Logging density diameter expansion correction system and method based on deep learning and physical constraints
By combining deep learning with physical constraints, the enlarged diameter section is identified and rock physical constraints are established. A convolutional neural network model is constructed, which solves the problems of existing logging density enlargement correction methods relying on experience and having poor generalization ability. This achieves high-precision, automated and highly adaptable correction results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHAANXI YANCHANG PETROLEUM GRP
- Filing Date
- 2026-04-23
- Publication Date
- 2026-07-21
AI Technical Summary
Existing logging density enlargement correction methods rely on experience and manual intervention, and the correction effect is greatly affected by the operator. Purely data-driven methods lack physical constraints, have poor generalization ability, and are difficult to adapt to different geological conditions.
By combining deep learning with physical constraints, the physical constraint relationship of rocks is established by identifying the expanded diameter section and the normal section, a convolutional neural network model is constructed, and a physical constraint loss function is introduced during the training process to optimize and correct the model.
It improves the accuracy and consistency of calibration, enhances the generalization ability of the model, has a high degree of automation, and the calibration results conform to geological laws, making them easy to interpret and apply, and applicable to different lithologies.
Smart Images
Figure CN122428889A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of well logging technology, specifically relating to a well logging density enlargement correction system and method based on deep learning and physical constraints. Background Technology
[0002] Well logging is a crucial method for obtaining subsurface geological information in oil and gas exploration and development. Density logging, in particular, is a key technology for determining formation porosity, identifying lithology, and assessing fluid properties. Conventional density logging uses a gamma-gamma density instrument to measure the Compton scattering of gamma rays by the formation to invert the bulk density. This method assumes a regular wellbore and a smooth wellbore wall. However, in actual drilling, wellbore enlargement is a common phenomenon, mainly caused by formation collapse, drilling fluid erosion, or low formation strength. The enlarged area is filled with low-density drilling mud, resulting in a large amount of mud signal within the instrument's detection volume. This leads to a significantly lower apparent density, causing severe distortion and affecting the accuracy of porosity calculations, lithology identification, rock mechanical parameters, and reservoir evaluation.
[0003] Currently, methods for well logging density enlargement correction mainly include empirical formulas, rock physics modeling, standard layer statistical analysis, and deep learning-based correction methods. Existing technologies have the following main shortcomings: First, they rely on experience and manual intervention, and the correction effect is greatly affected by the operator's experience; second, purely data-driven deep learning methods lack physical constraints, which may lead to correction results that violate basic geological laws; and third, they have poor generalization ability and are difficult to adapt to correction needs under different geological conditions.
[0004] Invention / Invention Content To address the problems of existing technologies, this invention provides a well logging density enlargement correction method based on deep learning and physical constraints, comprising the following steps: Step S1: Obtain logging data for the target well area, wherein the logging data includes at least density curves, well diameter curves, and auxiliary logging curves for correction; Step S2: Identify the enlarged section and the normal section based on the wellbore curve, and divide the logging data into normal section data and enlarged section data; Step S3: Based on rock physics theory, establish the physical constraint relationship between density and the auxiliary logging curve; Step S4: Construct a deep learning correction model, which takes the auxiliary logging curve as input and the corrected density value as output; Step S5: Train the deep learning correction model using the normal segment data, and introduce the physical constraint relationship as a regularization term into the model's loss function during the training process to simultaneously optimize the data fitting loss and the physical constraint loss. Step S6: Input the expanded diameter section data into the trained deep learning correction model to obtain the corrected density value of the expanded diameter section.
[0005] Furthermore, the step S2, which involves identifying the enlarged section and the normal section based on the caliper curve, specifically means that when the value of the caliper curve is greater than a preset threshold for the drill bit diameter, it is determined to be the enlarged section; otherwise, it is the normal section.
[0006] Furthermore, the physical constraint relationship mentioned in step S3 includes the Gardner equation, which is used to establish the relationship between sound wave velocity and density, and its expression is: (1); in For density, denoted as sound wave velocity, and a and b are empirical parameters obtained by fitting well logging data from the normal section; Furthermore, the physical constraint relationship described in step S3 also includes the Archie equation, used to establish the relationship between resistivity and porosity, and its expression is: (2); in True resistivity For formation water resistivity, For porosity, , These are empirical parameters.
