Artificial intelligence interpretation method for multi-arm hole diameter logging data

Abstract

The invention provides an artificial intelligence interpretation method for multi-arm hole diameter logging data. The artificial intelligence interpretation method for the multi-arm hole diameter logging data comprises the steps that 1, imaging processing is conducted on logging interpretation data; 2, building a working condition interpretation model based on the convolutional neural network; step 3, training the built working condition interpretation model; 4, the well trained working condition interpretation model is utilized, and according to the collected actual multi-arm hole diameter logging data, the multi-arm hole diameter logging working condition is judged. According to the artificial intelligence interpretation method for the multi-arm hole diameter logging data, by combining the high-precision data acquisition capability of multi-arm hole diameter logging and the efficient intelligent analysis advantage of artificial intelligence, the interpretation efficiency and interpretation precision of multi-arm hole diameter logging interpretation work are effectively improved, and the labor intensity of workers is reduced.

Description

Technical Field

[0001] This invention relates to the field of oil and gas exploration and development technology, and in particular to an artificial intelligence interpretation method for multi-arm caliper logging data. Background Technology

[0002] Currently, the interpretation of multi-arm caliper logging relies on the experience of operational personnel for manual interpretation, which mainly presents the following two problems:

[0003] (1) Manual interpretation of multi-arm caliper logging results is labor-intensive and inefficient.

[0004] In practical applications, multi-arm caliper logging results often require manual interpretation and analysis, which is not only labor-intensive but also inefficient. As one of the most commonly used casing quality evaluation logging projects in oilfields, multi-arm caliper logging has an average interpretation time of about 3 hours per well, accounting for 43.3% of the total interpretation time of all logging operations.

[0005] Manually interpreting multi-arm caliper logging results requires experts or professionals to review, analyze, and judge a large amount of multi-arm caliper logging data one by one. This not only consumes a lot of time and energy, but is also easily affected by human factors, such as fatigue and subjective judgment, which can lead to inaccurate or inconsistent interpretation results.

[0006] As multi-arm caliper logging data continues to increase in quantity and complexity, the difficulty and workload of manual interpretation are also constantly increasing. This not only makes the entire multi-arm caliper logging interpretation process more cumbersome and time-consuming, but may also lead to the omission or misjudgment of some important information, thereby affecting the final analysis results and application effectiveness.

[0007] (2) Interpretation of multi-arm caliper logging relies on human experience, and the efficiency of identification needs to be further improved.

[0008] The interpretation of multi-arm caliper logging results often relies on human experience. This means that each time new multi-arm caliper logging data or situations arise, professionals must rely on their extensive knowledge and experience to make judgments. However, this approach is not only inefficient but also prone to overlooking data or situations with subtle changes. Because everyone's experience and understanding are limited, it's difficult to guarantee accurate judgments every time.

[0009] Multi-arm caliper logging data consists of more than forty curves, relying on manual experience and visual observation, which is labor-intensive, prone to fatigue, and highly susceptible to overlooking problems. Currently, field verification shows that the interpretation accuracy of multi-arm caliper logging is 80%.

[0010] Chinese patent application CN202311118038.4 discloses a method, system, device, and storage medium for predicting casing deformation conditions in horizontal wells. The method includes: constructing a large casing deformation data table using engineering parameters and geological parameters as feature data and casing deformation condition information as labeled data; performing correlation matrix analysis on the relationships between continuous feature parameters based on the large casing deformation data table to generate analysis results; performing feature dimensionality reduction processing on the analysis results to calculate feature parameters with autocorrelation below a set threshold; performing quantitative analysis on the relationship between casing deformation condition parameters and continuous feature parameters to obtain sensitive feature parameters; and performing similarity relationship analysis between casing deformation condition parameters and discrete feature parameters to obtain the main controlling factors affecting casing deformation conditions. Using this exemplary embodiment, a machine learning intelligent classification decision system can be used to minimize the uncertainty of casing deformation prediction and improve the accuracy and generalization ability of the casing deformation prediction model.

