Method for permeability prediction based on heuristic convolutional neural network

Abstract

The present application provides a kind of permeability prediction method based on heuristic convolutional neural network, the method is by lithology interpretation analysis to geological exploration data, to obtain including the first rock physical elastic parameter and porosity parameter sandstone sample set;And the parameter of sandstone sample set is analyzed, to obtain including the second rock physical elastic parameter and the high pore sandstone sample set of permeability parameter;Second rock physical elastic parameter and permeability are analyzed, to obtain including the third rock physical elastic parameter, shear Lee factor parameter and the sensitive parameter combination set of permeability;The intersection analysis of shear Lee factor parameter and permeability parameter of sensitive parameter combination set is carried out, according to the relationship between shear Lee factor parameter and permeability parameter, obtain classification standard;The sample after classification is input into prediction permeability model, and the final prediction permeability data is obtained.The present application introduces shear Lee factor parameter, and it is convenient for the prediction of permeability.

Description

Technical Field

[0001] This invention relates to the field of petroleum exploration and development, and in particular to a method for permeability prediction based on heuristic convolutional neural networks. Background Technology

[0002] Permeability is a crucial parameter reflecting the characteristics of underground reservoirs and fluid flow. Accurate permeability prediction is essential for forecasting the distribution and development of underground oil and gas. Currently, reservoir permeability prediction in the oil exploration field mainly relies on core mercury intrusion porosimetry and nuclear magnetic resonance logging analysis. However, these methods are costly, and the data obtained is limited. If permeability could be predicted using conventional logging data, such as P-wave and S-wave velocities and density, it would not only reduce the cost of permeability prediction based on heuristic convolutional neural networks but also provide a wealth of permeability data.

[0003] There are two commonly used methods for permeability prediction: multiple linear fitting based on well logging data and neural network prediction. Multiple linear fitting typically uses multiple parameters intersecting with permeability, prioritizing parameters that significantly affect permeability. Statistical methods such as correlation analysis and principal component analysis (PCA) are used to reduce dimensionality, and permeability is predicted using a multiple linear regression equation. Neural network prediction uses an intersection method to filter permeability-sensitive data, removes outliers, normalizes the data, and divides the data into training and validation sets. The training set is used to adjust the training model using a loss function and optimization algorithms, continuously adjusting hyperparameters (such as learning rate, number of layers, number of neurons, etc.) to improve prediction performance. However, conventional multiple linear fitting usually performs poorly in low-permeability areas, and there may not be a good relationship between permeability and various parameters. Permeability varies significantly across different pore types, and conventional neural network fitting typically uses rock physical parameters such as P-wave and S-wave velocity and density, neglecting the influence of pore structure parameters on permeability.

[0004] Currently, there are few patents and literature on permeability prediction based on heuristic convolutional neural networks. Specific technical methods mainly fall into two categories: (I) predicting permeability through core analysis using multiple regression methods. Patent application number CN202410794002.6, "A Method for Predicting Reservoir Permeability Based on Digital Core," is a representative patent of this type. This method acquires core CT scan images, establishes a three-dimensional digital core model, extracts pore structure parameters, classifies pore space types, and uses the fitting relationship between weighted pore structure parameters, weighted fracture structure parameters, and permeability to establish an optimized multiple regression model, obtaining a permeability prediction model. The data to be measured is then imported into the prediction model to obtain the predicted permeability. This type of method is usually costly and difficult to implement on a large scale; (II) predicting permeability using traditional rock physical parameters and neural networks. Patent application number CN202410837028.4, "A Method for Predicting Shale Oil Reservoir Permeability Based on Artificial Intelligence," is a representative patent of this type. This type of method extracts permeability-influencing features from the data, using porosity, fracture development degree, and formation pressure. After anomaly removal, noise reduction, and normalization, the data is input into a neural network for prediction. This method ignores the influence of pore structure on permeability. While fracture development degree and formation pressure have a significant impact on the permeability of tight shale, the influence of pore structure parameters on permeability is more important for traditional sandstone and mudstone reservoirs. The article "Dual Porosity Structure Factor and Its Application in Seismic Prediction of Reservoir Permeability" in the *Acta Petrolei Sinica* mentions using pore structure parameters to predict permeability. This article introduces the classification of lithofacies using porosity, P-wave and S-wave impedance, and dual pore structure factors (shear Lee factor and pore aspect ratio), and uses a GA-BP neural network to predict permeability. Although this article utilizes pore structure parameters, it only applies to lithofacies classification. When lithofacies classification is clear, classifying samples in favorable lithologies using pore structure parameters to inspire deep learning, based on heuristic convolutional neural networks, will improve the accuracy of permeability prediction.

[0005] Based on current research of patents and literature, no existing technology utilizes pore structure parameters to classify favorable lithological samples using heuristic deep learning for permeability prediction. Therefore, proposing a permeability prediction method based on heuristic convolutional neural networks (CNNs) using pore structure parameters would be beneficial for breakthroughs in permeability prediction in low-permeability areas and would improve the economic efficiency of oil and gas exploration. Summary of the Invention

[0006] This invention provides a method for permeability prediction based on heuristic convolutional neural networks, which can solve the problem that the accuracy of prediction needs to be improved due to the failure to fully utilize pore structure parameters for heuristic deep learning prediction.

[0007] To solve the above-mentioned technical problems, the method includes the following steps:

[0008] S1: By performing lithological interpretation and analysis on geological exploration data, a sandstone sample set including the first rock physical elastic parameters and porosity parameters is obtained;

[0009] S2: Perform a correlation analysis between the first rock physical elastic parameter and the porosity to obtain a high-porosity sandstone sample set including the second rock physical elastic parameter and permeability parameter;

[0010] S3: Based on rock physics analysis, perform correlation analysis on the second rock physics elastic parameter and the permeability to obtain a set of sensitive parameter combinations including the third rock physics elastic parameter, the shear Lee factor parameter and the permeability;

[0011] S4: Perform cross-analysis on the shear Lee factor parameter and the permeability parameter of the sensitive parameter combination set, and obtain the classification criteria based on the relationship between the shear Lee factor parameter and the permeability parameter;

[0012] S5: Based on the classification criteria, the high-porosity sandstone sample set is classified according to the classification criteria, and the classified samples are input into the predicted permeability model to obtain the final predicted permeability data.

