While-drilling geosteering data set establishment method for deep learning

Abstract

The invention discloses a while-drilling geosteering data set establishment method for deep learning, and relates to the technical field of petroleum and natural gas exploration and development, and the method comprises the steps: controlling a while-drilling gamma measurement instrument, carrying out the detection of a drilled well in a target region, and determining while-drilling gamma data; statistical analysis and data reconstruction are carried out, and a first statistical index matrix is constructed; performing multiple rounds of matrix correction calculation by introducing a row vector, and constructing a first power law dimension raising matrix; and integrating the first statistical index matrix and the first power law dimension raising matrix, establishing a while-drilling geosteering data set, and storing the while-drilling geosteering data set to a geological data center. The technical problems that in the prior art, in a horizontal well geosteering scene of oil exploitation, the data processing difficulty is large, the technology is complex, the real-time requirement is high, and the cost is high are solved, and the technical effects that the while-drilling geosteering data set is established, the data processing difficulty is reduced, the technical process is simplified, and the real-time requirement is met are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas exploration and development, and in particular to a method for establishing a while-drilling geosteering dataset for deep learning. Background Art

[0002] In the process of oil production, geosteering while drilling is crucial for precise production. Currently, in geosteering while drilling technology, there are difficulties in establishing a data set for deep learning. When dealing with three-dimensional fine geological modeling and unconventional, thin-layer, and heterogeneous reservoirs, existing technologies have problems such as complex data processing, high technical difficulty, high real-time requirements, and high costs. Although intelligence is an important direction to improve guidance accuracy, and there are related technologies such as the deep learning-based azimuthal electromagnetic wave resistivity inversion method for drilling, these technologies cannot solve the key problem of how to establish a data set. Due to the lack of an effective data set establishment method, the application of deep learning in geosteering while drilling is severely limited, making it difficult to achieve accurate and efficient guidance goals, and unable to meet the oil production industry's growing demand for precise production technology. Summary of the Invention

[0003] This application addresses the technical issues faced by existing technologies in horizontal well geosteering scenarios in oil production, such as difficulty in data processing, technical complexity, high real-time requirements, and high costs. This application processes the gamma data acquired by a downhole gamma meter, calculates the gamma amplitude, constructs a first statistical indicator matrix and a first power-law dimension-increasing matrix, and then establishes a downhole geosteering dataset through traversal calculation, matrix segmentation, and integration operations. This solves the problem of deep learning lacking high-quality datasets in downhole geosteering, and provides data support for accurate downhole geosteering.

[0004] In response to the above technical problems, the present application proposes a technical solution for a method for establishing a downhole geosteering dataset for deep learning, wherein the method includes: controlling a downhole gamma measuring instrument to detect drilling in a target area and determine downhole gamma data, wherein the downhole gamma data is in matrix form, and taking the difference between the maximum and minimum values ​​in the downhole gamma data as the gamma amplitude; for the first downhole gamma data, introducing statistical indicators, and constructing a first statistical indicator matrix by performing statistical analysis and data reconstruction, wherein the first downhole gamma data is any one of the downhole gamma data; for the first statistical indicator matrix, introducing row vectors, performing multiple rounds of matrix correction calculations, and constructing a first power-law dimensionality-raising matrix; integrating the first statistical indicator matrix and the first power-law dimensionality-raising matrix to establish a downhole geosteering dataset, and storing the downhole geosteering dataset in a geological data center.

[0005] This application proposes one or more technical solutions, which have at least the following technical effects:

[0006] This application controls a downhole gamma meter to acquire downhole gamma data and calculate gamma amplitudes. A first statistical indicator matrix is ​​constructed for the downhole gamma data, followed by a first power-law dimension-increasing matrix. The downhole gamma data is traversed, correlation matrix calculations and associations are performed on each data point, matrix segmentation is performed according to matrix segmentation rules, and a set of segmented sub-matrices is integrated to establish and store downhole geosteering datasets for deep learning. This achieves the technical effect of establishing a downhole geosteering dataset, reducing data processing difficulty, simplifying the technical process, and meeting real-time requirements.

[0007] The above content summarizes the present application's solution to a method for establishing a while-drilling geosteering dataset for deep learning. The present application will describe the steps of the technical solution in detail in the following specific implementation methods to facilitate a clear and complete understanding of the present application by technical personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 This is a flow chart of a method for establishing a while-drilling geosteering dataset for deep learning provided in an embodiment of the present application.

[0010] Figure 2 This is a flow chart of constructing a first statistical indicator matrix in a method for establishing a while-drilling geological guidance dataset for deep learning provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] This application controls a downhole gamma meter to obtain downhole gamma data in matrix form and calculates the gamma amplitude. For the first downhole gamma data in the downhole gamma data, statistical indicators are introduced to construct a first statistical indicator matrix, and then a first power-law dimensionality-increasing matrix is ​​constructed by introducing row vectors. The downhole gamma data are traversed, and correlation matrix calculations and associations are performed on each data point. Matrix segmentation is performed according to preset matrix segmentation rules, and a set of segmented sub-matrices is integrated. Finally, a downhole geosteering dataset for deep learning is established and stored in a geological data center. This achieves the technical effect of establishing a downhole geosteering dataset, reducing the difficulty of data processing, simplifying the technical process, and meeting real-time requirements.

