Intelligent prediction method for shale content logging curve of oil and gas reservoir

Abstract

The invention discloses an oil and gas reservoir shale content logging curve intelligent prediction method, and belongs to the technical field of rock physics and logging curve processing and interpretation, and the method comprises the steps: in step 101, determining Gardner formula parameters through combining a determination coefficient and a grid optimization method, and determining a Gardner formula parameter; and then generating a virtual density data set capable of covering the actually measured density data set by using a Gardner formula. According to the method, the optimal parameters of the Gardner rock physical formula and the Larionov rock physical formula are determined by using the decision coefficient and the grid optimization method, so that the abnormal value of the actually measured logging data can be effectively corrected, and a fusion data set for training various machine learning models can be generated; a cross validation method, a correlation coefficient, a determination coefficient and a multi-dimensional SHAP method are comprehensively utilized to more comprehensively evaluate and explain a prediction result, compared with a traditional rock physics method, the shale content prediction precision is improved by 12%, a logging curve data set can be effectively corrected and expanded, and the prediction accuracy of the shale content is improved by 12%. And the accuracy and the reliability of predicting the shale content of the oil and gas reservoir can be obviously improved.

Description

Technical Field

[0001] The invention belongs to the technical field of rock physics and logging curve processing and interpretation, and in particular relates to an intelligent prediction method for logging curves of mud content in oil and gas reservoirs. Background Art

[0002] Oil and natural gas are not only non-renewable energy resources, but also strategic pillars for a country's survival and development. They play a vital role in ensuring energy security and promoting economic and social development. However, due to the limitations of my country's oil and gas exploration technology and costs, the exploration and development of deep, ultra-deep, thin, complex and harsh natural conditions on the ground in oil and gas reservoirs faces great technical challenges. Well logging data, as a more direct reflection of the physical properties of oil and gas formations, contains rich geological information and provides a reliable basis for accurately obtaining geological parameters. Through professional analysis of well logging curves, geologists can understand the geological structure and physical properties of formation rocks in detail, so as to better understand the underground geological conditions and establish a more accurate geological model based on this. Therefore, in oil and gas exploration and development, well logging curves are an indispensable and important source of data.

[0003] Due to the complexity of the three-dimensional structure of the formation and the limitations of logging methods and equipment, logging curves can usually only reflect the formation parameters around the wellbore. When analyzing the formation farther away from the wellbore, the parameters may change significantly. Considering that drilling activities are costly and require huge manpower and material resources, logging wells are usually arranged in specific research areas. This layout may lead to blank formation information in areas where wells are not laid, and even ignore the existence of hidden reservoirs, resulting in a waste of resources. In addition, during the logging process, some logging data may be distorted or missing due to equipment limitations, mud invasion or well wall collapse. Therefore, using existing logging data to correct and complete logging data or predict logging curves in unknown areas has become an important means of reservoir prediction, which can effectively improve the reliability and safety of drilling while reducing exploration costs.

[0004] In the early stage of oil logging, researchers mostly used mathematical statistics methods to process and interpret logging data. For example, Archie (1942) could infer porosity from measured resistivity and known formation water resistivity; Larionov (1969) proposed a method to estimate the clay content in reservoir rocks through the relationship between shale volume and gamma ray index; Gardner GHF, Gardner LW et al. (1974) described the relationship between acoustic wave velocity and density in rocks through their equations; Castagna, Batzle et al. (1985) derived an empirical formula for shear wave velocity based on longitudinal wave velocity; Smith (2007) studied the relationship between resistivity logging and acoustic logging, with the main purpose of improving the interpretation of formation characteristics through the correlation between acoustic wave velocity (compressional or shear wave velocity) and resistivity.

[0005] This method based on mathematical statistics and rock physics models is the main means of logging data processing and interpretation. Although this traditional logging data processing and interpretation method has laid a theoretical foundation and provided technical support for subsequent research, they still have limitations. This method relies on rock physics models for mathematical and statistical analysis, but fails to fully consider the potential nonlinear correlation between data, resulting in certain limitations on the accuracy and applicability of the prediction results. In practical applications, this method based on mathematical statistics and rock physics models is difficult to provide sufficiently accurate geological information. First, empirical models and statistical methods are often verified based on specific assumptions or limited data, resulting in insufficient generalization ability. Secondly, these methods rely on artificial selection and adjustment of model parameters, which can easily introduce subjective biases. Finally, when dealing with large-scale or high-dimensional data, traditional methods often find it difficult to effectively deal with complex data structures.