[0013] In one embodiment, the geological exploration data includes permeability curves;

[0014] Step S1 includes:

[0015] The geological exploration data is cleaned, and the logarithm of the permeability curve is taken to obtain a preprocessed set of well logging data.

[0016] The preprocessed well logging data set is filtered according to preset sandstone screening conditions to obtain a sandstone sample set containing the first rock physical elastic parameter and porosity parameter.

[0017] In one embodiment, the sensitive parameter combination set includes a first sensitive parameter combination set and a second sensitive parameter combination set;

[0018] Step S3 includes:

[0019] By performing cross-analysis on the second rock physical elastic parameters and the permeability parameters, the first set of sensitive parameter combinations is obtained;

[0020] The porosity and density parameters in the first set of sensitive parameters are calculated using a preset rock physics model to obtain the shear Lee factor parameters.

[0021] The shear Lee factor parameter and the first sensitive parameter set are combined to obtain a second sensitive parameter set including the third rock physical elastic parameter, the shear Lee factor parameter and the permeability parameter.

[0022] In one embodiment, the step of calculating the porosity parameter and density parameter from the first set of sensitive parameters using a preset rock physics model to obtain the shear Lee factor parameter includes:

[0023] Obtain a preset rock physics model;

[0024] Based on the boundary theory method, the shear modulus parameters of the dry rock in the preset rock physics model are calculated;

[0025] The initial shear compliance factor of the preset rock physics model is adaptively adjusted according to the particle swarm optimization algorithm to obtain the shear compliance factor parameters.

[0026] Substituting the shear modulus parameter, the shear compliance factor parameter, and the porosity parameter of the dry rock into the first formula, the shear modulus parameter of the rock matrix is ​​obtained.

[0027] According to the Gassmann equation, the shear modulus parameter of the rock matrix is ​​converted into the modulus parameter of the fluid-saturated rock, and the shear wave velocity parameter is calculated.

[0028] Substituting the shear modulus parameter, porosity parameter, density parameter, and shear wave velocity parameter of the rock matrix into the second formula yields the shear Lee factor.

[0029] In one embodiment, step S4 includes:

[0030] The high-porosity sandstone samples are divided into high-porosity high-permeability samples and high-porosity low-permeability samples according to the preset high-permeability and low-permeability boundaries.

[0031] The cross-analysis of the shear Lee factor and the permeability of the sensitive parameter set, and the classification criteria obtained based on the relationship between the shear Lee factor and the permeability, include:

[0032] Based on a preset threshold, the high-porosity sandstone sample set is divided into high-porosity and high-permeability samples and high-porosity and low-permeability samples.

[0033] The third rock physical elastic parameter and the shear Lee factor in the second sensitive parameter combination set are intersected with the permeability parameter to determine the first classification threshold and the second classification threshold.

[0034] Using the particle swarm optimization algorithm, the boundary between the high-porosity high-permeability sample and the high-porosity low-permeability sample is dynamically optimized within a preset floating range, with the goal of minimizing the permeability prediction error, to obtain a first floating threshold and a second floating threshold.

[0035] The first classification standard is obtained based on the first classification threshold and the first floating threshold;

[0036] The second classification criterion is obtained based on the second classification threshold and the second floating threshold.

[0037] In one embodiment, the step of dynamically optimizing the boundary between the high-porosity, high-permeability sample and the high-porosity, low-permeability sample within a preset floating range using a particle swarm optimization algorithm to obtain a first floating threshold and a second floating threshold includes:

[0038] Step 1: Based on the shearing Lee factor, perform a classification and cross-analysis of the parameters in the second set of sensitive parameter combinations;

[0039] Step 2: Adjust the origin of intersection and the rotation angle so that the dividing line distinguishes the high-porosity high-permeability sample from the high-porosity low-permeability sample, thereby determining the first dividing line and the second dividing line.

[0040] Step 3: Adjust the first dividing line and the second dividing line to be perpendicular to each other in order to obtain the first parameter and the second parameter;

[0041] Step 4: Linearly fit the first parameter or the second parameter to the permeability to obtain the first permeability prediction error, wherein the first permeability prediction error is the permeability prediction error obtained based on the coordinate axis rotation method;

[0042] Step 5: Dynamically optimize the boundary line within a preset floating range using the particle swarm optimization algorithm, and obtain the first floating threshold and the second floating threshold based on the principle of minimizing the permeability prediction error.

[0043] In one embodiment, the step of linearly fitting the first parameter or the second parameter to the permeability to obtain the first permeability prediction error includes:

[0044] When the third rock physical elastic parameter and the shear Lee factor are negatively correlated with the permeability parameter, the first permeability prediction error is obtained by linearly fitting the second parameter with the permeability.

[0045] When the third rock physical elastic parameter and the shear Lee factor are positively correlated with the permeability parameter, the first parameter is linearly fitted with the permeability to obtain the first permeability prediction error.

[0046] In one embodiment, step S5 includes:

[0047] The high-porosity sandstone sample set is classified according to the classification criteria to obtain different categories of sandstone subsets; the sandstone subsets include the third rock physical elastic parameter and the shear Lee factor;

[0048] The third rock physical elastic parameter and shear Lee factor in the sandstone subset are used to divide the dataset into training and test sets.

[0049] The training set is used to train the penetration rate prediction model to obtain the trained penetration rate prediction model;

[0050] The test set is based on a trained penetration rate prediction model to obtain the predicted penetration rate.

[0051] In one embodiment, the predicted penetration rate model includes a convolutional neural network;

[0052] The step of training the penetration rate prediction model using the training set to obtain the trained penetration rate prediction model includes:

[0053] Step 1: Input the training set into the convolutional neural network and calculate the actual output value;

[0054] Step 2: Calculate the error between the actual output value and the expected output value using a preset loss function;

[0055] Step 3: Repeat steps 1 and 2 until the preset loss function converges or the preset number of training rounds is reached.