[0011] Chinese patent application CN202211204512.0 discloses a method for establishing and applying a paleowater depth prediction model based on well logging curves. The method includes: acquiring well logging curves and paleowater depth data along the well depth direction of a known well; correcting the paleowater depth data using a data correction method to obtain corrected paleowater depth data along the well depth direction; obtaining a paleowater depth-sensitive well logging curve whose correlation with the paleowater depth data exceeds a predetermined level based on the correlation between the corrected paleowater depth data and the well logging curves; using the paleowater depth-sensitive well logging curve as sample parameters and the corrected paleowater depth data as sample labels to train a machine learning model, thereby obtaining a prediction model for predicting paleowater depth data. The prediction model described in this paper can improve prediction accuracy and generalization ability.

[0012] Chinese patent application CN200910307202.X discloses a system and method for interpreting oil logging data. This system includes a data preprocessing module, a data import module, a data editing module, a data interpretation module, and an MS-SQL Server relational database. This invention overcomes the technical problems of complex processing steps and inconvenient data retrieval, comparison, and verification in existing oil logging data interpretation systems. This invention offers advantages such as enabling information management of oil wells, storing logging data in any format currently available in the field of logging technology, and obtaining information about any oil well through well identification.

[0013] The existing technologies described above differ significantly from this invention and do not solve the problem of interpreting multi-arm caliper logging curves. With the continuous development of petroleum exploration technology, the requirements for interpreting multi-arm caliper logging curves are becoming increasingly stringent. Interpretation methods also require more advanced and efficient technical means to support them, further improving the consistency and timeliness of multi-arm caliper logging interpretation results, enhancing the accuracy of curve identification, and reducing reliance on human experience. Therefore, we have invented a new artificial intelligence interpretation method for multi-arm caliper logging data. Summary of the Invention

[0014] The purpose of this invention is to provide an artificial intelligence interpretation method for multi-arm caliper logging data, which enables intelligent recognition and analysis of multi-arm caliper images and achieves rapid and accurate interpretation of multi-arm caliper logging data.

[0015] The objective of this invention can be achieved through the following technical measures: an artificial intelligence interpretation method for multi-arm caliper logging data, comprising:

[0016] Step 1: Graphicalize the well logging interpretation data;

[0017] Step 2: Build a working condition interpretation model based on convolutional neural networks;

[0018] Step 3: Train the constructed working condition interpretation model;

[0019] Step 4: Using the trained operating condition interpretation model, determine the operating conditions of the multi-arm caliper logging based on the collected actual multi-arm caliper logging data.

[0020] The objective of this invention can also be achieved through the following technical measures:

[0021] In step 1, through

[0022]

[0023] The grid data signal is converted into a two-dimensional grayscale image with pixel values ​​between 0 and 255; where P(i,j) and P′(i,j) represent the values ​​of the grid data in the i-th row and j-th column before and after normalization, respectively, and R i This represents the data in the i-th row; the round function rounds the value to the nearest integer by the specified number of decimal places, and by default retains 0 decimal places.

[0024] Step 2 includes:

[0025] Step 21: Construct three convolutional feature extraction layers of different sizes to process graphic features of different scales;

[0026] Step 22: Each convolutional feature extraction layer consists of three layers of convolution-pooling-activation network;

[0027] Step 23: Repeat step 22, feeding in convolutional kernels of different sizes in sequence to extract the final feature map;

[0028] Step 24: Use the Softmax classifier to classify the multi-arm caliper logging interpretation conditions.

[0029] In step 21, the first layer uses convolutional kernels of three sizes: 3*3, 5*5, and 7*7 to extract the features of the current layer; the second layer uses convolutional kernels of three sizes: 7*7, 9*9, and 11*11 to extract the features of the current layer; and the third layer uses convolutional kernels of three sizes: 9*9, 11*11, and 13*13 to extract the features of the current layer.

[0030] Step 22 includes:

[0031] a. The features of each convolutional layer are extracted using the following formula.

[0032] S(i,j)=(K*I)(i,j)=Σ m Σ n I(i+m, j+n)K(m, n);

[0033] Where I represents the input grid data; K represents the convolution kernel grid data; S represents the data after feature mapping, i.e., the extracted feature grid data; S(i,j) represents the processed feature mapping (output) grid data in the i-th row and j-th column; (K*I) represents the matrix multiplication operation between the convolution kernel and the input grid data, its expansion is ∑ m ∑ n I(i+m,j+n)K(m,n); m and n represent the width and height of the convolution kernel, respectively;

[0034] b. After extracting grid features, perform batch standardization to overcome the problem of difficulty in training caused by increasing the number of neural network layers;

[0035] c. The convolutional layer uses the ReLU activation function, which is a linear rectified function, and its formula is:

[0036] σ(x) = max(0, x)

[0037] x is the input value sent to the activation function, and the max function returns the maximum value among a set of values ​​or specified parameters.

