it still cannot accurately depict the laminae. The laminae seen in electrical imaging are actually a collection of many laminae. The present invention intends to define the laminae in electrical imaging as laminae groups, ranging from (0.254-10) cm. The laminae group interface often presents a group of parallel or nearly parallel conductivity anomaly areas. If the downhole strata are tilted, a sinusoidal wave distribution will appear in the image;

[0146] Using the results of identification of laminae development types by cores and thin sections, the idea of ​​calibrating imaging logging images with cores was adopted to determine the response characteristics of the same laminae group on the imaging images. Through multi-well clustering statistical analysis, it can be found that the response characteristics of organic-clay laminae groups on electrical imaging are generally black, with a gray value range of [16.67-81.03] and a cluster center value of 48.85.

[0147] The response characteristics of the siliceous-clay laminae group in electrical imaging are generally brown, with a gray value range of [95.98-169.66] and a cluster center value of 132.82;

[0148] The response characteristics of the calcareous-clay lamina group in electrical imaging are generally bright yellow, with a grayscale value range of [177.86-239.53] and a cluster center value of 208.69.

[0149] Finally, the lamination group types were divided: based on the range of characteristic values ​​of the electrical imaging logging clustering response of different lamination groups, the shale gas reservoir laminae were divided into three major categories: organic-clay laminae, siliceous-clay laminae, and calcareous-clay laminae, and the corresponding laminae group development density parameters were calculated.

[0150] What is disclosed above is only one or more preferred embodiments of the present application, and cannot be used to limit the scope of rights of the present application. Ordinary technicians in this field can understand that all or part of the processes of implementing the above embodiments and equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A method for identifying shale laminae types based on electrical imaging logging, characterized in that: The steps include: Load and decode the electrical imaging data, and generate electrical imaging plate image data after preprocessing; Convert the data from floating point to unsigned integer, then set the blank strips to null values ​​on the dynamic graph, and finally equalize the full-well segment image by setting the window length and step length to obtain a dynamic grayscale image. Initialize the image center cluster using the K-means++ algorithm, calculate the Euclidean distance and iterate the new cluster center according to the mean, and complete the clustering after reaching the threshold or the number limit; After the clustered N values ​​are binarized, the matrix is ​​connected end to end to better conform to the principle of wellbore imaging; Then, the circular connected region marking algorithm is performed, and the head-to-tail connected regions are merged to obtain the coordinate points of each closed laminae; After extracting the pixel points of the laminae through the bedding clustering algorithm, the sinusoidal bedding interface fitting can be performed; Based on the characteristic value range of electrical imaging logging clustering response of different laminae groups, the shale gas reservoir laminae are divided, and the corresponding laminae group development density parameters are calculated.

2. The method for identifying shale laminae types based on electrical imaging logging according to claim 1, characterized in that: In the dynamic grayscale image, a data point is randomly selected from the data set as the initial center point. For the subsequent cluster centers, the shortest distance from each data point to the currently selected cluster center set is calculated, that is, the square of the distance to the nearest cluster center point. The calculation model is: D(x) 2 =min i ||xc i || 2 ; Where D(x) 2 : represents the square of the distance from the data point x to the cluster center point, and D represents the distance; min i : Indicates that the minimum value is selected after calculating the distance of all cluster center points, min i The i in is an index, representing the center point of the i-th cluster; ||xc i ||: represents the distance from data point x to the center point c of the i-th cluster i The Euclidean distance of || || represents the norm or distance of the vector; Calculate the probability distribution, which represents the probability of selecting the next cluster center point. The probability is inversely proportional to the square of the distance to ensure that the data points farther away from the selected center point have a greater probability of being selected. The calculation model is: The next cluster center point is randomly selected from the data set using probability distribution, and the above process is repeated until K initial cluster centers are selected, completing the K-means++ initialization process; Where, P(x): represents the probability of selecting data point x as the next cluster center, P represents the probability; D(x) 2 : represents the square of the distance from the data point x to the cluster center, which is mentioned earlier; ∑ x′ D(x′) 2 : represents the sum of the squares of the distances from all data points x′ to the cluster center, ∑ represents the sum; For each data point x i , calculate the distance to all cluster centers, and select the nearest cluster center c j , which can be expressed by the Euclidean distance metric, the data point x i Assigned to cluster C j In the calculation model, In the formula, C j : represents the data point x i is assigned to cluster j, C represents the cluster; x i : represents the i-th data point, x is the data point; c j : represents the center point of cluster j, c is the center point of the cluster; ||x i -c j ||: represents data point x i To cluster center c j The Euclidean distance of For all cluster center points c k , where K is the total number of clusters; For each cluster C j , recalculate its center point c j , that is, take the mean of all data points in the cluster, which will become the new cluster center point for the next round of iteration. The calculation model is: In the formula, c j : represents the center point of cluster j, c is the center point of the cluster; C j : represents the set of all data points in cluster j, C is the set of data points; |C j ∣: represents the number of data points in cluster j, ∣·∣ represents the size of the set; x i ∈C j ∑x i :For all data points x in cluster j i , sum their coordinates; 1 / |C j ∣: Represents the inverse of the number of data points, used to calculate the mean of the data points.

