Methods and devices for filling missing logging data

By performing wavenumber domain low-pass filtering and model training on well logging data, the accuracy and reliability issues of missing well logging data were resolved, enabling more accurate reservoir prediction.

CN119066411BActive Publication Date: 2026-11-17CHINA NAT PETROLEUM CORP +2
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310630714.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-05-30
Publication Date
2026-11-17
Estimated Expiration
2043-05-30

Smart Images

  • Figure CN119066411B_ABST
    Figure CN119066411B_ABST
Patent Text Reader

Abstract

This application discloses a method and apparatus for filling missing logging data. The method includes: performing wavenumber domain low-pass filtering on multiple original logging curves within the study area to obtain multiple low-frequency logging curves, wherein each original logging curve and each low-frequency logging curve contains missing data; training a model on the multiple low-frequency logging curves to predict multiple low-frequency logging curves without missing data; for the portion of each original logging curve with missing data, filling it with data at the same depth position from the corresponding low-frequency logging curve without missing data to obtain multiple transition logging curves; training a model on the multiple transition logging curves to predict multiple complete logging curves without missing data. This method can improve the reliability and accuracy of filling missing logging data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of petroleum exploration technology, and in particular to a method and apparatus for filling in missing well logging data. Background Technology

[0002] In oil and gas exploration technology, well logging curves constructed from well logging data are used to predict reservoir locations, and their quality directly affects the prediction results. In actual production, factors such as enlargement, instrument malfunction, and improper operation can easily lead to missing well logging data and incomplete well logging curves, posing significant challenges to subsequent reservoir prediction work such as modeling and inversion.

[0003] In related technologies, the main method for filling missing logging data is the single-method filling, which fills in the missing parts using default values, mean, or mode. Although this method is fast, it has low filling accuracy, poor reliability, and is prone to serious data deviations. Summary of the Invention

[0004] Therefore, this application provides a method and apparatus for filling missing logging data, which can improve the reliability and accuracy of filling missing logging data.

[0005] Specifically, the following technical solutions are included:

[0006] In a first aspect, embodiments of this application provide a method for filling in missing well logging data, the method comprising:

[0007] Wavenumber domain low-pass filtering was performed on multiple original logging curves in the study area to obtain multiple low-frequency logging curves. Each original logging curve and each low-frequency logging curve had missing data.

[0008] Model training was performed on the multiple low-frequency logging curves to predict multiple low-frequency logging curves without missing data;

[0009] For the part of each original logging curve with missing data, data at the same depth position on the corresponding low-frequency logging curve without missing data is used to fill in the missing data, resulting in multiple transition logging curves.

[0010] The model was trained on the multiple transition logging curves to predict multiple complete logging curves without missing data.

[0011] In some embodiments, before performing wavenumber domain low-pass filtering on multiple original logging curves within the study area to obtain multiple low-frequency logging curves, the method further includes:

[0012] Acquire well logging data for the area to be studied, wherein the well logging data includes multiple raw well logging curve data;

[0013] The multiple original logging curve data are preprocessed to obtain multiple original logging curves.

[0014] In some embodiments, the preprocessing of the plurality of original well logging curve data to obtain a plurality of original well logging curves includes:

[0015] Each raw logging curve data is inspected, and invalid values ​​in each raw logging curve are marked, where the invalid values ​​represent missing data;

[0016] For each original logging curve data, the data point corresponding to the current depth position is marked as invalid, and the data point at the current depth position of each original logging curve is deleted.

[0017] In some embodiments, the step of detecting and marking invalid values ​​in each original well logging curve includes:

[0018] For each raw well logging curve data, in response to a data value equaling a preset value, the data value is marked as invalid; or...

[0019] For each original logging curve data, in response to a data value greater than a first threshold or less than a second threshold, the data value is marked as invalid.

[0020] In some embodiments, the preprocessing of the plurality of original well logging curve data to obtain a plurality of original well logging curves further includes:

[0021] Based on the invalid values ​​in each original logging curve, the missing proportion of each original logging curve is calculated;

[0022] Based on the missing proportion of each original logging curve, multiple original logging curves are obtained after missing proportion sorting. The missing proportion sorting includes sorting from high to low and from low to high.