[0012] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0013] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.

[0014] like Figure 1 As shown, a method for establishing a while-drilling geosteering dataset for deep learning, wherein the method includes:

[0015] Step A100: Detecting the target area by controlling the while-drilling gamma measurement instrument to determine while-drilling gamma data, wherein the while-drilling gamma data is in matrix form, and taking the difference between the maximum and minimum values ​​in the while-drilling gamma data as the gamma amplitude.

[0016] In the embodiment of the present application, the gamma-ray while drilling data is presented in a matrix form, which is data obtained by the gamma-ray while drilling instrument during drilling detection of the target area.

[0017] Specifically, using a shale gas block in a specific location as the target area, we used gamma-ray geosteering data to construct a geosteering dataset for deep learning. First, we controlled the gamma-ray geosteering instrument to drill wells in the target area. Assuming the target area for this study includes 21 wells, we collected gamma-ray geosteering data from all 21 wells. After acquiring this large amount of gamma-ray geosteering data, we organized it into a matrix format for subsequent processing. The specific steps are detailed in A110-A130.

[0018] Next, find the maximum gamma value from this data and minimum value And calculate the gamma amplitude, the specific steps are described in detail in A140.

[0019] The calculated gamma amplitude can be used to construct more complex data structures, laying the foundation for establishing a while-drilling geosteering dataset for deep learning.

[0020] Step A200: introducing statistical indicators for first while-drilling gamma data, and constructing a first statistical indicator matrix by performing statistical analysis and data reconstruction, wherein the first while-drilling gamma data is any item of the while-drilling gamma data.

[0021] In the embodiment of the present application, the first gamma-ray while drilling data is the gamma-ray while drilling data obtained from any of the 21 wells in the target area. The first statistical indicator matrix is ​​the matrix obtained after a series of statistical analysis and data reconstruction operations on the first gamma-ray while drilling data obtained from a well in the target area.

[0022] Alternatively, assume that a person skilled in the art first selects one item from the gamma-ray while drilling data of 21 wells as the first gamma-ray while drilling data. Next, determine S (S ≥ 1 and a positive integer) statistical indicators, use Y as the number of matrix row vectors and the number of statistical indicators to determine the number of matrix column vectors, and perform statistical analysis on the first gamma-ray while drilling data within a statistical range of a positive integer n greater than 9 to calculate a first statistical indicator matrix. The specific steps are described in detail in A210-A220.

[0023] By constructing the first statistical indicator matrix, we can more deeply explore the statistical characteristics of while-drilling gamma data, providing strong support for subsequent dataset construction and deep learning applications.

[0024] Step A300: For the first statistical indicator matrix, by introducing row vectors, performing multiple rounds of matrix correction calculations, and constructing a first power-law dimension-raising matrix.

[0025] In the embodiment of the present application, the first power-law dimension-raising matrix is ​​a matrix constructed based on the first while-drilling gamma data, with the purpose of mining data features and improving data dimensions.

[0026] In one embodiment of the present application, the first statistical indicator matrix is ​​constructed based on the gamma data of a well in the target area. Next, the first power law dimension-raising matrix is ​​constructed. The row vector of the elements with positive serialized distribution is replicated the same number of times as the first while-drilling gamma data detection and distributed in columns to construct an initial power-law dimension-raising matrix of the corresponding size. The initial matrix is ​​then corrected and calculated to determine the first power-law dimension-raising matrix. The specific steps are described in detail in A310-A330.

[0027] Step A400: Integrate the first statistical indicator matrix and the first power-law dimension-raising matrix to establish a while-drilling geosteering dataset, and store the while-drilling geosteering dataset in a geological data center.

[0028] In the embodiment of the present application, the geosteering while drilling dataset is used for geosteering while drilling in deep learning. The geological data center is an area that stores geosteering while drilling datasets and provides data support for subsequent geosteering while drilling related work.

[0029] Specifically, the above steps acquire the first LWD gamma data for a specific well, and construct the corresponding first statistical indicator matrix and first power-law dimensionality-increasing matrix. Next, the LWD gamma data is traversed, and correlation matrices are calculated and associated for each target area. A matrix segmentation rule, determined by a preset segmentation method and size, is introduced. The mapped associated matrices are traversed and segmented according to this rule to construct a set of segmentation submatrices. Finally, the set of segmentation submatrices is integrated to generate a LWD geosteering dataset. The specific steps are detailed in A410-A440. After the dataset is established, it is stored in the geological data center for easy access at any time.

[0030] By generating a while-drilling geosteering dataset, it is convenient to train deep learning models later to achieve more accurate while-drilling geosteering and improve the efficiency and accuracy of oil production.

[0031] Furthermore, step A100 in the method provided in the embodiment of the present application includes:

[0032] A110: For the first drilling, determine Y items of gamma data through detection and perform data preprocessing, where Y is the number of detections.

[0033] A120: Using the amount of detection data as the matrix row vector, the detection depth as the first matrix column vector, and the gamma data as the second matrix column vector, perform matrix conversion on the Y-item gamma data to serve as the first while-drilling gamma data, wherein the first well is any well in the target area.