[0006] With the rapid development of big data technology and the widespread application of deep learning in science and engineering, more and more researchers and production personnel have begun to use data-driven methods such as artificial neural networks to process and interpret well logging curves. For example, Rolon, Mohaghegh et al. (2009) used a neural network model to synthesize resistivity curves through natural gamma, density and neutron curve data, showing higher prediction accuracy than traditional methods. In addition, researchers have also tried to use advanced artificial intelligence algorithms such as deep learning, transfer learning, and small sample learning to construct prediction models for well logging curves. These algorithms perform well under specific conditions. However, due to the complex network structure of these models, it is difficult to accurately interpret and capture the rock physical relationship between the input and output curves. They may still face challenges when dealing with nonlinear problems in new work areas and more complex formations.

[0007] In summary, the traditional methods based on mathematical statistics and rock physics models have the problem of limited generalization ability and scope of application. Artificial adjustment of rock physics model parameters will lead to large errors, and the machine learning method is a black box operation, which cannot determine the rock physics relationship between output data and input data. To this end, the present invention proposes an intelligent prediction method for logging curves of shale content in oil and gas reservoirs based on rock physics constraints. The method uses the determination coefficient and grid optimization method to determine the optimal parameters of the Gardner rock physics formula and the Larionov rock physics formula, which can effectively correct the abnormal values ​​of logging data and generate virtual data sets for training multiple machine learning models. The cross-validation method, correlation coefficient, determination coefficient and multi-dimensional SHAP method are comprehensively used to evaluate and interpret the prediction results more comprehensively. Compared with the traditional rock physics method, the prediction accuracy of shale content is improved by 12%. This method can not only effectively correct and expand the logging curve data set, but also significantly improve the accuracy and reliability of shale content prediction.

[0008] Based on this, the present invention designs an intelligent prediction method for logging curves of mud content in oil and gas reservoirs to solve the above problems. Summary of the invention

[0009] The purpose of the present invention is to solve the problem that the generalization ability and application scope of the traditional method based on mathematical statistics and rock physics model are limited, and to propose an intelligent prediction method for the logging curve of the mud content of oil and gas reservoirs.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions: An intelligent prediction method for logging curves of mud content in oil and gas reservoirs, comprising: In step 101, the parameters of the Gardner formula are determined by combining the coefficient of determination and the grid optimization method, and then the Gardner formula is used to generate a virtual density data set that can cover the measured density data set, and the virtual data set is combined with the measured data set to form a comprehensive density data set; In step 102, the determination coefficient and grid optimization method are also used to determine the parameters of the Larionov formula, and a virtual gamma data set that can cover the actual measured gamma data set is created based on this formula. This virtual data set is merged with the measured data set to construct a complete gamma data set; In step 103, the fused density data set and gamma data set are used to apply a variety of machine learning algorithms such as random forest (RFs) algorithm, gradient boost decision tree (XGBoost) algorithm, multi-layer perceptron (MLP) algorithm and support vector machine (SVM) to perform regression training of the model to intelligently predict the mud content. The prediction accuracy of each algorithm is evaluated by the cross-validation method; In step 104, correlation coefficient, determination coefficient and multi-dimensional SHAP method are used to comprehensively interpret the prediction results of machine learning, and compare them with traditional rock physics methods to analyze the improvement effect of the fused data set on the intelligent prediction performance.

[0011] As a further description of the above technical solution: In step 101, the variable involved in the Gardner empirical formula is only the sound wave velocity, and the formula is as follows: (1) Where ρ is the rock density (g / cc); Vt is the speed of sound waves (us / ft); a and b are constants and are regional; The Gardner empirical formula (1) fits the best parameters a=0.0665 and b=0.3923 for this work area based on the determination coefficient and grid optimization comprehensive parameter adjustment. The highest determination coefficient of the fitting effect can reach 0.998, which shows the reliability and accuracy of this method. Then, the virtual velocity-density data set is generated using the Gardner formula.