[0023] In some embodiments, the plurality of low-frequency logging curves include a target low-frequency logging curve, the target low-frequency logging curve including a first segment and a second segment, wherein the first segment is the segment containing valid values, and the second segment is the segment containing invalid values, and the step of training a model on the plurality of low-frequency logging curves to predict a plurality of low-frequency logging curves without missing data includes:

[0024] For the target low-frequency logging curve, the following first processing procedure is performed:

[0025] The data of the first segment of the target low-frequency logging curve and the data of the corresponding segments of other low-frequency logging curves at the same depth as the first segment are used as the output and input of the model training to obtain the first training model. The missing data of the corresponding segments of other low-frequency logging curves at the same depth as the first segment are obtained by polynomial interpolation.

[0026] Substitute the data of the corresponding segments of other low-frequency logging curves (excluding the target low-frequency logging curve) that are at the same depth as the second segment into the first training model to obtain the predicted data of the second segment of the target low-frequency logging curve.

[0027] The predicted data of the second segment of the target low-frequency logging curve is filled into the second segment of the target low-frequency logging curve to obtain a low-frequency logging curve without missing data.

[0028] Each low-frequency logging curve is taken as the target low-frequency logging curve, and the first processing procedure described above is performed to obtain the multiple low-frequency logging curves without missing data.

[0029] In some embodiments, the step of taking each low-frequency logging curve as a target low-frequency logging curve and performing the first processing procedure as described above includes:

[0030] According to the missing data sorting, each low-frequency logging curve is taken as the target low-frequency logging curve, and the first processing procedure described above is performed, wherein the missing data sorting is obtained based on the missing proportion of the original logging curve, and the missing proportion of the original logging curve is obtained based on the invalid values ​​in the original logging curve.

[0031] The missing information sorting is based on the missing information ratio from low to high.

[0032] In some embodiments, the multiple transition logging curves include a target transition logging curve, which includes a third segment and a fourth segment, wherein the third segment is the segment containing valid values ​​and the fourth segment is the segment containing invalid values. The step of training a model on the multiple transition logging curves to predict multiple complete logging curves without missing data includes:

[0033] For the target transition logging curve, the following second processing procedure is performed:

[0034] The data of the third segment of the target transition logging curve and the data of the corresponding segments of other transition logging curves at the same depth as the third segment are used as the output and input of the model training, respectively, to obtain the second training model.

[0035] Substitute the data of the corresponding segments of other transition logging curves (excluding the target transition logging curve) that are at the same depth as the fourth segment into the second training model to obtain the predicted data of the fourth segment of the target transition logging curve.

[0036] The predicted data of the fourth segment of the target transition logging curve is filled into the fourth segment of the target transition logging curve to obtain a complete logging curve without missing data.

[0037] Each transition logging curve is used as a target transition logging curve, and the second processing procedure described above is performed to obtain the multiple complete logging curves without missing data.

[0038] In some embodiments, the step of taking each transition logging curve as a target transition logging curve and performing the second processing procedure as described above includes:

[0039] According to the missing data sorting, each transition logging curve is taken as the target transition logging curve, and the second processing process described above is performed, wherein the missing data sorting is obtained based on the missing proportion of the original logging curve, and the missing proportion of the original logging curve is obtained based on the invalid values ​​in the original logging curve.

[0040] The missing information sorting is based on the missing information ratio from low to high.

[0041] Secondly, embodiments of this application provide a device for filling missing logging data, the device comprising:

[0042] The low-pass filtering module is used to perform wavenumber domain low-pass filtering on multiple original logging curves in the area to be studied, resulting in multiple low-frequency logging curves. Each original logging curve and each low-frequency logging curve has missing data.

[0043] The first prediction module is used to train a model on the multiple low-frequency logging curves and predict multiple low-frequency logging curves without missing data.

[0044] The filling module is used to fill in the missing data in each original logging curve by using data from the same depth position on the corresponding low-frequency logging curve that does not have missing data, thus obtaining multiple transition logging curves.

[0045] The second prediction module is used to train a model on the multiple transition logging curves and predict multiple complete logging curves without missing data.

[0046] The well logging data missing data filling method provided in this application involves performing wavenumber domain low-pass filtering on multiple original well logging curves within the study area to obtain multiple low-frequency well logging curves that display low-frequency trend patterns. Then, by training a model on these multiple low-frequency well logging curves, multiple complete low-frequency well logging curves without missing data can be predicted. Data from these corresponding low-frequency well logging curves without missing data is used to fill in the missing data portions of the original well logging curves, resulting in multiple transitional well logging curves. Based on model training on these transitional well logging curves, multiple well logging curves without missing data can be predicted. These multiple well logging curves without missing data are well logging curves with completed data missing data filling. This method fully utilizes existing effective well logging data, considers the overall trend of missing data, and combines model training to predict well logging curves with complete data, thereby improving both the reliability and accuracy of well logging data missing data filling. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 A flowchart illustrating a method for filling in missing well logging data, provided as an embodiment of this application;

[0049] Figure 2 This application provides a real well logging curve without missing well logging data.