[0056] In one embodiment, the first formula includes:

[0057]

[0058] Where: μ d μ is the shear modulus parameter of dry rock. m φ represents the shear modulus of the rock matrix; φ represents porosity; γ represents... μ It is the shear compliance factor;

[0059] The second formula includes:

[0060]

[0061] In the formula: c μ For the cleavage Lee factor; v S ρ is the transverse wave velocity; φ is the porosity; s Density; μ m This represents the shear modulus parameter of the rock matrix.

[0062] The present invention has the following beneficial effects: The present invention relates to a method for permeability prediction based on a heuristic convolutional neural network, wherein the method comprises the following steps: obtaining a sandstone sample set including a first rock physical elastic parameter and a porosity parameter by performing lithological interpretation analysis on geological exploration data; further performing correlation analysis on these parameters to screen out a second rock physical elastic parameter and a porosity parameter that are highly correlated with permeability, forming a high-porosity sandstone sample set; then, based on rock physical analysis, exploring in depth the relationship between these parameters and permeability, especially introducing a shear Lee factor parameter to form a sensitive parameter combination set including a third rock physical elastic parameter, a shear Lee factor parameter, and permeability; next, using intersection analysis to determine the classification criteria between the shear Lee factor parameter and the permeability parameter, and classifying the high-porosity sandstone sample set according to this criterion; inputting the classified samples into a heuristic convolutional neural network permeability prediction model, which is trained by a deep learning algorithm and can make full use of the classification information of favorable lithological samples by pore structure parameters, thereby improving the accuracy of permeability prediction based on the heuristic convolutional neural network and obtaining the final predicted permeability data. This invention improves the prediction of permeability and enhances the economic benefits of oil and gas exploration by introducing a pore structure parameter: the shear Lee factor parameter. Attached Figure Description

[0063] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the present invention will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 This is a flowchart illustrating the penetration rate prediction method based on heuristic convolutional neural networks proposed in this application.

[0065] Figure 2 This is a schematic diagram of sandstone sensitive parameters in one embodiment of this application;

[0066] Figure 3 This is a schematic diagram comparing the predicted P-wave and S-wave velocities with the actual P-wave and S-wave velocities in one embodiment of this application;

[0067] Figure 4 This is a schematic diagram of the intersection of permeability and shear Lee factor in one embodiment of this application;

[0068] Figure 5 This is a schematic diagram of the prediction results of the rotation of the shear modulus and porosity coordinate axes in one embodiment of this application;

[0069] Figure 6 This is a schematic diagram comparing the results of heuristic penetration rate prediction with those of traditional penetration rate prediction in one embodiment of this application;

[0070] Figure 7 This is a schematic diagram comparing the heuristic prediction of penetration rate and the traditional prediction of penetration rate in one embodiment of this application. Detailed Implementation

[0071] To make the objectives, features, and advantages of this invention more apparent and understandable, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0072] Addressing the technical problems of existing permeability prediction methods, namely their failure to fully utilize pore structure parameters for accurate prediction, this invention presents a permeability prediction method based on a heuristic convolutional neural network. The method involves the following steps: First, lithological interpretation and analysis are performed on geological exploration data to extract key rock physical parameters. Second, correlation analysis and rock physical analysis are used to screen for sensitive parameter combinations highly correlated with permeability, particularly introducing pore structure parameters such as the shear Lee factor. Third, based on these sensitive parameters, cross-reference analysis is used to determine classification criteria and perform fine-grained classification of the samples. Finally, the classified samples are input into a heuristic convolutional neural network model for training and optimization to improve the accuracy of permeability prediction. Through these steps, this invention effectively improves the performance of the permeability prediction model and enhances the economic benefits of oil and gas exploration by introducing pore structure parameters, specifically the shear Lee factor, and performing heuristic classification.

[0073] like Figure 1 and Figure 2 As shown, the steps of the penetration rate prediction method based on heuristic convolutional neural networks include:

[0074] S1: By performing lithological interpretation and analysis on geological exploration data, a sandstone sample set including the first rock physical elastic parameters and porosity parameters is obtained;

[0075] It should be noted that the geological exploration data was obtained through professional geological exploration methods and technologies, including but not limited to well logging, seismic exploration, and geological surveys, ensuring the accuracy and reliability of the data. Lithology interpretation and analysis, based on the geological exploration data, aims to accurately identify sandstone, mudstone, shale, and other lithologies within the formation through in-depth processing of well logging data. Differences in parameters such as density, sonic velocity, and natural gamma ray can effectively distinguish between different lithologies.

[0076] S2: Perform correlation analysis on the first rock physical elastic parameters and porosity to obtain a high-porosity sandstone sample set including the second rock physical elastic parameters and permeability parameters;

[0077] It should be noted that, through meticulous data analysis, key parameters closely related to the characteristics of high-porosity sandstone were identified from a wide range of rock physical parameters. The initial strategy employed was to perform cross-analysis of the primary rock physical elastic parameter and other relevant rock physical parameters with porosity. Cross-analysis, by plotting scatter plots or relationship diagrams between the parameters, can visually demonstrate the correlations between them.

[0078] In one embodiment, these rock physical parameters were first intersected with porosity. By comparing the correlation coefficients between different parameters and porosity, density was selected as the sensitive parameter for porosity. Next, a high porosity threshold (10% in this example) was set based on the actual conditions of different regions. Based on this threshold, a high-porosity sandstone sample set (S high-porosity sandstone) was successfully further screened from the previously selected sandstone samples, providing a more accurate data foundation for subsequent analysis and prediction.

[0079] S3: Based on rock physics analysis, a correlation analysis is performed on the second rock physics elastic parameter and permeability to obtain a set of sensitive parameter combinations including the third rock physics elastic parameter, shear Lee factor parameter and permeability;

[0080] It should be noted that a correlation analysis was performed between the second rock physical elastic parameter and permeability, which is a statistical method for assessing the relationship between variables. The aim was to identify a sensitive parameter combination that includes the third rock physical elastic parameter, the shear Lee factor parameter, and permeability.