[0007] Furthermore, the deep learning correction model mentioned in step S4 is a convolutional neural network model, which includes an input layer, a one-dimensional convolutional layer, a pooling layer, a fully connected layer and an output layer connected in sequence. Furthermore, the loss function described in step S5 Represented as: ;in Data fitting loss, For physical constraint loss, This is the balance coefficient; The data fitting loss mentioned above Using mean square error, the physical constraint loss The error between the model output and the physical constraint prediction value is calculated based on the physical constraint relationship. Furthermore, it also includes step S7: comparing the corrected density curve with the normal segment density curve, using the correlation coefficient R. 2 Verify the correction effect using the root mean square error (RMSE) and / or root mean square error (RMSE).
[0008] Furthermore, the auxiliary logging curves include one or more of gamma curves, resistivity curves, and sonic transit time curves; Furthermore, the correction method is applicable to well logging density enlargement correction for sandstone, mudstone, and carbonate rocks of all lithologies; This invention also provides a logging density enlargement correction system based on deep learning and physical constraints, comprising: The data acquisition module is used to acquire logging data for the target well area; The enlargement identification module is used to identify the enlarged section and the normal section based on the wellbore curve; The physical constraint construction module is used to establish the physical constraint relationship between density and auxiliary logging curves; The model building and training module is used to build a deep learning correction model and train it using normal segment data combined with physical constraints. The correction module is used to input the expanded diameter section data into the trained model to obtain the corrected density value. The beneficial effects that this application can produce include: 1) This application combines deep learning with physical constraints, which not only utilizes the strong fitting ability of deep learning to nonlinear relationships, but also ensures that the correction results conform to geological laws through physical constraints, thus avoiding the physically unreasonable results that may be produced by pure data-driven methods and improving the correction accuracy. 2) The entire calibration process of this application is highly automated, eliminating the need for manual selection of standard layers or empirical parameters, thereby improving calibration efficiency and consistency and reducing manual intervention; 3) This invention does not depend on the geological conditions of a specific well area. By introducing a rock physics equation with strong universality as a constraint, the model has good generalization ability in different regions and under different geological conditions, thus enhancing the generalization ability. The introduction of physical constraints makes the model output have clear physical meaning, which is convenient for geologists to understand and apply the correction results, thereby improving interpretability. 4) This invention is applicable to diameter correction of various lithologies, and is not limited to mudstone sections. Attached Figure Description
[0009] Figure 1 This is an overall flowchart of the method of the present invention; Figure 2 This is a graph showing the correspondence between density curve anomalies in this invention, illustrating the relationship between the expanded diameter section and density distortion. Figure 3 This is a comparison diagram of the density curve before and after correction in this invention (first example well section); Figure 4 This is a comparison diagram of the density curve before and after correction in this invention (second example well section); Figure 5 This is a comparison chart of calculated porosity and core porosity before and after density curve correction in this invention. Detailed Implementation
[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0011] Please see Figure 1 This invention provides a well logging density enlargement correction method based on deep learning and physical constraints, comprising the following steps: Step S1: Obtain logging data for the target well area. The logging data includes at least density curves, caliper curves, and auxiliary logging curves, including gamma curves, resistivity curves, and sonic transit time curves. Step S2: Based on the wellbore curve, identify the enlarged section and the normal section, and divide the logging data into normal section data and enlarged section data; Step S3: Construct a rock physical constraint model and establish the physical equations between density, sound wave velocity, and resistivity; Step S4: Construct a deep learning correction model. The model takes the auxiliary logging curve as input, the corrected density value as output, and introduces a physical constraint term based on the rock physics constraint model into the model loss function. Step S5: Use the normal segment data to jointly train the deep learning correction model, and optimize the data fitting loss and physical constraint loss to obtain the trained deep learning correction model. Step S6: Input the expanded diameter section data into the trained deep learning correction model and output the corrected density value of the expanded diameter section; Step S7: Combine the density values of the normal section with the corrected density values of the enlarged section to obtain the corrected density curve for the entire well section.
[0012] Furthermore, the step S2, which involves identifying the enlarged section and the normal section based on the caliper curve, specifically means that when the value of the caliper curve is greater than the preset threshold for the drill bit diameter, it is determined to be the enlarged section; otherwise, it is the normal section.
[0013] Furthermore, the physical constraint relationship mentioned in step S3 includes the Gardner equation, which is used to establish the relationship between sound wave velocity and density, and its expression is: (1); in For density, denoted as sound wave velocity, and a and b are empirical parameters obtained by fitting well logging data from the normal section; Furthermore, the physical constraint relationship described in step S3 also includes the Archie equation, used to establish the relationship between resistivity and porosity, and its expression is: (2); in True resistivity For formation water resistivity, For porosity, , These are empirical parameters.