[0050] Figure 3 This application provides a logging curve diagram with missing logging data as an embodiment of the present application;

[0051] Figure 4 A well logging curve image filled with missing well logging data provided in this application embodiment;

[0052] Figure 5 This is a structural block diagram of a well logging data filling device provided in an embodiment of this application. Detailed Implementation

[0053] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] Well logging uses geophysical methods to measure geophysical parameters of rock formations, such as electrical conductivity, acoustic properties, radioactivity, and electrochemical properties. By interpreting and applying well logging curves that represent these parameters, subsurface reservoirs can be identified, which is of great significance for oil and gas exploration and development. The quality of the well logging curves directly affects the results of reservoir prediction.

[0055] However, in actual production, factors such as enlargement, instrument malfunction, improper operation, and cost control often lead to missing sections of logging curves and multiple types of curves in some wells. This results in reduced reliability and fewer types of logging curves, posing a significant challenge to subsequent reservoir prediction work such as modeling and inversion. Therefore, it is necessary to fill in the missing parts of the logging data to ensure the integrity and richness of the data.

[0056] Currently, the following two problems are common in the process of filling missing logging data: (1) Different logging curves are essentially manifestations of different physical characteristics of the same reservoir. There is a certain mapping relationship between logging data. However, due to the complexity of the formation and the existence of anisotropy, traditional physical models or empirical models are generally linear or simple regression, which are difficult to accurately describe the relationship between logging curves. The filling accuracy of missing values ​​is not high. (2) Because the missing layers of various curves in the same well are different, or the missing curve types of different wells are different, it is difficult to use the same model / method to fill missing data of multiple wells in the work area. The variability of missing data types and missing areas increases the difficulty of applying traditional models. The commonly used single filling method, although fast in processing speed, has low filling accuracy, poor reliability, and is prone to serious data deviation.

[0057] To address the problems existing in the prior art, this application provides a method for filling in missing well logging data. Figure 1 A flowchart illustrating a method for filling missing logging data provided in this application embodiment is shown below. Figure 1 The method includes the following steps:

[0058] Step 101: Wavenumber domain low-pass filtering is performed on multiple original logging curves in the study area to obtain multiple low-frequency logging curves, in which each original logging curve and each low-frequency logging curve has missing data.

[0059] By performing wavenumber domain low-pass filtering on multiple original logging curves within the study area, multiple low-frequency logging curves can be obtained, and these low-frequency logging curves can reflect the overall trend.

[0060] The original logging curves and low-frequency logging curves correspond one-to-one, and their trends and changes are consistent. Because each original logging curve contains missing data, each low-frequency logging curve also contains missing data. The presence of missing data in each original logging curve indicates that the logging data from multiple original logging curves at the same depth is incomplete; that is, at the same depth, at least one original logging curve has missing data.

[0061] For example, the number of original logging curves can be seven.

[0062] Prior to this step, the method for filling missing logging data provided in this application embodiment further includes: acquiring logging data in the area to be studied, wherein the logging data includes multiple original logging curve data; and preprocessing the multiple original logging curve data to obtain multiple original logging curves.

[0063] It is understandable that each data point in the well logging curve data corresponds to a depth value and multiple physical property values.

[0064] Furthermore, the procedure for performing wavenumber domain low-pass filtering on multiple original logging curves within the study area can be as follows:

[0065] (1) Perform wavenumber domain transformation on the logging data within the study area in the depth domain, so that the depth domain logging data is transformed into the wavenumber domain:

[0066] U(k)=∫U(z)e ikz dz ①

[0067]

[0068] In the formula: U(k) represents wavenumber domain logging data, z represents well depth, i indicates the complex domain, k represents wavenumber, v represents sonic velocity, and w represents the dominant frequency of the logging data. It can be understood that the logging data acquired within the study area are all depth domain data.

[0069] (2) Determine the dominant frequency w2 of the seismic data at the location of the well to be studied within the study area, and calculate the wave number k2 corresponding to the dominant seismic frequency using the above formula ②.

[0070] (3) Use Chebyshev low-pass filter to perform low-pass filtering on the wavenumber domain data. Set the cutoff frequency to the wavenumber k2 corresponding to the seismic main frequency to obtain the filtered low-frequency wavenumber domain logging data U(k).