[0034] A130: Add the first while-drilling gamma data to the while-drilling gamma data.

[0035] Specifically, a well in the target area is randomly selected as the first well to be drilled. During the drilling process, the first well is surveyed using a gamma-ray while drilling instrument. The number of surveys is set to Y, for example, Y = 100. Each survey will generate corresponding gamma data, and a total of Y gamma data items are determined. The collected raw gamma data may contain noise or inconsistent data formats. Data preprocessing is required for these 100 gamma data items to improve data quality, such as removing outliers and filling missing values:

[0036] Step a: Calculate the mean of this set of gamma data (the mean represents the central tendency of the data). Add up all the gamma data and divide it by the total number of data to get the mean.

[0037] Step b: Calculate the standard deviation (the standard deviation measures the degree of dispersion of the data) by calculating the sum of the squares of the differences between each data point and the mean, dividing it by the number of data points, and finally taking the square root of the result to obtain the standard deviation.

[0038] Step c: Based on relevant experience and actual mine conditions, skilled artisans set a standard deviation multiple (e.g., 3 standard deviations). Data that deviates beyond 3 standard deviations from the mean are considered outliers. These outliers can be directly deleted or replaced with a reasonable value (such as the mean of the adjacent data) based on the data characteristics.

[0039] Step d: Then fill the missing values ​​of the gamma data. Mean filling is to use the mean of this set of data as the filling value of the missing value (the calculation method is the same as step a). This method is direct but easily affected by extreme values; median filling is to sort the data from small to large, and take the value in the middle position (if the number of data is an odd number) or the average of the two middle numbers (if the number of data is an even number) as the filling value. This method is not sensitive to extreme values; in addition, interpolation is performed based on the time series characteristics of the data, such as linear interpolation. It is assumed that the data changes continuously in time. The estimated value of the missing value is calculated according to the linear relationship based on the data before and after the missing value. Technical personnel in this field will make a choice based on the actual situation.

[0040] After preprocessing, the amount of detection data (i.e., Y = 100 detections) is used as the matrix row vector, the depth corresponding to each detection is used as the first matrix column vector, and the corresponding gamma data is used as the second matrix column vector. Matrix transformation is performed on these 100 gamma data.

[0041] When performing matrix conversion, first determine the number of matrix rows based on the amount of data from 100 detections. Next, fill the depth corresponding to each detection into the first matrix column vector in sequence. According to the detection order, the depth corresponding to the first detection is placed in the first row and first column, the depth corresponding to the second detection is placed in the second row and first column, and so on, until the depth corresponding to the 100th detection is filled in the 100th row and first column. Then, fill the corresponding gamma data into the second matrix column vector. Also according to the detection order, the gamma data from the first detection is placed in the first row and second column, the gamma data from the second detection is placed in the second row and second column, and so on, until the gamma data from the 100th detection is located in the 100th row and second column. This organizes the originally scattered 100 gamma data items and their corresponding detection depths into a structured matrix with 100 rows and 2 columns.

[0042] This process organizes the previously scattered exploration data into a structured matrix, yielding the first LWD gamma data set. The first well is any well within the target area, meaning the above process can be applied to every well within the area. Finally, the processed first LWD gamma data acquired from the first well is added to the entire LWD gamma data set.

[0043] By performing the above operations on all wells in the target area, the gamma-ray data while drilling will be continuously enriched and improved, providing a solid data foundation for the subsequent establishment of a geosteering while drilling dataset for deep learning, thereby improving the accuracy and reliability of geosteering while drilling.

[0044] Furthermore, step A100 in the method provided in the embodiment of the present application includes:

[0045] A140: Calculate the difference between the maximum gamma data and the minimum gamma data of L while-drilling gamma data of the target area, as the gamma amplitude, where L is the number of wells drilled in the target area.

[0046] In the embodiment of the present application, the number of wells drilled in the target area is 21, that is, L=21.

[0047] Optionally, first use a while-drilling gamma measurement instrument to survey these 21 wells and obtain the while-drilling gamma data for each well (step A100). Each well obtains Y (Y is 100 times) while-drilling gamma data points. Then, from all the while-drilling gamma data of these 21 wells, the data with the largest gamma value is selected and recorded as , and the smallest data, recorded as This process requires technicians to compare a large amount of data point by point to ensure that the extreme value is found accurately. Finally, by Subtract the minimum gamma value , that is, amplitude = - , obtaining the gamma amplitude. This gamma amplitude reflects the fluctuation range of the target area's gamma data while drilling, providing key parameters for subsequent operations such as constructing the first power-law dimensionality-increasing matrix. This helps improve the quality of while-drilling geosteering datasets, thereby enhancing the effectiveness of deep learning in geosteering while drilling, achieving more accurate geosteering and reducing mining costs and risks.

[0048] Further, if Figure 2 As shown, step A200 in the method provided in the embodiment of the present application includes:

[0049] A210: Determine a statistical indicator, where the statistical indicator includes S items, and S is a positive integer greater than or equal to 1.

[0050] A220: With Y as the number of matrix row vectors, with statistical indicators determining the number of matrix column vectors, and with the statistical range as a constraint, perform statistical analysis on the first downhole gamma data and calculate the first statistical indicator matrix, wherein the statistical range is set to n, where n is a positive integer greater than 9.