[0014] Furthermore, the deep learning correction model mentioned in step S4 is a convolutional neural network model, which includes an input layer, a one-dimensional convolutional layer, a pooling layer, a fully connected layer and an output layer connected in sequence. Furthermore, the loss function described in step S5 Represented as: ;in Data fitting loss, For physical constraint loss, This is the balance coefficient; The data fitting loss mentioned above Using mean square error, the physical constraint loss The error between the model output and the physical constraint prediction value is calculated based on the physical constraint relationship. Furthermore, it also includes step S7: comparing the corrected density curve with the normal segment density curve, using the correlation coefficient R. 2 Verify the correction effect using the root mean square error (RMSE) and / or root mean square error (RMSE).
[0015] Furthermore, the auxiliary logging curves include one or more of gamma curves, resistivity curves, and sonic transit time curves; Furthermore, the correction method is applicable to well logging density enlargement correction for sandstone, mudstone, and carbonate rocks of all lithologies; This invention also provides a logging density enlargement correction system based on deep learning and physical constraints, comprising: The data acquisition module is used to acquire logging data for the target well area; The enlargement identification module is used to identify the enlarged section and the normal section based on the wellbore curve; The physical constraint construction module is used to establish the physical constraint relationship between density and auxiliary logging curves; The model building and training module is used to build a deep learning correction model and train it using normal segment data combined with physical constraints. The correction module is used to input the expanded diameter section data into the trained model to obtain the corrected density value. Example 1 This embodiment provides a well logging density enlargement correction system and method based on deep learning and physical constraints, specifically including the following steps: I. Data Collection and Preprocessing Collect logging data for the target well area, including density (DEN), natural gamma (GR), deep lateral resistivity (RT), sonic transit time (AC), and wellbore caliper (CAL) curves. The raw logging data may contain outliers and noise; therefore, data cleaning is performed first: outliers that are significantly outside the physical range are removed, such as density values that are negative or greater than 3.0 g / cm³. 3 The points were then standardized using the Z-score normalization method, which normalized the mean and variance of each curve to zero to eliminate the influence of dimensions and improve the stability of model training.
[0016] II. Diameter Enlargement Identification The enlarged section is identified using the caliper curve. In this embodiment, the drill bit diameter is 8.5 inches. When the CAL curve value is greater than 9.5 inches, it is determined to be an enlarged section. Figure 2 As shown in the figure, the left side represents the caliper curve (CAL), and the right side represents the density curve (DEN). The horizontal axis represents the logging values, and the vertical axis represents the depth (in meters). Within the depth range of X150–X200 meters, the caliper curve value increases significantly, from 8.5 inches to over 10 inches, indicating significant enlargement in this well section. Simultaneously, the corresponding density curve shows a significant low-value anomaly in this depth range, with density values increasing from the normal range of 2.5–2.7 g / cm³. 3 Decreased to 2.0–2.3 g / cm³ 3 This phenomenon is due to the fact that after the wellbore is enlarged, low-density drilling mud fills the space around the well wall, causing a large amount of mud signal to be mixed into the detection volume of the density logging instrument, resulting in a lower measured apparent density. Figure 2 This study visually demonstrates the correlation between borehole enlargement and density distortion, validating the rationale for using borehole diameter curves to identify enlargement sections and to divide training and calibration samples. Based on this identification rule, well logging data is divided into normal sections (CAL ≤ 9.5 inches) and enlargement sections (CAL > 9.5 inches), with the normal sections serving as training samples and the enlargement sections as samples to be calibrated.
[0017] III. Physical Constraint Modeling Based on rock physics theory, the following physical constraint relationships are established: Gardner equations: used to establish the relationship between sound wave velocity and density. (1); in For density, The speed of sound can be determined by the sound wave time difference curve. =1 / AC is calculated; a and b are empirical parameters obtained by fitting the sound wave velocity and density data of the normal segment. In this embodiment, the fitting results are a=0.31 and b=0.25. The Archie equation, as an auxiliary physical constraint, establishes the relationship between resistivity and porosity, enhancing the model's physical consistency with reservoir properties. Its expression is: (2); in True resistivity For formation water resistivity, For porosity, , These are empirical parameters; IV. Deep Learning Model Construction A well logging density enlargement correction model based on a one-dimensional convolutional neural network (1D-CNN) is constructed. The model structure is as follows: Input layer: Receives three auxiliary logging curves, GR, RT, and AC, as input features. The input shape is (number of samples, time step, 3), where the time step is set to 25 sampling points (corresponding to a depth window of approximately 1.25 meters).