[0071] (4) Transform the filtered low-frequency wavenumber domain logging data back into the depth domain to obtain the depth domain low-frequency logging data:

[0072] U(z)=∫U(k)e -ikz dk

[0073] In the formula: U(z) represents the depth domain logging data.

[0074] In some embodiments, preprocessing multiple raw logging curve data to obtain multiple raw logging curves includes: detecting each raw logging curve data and marking invalid values ​​in each raw logging curve, wherein the data corresponding to the invalid values ​​represents missing data; for each raw logging curve data, all data points corresponding to the current depth position are marked as invalid values, and deleting the data points at the current depth position of each raw logging curve.

[0075] Here, invalid values ​​can also be called null values, indicating that the data is missing.

[0076] When all data points at the current depth position of each original logging curve are marked as invalid, it means that there are no valid values ​​at the current depth position. Therefore, the data points at the current depth position of each original logging curve can be deleted to facilitate subsequent model training.

[0077] In some embodiments, invalid values ​​include system outliers and invalid values ​​within the response range. Therefore, detecting and identifying invalid values ​​in each raw logging curve includes:

[0078] In the case of invalid values ​​being system outliers, for each original logging curve data, in response to the data value being equal to a preset value, the data value is marked as invalid; or, in the case of invalid values ​​being within the response range, for each original logging curve data, in response to the data value being greater than a first threshold or less than a second threshold, the data value is marked as invalid.

[0079] In other words, when a data value is a preset value that represents an abnormal value in the system, the data value is marked as invalid; when a data value does not fall within the threshold range formed by the first threshold and the second threshold, the data value is marked as invalid.

[0080] In some embodiments, preprocessing multiple original logging curve data to obtain multiple original logging curves further includes: calculating the missing proportion of each original logging curve based on invalid values ​​in each original logging curve; and obtaining multiple original logging curves sorted by missing proportion based on the missing proportion of each original logging curve, wherein the missing proportion sorting includes sorting by missing proportion from high to low and sorting by missing proportion from low to high.

[0081] By calculating the missing proportion of the original well logging curves, data is prepared for subsequent model training.

[0082] Understandably, for a given original logging curve, the missing percentage is the ratio of the number of invalid values ​​in the original logging curve to the total number of all values ​​(invalid and valid values) in the original logging curve.

[0083] Step 102: Train the model on multiple low-frequency logging curves to predict multiple low-frequency logging curves without missing data.

[0084] Among them, multiple low-frequency logging curves include target low-frequency logging curves, which include a first segment and a second segment. The first segment is the segment where the effective value is located, and the second segment is the segment where the invalid value is located.

[0085] It is understandable that the target low-frequency logging curve can be any one of multiple low-frequency logging curves.

[0086] This step may include the following sub-steps:

[0087] Step 1: For the target low-frequency logging curve, perform the following first processing procedure:

[0088] (1) The data of the first segment of the target low-frequency logging curve and the data of the corresponding segments of other low-frequency logging curves at the same depth as the first segment are used as the output and input of the model training to obtain the first training model. The missing data of the corresponding segments of other low-frequency logging curves at the same depth as the first segment are obtained by polynomial interpolation.

[0089] The first training model can be a linear regression model.

[0090] For example, when there are seven low-frequency logging curves and a well depth of 1000 meters, if the first segment of the target low-frequency logging curve is the 0-800 meter segment and the second segment is the 800-1000 meter segment, the input for training the model can be the data segments corresponding to the 0-800 meters of the other six low-frequency logging curves besides the target low-frequency logging curve, and the output can be the data segments corresponding to the 0-800 meters of the target low-frequency logging curve.

[0091] (2) Substitute the data of the corresponding segments of other low-frequency logging curves other than the target low-frequency logging curve at the same depth as the second segment into the first training model to obtain the predicted data of the second segment of the target low-frequency logging curve.

[0092] For example, when the first segment of the target low-frequency logging curve is the 0-800 meter segment and the second segment is the 800-1000 meter segment, the data segments corresponding to the 800-1000 meters of the other six low-frequency logging curves (excluding the target low-frequency logging curve) can be used as input, and the output can be the data segments corresponding to the 800-1000 meters of the target low-frequency logging curve.

[0093] (3) Fill the second segment of the target low-frequency logging curve with the predicted data of the second segment of the target low-frequency logging curve to obtain a low-frequency logging curve without missing data.