[0051] Specifically, after the above steps, the first gamma-ray while drilling data has been acquired. First, statistical indicators are determined. For example, based on geological analysis requirements and past experience, those skilled in the art have determined S = 5 statistical indicators (such as mean, standard deviation, skewness, kurtosis, and correlation coefficient, where S is a positive integer greater than or equal to 1). These indicators are used to reflect different characteristics of the data, such as central tendency, dispersion, and trend.

[0052] Next, let's use the number of detections, Y, as the number of matrix row vectors. Assume that during the first well's detection, a total of Y = 100 detections were performed. These 100 detections constitute the matrix's row vectors. Then, based on the five statistical indicators, we determine the number of matrix column vectors, resulting in a matrix with 100 rows and 5 columns.

[0053] Then, set the statistical range to n, for example, n=15 (n is a positive integer greater than 9). The matrix column vector corresponding to the first statistical indicator For example, when calculating the first matrix column vector, for the matrix column vector corresponding to the first statistical indicator, any element thereof and the calculated value of the first n-1 elements based on the statistical indicator are used to reconstruct the first matrix column vector; then all statistical indicators are traversed, corresponding calculations are performed on each matrix column vector, the Sth matrix column vector is determined, and all matrix column vectors are integrated to obtain the first statistical indicator matrix, which can deeply explore the intrinsic characteristics of the first while drilling gamma data. The specific steps are described in detail in A221-A222.

[0054] By constructing the first statistical indicator matrix, it is helpful to improve the application effect of deep learning in geological guidance while drilling, achieve more accurate geological guidance, and improve oil production efficiency.

[0055] Furthermore, step A220 in the method provided in the embodiment of the present application includes:

[0056] A221: For the matrix column vector V1 corresponding to the first statistical indicator, reconstruct the first matrix column vector of the first statistical indicator matrix based on the calculated value of the first statistical indicator using the first element and the first n-1 elements, wherein the first element is any matrix column vector element in the matrix column vector V1.

[0057] A222: Traverse the statistical indicators, perform statistical indicator calculations on each matrix column vector until the Sth matrix column vector is determined, and integrate the first matrix column vectors until the Sth matrix column vector as the first statistical indicator matrix.

[0058] Specifically, according to the five statistical indicators determined in step A210, when detecting the first well, a total of Y=100 detections (Y is the number of detections) are performed, and the first while-drilling gamma data is obtained. At the same time, the statistical range n=15 (n is a positive integer greater than 9) is set.

[0059] For the matrix column vector corresponding to the first statistical indicator , selecting any matrix column vector element as the first element. For example, when calculating the mean statistical indicator, if the first element is the gamma data corresponding to the 10th detection, it is added to the first n-1 (i.e., 14) elements and then divided by 15. The resulting value is used to reconstruct the first matrix column vector of the first statistical indicator matrix.

[0060] After reconstructing the first matrix column vector, we iterate through all five statistical indicators. A similar method is used to calculate the matrix column vector corresponding to each statistical indicator. To calculate the standard deviation, we take the sum of the squares of the differences between each matrix column vector element and the mean, divide it by n (here, 15), and then take the square root to obtain the standard deviation. This is used to determine the value of the corresponding matrix column vector and the second matrix column vector of the first statistical indicator matrix.

[0061] Next, calculate the skewness to determine the third matrix column vector of the first statistical indicator matrix. First, calculate the mean of the matrix column vector elements. For each matrix column vector element, calculate the difference between it and the mean, cube the difference, and then add up all the cubed differences. Divide this sum by the number of elements (assuming it is m), and then divide it by the cube of the standard deviation (the standard deviation is calculated in the same way as above). That is, skewness = ,in are the elements in the matrix column vector, is the mean, s is the standard deviation. The value of the matrix column vector corresponding to the skewness is determined by this calculation result.

[0062] Then, calculate the kurtosis to determine the fourth matrix column vector of the first statistical indicator matrix. Similarly, calculate the mean first. For each matrix column vector element, calculate the difference between it and the mean and raise it to the fourth power. Add up all the fourth-powered differences. Divide this sum by the number of elements m, then by the fourth power of the standard deviation, and finally subtract 3. That is, kurtosis = .

[0063] Finally, the correlation coefficient is calculated to determine the fifth matrix column vector of the first statistical indicator matrix. Suppose that the correlation coefficient of two matrix column vectors (such as the first matrix column vector and the second matrix column vector) is calculated. The mean of the two columns is calculated separately. For each element of the first column and the elements corresponding to the second column ,calculate( − )( − ) and add all these products to get the numerator. At the same time, calculate the standard deviation of the first and second columns separately and , the numerator is divided by the product of the two column standard deviations and the number of elements m, that is, the correlation coefficient = In this way, statistical index calculations are performed on each matrix column vector in turn until the fifth matrix column vector is determined.

[0064] The data from the first matrix column vector to the fifth matrix column vector are integrated to form a matrix containing 100 rows and 5 columns. This matrix is ​​the first statistical indicator matrix, which integrates various statistical features of the while drilling gamma data.

[0065] By integrating and forming the first statistical indicator matrix, it provides important data support for the subsequent establishment of a while-drilling geosteering dataset, helping the deep learning model to more accurately analyze geological conditions and achieve more precise while-drilling geosteering.