[0012] When generating a virtual velocity-density data set, the following principles need to be followed: the generated virtual data set needs to cover the measured velocity-density data set. At the same time, in order to maintain the characteristics of the measured data set, the measured data set cannot be expanded on a large scale. It is optimal to just cover the measured data set. Virtual density data is generated by selecting the acoustic velocity and coefficient within a certain range. The range of the acoustic velocity Vt is 115 to 150, and the step length is 0.01, which represents the range of variation of the acoustic velocity. The value range of the coefficient a in the density formula is 0.0653 to 0.0675, and the step length is 0.00005. The parameter b=0.3923 is the constant exponent in the density formula, which represents the exponential relationship between the acoustic velocity and density. Through these parameters, the code is used to generate density curves under different a values ​​to simulate the virtual data relationship between the acoustic velocity and density. We simply merge the measured data set and the virtual data set to obtain the fused data set.

[0013] As a further description of the above technical solution: In step 102, based on the Larionov correction formula of the determination coefficient and grid optimization, the optimal parameter a=3.2 is fitted for the work area. There is a relationship between VSh and IRA: (2) Where IRA is the gamma ray or radioactivity index; VSh is the mud content.

[0014] (3) Where GR is the true gamma log reading; GRcs is the gamma ray reading of clean sandstone (no shale), which is 20; GRSh is the gamma ray reading of shale, which is 140.

[0015] The following principles need to be followed when using the Larionov formula to generate a virtual gamma-shale content dataset: the generated virtual dataset needs to cover the measured gamma-shale content dataset. At the same time, in order to maintain the characteristics of the measured dataset, the measured dataset cannot be expanded on a large scale. It is optimal to just cover the measured dataset. The virtual gamma curve (GR) data is generated by selecting the shale content (VSh) value and parameters within a certain range. We defined the range of shale content from 0.01 to 0.3 with a step size of 0.0001. The parameter a in the Larionov formula ranges from 1 to 5.5 with a step size of 0.03. Through this parameter, the code was used to generate gamma-ray curves under different a values ​​to simulate the virtual data relationship between shale content and gamma. The fused dataset can be obtained by simply merging the measured dataset and the virtual dataset.

[0016] As a further description of the above technical solution: In step 103, algorithms such as random forest (RFs) algorithm, gradient boosting decision tree (XGBoost) algorithm, multi-layer perceptron (MLP) algorithm and support vector machine (SVM) are used to perform machine learning training based on the fusion data set to intelligently predict the mud content.

[0017] As a further description of the above technical solution: In step 103, during the model training process, we selected three wells, J032, J061 and J034, as training wells, used gamma, acoustic wave, density, impedance and velocity curves as input features, and used shale content as the prediction target to train the model. Well J021 was used for model verification, and the algorithm prediction results were evaluated by the determination coefficient (R²) and correlation coefficient (r). The results showed that the RFs algorithm performed best, with a determination coefficient of 0.999 and a correlation coefficient of 0.998.

[0018] As a further description of the above technical solution: In step 104, the accuracy of the intelligent prediction results of reservoir physical properties is evaluated using the cross-validation method, correlation coefficient, determination coefficient (R2) and SHAP value method.