[0094] The predicted data of the second segment of the target low-frequency logging curve, obtained through model training, can reflect the trend of the target low-frequency logging curve. Filling it into the second segment of the target low-frequency logging curve can make the target low-frequency logging curve a logging curve with complete data, that is, a low-frequency logging curve without missing data.

[0095] Step 2: Take each low-frequency logging curve as the target low-frequency logging curve and perform the first processing procedure in Step 1 above to obtain multiple low-frequency logging curves without missing data.

[0096] It is understandable that after the first processing step in step one, a target low-frequency logging curve without missing data can be obtained. By performing the first processing step in step one on each low-frequency logging curve, multiple low-frequency logging curves without missing data can be obtained.

[0097] In some embodiments, the step includes: sorting each low-frequency logging curve as a target low-frequency logging curve according to the missing data, and performing the first processing procedure as described above, wherein the missing data sorting is obtained based on the missing proportion of the original logging curve, and the missing proportion of the original logging curve is obtained based on the invalid values ​​in the original logging curve; wherein the missing data sorting is sorted from low to high missing proportion.

[0098] By performing the first processing step above on multiple low-frequency logging curves according to the missing data sorted from low to high, the predicted low-frequency logging curves can better match the actual low-frequency trend.

[0099] Once this step is completed, the low-frequency regression of all invalid values ​​in the low-frequency logging curves is finished.

[0100] Step 103: For the part of each original logging curve with missing data, fill it with data at the same depth position on the corresponding low-frequency logging curve without missing data to obtain multiple transition logging curves.

[0101] Based on step 102, multiple complete low-frequency logging curves without missing data can be obtained. This step is equivalent to filling in the missing data portion of the original logging curve using the corresponding portion of the obtained low-frequency logging curve without missing data, resulting in multiple complete transition logging curves.

[0102] It is understandable that this transitional logging curve is equivalent to a logging curve that has undergone preliminary patching of the original logging curve, and is still not the final logging curve.

[0103] Step 104: Train the model on multiple transition logging curves to predict multiple complete logging curves without missing data.

[0104] Among them, multiple transition logging curves include target transition logging curves, which include the third and fourth segments. The third segment is the segment where the effective value is located, and the fourth segment is the segment where the invalid value is located.

[0105] Understandably, the target transition logging curve can be any one of multiple past logging curves.

[0106] This step may include the following sub-steps:

[0107] Step 1: For the target transition logging curve, perform the following second processing procedure:

[0108] (1) The data of the third segment of the target transition logging curve and the data of the corresponding segments of other transition logging curves at the same depth as the third segment are used as the output and input of the model training to obtain the second training model.

[0109] The second training model can be an extreme tree model.

[0110] The training method is the same as the training method in step 1(1) of step 102, and will not be repeated here.

[0111] (2) Substitute the data of the corresponding segments of other transition logging curves other than the target transition logging curve at the same depth as the fourth segment into the second training model to obtain the predicted data of the fourth segment of the target transition logging curve.

[0112] The substitution method is the same as the substitution method in step 1(2) of step 102, and will not be repeated here.

[0113] (3) Fill the fourth segment of the target transition logging curve with the predicted data of the fourth segment of the target transition logging curve to obtain a complete logging curve without missing data.

[0114] The filling method is the same as the filling method in step 1(3) of step 102, and will not be repeated here.

[0115] Step 2: Take each transition logging curve as the target transition logging curve and perform the second processing procedure above to obtain multiple complete logging curves without missing data.

[0116] It is understandable that after the second processing step in step one, a complete logging curve without missing data can be obtained. By performing the second processing step in step one on each transition logging curve, multiple complete logging curves without missing data can be obtained.

[0117] In some embodiments, the step includes: sorting each transition logging curve as a target transition logging curve according to the missing data, and performing the second processing procedure as described above, wherein the missing data sorting is obtained based on the missing proportion of the original logging curve, and the missing proportion of the original logging curve is obtained based on the invalid values ​​in the original logging curve; wherein the missing data sorting is sorted from low to high missing proportion.

[0118] Therefore, the well logging data missing completion method provided in this application involves performing wavenumber domain low-pass filtering on multiple original well logging curves within the study area to obtain multiple low-frequency well logging curves that display low-frequency trend patterns. Then, by training a model on these multiple low-frequency well logging curves, multiple complete low-frequency well logging curves without missing data can be predicted. Data from these corresponding low-frequency well logging curves without missing data is used to fill in the missing data portions of the original well logging curves, resulting in multiple transitional well logging curves. Based on model training on these transitional well logging curves, multiple well logging curves without missing data can be predicted, and these curves without missing data represent well logging curves with completed data missing completion. Compared to existing technologies, this method fully utilizes existing effective well logging data, considers the overall trend of missing data, and combines model training to predict complete well logging curves, thereby improving both the reliability and accuracy of well logging data missing completion.