[0038] In step 23, repeat steps a, b, and c above, feeding in convolutional kernels of different sizes in sequence to extract the final feature map; the obtained feature map is then transformed into a one-dimensional output vector y = [y1, y2, ..., y3] of length L through a fully connected layer. l ].

[0039] In step 24, a softmax operation is performed on the output vector using the following formula:

[0040]

[0041] A probability distribution is obtained, and the working condition category corresponding to the maximum probability value is output.

[0042] In step 3, the working condition interpretation model is trained using a large amount of labeled multi-arm caliper logging interpretation sample data, so that the model can learn the feature representation and classification rules of caliper logging data maps of different working condition types, including: normal, reduced diameter, enlarged diameter, corrosion, perforation, faulting and bending.

[0043] In step 3, the criterion for judging normal working conditions is: compared with the standard size value, all measuring arm values ​​decrease, and the average reduction is within the allowable error, which is judged as normal;

[0044] The criteria for judging the reduced diameter condition are: compared with the standard size value, all measuring arm values ​​decrease, and the average reduction exceeds the allowable error, which is judged as a reduced diameter condition;

[0045] The criteria for judging the expansion condition are: compared with the standard size value, all measuring arm values ​​increase, and the average increase exceeds the allowable error, which is judged as expansion.

[0046] The criteria for judging corrosion conditions are: compared with the standard size value, if the increase of some measuring arm values ​​is close to or exceeds the sleeve wall thickness, and the increase of the average value exceeds the allowable error, it is judged as corrosion;

[0047] The criteria for judging the perforation condition are: if, compared with the standard size value, the increase of some measuring arm values ​​is close to or exceeds the sleeve wall thickness, it is judged as perforation;

[0048] The criteria for judging the faulty working condition are: at the same depth, all measuring arm values ​​change abruptly, all measuring arm values ​​increase, and the increase in all arm values ​​is close to or exceeds the casing wall thickness, which is judged as faulty working condition;

[0049] The criteria for judging bending conditions are: compared with the standard dimension value, some measuring arm values ​​increase, the measuring arm values ​​in the relative orientation decrease, and the deformation is continuous, which is judged as bending.

[0050] In step 4, the preprocessed actual data is used as test data and input into the pre-trained working condition interpretation model, which outputs the working condition category corresponding to the maximum probability value.

[0051] The objective of this invention can also be achieved through the following technical measures: an artificial intelligence interpretation system for multi-arm caliper logging data, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps corresponding to the method described.

[0052] This invention presents an artificial intelligence interpretation method for multi-arm caliper logging data. It involves AI-based multi-arm caliper logging interpretation technology, employing a multi-arm caliper logging interpretation method based on convolutional neural network (CNN) technology. This method enables intelligent interpretation of raw multi-arm caliper logging data. The invention collects multi-arm caliper logging data, constructs a sample library categorized by multi-arm caliper logging interpretation type, and uses CNN technology to build an intelligent interpretation model based on multi-arm caliper logging data images. This model intelligently identifies and analyzes multi-arm caliper images, achieving rapid and accurate interpretation of multi-arm caliper logging data. This technical solution effectively improves the interpretation efficiency and accuracy of multi-arm caliper logging work and reduces manual labor intensity by combining the high-precision data acquisition capabilities of multi-arm caliper logging with the efficient and intelligent analysis advantages of artificial intelligence.

[0053] This invention presents an artificial intelligence interpretation method for multi-arm caliper logging, addressing the current challenges in interpreting multi-arm caliper logging data in reservoir development. By introducing artificial intelligence technology, it achieves automated and intelligent interpretation of multi-arm caliper logging data. Through the construction of an efficient artificial intelligence interpretation model, and utilizing deep learning, machine learning, and other technologies, it intelligently analyzes and processes multi-arm caliper logging data, automatically extracting useful information and providing accurate interpretation of logging conditions. These conditions can be categorized into several types, including reduced diameter, increased diameter, corrosion, perforation, and faulting.