[0066] Furthermore, step A223 in the method provided in the embodiment of the present application includes:

[0067] A223-1: When the first element is less than n, use a unified symbol to replace the reconstructed matrix elements of the first matrix column vector.

[0068] In an embodiment of the present application, reconstructing the matrix elements is to use the first element and the first n-1 elements in the matrix column vector based on the calculated value of the first statistical indicator to reconstruct the elements in the matrix column vector, thereby forming the first matrix column vector of the first statistical indicator matrix.

[0069] Specifically, assuming that the statistical range n=15 (n is a positive integer greater than 9), when processing a matrix column vector If the first element is the fifth data point acquired during the exploration process (assuming the total number of explorations Y is sufficiently large), that is, the first element is less than n, then using this first element and the first n-1 (i.e., nine) elements to calculate the first statistical indicator using the conventional calculation method described above may result in inaccurate and ineffective results due to missing or illogical data. In this case, to ensure consistency and effectiveness in data processing, a unified symbol is used to replace the reconstructed matrix elements of the first matrix column vector. For example, "X" is used to represent them uniformly. This prevents errors or deviations caused by these special data points when processing the entire matrix column vector and integrating the matrix column vectors to construct the first statistical indicator matrix. This ensures a smooth data processing process and allows the final constructed first statistical indicator matrix to more accurately reflect the characteristics of the while-drilling gamma data, providing a reliable data foundation for the subsequent construction of a while-drilling geosteering dataset.

[0070] Furthermore, step A300 in the method provided in the embodiment of the present application includes:

[0071] A310: Construct a row vector V2, wherein the row vector V2 includes 2 8 Item elements are values ​​1 to 2 with positive serial distribution 8 .

[0072] A320: Copy the row vector V2 Y times, perform column distribution, and construct an initial power-law dimension-raising matrix, wherein the matrix size of the initial power-law dimension-raising matrix is ​​Y×2 8 .

[0073] A330: Perform correction calculation on the initial power-law dimension-raising matrix to determine a first power-law dimension-raising matrix.

[0074] In the embodiment of the present application, the positive serialization distribution refers to the element distribution mode of the row vector V2, that is, the row vector V2 includes 2 8 Item elements, these elements are from 1 to 2 8 Arranged in sequence, they show a pattern of increasing order and the difference between adjacent elements is 1.

[0075] In one embodiment, oil production is carried out in the target area, the first gamma data while drilling has been obtained, and the number of detections is Y times (such as Y=100 times). First, construct a row vector V2. V2 contains 2 8 Item elements, from 1 to 2 8 It is positively serialized and distributed. This ordered vector provides a basic numerical sequence for subsequent matrix construction.

[0076] Next, the row vector V2 is replicated Y times and distributed in columns to construct the initial power-law dimension-raising matrix. The size of the initial power-law dimension-raising matrix is ​​Y×2 8 (i.e. 100×2 8 ), each row of which is the same vector V2, initially forming a data matrix with a certain dimensional structure.

[0077] However, the initial matrix cannot directly meet the requirements, and it needs to be modified and calculated to determine the first power-law dimension-upgraded matrix. The specific steps are described in detail in A331-A335.

[0078] Through the above steps, the first power-law dimensionality-increasing matrix is ​​constructed, which increases the dimension of the data and also mines the potential characteristics of the data through a variety of mathematical transformations.

[0079] Furthermore, step A330 in the method provided in the embodiment of the present application includes:

[0080] A331: For the first while-drilling gamma data, obtain a second matrix column vector, perform mathematical calculations on the elements in the second matrix column vector, and obtain the vector V3.

[0081] A332: Use the first vector element in vector V3 to correct the first row of the first power-law dimensionality-raising matrix until the Yth vector element in V3 corrects the Yth row of the first power-law dimensionality-raising matrix, and determine a corrected power-law dimensionality-raising matrix.

[0082] A333: Subtract 2 from each matrix element in the first-order modified power-law dimension-raising matrix 8 And take the absolute value, perform the power operation, and determine the quadratic modified power law dimensionality-raising matrix, where the exponent of the power operation is a real number greater than 1.

[0083] A334: Introduce coefficient α, calculate the product of coefficient α and the quadratic corrected power-law dimension-raising matrix as the cubic corrected power-law dimension-raising matrix.

[0084] A335: Subtract 2 from the cubic modified power law dimension-raising matrix 8 Then take the absolute value as the first power-law dimension-increased matrix.

[0085] Optionally, first, a second matrix column vector is taken from the first while-drilling gamma data, and a specific mathematical calculation is performed on each element in the second matrix column vector to obtain a vector V3. The specific calculation method is to subtract the minimum gamma data from each matrix element, divide it by the gamma amplitude, and then multiply it by 2 8 , then round the result to the nearest integer, and at the same time, the value of the element in vector V3 that exceeds 2 8 The correction is 2 8 , element values ​​of 0 are corrected to 1.