[0119] Meanwhile, this method does not depend on the type and correlation of logging curves, has low requirements for users' professional background, and the prediction results ensure the integrity of logging data, laying the foundation for subsequent reservoir prediction work such as logging interpretation, modeling, and inversion, and has a broad application market.

[0120] In one possible example, to verify the effectiveness of the well logging data missing filling method provided in the embodiments of this application, well logging data from a carbonate rock research area is used as an example. The data types include gamma, neutron, density, porosity, water saturation, and P-wave velocity, and the curves are displayed sequentially from top to bottom as follows: Figure 2 ,at this time, Figure 2 This is a true well logging curve without missing data. Several segments of well logging data were randomly deleted from each well logging curve, such as... Figure 3 As shown, Figure 3 For well logging curves with missing data, where multiple types of data are missing simultaneously, data filling is required. After predicting the well logging curve according to the data filling method provided in this application, as follows... Figure 4 As shown, Figure 4 The well logging curves after filling in the missing logging data using the method for filling in missing logging data show that the predicted results are highly consistent with the actual results, indicating that the method for filling in the missing logging data provided in this application is highly practical and the prediction results are reasonable.

[0121] Figure 5 A structural block diagram of a well logging data missing filling device provided in an embodiment of this application. See also... Figure 5 The device includes:

[0122] The low-pass filter module 501 is used to perform wavenumber domain low-pass filtering on multiple original logging curves in the area to be studied to obtain multiple low-frequency logging curves, in which each original logging curve and each low-frequency logging curve has missing data;

[0123] The first prediction module 502 is used to train a model on multiple low-frequency logging curves and predict multiple low-frequency logging curves without missing data.

[0124] The filling module 503 is used to fill in the missing data in each original logging curve by using data at the same depth position on the corresponding low-frequency logging curve that does not have missing data, so as to obtain multiple transition logging curves.

[0125] The second prediction module 504 is used to train a model on multiple transition logging curves and predict multiple complete logging curves without missing data.

[0126] In some embodiments, the well logging data missing filling device provided in this application further includes:

[0127] The acquisition module is used to acquire well logging data within the area to be studied. The well logging data includes multiple raw well logging curves.

[0128] The preprocessing module is used to preprocess multiple raw logging curve data to obtain multiple raw logging curves.

[0129] In some embodiments, the preprocessing module includes:

[0130] The marking unit is used to detect each raw logging curve data and mark the invalid values ​​in each raw logging curve. The data corresponding to the invalid values ​​represent missing data.

[0131] The deletion unit is used to mark the data points at the current depth position as invalid values ​​for each original well logging curve data and delete the data points at the current depth position of each original well logging curve.

[0132] In some embodiments, the marking unit includes:

[0133] The first marking subunit is used to mark the data value as invalid if the data value equals a preset value for each original well logging curve data; or...

[0134] The second marking subunit is used to mark the data value as invalid for each original logging curve data in response to the data value being greater than the first threshold or less than the second threshold.

[0135] In some embodiments, the preprocessing module further includes:

[0136] The calculation unit is used to calculate the missing proportion of each original well logging curve based on the invalid values ​​in each original well logging curve;

[0137] The sorting unit is used to obtain multiple original logging curves after missing data sorting based on the missing data ratio of each original logging curve. The missing data sorting includes sorting from high to low missing data ratio and sorting from low to high missing data ratio.

[0138] In some embodiments, the multiple low-frequency logging curves include a target low-frequency logging curve, which includes a first segment and a second segment, wherein the first segment is the segment containing valid values ​​and the second segment is the segment containing invalid values. The first prediction module 502 includes:

[0139] The first processing unit performs the following first processing procedure on the target low-frequency logging curve: It uses the data of the first segment of the target low-frequency logging curve and the data of the corresponding segments of other low-frequency logging curves at the same depth as the first segment as the output and input of the model, respectively, to train the model and obtain a first training model. The missing data of the corresponding segments of other low-frequency logging curves at the same depth as the first segment are obtained using polynomial interpolation. The data of the corresponding segments of other low-frequency logging curves at the same depth as the second segment are substituted into the first training model to obtain the predicted data of the second segment of the target low-frequency logging curve. The predicted data of the second segment of the target low-frequency logging curve is then filled into the second segment of the target low-frequency logging curve to obtain a low-frequency logging curve without missing data.