[0019] As a further description of the above technical solution: Cross-validation is a statistical method widely used in machine learning model performance evaluation. Its core idea is to verify the stability and generalization ability of the model by dividing the data set multiple times. In specific operations, cross-validation divides the data set into several equal-sized parts (called "folds"), selects one of the folds as the test set each time, and the remaining folds as the training set, and repeatedly trains and tests until each fold is used as a test set. The most common K-fold cross-validation is to divide the data into K subsets, and perform K training and testing respectively. The final evaluation result is obtained by the average of multiple tests: The correlation coefficient is a statistic used to measure the strength of the linear relationship between two variables, usually expressed as: (4) In the formula, xi refers to the actual measured curve value (ie, the true value), yi is the curve value predicted by the model (ie, the predicted value), and x and y are the average values ​​of x and y, respectively. It is a statistical indicator used to measure the degree of correlation between two variables, usually represented by the Pearson correlation coefficient. Its value is between -1 and 1: 1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 indicates no linear correlation. The calculation method is the covariance of the two variables divided by the product of their standard deviations, which reflects the strength and direction of the linear relationship between the variables. The correlation coefficient can be used to understand the linear consistency between the model predicted value and the actual value; For the coefficient of determination, it is defined as follows: (5) Among them, xi and yi are the original value and the predicted value of the machine learning model respectively, X is the average value of xi, and n is the total number of samples. In R2, the sum of the squares of the difference between the actual value and the predicted value, that is, the total amount of information that our model does not capture, and the denominator is the amount of information carried by the true label. So both measure 1-the ratio of the amount of information that our model does not capture to the amount of information carried by the true label, so the closer to 1, the better. If the result is 0, it means that the model fit is very poor; if the result is 1, it means that the model is error-free; SHAP is based on the Shapley value in game theory and provides a fair and consistent feature attribution method. In this framework, the contribution of each feature is calculated by considering all possible feature combinations and their average impact under these combinations, thus ensuring that the result is not affected by the order of feature input.

[0020] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: The intelligent prediction method for logging curves of shale content in oil and gas reservoirs under rock physics constraints provided by the present invention comprehensively utilizes the determination coefficient and grid optimization method to determine the optimal parameters of the Gardner rock physics formula and the Larionov rock physics formula, can effectively correct the abnormal values ​​of the measured logging data and generate a fusion data set for training multiple machine learning models, and comprehensively utilizes the cross-validation method, correlation coefficient, determination coefficient and multi-dimensional SHAP method to more comprehensively evaluate and interpret the prediction results. Compared with the traditional rock physics method, the prediction accuracy of the shale content is improved by 12%. This method can not only effectively correct and expand the logging curve data set, but also significantly improve the accuracy and reliability of the prediction of the shale content in oil and gas reservoirs. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flowchart of a specific embodiment of a method for intelligent prediction of well logging curves of shale content in oil and gas reservoirs based on rock physics constraints of the present invention; Figure 2 A sonic and density cross-plot of a fused data set consisting of measured data and rock physics generated data of the present invention; Figure 3 Comparison between the density data generated by the Gardner formula based on the coefficient of determination and grid optimization of the present invention and the actual well logging data set; Figure 4 Gamma and shale content cross-plots of the measured data of the present invention and the fused data set consisting of data generated based on Larionov rock physics relations; Figure 5 A comparison chart of the results of intelligent prediction of shale content based on the rock physics method and the RFs algorithm of the present invention; Figure 6 This is a process diagram for explaining the importance of features of the random forest algorithm using the multi-dimensional SHAP method of the present invention. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Please see attached Figure 1 -Attached Figure 6 The present invention provides a technical solution: an intelligent prediction method for logging curves of mud content in oil and gas reservoirs, comprising: In step 101, the parameters of the Gardner formula are determined by combining the coefficient of determination with the grid optimization method, and then the Gardner formula is used to generate a virtual density data set that can cover the measured density data set. This virtual data set is combined with the measured data set to form a comprehensive density data set.

[0024] The scatter plot intersection plot can show the scatter plot between each pair of variables in the data set, which helps to observe the relationship between different logging parameters. It can be observed whether there is a linear relationship between two parameters and whether there is a clustering trend. By setting different shapes for different well categories, the scatter plot can clearly show the distribution differences of data of different categories. This is particularly useful when studying the performance of different regions of the formation. By observing the scatter plot intersection plot, researchers can quickly identify possible patterns, trends and outliers in the data, which is very valuable for further geological analysis and decision support. It can be seen from the intersection plot that there is an anomaly in the density curve of the J021 well. Then, the velocity density logging scatter data intersection plot is corrected based on the Gardner formula of the determination coefficient and grid optimization, and the density data set is generated using the acoustic logging data, such as Figure 3 As shown. The only variable involved in the Gardner empirical formula is the speed of sound waves, and the formula is as follows: (1) Where ρ is the rock density (g / cc); Vt is the speed of sound waves (us / ft); a and b are constants and are regional in nature.