[0054] This approach improves the efficiency and accuracy of multi-arm caliper logging data interpretation, reduces the labor intensity and error rate of manual interpretation, and provides more reliable and efficient data support for reservoir development. Simultaneously, the project's implementation will promote intelligent development in the reservoir development field, injecting new momentum into the industry's sustainable development. Attached Figure Description

[0055] Figure 1 This is a diagram of the original 40-arm logging curve within a certain depth range in a specific embodiment of the present invention;

[0056] Figure 2 This is a grayscale processed 40-arm logging curve image within a certain depth range in a specific embodiment of the present invention;

[0057] Figure 3 This is a schematic diagram illustrating the working condition interpretation model of three convolutional feature extraction layers of different sizes in a specific embodiment of the present invention;

[0058] Figure 4 This is a schematic diagram of the interface of the multi-arm caliper logging interpretation sample label processing tool program in a specific embodiment of the present invention;

[0059] Figure 5This is a flowchart of a specific embodiment of the artificial intelligence interpretation method for multi-arm caliper logging data of the present invention;

[0060] Figure 6 This is a typical illustration of a normal operating condition in a specific embodiment of the present invention;

[0061] Figure 7 This is a typical illustration of a reduced-diameter working condition in a specific embodiment of the present invention;

[0062] Figure 8 This is a typical illustration of a diameter expansion working condition in a specific embodiment of the present invention;

[0063] Figure 9 This is a typical illustration of a corrosion condition type in a specific embodiment of the present invention;

[0064] Figure 10 This is a typical illustration of a perforation condition type in a specific embodiment of the present invention;

[0065] Figure 11 This is a typical illustration of a fault condition type in a specific embodiment of the present invention;

[0066] Figure 12 This is a typical illustration of a bending condition type in a specific embodiment of the present invention. Detailed Implementation

[0067] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0068] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, and / or combinations thereof.

[0069] like Figure 5 As shown, Figure 5 This is a flowchart of the artificial intelligence interpretation method for multi-arm caliper logging data according to the present invention. The artificial intelligence interpretation method for multi-arm caliper logging data includes:

[0070] Step 1: Graphicalize the well logging interpretation data.

[0071] Because different wells have different casing inner diameters, the logging data from different wells are not comparable laterally. To ensure that the data input to the convolutional neural network has consistent dimensions and format, the raw data needs to be graphically processed.

[0072] That is, through This function converts grid data signals into a 2D grayscale image with pixel values ​​between 0 and 255. In the formula, P(i,j) and P′(i,j) represent the values ​​of the grid data in the i-th row and j-th column before and after normalization, respectively, and R represents the data in the i-th row. The round function rounds the value to the nearest integer by the specified number of decimal places, retaining 0 decimal places by default.

[0073] Step 2: Build a working condition interpretation model based on convolutional neural networks

[0074] 1) Construct three convolutional feature extraction layers of different sizes to process image features at different scales. The first layer uses convolutional kernels of 3*3, 5*5, and 7*7 sizes to extract features of the current layer; the second layer uses convolutional kernels of 7*7, 9*9, and 11*11 sizes to extract features of the current layer; and the third layer uses convolutional kernels of 9*9, 11*11, and 13*13 sizes to extract features of the current layer.

[0075] 2) Each convolutional feature extraction layer consists of three layers of convolution-pooling-activation network.

[0076] a. The features of each convolutional layer are extracted using the following formula.

[0077] S(i,j)=(K*I)(i,j)=Σ m Σ n I(i+m,j+n)K(m,n)

[0078] Where I represents the input grid data; K represents the convolution kernel (e.g., 3*3) grid data; S represents the data after feature mapping, i.e., the extracted feature grid data; S(i,j) represents the processed feature mapping (output) grid data in the i-th row and j-th column; (K*I) represents the matrix multiplication operation between the convolution kernel and the input grid data, and its expansion is:

[0079] ∑ m ∑ n I(i+m,j+n)K(m,n);

[0080] m and n represent the width and height of the convolution kernel (e.g., 3*3), respectively. In this example, m is 3 and n is 3.

[0081] b. After extracting grid features, batch standardization is performed to overcome the problem of difficulty in training caused by increasing the number of neural network layers.