[0140] The first execution unit is used to take each low-frequency logging curve as the target low-frequency logging curve and perform the first processing procedure as described above to obtain multiple low-frequency logging curves without missing data.

[0141] In some embodiments, the first execution unit includes:

[0142] The first execution subunit is used to sort by missing values, take each low-frequency logging curve as the target low-frequency logging curve, and perform the first processing procedure as described above. The missing value sorting is obtained based on the missing ratio of the original logging curve, and the missing ratio of the original logging curve is obtained based on the invalid values ​​in the original logging curve.

[0143] The missing information is sorted by the missing percentage from low to high.

[0144] In some embodiments, the multiple transition logging curves include a target transition logging curve, which includes a third segment and a fourth segment, wherein the third segment is the segment containing valid values ​​and the fourth segment is the segment containing invalid values. The second prediction module 504 includes:

[0145] The second processing unit performs the following second processing procedure on the target transition logging curve: It uses the data of the third segment of the target transition logging curve and the data of the corresponding segments of other transition logging curves (excluding the target transition logging curve) at the same depth as the third segment as the output and input of the model training, respectively, to obtain a second training model; it substitutes the data of the corresponding segments of other transition logging curves (excluding the target transition logging curve) at the same depth as the fourth segment into the second training model to obtain the predicted data of the fourth segment of the target transition logging curve; it fills the predicted data of the fourth segment of the target transition logging curve into the fourth segment of the target transition logging curve to obtain a complete logging curve without missing data.

[0146] The second execution unit is used to take each transition logging curve as the target transition logging curve and perform the second processing procedure as described above to obtain multiple complete logging curves without missing data.

[0147] In some embodiments, the second execution unit includes:

[0148] The second execution subunit is used to sort by missing data, take each transition logging curve as the target transition logging curve, and perform the second processing procedure as described above. The missing data sorting is obtained based on the missing proportion of the original logging curve, and the missing proportion of the original logging curve is obtained based on the invalid values ​​in the original logging curve.

[0149] The missing information is sorted by the missing percentage from low to high.

[0150] Therefore, the well logging data missing completion device provided in this application fully utilizes existing effective well logging data. Considering the overall trend of the missing data, and combined with model training, it predicts a complete well logging curve, thereby improving both the reliability and accuracy of well logging data missing completion.

[0151] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory including program code that can be executed by a processor in an electronic device to complete the well logging data missing filling method in the above embodiments. For example, the non-volatile computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0152] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0153] In this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The term "multiple" refers to two or more unless otherwise expressly defined.

[0154] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only.

[0155] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for filling in missing well log data, characterized by, The method includes: Wavenumber domain low-pass filtering was performed on multiple original logging curves in the study area to obtain multiple low-frequency logging curves. Each original logging curve and each low-frequency logging curve had missing data. Model training was performed on the multiple low-frequency logging curves to predict multiple low-frequency logging curves without missing data; For the part of each original logging curve with missing data, data at the same depth position on the corresponding low-frequency logging curve without missing data is used to fill in the missing data, resulting in multiple transition logging curves. The model was trained on the multiple transition logging curves to predict multiple complete logging curves without missing data.

2. The method of claim 1, wherein, Before performing wavenumber domain low-pass filtering on multiple original logging curves within the study area to obtain multiple low-frequency logging curves, the method further includes: Acquire well logging data for the area to be studied, wherein the well logging data includes multiple raw well logging curve data; The multiple original logging curve data are preprocessed to obtain multiple original logging curves.

3. The method of claim 2, wherein, The preprocessing of the multiple raw well logging curve data to obtain multiple raw well logging curves includes: Each raw logging curve data is inspected, and invalid values ​​in each raw logging curve are marked, where the invalid values ​​represent missing data; For each original logging curve data, the data point corresponding to the current depth position is marked as invalid, and the data point at the current depth position of each original logging curve is deleted.

4. The method of claim 3, wherein, The process of detecting and marking invalid values ​​in each original well logging curve includes: For each raw well logging curve data, in response to a data value equaling a preset value, the data value is marked as invalid; or... For each original logging curve data, in response to a data value greater than a first threshold or less than a second threshold, the data value is marked as invalid.