[0025] Gardner empirical formula (1) fits the best parameters a=0.0665 and b=0.3923 for this work area based on the determination coefficient and grid optimization comprehensive parameter adjustment. The highest determination coefficient of the fitting effect can reach 0.998, which shows the reliability and accuracy of this method.

[0026] Then, the virtual velocity-density data set is generated using the Gardner formula. When generating the virtual velocity-density data set, the following principles need to be followed: the generated virtual data set needs to cover the measured velocity density data set. At the same time, in order to maintain the characteristics of the measured data set, the measured data set cannot be expanded on a large scale. It is optimal to just cover the measured data set. The virtual density data is generated by selecting the acoustic wave velocity and coefficient within a certain range. The range of the acoustic wave velocity Vt is 115 to 150, and the step length is 0.01, which represents the range of variation of the acoustic wave velocity. The value range of the coefficient a in the density formula is 0.0653 to 0.0675, and the step length is 0.00005. The parameter b=0.3923 is the constant exponent in the density formula, which represents the exponential relationship between the acoustic wave velocity and density. Through these parameters, the code is used to generate density curves under different a values ​​to simulate the virtual data relationship between the acoustic wave velocity and density. We simply merge the measured data set and the virtual data set to obtain the fused data set. Compared with the measured data set, the fused data set has the following advantages: it can not only maintain the integrity of the measured data, but also make up for the shortcomings of the measured data. Effective expansion of the data set under rock physics constraints can lay a solid data foundation for improving the generalization ability and prediction effect of subsequent intelligent predictions.

[0027] In step 102, the determination coefficient and grid optimization method are also used to determine the parameters of the Larionov formula, and a virtual gamma data set that can cover the actual measured gamma data set is created based on this formula. This virtual data set is merged with the measured data set to construct a complete gamma data set; Based on the Larionov correction formula of the determination coefficient and grid optimization, the optimal parameter a=3.2 is fitted for this work area. There is a relationship between VSh and IRA: (2) Where IRA is the gamma ray or radioactivity index; VSh is the mud content.

[0028] (3) Where GR is the true gamma log reading; GRcs is the gamma ray reading of clean sandstone (no shale), which is 20; GRSh is the gamma ray reading of shale, which is 140.

[0029] Then, the Larionov formula is used to generate a virtual gamma-shale content data set. When generating a virtual gamma-shale content data set, the following principles need to be followed: the generated virtual data set needs to cover the measured gamma-shale content data set. At the same time, in order to maintain the characteristics of the measured data set, the measured data set cannot be expanded to a large extent. It is optimal to just cover the measured data set. Virtual gamma curve (GR) data is generated by selecting shale content (VSh) values ​​and parameters within a certain range. We define the range of shale content from 0.01 to 0.3 with a step size of 0.0001. The parameter a in the Larionov formula ranges from 1 to 5.5 with a step size of 0.03. Through this parameter, the code is used to generate gamma ray curves under different a values ​​to simulate the virtual data relationship between shale content and gamma. The measured data set and the virtual data set are simply merged to obtain a fused data set. Compared with the measured data set, the fused data set has the following advantages: it can not only maintain the integrity of the measured data, but also make up for the shortcomings of the measured data. Effective expansion of the data set under rock physics constraints can lay a solid data foundation for improving the generalization ability and prediction effect of subsequent intelligent predictions.

[0030] In step 103, the fused density data set and gamma data set are used to perform regression training of the model using a variety of machine learning algorithms such as random forest (RFs) algorithm, gradient boosting decision tree (XGBoost) algorithm, multi-layer perceptron (MLP) algorithm, and support vector machine (SVM) to intelligently predict the mud content. The prediction accuracy of each algorithm is evaluated by the cross-validation method; Using algorithms such as random forest (RFs), gradient boosting decision tree (XGBoost), multi-layer perceptron (MLP) and support vector machine (SVM), machine learning training is performed based on the fusion data set to intelligently predict the mud content. The RFs algorithm reduces the risk of overfitting the model to specific training data by constructing multiple decision trees and averaging or majority voting the results. It can handle data sets with a large number of features well and does not require excessive selection and preprocessing of features. XGBoost minimizes errors to improve prediction accuracy by constructing multiple step-by-step optimized decision trees. Its advantages are built-in regularization to prevent overfitting, support for parallel processing to increase speed, and strong processing capabilities for missing values. MLP uses a multi-layer neuron architecture and learns through a back-propagation algorithm to capture the nonlinear relationship of data. Its advantages are that it can simulate complex functions and is suitable for various tasks such as regression and classification. SVM classifies data by finding a hyperplane that maximizes the sample interval. Its advantages include good generalization performance, effective processing of high-dimensional data, and the ability to use kernel functions to solve nonlinear problems.