[0082] c. The ReLU activation function is used in the convolutional layer. It is a linear rectified function.

[0083] The formula is:

[0084] σ(x) = max(0, x)

[0085] x is the input value passed to the activation function. The max function is a widely used function in various programming environments and software. Its main function is to return the maximum value among a set of values ​​or specified parameters. For example, max(0,-0.3)=0; max(0,1.2)=1.2.

[0086] 3) Repeat steps a, b, and c above, feeding in convolutional kernels of different sizes in sequence to extract the final feature map. Transform the obtained feature map into a one-dimensional output vector y = [y1, y2, ..., y] of length L through a fully connected layer. l ].

[0087] 4) Use a Softmax classifier to classify multi-arm caliper logging interpretation conditions. Perform a Softmax operation on the output vector using the following formula:

[0088]

[0089] A probability distribution is obtained. The network will output the operating condition category corresponding to the maximum probability value.

[0090] Step 3: Train the established working condition interpretation model.

[0091] The model was trained using a large amount of labeled multi-arm caliper logging interpretation sample data, enabling it to learn the feature representations and classification rules of caliper logging data for different operating conditions. Typical examples of different operating conditions are shown below:

[0092] Normal: such as Figure 6 As shown, compared with the standard size value, all measuring arm values ​​decreased, and the average reduction was within the allowable error, which is considered normal.

[0093] Reduced diameter: such as Figure 7 As shown, compared with the standard size value, all measuring arm values ​​decreased, and the average shrinkage exceeded the allowable error, which was judged as a reduction in diameter.

[0094] Expanding diameter: such as Figure 8 As shown, compared with the standard size value, all measuring arm values ​​increased, and the average increase exceeded the allowable error, indicating that the diameter was expanded.

[0095] Corrosion: such as Figure 9As shown, compared with the standard size value, some measuring arm values ​​increased by a large amount close to or exceeding the sleeve wall thickness, and the average value increased by a large amount exceeding the allowable error, which was judged to be corrosion.

[0096] Perforation: such as Figure 10 As shown, compared with the standard size value, some measuring arm values ​​increased by a large amount, which is close to or exceeds the sleeve wall thickness, indicating perforation.

[0097] Misinterpretation: such as Figure 11 As shown, at the same depth, all measuring arm values ​​abruptly change, all measuring arm values ​​increase, and the increase in all arm values ​​approaches or exceeds the casing wall thickness, which is judged as a misalignment.

[0098] Bending: such as Figure 12 As shown, compared with the standard size value, some measuring arm values ​​are increased, while the measuring arm values ​​in the relative orientation are decreased, and the deformation is continuous, which is judged to be bending.

[0099] Step 4: Using the trained operating condition interpretation model, determine the operating conditions of the multi-arm caliper logging based on the collected actual multi-arm caliper logging data.

[0100] After preprocessing, the actual data is used as test data and input into the pre-trained working condition interpretation model. The network will output the working condition category corresponding to the maximum probability value. The working conditions can be divided into several categories such as diameter reduction, diameter expansion, corrosion, perforation, and faulting.

[0101] Compared with other methods, this method has the following advantages: (1) It starts entirely from process data, without requiring empirical knowledge and mechanism models of the process, and has strong applicability; (2) It has fast offline modeling and learning training speed, small online computation amount, and strong real-time performance; (3) It fully considers the correlation of multi-arm caliper logging data in depth and space, and uses convolution kernels of different sizes to better extract features related to the working conditions.

[0102] The artificial intelligence interpretation method for multi-arm caliper logging provided by this invention can effectively solve the problems of existing multi-arm caliper logging interpretation work relying on manual labor, resulting in a large workload, low efficiency, and interpretation quality dependent on experience and knowledge. By introducing artificial intelligence technology, the efficiency and accuracy of multi-arm caliper logging data interpretation can be improved, while reducing the cost and error rate of manual interpretation, alleviating the workload of frontline staff, and lowering labor costs. This method is easy to implement and requires no additional hardware equipment.

[0103] The present invention also provides an artificial intelligence interpretation system for multi-arm caliper logging, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps corresponding to the above-described method.

[0104] The following are several specific embodiments of the application of the present invention.