3. The method for identifying shale lamination types based on electrical imaging logging according to claim 2, characterized in that: The shale gas reservoir laminae are divided into organic-clay laminae, siliceous-clay laminae and calcareous-clay laminae.

4. The method for identifying shale lamination types based on electrical imaging logging according to claim 3, characterized in that: In the circularly connected region labeling, the calculation model is: Where K is the number of assumed clusters, and blank strips are ignored; C1, C2, …, C K is the point set obtained after clustering; C i is the point set of the i-th class; the points corresponding to these K classes are respectively formed into K binary images, denoted as B1, B2,…, B K , B K The first column of is appended to the end of the matrix; Label each binary image with eight connected regions to obtain a label image set L1, L2, ..., L K ; Calculate L K The unique value u in the last column is used to find the ordinate set corresponding to the element in the last column whose value is equal to u. The abscissa set X is the coordinate of the data in the first column, that is, all are 0. The calculation model is: Y = argwhere(L K:,-1 =u); Where, Y: represents the found vertical coordinate set, which is the index set; L K :, -1 : represents the last column of matrix L, L is a matrix, K: represents taking all rows, -1 represents taking the data of the last column; argwhere(·): represents finding the index set of elements that meet the conditions; Calculate L K The unique value u2 of the X,Y point value in L K The amplitude of all points in which the value is equal to u2 is u, and the calculation model is: L k [L k = =u2] = u; In the formula, u2: represents the unique value of the point X, Y, u2 is a unique value; L k [L k = = u2]: represents the matrix L k Find all elements whose values ​​are equal to u2; u: represents the value to be updated; Removing the last column implements the Circular Connected Component Labeling algorithm.

5. The method for identifying shale lamination types based on electrical imaging logging according to claim 4, characterized in that: After extracting the pixel points of the laminae through the bedding clustering algorithm, the sinusoidal bedding interface fitting can be performed, which includes the following steps: The three parameters of amplitude, phase and depth are randomly initialized. In this process, the calculation model is In the formula, y: represents the display value of the formation on the electrical image; A: represents the amplitude of the formation display, that is, the magnitude of the formation amplitude; x: represents the position coordinate along the cylindrical well wall; φ: represents the phase, that is, the phase offset of the sine function; h0: represents the base height of the formation; After the three parameters predicted by the random forest are input, they are input into the sine fitting algorithm as initialization parameters. After the sine fitting algorithm fits three new parameters, they are added to the parameter pool. After continuous iteration, the random forest parameter initialization can be realized. Vectorize the data to increase the computation speed. Then calculate the residual, Jacobian matrix and incremental direction in turn, and decide whether to continue fitting or output the results based on the conditions. Finally, append the output results to the parameter pool.

6. The method for identifying shale laminae types based on electrical imaging logging according to claim 5, characterized in that: In vectorizing data to increase computing speed, the following sub-steps are included: Assuming there are n data points, the calculation model is: (x i ,y i ) i=1,2,…,n ; In the formula, n: represents the number of data points, n is a positive integer; x i : represents the horizontal coordinate of the i-th data point, i ranges from 1 to n; y i : represents the ordinate of the i-th data point, i ranges from 1 to n; i: represents the index of the data point, i is an integer ranging from 1 to n; The goal is to fit a nonlinear function, and the calculation model is: In the formula, f(x; φ, A, h): represents a nonlinear function, where x is the independent variable, φ, A, and h are parameters; φ: represents the phase parameter, which affects the phase shift of the sine function; A: represents the amplitude parameter, which affects the amplitude of the sine function; h: represents the reference height parameter; The data points are expressed as vectors, and the calculation model is: X = [x1, x2, ..., x n ] T ,Y=[y1,y2,…,y n ] T ; Where X: represents the horizontal coordinate vector of the data point, which is a column vector containing the horizontal coordinates x1, x2,…, x n , where n is the number of data points; Y: represents the ordinate vector of the data points, which is also a column vector containing the ordinates of all data points y1, y2,…, y n , again n is the number of data points; The goal is to calculate the residual vector R, where each element is the difference between the actual observation and the model prediction. The calculation model is: Where, R: represents the residual vector, which is a column vector containing the residual of each data point, that is, the difference between the actual observed value and the model predicted value; Y: represents the ordinate vector of the data point, that is, the actual observed value; f(x; φ, A, h): represents the model predicted value, which is the nonlinear function value calculated based on the given parameters φ, A, h and the horizontal coordinate x of the data point; By vectorizing the residual computation, we are able to efficiently compute the residuals for the entire dataset and use these residuals for parameter updates and fit optimization in subsequent steps.