[0081] S4: Perform cross-analysis on the shear Lee factor parameter and permeability parameter of the sensitive parameter combination set, and obtain the classification criteria based on the relationship between the shear Lee factor parameter and the permeability parameter;

[0082] It is important to note that during cross-analysis, the shear Lee factor and permeability parameters within the sensitive parameter combination set must first be standardized to ensure they are compared on the same scale. Then, visualization tools such as scatter plots or histograms are used to display the distribution relationship and trends between the shear Lee factor and permeability parameters. By observing and analyzing these charts, the distribution characteristics of data points under different parameter combinations can be identified, such as dense areas, sparse areas, or clear boundaries. Based on these characteristics, classification criteria can be developed to categorize rock samples according to different combinations of shear Lee factor and permeability parameters. These classification criteria can be further used for reservoir evaluation, oil and gas reservoir development strategy formulation, and production optimization, providing more refined and accurate geological information for oil and gas exploration and development.

[0083] S5: Based on the classification criteria, the high-porosity sandstone sample set is classified according to the classification criteria, and the classified samples are input into the predicted permeability model to obtain the final predicted permeability data.

[0084] It should be noted that the accurate application of classification criteria and the precise classification of high-porosity sandstone samples are crucial to ensuring the quality of predicted permeability data. Simultaneously, the rational construction and effective application of the predicted permeability model are also important guarantees for achieving accurate permeability prediction. Through the rigorous execution of these steps, reliable permeability prediction results can be obtained, providing strong support for geological exploration and development.

[0085] This invention presents a permeability prediction method based on a heuristic convolutional neural network. First, geological exploration data is analyzed to identify rock physical parameters and porosity parameters highly correlated with permeability, forming a high-porosity sandstone sample set. Then, a shear Lee factor parameter is introduced to form a set of sensitive parameter combinations, and a classification criterion is determined through cross-analysis. Next, this criterion is used to classify the sample set, and the classified samples are input into a heuristic convolutional neural network model for training. This model can fully utilize pore structure parameters, improving the accuracy of permeability prediction and thus enhancing the economic benefits of oil and gas exploration.

[0086] Furthermore, geological exploration data includes permeability curves;

[0087] Step S1 includes:

[0088] The geological exploration data is cleaned, and the logarithm of the permeability curve is taken to obtain a preprocessed set of well logging data.

[0089] The pre-processed well logging data set is filtered according to preset sandstone screening conditions to obtain a sandstone sample set containing the first rock physical elastic parameters and porosity parameters.

[0090] In one embodiment, the logging curves (including density, P-wave and S-wave velocities, natural gamma, etc.) and lithological interpretation conclusions of well LF13-9-2 were first collected and organized, and outlier processing was performed on these logging curves to ensure data accuracy. Next, using the processed P-wave and S-wave velocities and density data, key petrophysical parameters such as bulk modulus, shear modulus, and P-wave / S-wave velocity ratio were calculated, and the permeability curve was logarithmically processed to facilitate subsequent cross-analysis and prediction. Then, combined with the lithological interpretation conclusions, these petrophysical parameters were analyzed in depth, and the sensitive parameters for sandstone reservoirs were determined to be the P-wave / S-wave velocity ratio and density. Based on these sensitive parameters, screening criteria were set (P-wave / S-wave velocity ratio less than 1.9 and density less than 2.53 g / cm³). 3 Based on this standard, a set of sandstone samples (S sandstone) that met the criteria was successfully selected from the original data, providing reliable data support for subsequent analysis and prediction.

[0091] Furthermore, the sensitive parameter combination set includes a first sensitive parameter combination set and a second sensitive parameter combination set;

[0092] Step S3 includes:

[0093] By cross-analyzing the second rock physical elastic parameters and permeability parameters, the first set of sensitive parameter combinations is obtained;

[0094] The porosity and density parameters in the first set of sensitive parameters are calculated using a preset rock physics model to obtain the shear Lee factor parameters.

[0095] By combining the shear Lee factor parameter and the first set of sensitive parameters, a second set of sensitive parameters is obtained, which includes the third rock physical elastic parameter, the shear Lee factor parameter, and the permeability parameter.

[0096] In one embodiment, the second rock physical elastic parameters include: clay content, natural gamma, porosity, density, and P-wave impedance. For high-porosity sandstone samples, the second rock physical elastic parameters and permeability parameters were cross-analyzed. The purpose of this step is to identify the potential relationship between the two, and based on this relationship, a first sensitive parameter combination set was successfully determined. Next, using a pre-defined rock physical model, the porosity and density parameters in the first sensitive parameter combination set were calculated in detail, thereby deriving the shear Lee factor parameter. This newly derived shear Lee factor parameter, combined with the first sensitive parameter combination set, constitutes the second sensitive parameter combination set.

[0097] Furthermore, the porosity and density parameters from the first set of sensitive parameters are calculated using a preset rock physics model to obtain the shear Lee factor parameters, including:

[0098] Obtain a preset rock physics model;

[0099] Based on the boundary theory method, the shear modulus parameters of dry rock in the pre-defined rock physics model are calculated.

[0100] The initial shear compliance factor of the preset rock physics model is adaptively adjusted according to the particle swarm optimization algorithm to obtain the shear compliance factor parameters.

[0101] Substituting the shear modulus, shear compliance factor, and porosity parameters of the dry rock into the first formula yields the shear modulus parameters of the rock matrix.

[0102] Based on the Gassmann equation, the shear modulus parameter of the rock matrix is ​​converted into the modulus parameter of the fluid-saturated rock, and the shear wave velocity parameter is calculated.

[0103] Substituting the shear modulus, porosity, density, and transverse wave velocity parameters of the rock matrix into the second formula yields the shear Lee factor.

[0104] In one embodiment, an automatic optimization and modeling method was employed in calculating the shear Lee factor. Since the shear modulus of the rock matrix was unknown, a low-permeability reservoir rock physics model based on adaptive dual compliance factors was established, and the moduli of different minerals in the study area were continuously adjusted. Boundary theory was used to calculate the bulk modulus and shear modulus of the rock matrix. Then, the particle swarm optimization algorithm was used again to adaptively adjust the bulk compliance factor and shear compliance factor to obtain more accurate results.