[0018] Convolutional layers: Two one-dimensional convolutional layers are used. The first layer has 32 kernels and a size of 3; the second layer has 64 kernels and a size of 3. The activation function is ReLU.
[0019] Pooling layer: A max pooling layer is added after each convolutional layer, with a pooling window size of 2.
[0020] Fully connected layer: The output of the pooling layer is flattened and then connected to two fully connected layers with 128 and 64 neurons respectively, and the activation function is ReLU.
[0021] Output layer: Outputs the corrected density values, with a linear activation function.
[0022] V. Model Training The deep learning model is trained using well logging data from normal sections as training samples. During training, physical constraints are introduced as a regularization term, and the loss function is defined as follows: ; in Data fitting loss, For physical constraint loss, This is the balance coefficient; The mean square error between the model's predicted density and the actual density of the normal segment is calculated as the data fitting loss. The physical constraint loss is the mean square error between the density predicted by the calculation model and the density calculated from the sound velocity using the Gardner equation; λ is the balance coefficient, which is taken as 0.1 in this embodiment.
[0023] The Adam optimizer was used for training, with a learning rate of 0.001, a batch size of 64, and 200 training epochs. 20% of the normal data was used as the validation set to prevent overfitting.
[0024] VI. Diameter Expansion Correction The well logging data (GR, RT, AC) of the enlarged section are input into the trained deep learning model, and the model automatically outputs the corrected density value. Figure 3 and Figure 4 The comparison between the density curves before and after correction is shown.
[0025] Specifically, Figure 3 This is a comparison chart of correction results for a certain depth range (X100–X250 meters). From left to right, the chart shows: caliper curve (CAL), density curve comparison, and natural gamma curve (GR). In the density curve comparison chart, the dashed line represents the original density curve before correction, and the solid line represents the density curve after correction using the method of this invention. Figure 3 It can be seen that in the enlarged well section (X150–X200 meters), the original density curve before correction shows a significant low-value depression due to the influence of drilling mud, with density values of approximately 2.0–2.2 g / cm³. 3 The density values of the upper and lower normal segments are approximately 2.5–2.7 g / cm³. 3 This creates discontinuous jumps. The corrected density curve recovers to approximately 2.4–2.6 g / cm³ in this expanded section. 3 The density curves are continuously connected to the density trends of the normal sections above and below, resulting in a smoother curve shape that conforms to geological patterns. In the non-expanded sections (X100–X150 meters, X200–X250 meters), the density curves before and after correction basically overlap, indicating that the model did not make unnecessary corrections to the normal sections and has good stability. Figure 3 The method of the present invention has been verified to effectively recover the true density value of the expanded diameter section while keeping the original information of the non-expanded diameter section unaffected.
[0026] Figure 4 This is a comparison chart of the correction results for another depth range (X300–X450 meters), which contains interbedded sandstone and mudstone, used to verify the applicability of the method of this invention to different lithologies. From left to right in the chart are: caliper curve (CAL), density curve comparison, and natural gamma curve (GR). In the density curve comparison chart, the dashed line is the original density curve before correction, and the solid line is the density curve after correction. Figure 4It can be seen that in the sandstone section (e.g., X350–X380 meters), the natural gamma curve exhibits low values. The original density curve is significantly lower in this section due to the diameter expansion. After correction, the density value is significantly increased, consistent with the regional sandstone density characteristics (approximately 2.5–2.7 g / cm³). 3 The results show good agreement. In the mudstone section (e.g., X380–X420 meters), the natural gamma curve exhibits high values, the original density curve is less affected by the diameter expansion, and the changes before and after correction are small. This indicates that the model can automatically identify lithological differences based on the characteristics of the auxiliary curves (GR, RT, AC) and avoid over-correction of the mudstone section. Figure 4 The invention verifies that the method is applicable not only to mudstone sections, but also to diameter correction of different lithologies such as sandstone, demonstrating the method's generalization ability and lithological adaptability.
[0027] VII. Result Verification The corrected density curve was used for porosity calculation and compared with the porosity from core analysis for verification. Figure 5 This is a comparison chart of the porosity calculated before and after density curve correction with the core porosity. The horizontal axis represents porosity (%), and the vertical axis represents depth. The diamond-shaped dots represent the porosity from core analysis; the gray curve represents the porosity calculated from density before correction, and the black curve represents the porosity calculated from density after correction.