5. The method of claim 2, wherein, The preprocessing of the multiple raw well logging curve data to obtain multiple raw well logging curves also includes: Based on the invalid values ​​in each original logging curve, the missing proportion of each original logging curve is calculated; Based on the missing proportion of each original logging curve, multiple original logging curves are obtained after missing proportion sorting. The missing proportion sorting includes sorting from high to low and from low to high.

6. The method of claim 1, wherein, The multiple low-frequency logging curves include a target low-frequency logging curve, which includes a first segment and a second segment. The first segment represents the segment containing valid values, and the second segment represents the segment containing invalid values. The model training performed on the multiple low-frequency logging curves to predict multiple low-frequency logging curves without missing data includes: For the target low-frequency logging curve, the following first processing procedure is performed: The data of the first segment of the target low-frequency logging curve and the data of the corresponding segments of other low-frequency logging curves at the same depth as the first segment are used as the output and input of the model training to obtain the first training model. The missing data of the corresponding segments of other low-frequency logging curves at the same depth as the first segment are obtained by polynomial interpolation. Substitute the data of the corresponding segments of other low-frequency logging curves (excluding the target low-frequency logging curve) that are at the same depth as the second segment into the first training model to obtain the predicted data of the second segment of the target low-frequency logging curve. The predicted data of the second segment of the target low-frequency logging curve is filled into the second segment of the target low-frequency logging curve to obtain a low-frequency logging curve without missing data. Each low-frequency logging curve is taken as the target low-frequency logging curve, and the first processing procedure described above is performed to obtain the multiple low-frequency logging curves without missing data.

7. The method for filling missing logging data according to claim 6, characterized in that, The step of taking each low-frequency logging curve as a target low-frequency logging curve and performing the first processing procedure as described above includes: According to the missing data sorting, each low-frequency logging curve is taken as the target low-frequency logging curve, and the first processing procedure described above is performed, wherein the missing data sorting is obtained based on the missing proportion of the original logging curve, and the missing proportion of the original logging curve is obtained based on the invalid values ​​in the original logging curve. The missing information sorting is based on the missing information ratio from low to high.

8. The method for filling missing logging data according to claim 1, characterized in that, The multiple transition logging curves include a target transition logging curve, which includes a third segment and a fourth segment, wherein the third segment is the segment containing valid values ​​and the fourth segment is the segment containing invalid values. The model training performed on the multiple transition logging curves to predict multiple complete logging curves without missing data includes: For the target transition logging curve, the following second processing procedure is performed: The data of the third segment of the target transition logging curve and the data of the corresponding segments of other transition logging curves at the same depth as the third segment are used as the output and input of the model training, respectively, to obtain the second training model. Substitute the data of the corresponding segments of other transition logging curves (excluding the target transition logging curve) that are at the same depth as the fourth segment into the second training model to obtain the predicted data of the fourth segment of the target transition logging curve. The predicted data of the fourth segment of the target transition logging curve is filled into the fourth segment of the target transition logging curve to obtain a complete logging curve without missing data. Each transition logging curve is used as a target transition logging curve, and the second processing procedure described above is performed to obtain the multiple complete logging curves without missing data.

9. The method for filling missing logging data according to claim 8, characterized in that, The step of taking each transition logging curve as a target transition logging curve and performing the second processing procedure as described above includes: According to the missing data sorting, each transition logging curve is taken as the target transition logging curve, and the second processing process described above is performed, wherein the missing data sorting is obtained based on the missing proportion of the original logging curve, and the missing proportion of the original logging curve is obtained based on the invalid values ​​in the original logging curve. The missing information sorting is based on the missing information ratio from low to high.

10. A device for filling missing well logging data, characterized in that, The device includes: The low-pass filtering module is used to perform wavenumber domain low-pass filtering on multiple original logging curves in the area to be studied, resulting in multiple low-frequency logging curves. Each original logging curve and each low-frequency logging curve has missing data. The first prediction module is used to train a model on the multiple low-frequency logging curves and predict multiple low-frequency logging curves without missing data. The filling module is used to fill in the missing data in each original logging curve by using data from the same depth position on the corresponding low-frequency logging curve that does not have missing data, thus obtaining multiple transition logging curves. The second prediction module is used to train a model on the multiple transition logging curves and predict multiple complete logging curves without missing data.

Citation Information

Patent Citations

  • Method and apparatus for estimating lithofacies by learning well log

    CN113762320A

  • High resolution seismic data derived from pre-stack inversion and machine learning

    US20190293818A1