[0105] After obtaining the precise shear modulus parameters, shear compliance factor, and porosity of the dry rock, these parameters were substituted into the previously mentioned first and third formulas to calculate the bulk modulus of the dry rock and the shear modulus parameters of the rock matrix. Then, the Gassmann equation was used again to calculate the modulus of the fluid-saturated rock, and based on this, the P-wave and S-wave velocities were calculated. To verify the accuracy of the model, such as... Figure 3 As shown, using the error between the predicted and measured P- and S-wave velocities as the objective function, suitable moduli for different minerals in the study area were determined through continuous adjustment and optimization. Finally, the shear Lee factor was calculated using the obtained shear modulus of the rock matrix and the second formula mentioned earlier. Notably, at locations with higher permeability, the values ​​of both the shear compliance factor and the shear Lee factor were observed to be significantly lower, indicating a certain negative correlation between them. This finding provides important clues and evidence for further understanding the petrophysical properties of high-porosity sandstones.

[0106] Furthermore, the first formula includes:

[0107]

[0108] Where: μ d The shear modulus of dry rock; μ m φ represents the shear modulus of the rock matrix; φ represents porosity; γ represents... μ It is the shear compliance factor;

[0109] The second formula includes:

[0110]

[0111] In the formula: c μ For the cleavage Lee factor; v S ρ is the transverse wave velocity; φ is the porosity; s Density; μ m This represents the shear modulus of the rock matrix.

[0112] The third formula includes:

[0113] K d =K m (1-φ) γ

[0114] In the formula: K d K represents the bulk modulus of dry rock. m φ is the bulk modulus of the rock matrix, φ is the porosity, and γ is the compliance factor.

[0115] Furthermore, step S4 includes:

[0116] Based on the preset high-permeability and low-permeability boundaries, the high-porosity sandstone samples were divided into high-porosity and high-permeability samples and high-porosity and low-permeability samples.

[0117] By performing a cross-analysis of the shear Lee factor and permeability of the sensitive parameter combination set, and based on the relationship between the shear Lee factor and permeability, the classification criteria are as follows:

[0118] Based on a preset threshold, the high-porosity sandstone sample set is divided into high-porosity and high-permeability samples and high-porosity and low-permeability samples.

[0119] The third rock physical elastic parameter and shear Lee factor in the second sensitive parameter combination set are interpolated with the permeability parameter to determine the first classification threshold and the second classification threshold.

[0120] The particle swarm optimization algorithm is used to dynamically optimize the boundary between high-porosity high-permeability samples and high-porosity low-permeability samples within a preset floating range, with the goal of minimizing the permeability prediction error, to obtain the first floating threshold and the second floating threshold.

[0121] The first classification standard is obtained based on the first classification threshold and the first floating threshold;

[0122] The second classification criterion is obtained based on the second classification threshold and the second floating threshold.

[0123] It should be noted that: First, based on the geological characteristics and actual data of the study area, a clear permeability threshold (Per threshold_high and Per threshold_low) needs to be set to classify the high-porosity sandstone sample set (S high-porosity sandstone) into a high-porosity high-permeability sandstone sample set (S high-porosity high-permeability sandstone) and a high-porosity low-permeability sandstone sample set (S high-porosity low-permeability sandstone). This step is the foundation for subsequent analysis and ensures the accuracy of sample classification. Second, the shear Lee factor and permeability curves are subjected to cross-analysis to determine the initial classification criteria for the shear Lee factor based on their relationship. In this example, the shear Lee factor is divided into three categories, and initial classification criterion thresholds LeeI and LeeII are set. The purpose of this step is to preliminarily classify the samples and provide a basis for subsequent optimization. Third, parameters A1 and A2 from the second sensitive parameter combination set for permeability prediction are selected for classification cross-analysis. By continuously adjusting the cross-analysis origin O... 12 0 and rotation angle α 12 0 To find the dividing line L that can effectively distinguish between high-porosity, high-permeability sandstone sample sets and high-porosity, low-permeability sandstone sample sets. 12 0 At the same time, the mutually perpendicular dividing line T was also determined. 12 0 This enhances the accuracy of classification. The fourth step involves determining the origin O in step 3. 12 0 and rotation angle α 12 0 The dividing line L 12 0 and T 12 0 Rotate to a vertical position and obtain the rotated parameter RX. 12 0 and RY 12 0 Based on the correlation between parameters A1, A2 and permeability, appropriate parameters are selected for linear fitting, and the permeability prediction error E is calculated. 12 0 This step provides an important basis for assessing the accuracy of penetration rate prediction. The fifth step involves setting a floating range and optimizing the classification criteria. To further optimize the classification criteria, a floating range (±L) was set for the initial classification criteria of the shear Lee factor. Imax and ±L IImax Then, using the particle swarm optimization algorithm, the classification criterion threshold (LeeI±L) of the shearing Lee factor is continuously adjusted. Imax With LeeII±L IImaxThe sixth step involves outputting the optimized results and the penetration rate prediction error. Following the methods in steps 3-4, a classification cross-analysis is performed on parameters A1 and A2 in the initial set of sensitive parameter combinations for penetration rate prediction, and the classification standard fluctuation value L is output. Imax 12 L IImax 12 And penetration rate prediction error E 12 0 The optimization results are obtained. This step provides optimized classification criteria and penetration rate prediction error, providing a basis for subsequent screening. The seventh step is to set a penetration rate prediction error threshold Ep. When the penetration rate prediction error E... 12 0 If the optimization result is less than Ep, retain the second sensitive parameter combination of parameters A1 and A2; otherwise, discard the combination. This step ensures that the selected parameter combination has high penetration rate prediction accuracy. In the eighth step, steps 3-7 are repeated to perform pairwise intersections of all parameters in the second sensitive parameter combination set for penetration rate prediction, and the classification criterion fluctuation value is continuously adjusted. Finally, the optimal set of classification criterion fluctuation values ​​is obtained, and the average of the two classification criterion fluctuation values ​​is taken to obtain the final classification criterion LeeI+L. Imax With LeeII+L IIma x. This step provides the final optimized classification criteria, offering strong support for subsequent research and applications. Through the detailed implementation of the above steps, the particle swarm optimization algorithm can be used to optimize the permeability classification criteria for high-porosity sandstone samples based on the correlation coefficient between predicted and measured permeability obtained by rotating the coordinate axes of the pairwise intersection of the second sensitive parameter combination set for permeability prediction. This method not only improves the accuracy of classification but also provides important references for subsequent reservoir evaluation and oil and gas reservoir development strategy formulation.