[0028] from Figure 5 It can be seen that in the enlarged diameter section (e.g., X150–X200 meters), the calculated porosity before correction was significantly higher (exceeding 30% in some cases), deviating considerably from the core porosity (approximately 15–20%). However, after correction, the calculated porosity matched the core porosity well, with the average absolute error decreasing from 6.5% to 1.8%, and the correlation coefficient increasing from 0.62 to 0.91. The verification results demonstrate that the method of this invention can effectively eliminate the influence of enlargement on density logging and significantly improve the accuracy of porosity evaluation.
[0029] The implementation of this embodiment achieves high-precision intelligent correction of the density curve of the expanded section. The correction result conforms to rock physics laws such as the Gardner equation. The whole process is highly automated, requiring no manual selection of standard layers or empirical parameters, and has good generalization ability and interpretability.
[0030] The above description is merely a few embodiments of this application and is not intended to limit this application in any way. Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any changes or modifications made by those skilled in the art without departing from the scope of the technical solution of this application using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.
Claims
1. A logging density enlargement correction system and method based on deep learning and physical constraints, characterized in that, Includes the following steps: Step S1: Obtain logging data for the target well area, wherein the logging data includes at least density curves, well diameter curves, and auxiliary logging curves for correction; Step S2: Identify the enlarged section and the normal section based on the wellbore curve, and divide the logging data into normal section data and enlarged section data; Step S3: Based on rock physics theory, establish the physical constraint relationship between density and the auxiliary logging curve; Step S4: Construct a deep learning correction model, which takes the auxiliary logging curve as input and the corrected density value as output; Step S5: Train the deep learning correction model using the normal segment data, and introduce the physical constraint relationship as a regularization term into the model's loss function during the training process to simultaneously optimize the data fitting loss and the physical constraint loss. Step S6: Input the expanded diameter section data into the trained deep learning correction model to obtain the corrected density value of the expanded diameter section.
2. The method according to claim 1, characterized in that, The step S2, which identifies the enlarged section and the normal section based on the caliper curve, specifically means that when the value of the caliper curve is greater than the preset threshold of the drill bit diameter, it is determined to be the enlarged section; otherwise, it is the normal section.
3. The method according to claim 1, characterized in that, The physical constraints mentioned in step S3 include the Gardner equation, which is used to establish the relationship between sound wave velocity and density. Its expression is: （1）； in For density, denoted as sonic velocity, and 'a' and 'b' are empirical parameters obtained by fitting logging data from the normal section.
4. The method according to claim 1, characterized in that, The physical constraint relationship mentioned in step S3 also includes the Archie equation, which is used to establish the relationship between resistivity and porosity. Its expression is: （2）； in True resistivity For formation water resistivity, For porosity, , These are empirical parameters.
5. The method according to claim 1, characterized in that, The deep learning correction model mentioned in step S4 is a convolutional neural network model, which includes an input layer, a one-dimensional convolutional layer, a pooling layer, a fully connected layer, and an output layer connected in sequence.
6. The method according to claim 1, characterized in that, The loss function described in step S5 Represented as: ;in Data fitting loss, For physical constraint loss, This is the balance coefficient; The data fitting loss mentioned above Using mean squared error, the physical constraint loss is calculated based on the physical constraint relationship to determine the error between the model output and the predicted physical constraint value.
7. The method according to claim 1, characterized in that, The process also includes step S7: comparing the corrected density curve with the normal segment density curve, using the correlation coefficient R. 2 The correction effect is verified by the root mean square error (RMSE) and / or the root mean square error (RMSE).
8. The method according to claim 1, characterized in that, The auxiliary logging curves include one or more of the following: gamma curves, resistivity curves, and sonic transit time curves.
9. The method according to claim 1, characterized in that, The correction method is applicable to well logging density enlargement correction for sandstone, mudstone, and carbonate rocks of all lithologies.
10. A logging density enlargement correction system based on deep learning and physical constraints, characterized in that, include: The data acquisition module is used to acquire logging data for the target well area; The enlargement identification module is used to identify the enlarged section and the normal section based on the wellbore curve; The physical constraint construction module is used to establish the physical constraint relationship between density and auxiliary logging curves; The model building and training module is used to build a deep learning correction model and train it using normal segment data combined with physical constraints. The correction module is used to input the expanded diameter section data into the trained model to obtain the corrected density value.