[0105] Example 1

[0106] In a specific embodiment 1 of the present invention, the artificial intelligence interpretation method for multi-arm caliper logging data of the present invention specifically includes:

[0107] A. Graphical processing of well logging interpretation data

[0108] 1) Wellbore testing instruments come in various types, such as 8-arm, 10-arm, 16-arm, and 40-arm. This embodiment uses the commonly used 40-arm wellbore testing instrument as an example for the invention.

[0109] Multi-arm logging data consists of data from 40 logging arms, forming a two-dimensional array denoted as R, with 40 columns horizontally and depth points vertically. (See appendix) Figure 1 .

[0110] 2) Due to the different casing inner diameters of different wells, the data from the 40 logging arms of different wells are not comparable laterally. To ensure that the data input to the convolutional neural network has consistent dimensions and format, the raw data needs to be graphically processed. That is, through... The grid data signal is converted into a two-dimensional grayscale image with pixel values ​​between 0 and 255. In the formula, P(i,j) and P′(i,j) represent the values ​​of the grid data in the i-th row and j-th column before and after normalization, respectively, and R... i This represents the data in the i-th row. See appendix. Figure 2 .

[0111] B. Build a working condition interpretation model based on convolutional neural networks

[0112] 1) Construct three convolutional feature extraction layers of different sizes to process image features at different scales. The first layer uses convolutional kernels of 3*3, 5*5, and 7*7 sizes to extract features; the second layer uses convolutional kernels of 7*7, 9*9, and 11*11 sizes; and the third layer uses convolutional kernels of 9*9, 11*11, and 13*13 sizes. (See appendix) Figure 3 .

[0113] 2) Each convolutional feature extraction layer consists of three layers of convolution-pooling-activation network.

[0114] a. The features of each convolutional layer are extracted using the following formula.

[0115] S(i,j)=(K*I)(i,j)=∑ m ∑ n I(i+m,j+n)K(m,n)

[0116] Where I represents the input grid data; K represents the convolution kernel (e.g., 3*3) grid data; and S represents the data after feature mapping, i.e., the extracted feature grid data.

[0117] b. After extracting grid features, batch standardization is performed to overcome the problem of difficulty in training caused by increasing the number of neural network layers.

[0118] c. The ReLU activation function is used in the convolutional layer. It is a linear rectified function.

[0119] Its formula is: σ(x)=max(0,x)

[0120] 3) Repeat steps a, b, and c above, feeding in convolutional kernels of different sizes in sequence to extract the final feature map. Transform the obtained feature map into a one-dimensional output vector y = [y1, y2, ..., y] of length L through a fully connected layer. l ].

[0121] 4) Use a Softmax classifier to classify multi-arm caliper logging interpretation conditions. Perform a Softmax operation on the output vector using the following formula:

[0122]

[0123] A probability distribution is obtained. The network will output the operating condition category corresponding to the maximum probability value.

[0124] C. Training and Application of the Operating Condition Interpretation Model

[0125] 1) The model was trained using a large amount of labeled multi-arm caliper logging interpretation sample data, enabling the model to learn the feature representation and classification rules of caliper logging data maps under different working conditions;

[0126] 2) After preprocessing the actual data, it is used as test data and input into the pre-trained network. The network will output the working condition category corresponding to the maximum probability value.

[0127] Example 2

[0128] In a specific embodiment 2 of the present invention, when using it, technicians first need to collect existing historical multi-arm caliper logging case information, parse and store the corresponding multi-arm caliper logging files, and standardize the management of the provided case information after parsing. Multi-arm caliper logging sample labeling work is carried out, using a combination of manual screening and self-learning technology to build a sample library. The collected historical multi-arm caliper logging case data is organized and analyzed, and manual sample calibration is performed (see attached). Figure 4Business experts, based on their relevant experience, reviewed existing multi-arm caliper logging case studies and formulated relevant business rules, such as those addressing issues like reduced diameter, increased diameter, deformation, and damage. Subsequently, model training and evaluation began, resulting in an intelligent diagnostic model for multi-arm caliper logging through multiple iterations. This model can then be used to automatically interpret logging data files requiring multi-arm caliper logging interpretation, derive interpretation conclusions, and perform graphical processing.