[0124] In one implementation, based on the specific conditions of the region, the boundary between high and low permeability was initially set at 1 mD; that is, samples with a permeability greater than 1 mD were considered high-permeability samples, while samples with a permeability less than 1 mD were considered low-permeability samples. To optimize the classification criteria for the shear Lee factor, such as... Figure 4As shown, shear Lee factors less than 1 and greater than 3 mainly concentrate in low-permeability reservoirs, while high-permeability and low-permeability reservoirs exist in the range of 1 to 3. Two initial classification thresholds, 1 and 3, were set with a fluctuation range of ±0.3. This means that during optimization, these two thresholds can be dynamically adjusted within the ranges of 1±0.3 and 3±0.3. A particle swarm optimization algorithm was used to continuously adjust the classification thresholds of the shear Lee factor to find the threshold combination that best distinguishes between high-permeability and low-permeability samples. Simultaneously, permeability prediction sensitive parameters (such as shear modulus and density) were used for classification, and samples were divided into different categories according to the classification criteria. Then, pairwise intersections of these parameters were performed, considering the influence of both parameters on permeability simultaneously. To maximize the correlation coefficient between the predicted and measured permeability based on these parameters, a coordinate axis rotation method was used, and the permeability prediction error, i.e., the difference between the predicted and measured permeability, was calculated. A 20% threshold is set for permeability prediction error. If the permeability prediction error of a given intersection exceeds this threshold, the prediction is considered inaccurate and must be discarded. Taking the intersection of shear modulus and density as an example... Figure 5 As shown, through continuous dynamic optimization, the classification standard fluctuation values ​​for shear modulus and density that minimize the permeability prediction error were found to be 0.1 and -0.15, respectively, with a permeability prediction error of 12%. The above steps were repeated for other permeability prediction sensitive parameters. If the cross-prediction result of a certain parameter remained above the 20% error threshold after exceeding the number of iterations, that cross-prediction result was discarded. Finally, the average of all obtained classification standard fluctuation values ​​yielded the final shear Lee factor classification standard thresholds of 1.1 and 2.95.

[0125] Furthermore, the particle swarm optimization algorithm is used to dynamically optimize the boundary between high-porosity, high-permeability samples and high-porosity, low-permeability samples within a preset floating range to obtain a first floating threshold and a second floating threshold, including:

[0126] Step 1: Based on the shearing Lee factor, classify and cross-analyze the parameters in the second sensitive parameter set;

[0127] Step 2: Adjust the origin of intersection and the rotation angle so that the dividing line distinguishes between high-porosity high-permeability samples and high-porosity low-permeability samples, thereby determining the first dividing line and the second dividing line.

[0128] Step 3: Adjust the first dividing line and the second dividing line to be perpendicular to obtain the first parameter and the second parameter;

[0129] Step 4: Linearly fit the first parameter or the second parameter to the permeability to obtain the first permeability prediction error. The first permeability prediction error is the permeability prediction error obtained based on the coordinate axis rotation method.

[0130] Step 5: Dynamically optimize the boundary line within a preset floating range using the particle swarm optimization algorithm, and obtain the first floating threshold and the second floating threshold based on the principle of minimizing the permeability prediction error.

[0131] It should be noted that the process begins with a classification and cross-analysis of parameters in the second set of sensitive parameter combinations based on the shear Lee factor. This aims to identify the parameter combinations most strongly correlated with high-porosity, high-permeability and high-porosity, low-permeability samples, laying the foundation for subsequent steps. Next, using a coordinate axis rotation method, the optimal boundary line position for accurately distinguishing between high-porosity, high-permeability and high-porosity, low-permeability samples was found through continuous trial and adjustment of the cross-analysis origin and rotation angle. This determined the first and second boundary lines. To simplify the analysis and facilitate parameter extraction, these two boundary lines were aligned vertically, resulting in the first and second parameters. Subsequently, these two parameters were linearly fitted to permeability to evaluate their accuracy in permeability prediction, and the permeability prediction error based on the coordinate axis rotation method was calculated. Finally, the particle swarm optimization algorithm was used to dynamically optimize the boundary lines within a preset floating range. By continuously adjusting the boundary line position (i.e., the first and second floating thresholds) and calculating the permeability prediction error, the boundary line position that minimized the prediction error was found, resulting in the final first and second floating thresholds.

[0132] Furthermore, by linearly fitting the first parameter or the second parameter to the permeability, the first permeability prediction error is obtained, including:

[0133] When the third rock physical elastic parameter and shear Lee factor are negatively correlated with the permeability parameter, the first permeability prediction error is obtained by linearly fitting the second parameter with the permeability.

[0134] When the third rock physical elastic parameter and shear Lee factor are positively correlated with the permeability parameter, the first permeability prediction error is obtained by linearly fitting the first parameter with the permeability.

[0135] It should be noted that:

[0136] In one embodiment, the method for obtaining the first permeability prediction error by linearly fitting a first parameter or a second parameter to permeability is further refined. Specifically, different correlations are observed between the third rock physical elastic parameter and the shear Lee factor and the permeability parameter. When these parameters are negatively correlated with the permeability parameter, i.e., when their values ​​increase, the permeability value decreases, the second parameter is chosen for linear fitting with permeability. This is because, in this case, the second parameter may better reflect the trend of permeability changes, thus obtaining a more accurate permeability prediction error. Conversely, when the third rock physical elastic parameter and the shear Lee factor are positively correlated with the permeability parameter, i.e., when their values ​​increase, the permeability value also increases, the first parameter is chosen for linear fitting with permeability. This is because, in this case, the first parameter may better represent the change in permeability, thus providing a more reliable permeability prediction. By selecting appropriate parameters for linear fitting based on the correlation between parameters and permeability, the accuracy and reliability of permeability prediction can be further improved.