[0031] During the model training process, we selected three wells, J032, J061 and J034, as training wells, used gamma, acoustic wave, density, impedance and velocity curves as input features, and used shale content as the prediction target to train the model. Well J021 was used for model validation, and the algorithm prediction results were evaluated by the determination coefficient (R²) and correlation coefficient (r). The results showed that the RFs algorithm performed best, with a determination coefficient of 0.999 and a correlation coefficient of 0.998.

[0032] In step 104, correlation coefficient, determination coefficient and multi-dimensional SHAP method are used to comprehensively interpret the prediction results of machine learning and compare them with traditional rock physics methods to analyze the improvement effect of the fused data set on the intelligent prediction performance.

[0033] For the intelligent prediction results of reservoir physical properties, the cross-validation method, correlation coefficient, determination coefficient (R2) and SHAP value method were used to evaluate its accuracy; Cross-validation is a statistical method widely used in machine learning model performance evaluation. Its core idea is to verify the stability and generalization ability of the model by dividing the data set multiple times. In specific operations, cross-validation divides the data set into several equal-sized parts (called "folds"), selects one of the folds as the test set each time, and the remaining folds as the training set, and repeatedly trains and tests until each fold is used as a test set. The most common K-fold cross-validation is to divide the data into K subsets and perform K training and testing respectively. The final evaluation result is obtained by the average of multiple tests. This method can not only avoid the model evaluation bias that may be caused by a single data partition, but also effectively balance the data distribution of training and testing, and improve the evaluation effect of the model's generalization ability. The advantage of this method is that it can make full use of the data set, especially when the amount of data is limited. Compared with the training-testing split of a single partition, cross-validation can provide a more robust performance evaluation and help reduce the risk of overfitting, so that the adaptability of the model to new data can be more accurately measured.

[0034] The correlation coefficient is a statistic used to measure the strength of the linear relationship between two variables, usually expressed as: (4) In the formula, xi refers to the actual measured curve value (ie, the true value), yi is the curve value predicted by the model (ie, the predicted value), and are the average values ​​of x and y, respectively. It is a statistical indicator used to measure the degree of correlation between two variables, usually represented by the Pearson correlation coefficient. Its value is between -1 and 1: 1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 indicates no linear correlation. The calculation method is the covariance of the two variables divided by the product of their standard deviations, which reflects the strength and direction of the linear relationship between the variables. Through the correlation coefficient, we can understand the linear consistency between the model predicted value and the actual value.

[0035] For the coefficient of determination, it is defined as follows: (5) Where xi and yi are the original value and the predicted value of the machine learning model, respectively, X is the average of xi, and n is the total number of samples. In R2, the sum of the squares of the difference between the actual value and the predicted value, that is, the total amount of information that our model does not capture, and the denominator is the amount of information carried by the true label. So both measure 1-the amount of information that our model does not capture as a proportion of the amount of information carried by the true label, so the closer both are to 1, the better. If the result is 0, it means that the model fit is very poor; if the result is 1, it means that the model is error-free. SHAP is based on the Shapley value in game theory and provides a fair and consistent feature attribution method. In this framework, the contribution of each feature is calculated by considering all possible feature combinations and their average impact under these combinations, thus ensuring that the result is not affected by the order of feature input. The advantage of SHAP is that it not only improves the transparency of the model, but also enables users to deeply understand the logic behind the prediction. In addition, SHAP can be used to optimize feature engineering, by quantitatively evaluating the importance of individual features, helping to screen out the features that have the greatest impact on model performance, thereby simplifying model input. It also plays a key role in model error diagnosis, helping to identify potential causes of inaccurate predictions, thereby guiding model adjustments and improvements. Through these capabilities, SHAP significantly enhances the interpretability and transparency of models, allowing users to not only see the prediction results, but also understand why the model reached these conclusions. This interpretability greatly improves the trust in the model, especially in practical applications, helping to make the decision-making process of machine learning models more transparent and reliable.