[0129] In practical implementation, the intelligent diagnostic model for multi-arm caliper logging can be customized and optimized according to actual needs and characteristics. For example, parameters such as the number of convolutional layers, the size of the convolutional kernel, and the stride can be adjusted to adapt to multi-arm caliper data of different sizes; a multi-channel input method can be adopted to consider multiple feature dimensions of the caliper data simultaneously; and advanced technologies such as attention mechanisms can be introduced to further improve the model's performance.

[0130] By introducing information technology and intelligent evaluation methods, the efficiency and accuracy of the multi-arm caliper logging interpretation process can be improved, the influence of human factors on the multi-arm caliper logging interpretation evaluation results can be reduced, the efficiency of multi-arm caliper logging interpretation can be improved, and the logging business technical service capabilities in reservoir development can be strengthened.

[0131] Example 3

[0132] In a specific embodiment 3 of the present invention, for cases where the features or operating conditions of the 40-arm multi-arm logging data input graphics are similar, constraints can be imposed based on operational experience to improve the accuracy and practicality of the multi-arm logging interpretation and evaluation results. The following example uses corrosion (…) Figure 9 ) and perforation ( Figure 10 The determination of the two interpretation results will be described using examples.

[0133] In distinguishing between corrosion and perforation interpretations, both exhibit a characteristic in their graphic features where one or more curves bulge to the right (shift or increase). For input data with similar graphic features, convolutional neural network (CNN) technology may output multiple results that meet the confidence requirements. For example, it might output an 83% probability of "corrosion" and an 87% probability of "perforation." Therefore, in such cases, the model output of the multi-arm caliper logging interpretation method based on CNN technology should be further optimized and improved, taking into account the actual field conditions and expert experience.

[0134] In terms of expert experience, the criteria for determining corrosion and perforation both involve the description that "compared to standard dimensions, some measured arm values ​​increase significantly, approaching or exceeding the casing wall thickness." In practice, corrosion occurs due to various factors, and as the corrosion deepens, perforation or cracks will appear in the tubing string. Corrosion should be a large longitudinal area, while perforation should be a series of small holes. Therefore, for feature depth points within a certain depth range whose output of the convolutional neural network concludes as corrosion or perforation, counting or distance-based judgments should be performed. By adding expert experience rules such as "if a continuous depth segment contains only corrosion or perforation conclusions, the interpretation conclusion should be modified to corrosion; if the depth distance between two adjacent corrosion or perforation conclusions exceeds 0.05 meters, the interpretation conclusion should be modified to perforation," the two interpretation results can be accurately distinguished, improving the accuracy and practicality of the multi-arm caliper logging interpretation model.

[0135] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0136] Except for the technical features described in the specification, all other technologies are known to those skilled in the art.

Claims

1. An artificial intelligence interpretation method for multi-arm caliper logging data, characterized in that, The artificial intelligence interpretation methods for this multi-arm caliper logging data include: Step 1: Graphicalize the well logging interpretation data; Step 2: Build a working condition interpretation model based on convolutional neural networks; Step 3: Train the constructed working condition interpretation model; Step 4: Using the trained operating condition interpretation model, determine the operating conditions of the multi-arm caliper logging based on the collected actual multi-arm caliper logging data.

2. The artificial intelligence interpretation method for multi-arm caliper logging data according to claim 1, characterized in that, In step 1, through Convert the grid data signal into a two-dimensional grayscale image with pixel values ​​between 0 and 255; In the formula, P(i,j) and P′(i,j) represent the values ​​of the grid data in the i-th row and j-th column before and after normalization, respectively, and R i This represents the data in the i-th row; the round function rounds the value to the nearest integer by the specified number of decimal places, and by default retains 0 decimal places.

3. The artificial intelligence interpretation method for multi-arm caliper logging data according to claim 1, characterized in that, Step 2 includes: Step 21: Construct three convolutional feature extraction layers of different sizes to process graphic features of different scales; Step 22: Each convolutional feature extraction layer consists of three layers of convolution-pooling-activation network; Step 23: Repeat step 22, feeding in convolutional kernels of different sizes in sequence to extract the final feature map; Step 24: Use the Softmax classifier to classify the multi-arm caliper logging interpretation conditions.