[0137] Furthermore, step S5 includes:

[0138] The high-porosity sandstone sample set was classified according to the classification criteria to obtain different categories of sandstone subsets; the sandstone subsets include the third rock physical elastic parameter and the shear Lee factor;

[0139] The third rock physical elastic parameter and shear Lee factor in the sandstone subset are used as the basis for dividing the dataset into training and test sets.

[0140] The training set is used to train the penetration rate prediction model to obtain the trained penetration rate prediction model.

[0141] The test set is based on a trained penetration prediction model to obtain the predicted penetration rate.

[0142] In one embodiment: Step S5 further details the specific process of permeability prediction using a high-porosity sandstone sample set. First, the high-porosity sandstone sample set is classified according to a preset classification standard, resulting in multiple sandstone subsets of different categories. These subsets contain key third-order rock physical elastic parameters and shear Lee factors. Subsequently, the third-order rock physical elastic parameters and shear Lee factors are extracted from these sandstone subsets, and they are divided into training and testing sets for subsequent model training and validation.

[0143] Next, the training set data is used to train the penetration rate prediction model. Machine learning methods are employed in this process, continuously adjusting the model parameters to minimize the loss function, thus obtaining the trained penetration rate prediction model. Then, the test set data is used to validate the model's predictive performance and obtain the predicted penetration rate values.

[0144] To further improve prediction accuracy, a convolutional neural network (CNN) was also used to build the model. Specifically, the shear modulus, density, P-wave velocity, and shear Lee factor of the samples were normalized and used as training data in the constructed CNN. Through heuristic training, the model was trained using the training data, and the model parameters were continuously adjusted to minimize the loss function, ultimately resulting in an optimized permeability prediction model.

[0145] like Figure 6 and Figure 7 As shown, a comparative experiment was conducted to verify the effectiveness of heuristic training. Without using the Lee factor for classification, all samples were directly fed into the convolutional neural network for training. By comparing the results of heuristic training with those of the fully fed network, it was found that the heuristic training results had a higher correlation coefficient and were closer to the actual penetration rate values ​​in detail. This result demonstrates that heuristic training has a significant advantage in improving the accuracy of penetration rate prediction.

[0146] In summary, through steps such as classifying and processing high-porosity sandstone sample sets, training a permeability prediction model using machine learning, and employing convolutional neural networks for heuristic training, a model capable of accurately predicting permeability was successfully obtained, providing strong support for geological exploration and development work.

[0147] Furthermore, the penetration rate prediction model includes convolutional neural networks;

[0148] The training set is used to train the penetration rate prediction model, resulting in a well-trained penetration rate prediction model including:

[0149] Step 1: Input the training set into the convolutional neural network and calculate the actual output value;

[0150] Step 2: Use a preset loss function to calculate the error between the actual output value and the expected output value;

[0151] Step 3: Repeat steps 1 and 2 until the loss function converges or the preset number of training rounds is reached.

[0152] In one embodiment, a convolutional neural network (CNN) was chosen as the core framework for constructing the penetration rate prediction model. This model framework comprises three parts: an input layer, hidden layers, and an output layer. Specifically, an input layer is first set up to receive penetration rate-sensitive parameters as training data; then, the number of hidden layers and the number of neurons in each layer are determined, as these parameter settings have a crucial impact on the model's performance; finally, an output layer is set up, whose output value is the predicted penetration rate.

[0153] To train this convolutional neural network model, the following steps were employed: First, the permeability-sensitive parameters from a set of classified high-porosity sandstone samples were input into the model as the training set, and the actual output value of the model was calculated. Then, a preset mean squared error (MSE) was used as the loss function to calculate the error between the actual output value and the expected output value (i.e., the true permeability value). Next, the model parameters were adjusted based on the error value, and the above steps were repeated until the loss function converged or a preset number of training epochs was reached. In this process, Adam was chosen as the optimization algorithm to more effectively adjust the model parameters and accelerate the training process.

[0154] Finally, after training and optimization, a convolutional neural network model capable of accurately predicting permeability was obtained. This model can accept new permeability-sensitive parameters as input and output predicted permeability values, providing strong support for geological exploration and development work. By comparing the results of heuristic training with those of full input, it was found that the results of heuristic training have a higher correlation coefficient and are closer to the actual permeability values ​​in detail, further validating the effectiveness and accuracy of the model.

[0155] This invention provides a pore structure factor for permeability prediction, a method for classifying samples using the shear Lee factor and particle swarm optimization algorithm, and a heuristic neural network method for predicting permeability.

[0156] It is understood that the above embodiments only illustrate preferred embodiments of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can freely combine the above technical features without departing from the concept of the present invention, and can also make several modifications and improvements, all of which fall within the protection scope of the present invention. Therefore, all equivalent transformations and modifications made with respect to the scope of the claims of the present invention should fall within the scope of the claims of the present invention.