[0086] After obtaining vector V3, the initial power law dimension-raising matrix (whose size is Y×2 8 , assuming it is 100×2 8 ). For example, if the value of the first vector element is 5 and an element in the first row of the initial power-law dimension-raising matrix is ​​8, according to the first-order correction method, if the row element in the first row is greater than the first vector element, it is corrected to 1, then this 8 will be corrected to 1. Following this step, the second row of the initial power-law dimension-raising matrix is ​​corrected using the second vector element in V3 until the Yth vector element in V3 corrects the Yth row of the initial power-law dimension-raising matrix, thereby determining the first-order corrected power-law dimension-raising matrix.

[0087] Next, the first modified power law dimension-raising matrix is ​​further processed. Each matrix element is subtracted by 2 8 And take the absolute value, and then perform a power operation (the exponent of the power operation is a real number greater than 1, assuming the exponent is 1.5) to obtain a quadratic modified power law dimensional matrix.

[0088] After that, the coefficient α is introduced, and the coefficient α is obtained by 2 8Subtract 1 and then divide by the maximum matrix element in the quadratic modified power law dimension-raising matrix to obtain. For example, if the maximum matrix element in the quadratic modified power law dimension-raising matrix is ​​35, then α=(2 8 -1)÷35. Calculate the product of the coefficient α and the quadratic modified power-law dimension-raising matrix to obtain the cubic modified power-law dimension-raising matrix.

[0089] Finally, subtract 2 from the cubic modified power law dimension-raising matrix 8 Then take the absolute value to obtain the final first power-law dimension-upgraded matrix.

[0090] Through the above series of correction calculation steps, the first power-law dimension-increasing matrix constructed can more accurately present the characteristics and laws of the while-drilling gamma data compared to the initial matrix.

[0091] Furthermore, step A336 in the method provided in the embodiment of the present application includes:

[0092] A336-1: The mathematical calculation method of vector V3 is: subtract the minimum gamma data from each matrix element in the second matrix column vector, divide by the gamma amplitude and multiply by 2 8 , round the calculated vector to the nearest integer, where the element value in vector V3 exceeds 2 8 The matrix element is corrected to 2 8 , matrix elements with element values ​​of 0 are corrected to 1.

[0093] A336-2: The primary correction method is: if the row element in the first row is greater than the first vector element, it is corrected to 1.

[0094] A336-3: coefficient α passes 2 8 Obtained by subtracting 1 and dividing by the maximum matrix element in the quadratic modified power law dimension-raising matrix.

[0095] Specifically, in an oil production project in a target area, a first while-drilling gamma data matrix has been obtained, wherein the second matrix column vector contains a series of gamma data. First, vector V3 is calculated in step A331.

[0096] Then, a correction is performed. Taking the first vector element in vector V3 as the benchmark, the first row of the first power-law dimension-raising matrix is ​​corrected. In actual operation, if a row element in the first row is larger than the first vector element, it will be corrected to 1. For example, if the first vector element is 5 and there is an element 8 in the first row, then this 8 will be corrected to 1. In this way, the second row of the first power-law dimension-raising matrix is ​​corrected with the second vector element in V3, until the Yth row is corrected with the Yth vector element, thus obtaining a corrected power-law dimension-raising matrix.

[0097] Finally, calculate the coefficient α. After obtaining the quadratic modified power law dimension-raising matrix, find the largest matrix element. 8 The result of subtracting 1 is then divided by the largest matrix element to obtain the coefficient α. For example, if the largest matrix element in the quadratic modified power law dimensioning matrix is ​​30, then α=(2 8 -1)÷30. This coefficient α is used to further adjust the quadratic corrected power-law dimension-raising matrix. Multiply it with the quadratic corrected power-law dimension-raising matrix to obtain the cubic corrected power-law dimension-raising matrix, and then subtract 2 from the cubic corrected power-law dimension-raising matrix. 8 And take the absolute value to finally determine the first power-law dimension-raising matrix.

[0098] Through the above steps, the characteristics of the while-drilling gamma data can be effectively mined, the quality and dimension of the data can be improved, and more valuable data support can be provided for the subsequent establishment of a while-drilling geosteering dataset for deep learning.

[0099] Furthermore, step A400 in the method provided in the embodiment of the present application includes:

[0100] A410: Traverse the gamma-ray data while drilling, and perform calculations and associations based on the statistical indicator matrix and the power-law dimension-raising matrix for drilling in each target area.

[0101] A420: Introducing a matrix partitioning rule, wherein the matrix partitioning rule is determined by a preset partitioning method and a preset partitioning size.

[0102] A430: According to the matrix partitioning rule, traverse the statistical indicator matrix and the power-law dimension-raising matrix associated with the mapping and perform matrix partitioning to construct a partition sub-matrix set.

[0103] A440: Integrate the segmentation sub-matrix set to generate the while-drilling geosteering dataset.

[0104] In the embodiment of the present application, matrix segmentation is to traverse the statistical indicator matrix and the power-law dimension-raising matrix associated with the mapping according to a preset segmentation method and a preset segmentation size, and perform a segmentation operation to finally construct a segmentation sub-matrix set.