4. The artificial intelligence interpretation method for multi-arm caliper logging data according to claim 3, characterized in that, In step 21, the first layer uses convolutional kernels of three sizes: 3*3, 5*5, and 7*7 to extract the features of the current layer; the second layer uses convolutional kernels of three sizes: 7*7, 9*9, and 11*11 to extract the features of the current layer; and the third layer uses convolutional kernels of three sizes: 9*9, 11*11, and 13*13 to extract the features of the current layer.

5. The artificial intelligence interpretation method for multi-arm caliper logging data according to claim 3, characterized in that, Step 22 includes: a. The features of each convolutional layer are extracted using the following formula. S(i,j)=(K*I)(i,j)=∑ m ∑ n I(i+m,j+n)K(m,n) Where I represents the input grid data; K represents the convolution kernel grid data; S represents the data after feature mapping, i.e., the extracted feature grid data; S(i,j) represents the processed feature-mapped grid data in the i-th row and j-th column; (K*I) represents the matrix multiplication operation between the convolution kernel and the input grid data, and its expansion is ∑m∑ n I(i+m,j+n)K(m,n); m and n represent the width and height of the convolution kernel, respectively; b. After extracting grid features, perform batch standardization to overcome the problem of difficulty in training caused by increasing the number of neural network layers; c. The convolutional layer uses the ReLU activation function, which is a linear rectified function, and its formula is: σ(x) = max(0, x) x is the input value sent to the activation function, and the max function returns the maximum value among a set of values ​​or specified parameters.

6. The artificial intelligence interpretation method for multi-arm caliper logging data according to claim 5, characterized in that, In step 23, repeat steps a, b, and c above, feeding in convolutional kernels of different sizes in sequence to extract the final feature map; the obtained feature map is then transformed into a one-dimensional output vector y = [y1y2, ..., y] of length L through a fully connected layer. l ].

7. The artificial intelligence interpretation method for multi-arm caliper logging data according to claim 6, characterized in that, In step 24, a softmax operation is performed on the output vector using the following formula: A probability distribution is obtained, and the working condition category corresponding to the maximum probability value is output.

8. The artificial intelligence interpretation method for multi-arm caliper logging data according to claim 1, characterized in that, In step 3, the working condition interpretation model is trained using a large amount of labeled multi-arm caliper logging interpretation sample data, so that the model can learn the feature representation and classification rules of caliper logging data maps of different working condition types, including: normal, reduced diameter, enlarged diameter, corrosion, perforation, faulting and bending.

9. The artificial intelligence interpretation method for multi-arm caliper logging data according to claim 8, characterized in that, In step 3, the criterion for judging normal working conditions is: compared with the standard size value, all measuring arm values ​​decrease, and the average reduction is within the allowable error, which is judged as normal; The criteria for judging the reduced diameter condition are: compared with the standard size value, all measuring arm values ​​decrease, and the average reduction exceeds the allowable error, which is judged as a reduced diameter condition; The criteria for judging the expansion condition are: compared with the standard size value, all measuring arm values ​​increase, and the average increase exceeds the allowable error, which is judged as expansion. The criteria for judging corrosion conditions are: compared with the standard size value, if the increase of some measuring arm values ​​is close to or exceeds the sleeve wall thickness, and the increase of the average value exceeds the allowable error, it is judged as corrosion; The criteria for judging the perforation condition are: if, compared with the standard size value, the increase of some measuring arm values ​​is close to or exceeds the sleeve wall thickness, it is judged as perforation; The criteria for judging the faulty working condition are: at the same depth, all measuring arm values ​​change abruptly, all measuring arm values ​​increase, and the increase in all arm values ​​is close to or exceeds the casing wall thickness, which is judged as faulty working condition; The criteria for judging bending conditions are: compared with the standard dimension value, some measuring arm values ​​increase, the measuring arm values ​​in the relative orientation decrease, and the deformation is continuous, which is judged as bending.

10. The artificial intelligence interpretation method for multi-arm caliper logging data according to claim 1, characterized in that, In step 4, the preprocessed actual data is used as test data and input into the pre-trained working condition interpretation model, which outputs the working condition category corresponding to the maximum probability value.

11. An artificial intelligence interpretation system for multi-arm caliper logging data, characterized in that, The artificial intelligence interpretation system for multi-arm caliper logging data includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps corresponding to the method described in any one of claims 1-10.

Images