Claims

1. A method for predicting penetration rate based on heuristic convolutional neural networks, characterized in that, The method includes: Step S1: Obtain a sandstone sample set including the first rock physical elastic parameters and porosity parameters by performing lithological interpretation and analysis on the geological exploration data; Step S2: Perform correlation analysis on the first rock physical elastic parameters and the porosity to obtain a high-porosity sandstone sample set including the second rock physical elastic parameters and permeability parameters; Step S3: Based on rock physical analysis, perform correlation analysis on the second rock physical elastic parameter and the permeability to obtain a set of sensitive parameter combinations including the third rock physical elastic parameter, the shear Lee factor parameter and the permeability; Step S4: Perform cross-analysis on the shear Lee factor parameter and the permeability parameter of the sensitive parameter set. Based on the relationship between the shear Lee factor parameter and the permeability parameter, obtain the classification criteria, including: The high-porosity sandstone samples are divided into high-porosity high-permeability samples and high-porosity low-permeability samples according to the preset high-permeability and low-permeability boundaries. The cross-analysis of the shear Lee factor and the permeability of the sensitive parameter set, and the classification criteria obtained based on the relationship between the shear Lee factor and the permeability, include: Based on a preset threshold, the high-porosity sandstone sample set is divided into high-porosity and high-permeability samples and high-porosity and low-permeability samples. The third rock physical elastic parameter and the shear Lee factor in the second sensitive parameter combination set are intersected with the permeability parameter to determine the first classification threshold and the second classification threshold; Using the particle swarm optimization algorithm, the boundary between the high-porosity high-permeability sample and the high-porosity low-permeability sample is dynamically optimized within a preset floating range, with the goal of minimizing the permeability prediction error, to obtain a first floating threshold and a second floating threshold. The first classification standard is obtained based on the first classification threshold and the first floating threshold; The second classification standard is obtained based on the second classification threshold and the second floating threshold; The step of using particle swarm optimization to dynamically optimize the boundary between the high-porosity, high-permeability samples and the high-porosity, low-permeability samples within a preset floating range to obtain a first floating threshold and a second floating threshold includes: Step 1: Based on the shearing Lee factor, perform a classification and cross-analysis of the parameters in the second set of sensitive parameter combinations; Step 2: Adjust the origin of intersection and the rotation angle so that the dividing line distinguishes the high-porosity high-permeability sample from the high-porosity low-permeability sample, thereby determining the first dividing line and the second dividing line. Step 3: Adjust the first dividing line and the second dividing line to be perpendicular to each other in order to obtain the first parameter and the second parameter; Step 4: Linearly fit the first parameter or the second parameter to the permeability to obtain the first permeability prediction error, wherein the first permeability prediction error is the permeability prediction error obtained based on the coordinate axis rotation method; Step 5: Dynamically optimize the boundary line within a preset floating range using the particle swarm optimization algorithm, and obtain the first floating threshold and the second floating threshold based on the principle of minimizing the permeability prediction error; Step S5: Based on the classification criteria, classify the high-porosity sandstone sample set according to the classification criteria, input the classified samples into the predicted permeability model, and obtain the final predicted permeability data.

2. The method for penetration rate prediction based on heuristic convolutional neural networks according to claim 1, characterized in that, The geological exploration data includes permeability curves; Step S1 includes: The geological exploration data is cleaned, and the logarithm of the permeability curve is taken to obtain a preprocessed set of well logging data; The preprocessed well logging data set is filtered according to preset sandstone screening conditions to obtain a sandstone sample set containing the first rock physical elastic parameter and porosity parameter.

3. The method for penetration rate prediction based on heuristic convolutional neural networks according to claim 1, characterized in that, The set of sensitive parameter combinations includes a first set of sensitive parameter combinations and a second set of sensitive parameter combinations; Step S3 includes: By performing cross-analysis on the second rock physical elastic parameters and the permeability parameters, the first set of sensitive parameter combinations is obtained; The porosity and density parameters in the first set of sensitive parameters are calculated using a preset rock physics model to obtain the shear Lee factor parameters. The shear Lee factor parameter and the first sensitive parameter set are combined to obtain a second sensitive parameter set including the third rock physical elastic parameter, the shear Lee factor parameter and the permeability parameter.

4. The method for penetration rate prediction based on heuristic convolutional neural networks according to claim 3, characterized in that, The step of calculating the porosity and density parameters from the first set of sensitive parameters using a preset rock physics model to obtain the shear Lee factor parameters includes: Obtain a preset rock physics model; Based on the boundary theory method, the shear modulus parameters of the dry rock in the preset rock physics model are calculated; The initial shear compliance factor of the preset rock physics model is adaptively adjusted according to the particle swarm optimization algorithm to obtain the shear compliance factor parameters. Substituting the shear modulus parameter, the shear compliance factor parameter, and the porosity parameter of the dry rock into the first formula, the shear modulus parameter of the rock matrix is ​​obtained. According to the Gassmann equation, the shear modulus parameter of the rock matrix is ​​converted into the modulus parameter of the fluid-saturated rock, and the shear wave velocity parameter is calculated. Substituting the shear modulus parameter, porosity parameter, density parameter, and shear wave velocity parameter of the rock matrix into the second formula yields the shear Lee factor.

5. The method for penetration rate prediction based on heuristic convolutional neural networks according to claim 1, characterized in that, The step of linearly fitting the first parameter or the second parameter with the permeability to obtain the first permeability prediction error includes: When the third rock physical elastic parameter and the shear Lee factor are negatively correlated with the permeability parameter, the first permeability prediction error is obtained by linearly fitting the second parameter with the permeability. When the third rock physical elastic parameter and the shear Lee factor are positively correlated with the permeability parameter, the first parameter is linearly fitted with the permeability to obtain the first permeability prediction error.

6. The method for penetration rate prediction based on heuristic convolutional neural networks according to claim 1, characterized in that, Step S5 includes: The high-porosity sandstone sample set is classified according to the classification criteria to obtain different categories of sandstone subsets; the sandstone subsets include the third rock physical elastic parameter and the shear Lee factor; The third rock physical elastic parameter and shear Lee factor in the sandstone subset are used to divide the dataset into training and test sets. The training set is used to train the penetration rate prediction model to obtain the trained penetration rate prediction model; The test set is based on a trained penetration rate prediction model to obtain the predicted penetration rate.

7. The method for penetration rate prediction based on heuristic convolutional neural networks according to claim 6, characterized in that, The predicted penetration rate model includes a convolutional neural network; The step of training the penetration rate prediction model using the training set to obtain the trained penetration rate prediction model includes: Step 1: Input the training set into the convolutional neural network and calculate the actual output value; Step 2: Calculate the error between the actual output value and the expected output value using a preset loss function; Step 3: Repeat steps 1 and 2 until the preset loss function converges or the preset number of training rounds is reached.

8. The method for penetration rate prediction based on heuristic convolutional neural networks according to claim 4, characterized in that, The first formula includes: In the formula: The shear modulus parameter for dry rock; The shear modulus parameter of the rock matrix; Porosity; It is the shear compliance factor; The second formula includes: In the formula: Lee's shear factor; S The transverse wave velocity; Porosity; Density; This represents the shear modulus parameter of the rock matrix.

Images