7. The method for identifying shale lamination types based on electrical imaging logging according to claim 6, characterized in that: The Jacobian matrix is ​​the partial derivative matrix of the model function with respect to the parameter vector. Each element of the matrix represents the partial derivative of the model function with respect to a certain parameter. The calculation model is: Wherein, ΔA: represents a small increment of amplitude A, which can be a small positive number and is used to calculate partial derivatives; Δf(x; φ, A, h): represents the change of model function f(x; φ, A, h), that is, the change of model function value when amplitude A increases by ΔA; f(x; φ, A+ΔA, h): represents the new model function value calculated after amplitude A increases by ΔA; f(x; φ, A, h): represents the original model function value, that is, the value of amplitude A; The model for calculating the partial derivative of amplitude A is: In the formula, Represents the partial derivative of amplitude A, that is, the rate of change of function f(x; φ, A, h) with respect to amplitude A; △f(x; φ, A, h) / △A: represents the ratio between the change of model function f(x; φ, A, h) and the change of amplitude A; The calculation model of the partial derivative of the model function with respect to the phase difference is: In the formula, It represents the partial derivative of the phase difference φ, that is, the rate of change of the function f(x; φ, A, h) with respect to the phase difference φ; A: represents the amplitude, the parameter in the cosine function; cos(x+φ): represents the cosine value of x+φ, where x is the independent variable; The calculation model for constructing the Jacobian matrix is: By calculating the Jacobian matrix, it can be used in each iteration of the Levenberg-Marquardt algorithm to update parameters and optimize the fit; Where, J: represents the Jacobian matrix, which is a row vector containing the partial derivatives of the model function f(x; φ, A, h) with respect to the parameters A, φ, h; It represents the partial derivative of the model function f(x; φ, A, h) with respect to the amplitude A, that is, the rate of change of the amplitude; represents the partial derivative of the model function f(x; φ, A, h) with respect to the phase difference φ, that is, the rate of change of the phase difference; It represents the partial derivative of the model function f(x;φ,A,h) with respect to the reference height h, that is, the rate of change of the reference height.

8. The method for identifying shale laminae types based on electrical imaging logging according to claim 7, characterized in that: The calculation model for calculating the incremental direction is: p = (J T *J+λ*diag(J T *J)) -1 *J T *R, where diag(J T *J) is used to adjust the effect of parameter changes on the incremental direction. λ is the adjustment factor used to control the size of the step size. If the fitting result becomes better after the parameter update, λ is reduced; if the fitting result becomes worse, λ is increased; Where Δp: represents the change in the parameter, which is a column vector indicating how to adjust the parameter to minimize the residual under the current parameter value; λ: represents the regularization parameter, which is used to control the strength of regularization; J T : represents the transpose of the Jacobian matrix J; J T *J: represents the product of the Jacobian matrix J transpose and itself; diag(J T *J): means converting the product of the Jacobian matrix J transpose and itself into a diagonal matrix, that is, only retaining the elements on the main diagonal; J T *R: represents the product of the transposed Jacobian matrix J and the residual vector R; (J T *J+λ*diag(J T *J)) -1 :Represents the matrix (J T *J+λ*diag(J T *J))'s inverse matrix.

9. The method for identifying shale laminae types based on electrical imaging logging according to claim 8, characterized in that: After updating the parameter estimate, the sum of squares of the new residual vector is calculated and compared with the previous value. If it is smaller and reaches the preset convergence threshold, the algorithm stops iterating and the parameter estimate is considered to have converged. If it is smaller but does not reach the convergence threshold, the iteration continues, the parameter value is updated and the residual is calculated again. If it is larger, the step size of the parameter update is adjusted according to the current parameter estimate and the residual vector, and the iteration continues. The calculation model is: In the formula, Represents the sum of squares of the residual vector of the previous iteration; Yf(x; p+Δp) 2 : represents the sum of squares of the new residual vector calculated after updating the parameter value; threshold: represents the preset convergence threshold, which is used to determine whether the change in the residual is small enough. If it is less than the threshold, the parameter estimation is considered to have converged; condition5: represents the stopping condition of the iteration. If If the sum of the squares of the residual vector minus the new residual vector is greater than the threshold, the iteration continues; otherwise, the iteration stops.