[0036] Using multiple evaluation indicators to evaluate the prediction model at the same time can not only analyze the specific contribution of features to model predictions, but also intuitively see the accuracy of the predicted values. This greatly enhances the transparency and credibility of the model, while providing insights into actual operations, so as to better understand and evaluate these models. Figure 6As shown in the figure, gamma plays a decisive role in the process of predicting shale content. From the comprehensive analysis of SHAP diagram, R² and r, it can be seen that the key to the successful prediction of shale content is to modify and generate a virtual training data set with a certain rock physics relationship with the prediction target based on the Gardner formula and Larionov formula based on the determination coefficient and grid optimization.

[0037] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent prediction method for logging curves of mud content in oil and gas reservoirs, characterized in that: include: In step 101, the parameters of the Gardner formula are determined by combining the coefficient of determination and the grid optimization method, and then the Gardner formula is used to generate a virtual density data set that can cover the measured density data set, and the virtual data set is combined with the measured data set to form a comprehensive density data set; In step 102, the determination coefficient and grid optimization method are also used to determine the parameters of the Larionov formula, and a virtual gamma data set that can cover the actual measured gamma data set is created based on this formula. This virtual data set is merged with the measured data set to construct a complete gamma data set; In step 103, the fused density data set and gamma data set are used to apply a variety of machine learning algorithms such as random forest (RFs) algorithm, gradient boost decision tree (XGBoost) algorithm, multi-layer perceptron (MLP) algorithm and support vector machine (SVM) to perform regression training of the model to intelligently predict the mud content. The prediction accuracy of each algorithm is evaluated by the cross-validation method; In step 104, correlation coefficient, determination coefficient and multi-dimensional SHAP method are used to comprehensively interpret the prediction results of machine learning, and compare them with traditional rock physics methods to analyze the improvement effect of the fused data set on the intelligent prediction performance.

2. The intelligent prediction method of well logging curve of oil and gas reservoir shale content according to claim 1 is characterized in that: In step 101, the variable involved in the Gardner empirical formula is only the speed of sound waves, and the formula is as follows: (1) Where ρ is the rock density (g / cc); Vt is the speed of sound waves (us / ft); a and b are constants and are regional; The Gardner empirical formula (1) fits the best parameters a=0.0665 and b=0.3923 for this work area based on the determination coefficient and grid optimization comprehensive parameter adjustment. The highest determination coefficient of the fitting effect can reach 0.998, which shows the reliability and accuracy of this method. Then, the virtual velocity-density data set is generated using the Gardner formula.

3. When generating a virtual velocity-density data set, the following principles need to be followed: the generated virtual data set needs to cover the measured velocity-density data set. At the same time, in order to maintain the characteristics of the measured data set, the measured data set cannot be expanded on a large scale. It is optimal to just cover the measured data set. Virtual density data is generated by selecting the acoustic wave velocity and coefficient within a certain range. The range of the acoustic wave velocity Vt is 115 to 150, and the step length is 0.01, which represents the range of variation of the acoustic wave velocity. The value range of the coefficient a in the density formula is 0.0653 to 0.0675, and the step length is 0.00005. The parameter b=0.3923 is the constant exponent in the density formula, which represents the exponential relationship between the acoustic wave velocity and density. Through these parameters, the code is used to generate density curves under different a values ​​to simulate the virtual data relationship between the acoustic wave velocity and density. We simply merge the measured data set and the virtual data set to obtain the fused data set.

4. The intelligent prediction method of well logging curve of oil and gas reservoir shale content according to claim 1 is characterized in that: In step 102, based on the Larionov correction formula of the determination coefficient and grid optimization, the optimal parameter a=3.2 is fitted for the work area. There is a relationship between VSh and IRA: (2) Where IRA is the gamma ray or radioactivity index; VSh is the mud content.