[0105] Optionally, when oil production is carried out in the target area, a large amount of gamma data while drilling has been obtained according to the above steps, and a first statistical indicator matrix and a first power-law dimension-raising matrix are constructed for these data. First, the gamma data while drilling of all wells in the target area are traversed, and the corresponding first statistical indicator matrix and the first power-law dimension-raising matrix are extracted for each well (such as the first well), and data association calculation is performed. For example, assuming that the number of explorations of a well is Y=100, the first statistical indicator matrix is ​​100 rows and 5 columns (S=5 statistical indicators), and the first power-law dimension-raising matrix is ​​100 rows and 2 8By matching the statistical features (such as mean and standard deviation) corresponding to the detection depth and the features after dimensionality increase (such as dimensional features after power law transformation) by row, the preliminary fusion of multi-dimensional data is achieved.

[0106] Next, we introduce the matrix partitioning rule and perform structured partitioning on the associated matrix using the preset partitioning method (such as row-by-row sliding window, fixed block partitioning, the selection method is described below) and size (such as 10 rows × 10 columns of submatrices). For example, set the preset partitioning size to 10 rows × 30 columns (covering 5 columns of the statistical indicator matrix and 2 columns of the power law dimension matrix). 8 The segmentation window is slid along the detection depth direction to ensure that each sub-matrix contains the complete feature combination of continuous depth intervals.

[0107] Step e: Segmentation by row sliding window. This approach can be used to capture the continuous variation of LWD gamma data across depth (e.g., the gradual change in formation lithology with depth). For example, assuming the number of detections is Y = 100, the preset sliding window size is 10 rows, and the step size is 5 rows. Starting from rows 1-10, the data is segmented by sliding backward 5 rows at a time to produce sub-matrices containing continuous detection data (e.g., rows 1-10, 6-15, etc.). This approach preserves the time / depth sequence correlation of the data, facilitating deep learning models to learn the feature associations between adjacent detection points. It is suitable for scenarios where geological features vary continuously.

[0108] Step f: Fixed block segmentation. If the data needs to be divided into independent, standardized units (e.g., to adapt to model input size or parallel computing requirements), fixed block segmentation is used. For example, the matrix is ​​segmented into fixed dimensions (e.g., 20 rows × 30 columns), with each block containing a fixed amount of detection data (20 detections) and its corresponding statistical indicators and power-law dimensionality-increasing features. This approach is suitable for structured data processing, ensuring that each submatrix has a uniform dimension, facilitating batch model input and feature extraction. It is suitable for scenarios where geological data features are relatively stable and require standardization.

[0109] Then, the mapping matrix is ​​partitioned according to the partitioning rule to construct a partition sub-matrix set. Assume that the original matrix is ​​100 rows and 2 8 +5 columns (5 columns of statistical indicators +2 8 Column power law characteristics), according to 10 rows × (2 8 After column segmentation, 10 sub-matrices can be obtained. Each sub-matrix corresponds to the complete feature vectors of 10 detection data, forming a set of sub-matrices containing local geological characteristics.

[0110] Finally, these sub-matrices are integrated. During the integration process, the sub-matrices are merged and filtered to integrate the effective information from each sub-matrix, ultimately generating a geosteering while drilling dataset. This dataset retains the statistical regularities of the original gamma data (such as the mean and variance reflecting formation stability) while incorporating the nonlinear characteristics of power-law dimensionality increase (such as the spatial distribution of depth and gamma values), providing multi-dimensional, structured input data for deep learning models.

[0111] By generating a geosteering while drilling dataset, we provide high-quality data support for subsequent deep learning models. Using this data, deep learning models can more accurately analyze geological conditions and achieve more precise geosteering while drilling, thereby improving the efficiency and success rate of oil production.

[0112] In summary, the method for establishing a while-drilling geosteering dataset for deep learning provided by the embodiments of the present application has the following technical effects:

[0113] This application controls a downhole gamma meter to acquire downhole gamma data and calculates gamma amplitudes. A first statistical indicator matrix and a first power-law dimension-increasing matrix are constructed for the first downhole gamma data. After matrix correlation calculations and segmentation according to preset rules, the segmented sub-matrices are integrated to establish a downhole geosteering dataset and store it in a geological data center. This achieves the technical effect of establishing a downhole geosteering dataset based on different data processing steps and matrix operations, reducing the difficulty of data processing, simplifying the technical process, and meeting real-time requirements.

[0114] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0115] Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A method for establishing a while-drilling geosteering dataset for deep learning, characterized in that: The method comprises: By controlling a while-drilling gamma measurement instrument, drilling in a target area is detected to determine while-drilling gamma data, wherein the while-drilling gamma data is in matrix form, and the difference between the maximum and minimum values ​​in the while-drilling gamma data is taken as the gamma amplitude; For the first while-drilling gamma data, statistical indicators are introduced, and a first statistical indicator matrix is ​​constructed by performing statistical analysis and data reconstruction, wherein the first while-drilling gamma data is any one of the while-drilling gamma data; For the first statistical indicator matrix, by introducing row vectors, performing multiple rounds of matrix correction calculations to construct a first power-law dimension-raising matrix; The first statistical indicator matrix and the first power-law dimension-raising matrix are integrated to establish a while-drilling geosteering dataset, and the while-drilling geosteering dataset is stored in a geological data center.