5. (3) Where GR is the true gamma log reading; GRcs is the gamma ray reading of clean sandstone (no shale), which is 20; GRSh is the gamma ray reading of shale, which is 140.

6. The following principles need to be followed when using the Larionov formula to generate a virtual gamma-shale content dataset: the generated virtual dataset needs to cover the measured gamma-shale content dataset. At the same time, in order to maintain the characteristics of the measured dataset, the measured dataset cannot be expanded on a large scale. It is optimal to just cover the measured dataset. The virtual gamma curve (GR) data is generated by selecting the shale content (VSh) value and parameters within a certain range. We defined the range of shale content from 0.01 to 0.3 with a step size of 0.0001. The parameter a in the Larionov formula ranges from 1 to 5.5 with a step size of 0.

03. Through this parameter, the code was used to generate gamma-ray curves under different a values ​​to simulate the virtual data relationship between shale content and gamma. A simple merger of the measured dataset and the virtual dataset can obtain a fused dataset.

7. The intelligent prediction method of well logging curve of mud content in oil and gas reservoirs according to claim 1 is characterized in that: In step 103, algorithms such as random forest (RFs) algorithm, gradient boosting decision tree (XGBoost) algorithm, multi-layer perceptron (MLP) algorithm and support vector machine (SVM) are used to perform machine learning training based on the fusion data set to intelligently predict the mud content.

8. The intelligent prediction method of well logging curve of oil and gas reservoir shale content according to claim 4 is characterized in that: In step 103, during the model training process, we selected three wells, J032, J061 and J034, as training wells, used gamma, acoustic wave, density, impedance and velocity curves as input features, and used shale content as the prediction target to train the model. Well J021 was used for model verification, and the algorithm prediction results were evaluated by the determination coefficient (R²) and correlation coefficient (r). The results showed that the RFs algorithm performed best, with a determination coefficient of 0.999 and a correlation coefficient of 0.

998.

9. The intelligent prediction method of well logging curve of oil and gas reservoir shale content according to claim 4 is characterized in that: In step 104, the accuracy of the intelligent prediction results of reservoir physical properties is evaluated using the cross-validation method, correlation coefficient, determination coefficient (R2) and SHAP value method.

10. The intelligent prediction method of well logging curve of oil and gas reservoir shale content according to claim 3, characterized in that: Cross-validation is a statistical method widely used in machine learning model performance evaluation. Its core idea is to verify the stability and generalization ability of the model by dividing the data set multiple times. In specific operations, cross-validation divides the data set into several equal-sized parts (called "folds"), selects one of the folds as the test set each time, and the remaining folds as the training set, and repeatedly trains and tests until each fold is used as a test set. The most common K-fold cross-validation is to divide the data into K subsets, and perform K training and testing respectively. The final evaluation result is obtained by the average of multiple tests: The correlation coefficient is a statistic used to measure the strength of the linear relationship between two variables, usually expressed as: (4) In the formula, xi refers to the actual measured curve value (ie, the true value), yi is the curve value predicted by the model (ie, the predicted value), and x and y are the average values ​​of x and y, respectively. It is a statistical indicator used to measure the degree of correlation between two variables, usually represented by the Pearson correlation coefficient. Its value is between -1 and 1: 1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 indicates no linear correlation. The calculation method is the covariance of the two variables divided by the product of their standard deviations, which reflects the strength and direction of the linear relationship between the variables. The correlation coefficient can be used to understand the linear consistency between the model predicted value and the actual value; For the coefficient of determination, it is defined as follows: (5) Among them, xi and yi are the original value and the predicted value of the machine learning model respectively, X is the average value of xi, and n is the total number of samples. In R2, the sum of the squares of the difference between the actual value and the predicted value, that is, the total amount of information that our model does not capture, and the denominator is the amount of information carried by the true label. So both measure 1-the ratio of the amount of information that our model does not capture to the amount of information carried by the true label, so the closer to 1, the better. If the result is 0, it means that the model fit is very poor; if the result is 1, it means that the model is error-free; SHAP is based on the Shapley value in game theory and provides a fair and consistent feature attribution method. In this framework, the contribution of each feature is calculated by considering all possible feature combinations and their average impact under these combinations, thus ensuring that the result is not affected by the order of feature input.

Images