2. The method for establishing a while-drilling geosteering dataset for deep learning according to claim 1, wherein: Drilling in the target area to determine the gamma data while drilling, including: For the first well, detect and determine Y gamma data and perform data preprocessing, where Y is the number of detections; Using the amount of detection data as a matrix row vector, the detection depth as a first matrix column vector, and the gamma data as a second matrix column vector, matrix conversion is performed on the Y-item gamma data to obtain first while-drilling gamma data, wherein the first well is any well in the target area; The first while drilling gamma data is added to the while drilling gamma data.

3. The method for establishing a while-drilling geosteering dataset for deep learning according to claim 1, wherein: The gamma amplitude is obtained by calculating the difference between the maximum gamma data and the minimum gamma data of L while-drilling gamma data of the wells drilled in the target area, where L is the number of wells drilled in the target area.

4. The method for establishing a while-drilling geosteering dataset for deep learning according to claim 1, wherein: Statistical indicators are introduced, and the first statistical indicator matrix is ​​constructed through statistical analysis and data reconstruction, including: Determining a statistical indicator, wherein the statistical indicator includes S items, where S is a positive integer greater than or equal to 1; With Y as the number of matrix row vectors, the number of matrix column vectors determined by statistical indicators, and the statistical range as a constraint, a statistical analysis is performed on the first while-drilling gamma data to calculate the first statistical indicator matrix, wherein the statistical range is set to n, where n is a positive integer greater than 9.

5. The method for establishing a while-drilling geosteering dataset for deep learning according to claim 4, wherein: Calculating the first statistical indicator matrix includes: For the matrix column vector V1 corresponding to the first statistical indicator, reconstruct a first matrix column vector of the first statistical indicator matrix using the first element and the first n-1 elements based on the calculated value of the first statistical indicator, wherein the first element is any matrix column vector element in the matrix column vector V1; The statistical indicators are traversed, and statistical indicator calculation is performed on each matrix column vector until the S-th matrix column vector is determined, and the first matrix column vectors are integrated until the S-th matrix column vector is determined as the first statistical indicator matrix.

6. The method for establishing a while-drilling geosteering dataset for deep learning according to claim 5, wherein: When the first element is smaller than n, a uniform symbol is used to replace the reconstructed matrix elements of the first matrix column vector.

7. The method for establishing a while-drilling geosteering dataset for deep learning according to claim 1, wherein: Construct the first power-law dimension-raising matrix, including: Construct a row vector V2, wherein the row vector V2 includes 2 8 Item elements are values ​​1 to 2 with positive serial distribution 8 ; The row vector V2 is copied Y times and distributed in columns to construct an initial power-law dimension-raising matrix, where the matrix size of the initial power-law dimension-raising matrix is ​​Y×2 8 ; The initial power-law dimension-raising matrix is ​​corrected and calculated to determine a first power-law dimension-raising matrix.

8. The method for establishing a while-drilling geosteering dataset for deep learning according to claim 7, wherein: Performing a correction calculation on the initial power-law dimension-raising matrix to determine a first power-law dimension-raising matrix includes: For the first while-drilling gamma data, obtain a second matrix column vector, perform mathematical calculations on the elements in the second matrix column vector, and obtain the vector V3; Using the first vector element in vector V3, correcting the first row of the first power-law dimension-raising matrix until the Yth vector element in V3 corrects the Yth row of the first power-law dimension-raising matrix, thereby determining a corrected power-law dimension-raising matrix. Subtract 2 from each matrix element in the first modified power law dimension-raising matrix 8 And take the absolute value, perform the power operation, and determine the quadratic modified power law dimension-raising matrix, where the exponent of the power operation is a real number greater than 1; Introducing a coefficient α, and calculating the product of the coefficient α and the quadratic corrected power-law dimension-raising matrix as a cubic corrected power-law dimension-raising matrix; Subtract 2 from the cubic modified power law dimension-raising matrix 8 Then take the absolute value as the first power-law dimension-increased matrix.

9. The method for establishing a while-drilling geosteering dataset for deep learning according to claim 8, wherein: include: The mathematical calculation method of vector V3 is: subtract the minimum gamma data from each matrix element in the second matrix column vector, divide by the gamma amplitude and multiply by 2 8 , round the calculated vector to the nearest integer, where the element value in vector V3 exceeds 2 8 The matrix element is corrected to 2 8 , the matrix elements with element values ​​of 0 are corrected to 1; The primary correction method is: if the row element in the first row is greater than the first vector element, it is corrected to 1; The coefficient α is 2 8 Obtained by subtracting 1 and dividing by the maximum matrix element in the quadratic modified power law dimension-raising matrix.

10. The method for establishing a while-drilling geosteering dataset for deep learning according to claim 1, wherein: Integrating the first statistical indicator matrix and the first power-law dimension-raising matrix to establish a while-drilling geosteering data set includes: Traversing the gamma-ray data while drilling, and performing calculation and correlation based on the statistical indicator matrix and the power law dimension-raising matrix for drilling in each target area; Introducing a matrix partitioning rule, wherein the matrix partitioning rule is determined by a preset partitioning method and a preset partitioning size; According to the matrix partitioning rule, traverse the statistical indicator matrix and the power-law dimension-raising matrix associated with the mapping and perform matrix partitioning to construct a partition submatrix set; The segmentation sub-matrix set is integrated to generate the while-drilling geosteering data set.

Images