Factor analysis-based post-compression effect intelligent evaluation method

Abstract

The invention discloses an intelligent post-fracturing effect evaluation method based on factor analysis, and relates to the technical field of oil and gas reservoir intelligent algorithms. The method comprises the following steps: obtaining yield data of fractured wells in the same block and parameters influencing the yield, and preprocessing the yield data and the parameters to obtain evaluation parameters; based on a CatBoost model and a cross validation theory, performing preliminary screening on the evaluation parameters to obtain candidate parameters; based on a factor analysis method, the number of factors participating in analysis is obtained through the variance contribution rate of each factor and an accumulated variance contribution rate threshold value, comprehensive scores of different wells are obtained through a comprehensive score calculation formula in factor analysis, and the higher the comprehensive score is, the better the post-fracturing effect is. Compared with the prior art, the method has higher intelligent analysis capability, key factors can be accurately identified, correlation among multiple variables can be reduced, and more comprehensive and stable analysis results can be provided.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent algorithms for oil and gas reservoirs, and in particular to an intelligent evaluation method for post-compression effects based on factor analysis. Background Art

[0002] Due to its significant anisotropy and complex geological conditions, tight conglomerate reservoirs transfer stress in different ways in different directions, increasing the complexity and uncertainty of the stress field. At the same time, the heterogeneity and structural characteristics inside the reservoir lead to local stress concentration, further exacerbating the unevenness of stress distribution. Many traditional models are based on static conditions or simplified assumptions, and it is difficult to accurately capture the stress evolution under dynamically changing environments. Especially when dealing with complex geological conditions, it brings major challenges to the prediction of ground stress.

[0003] Existing machine learning relies heavily on a large amount of high-quality training data. Insufficient or unevenly distributed data can lead to performance degradation. The ability to capture the relationship between complex anisotropic behavior and nonlinear stress is limited, and a simple mapping relationship between input and output is usually assumed. It is difficult to handle the spatial heterogeneity and anisotropy that are common in tight conglomerate reservoirs, which makes the prediction results less accurate under complex reservoir conditions. In addition, there are limitations in interpretability and accuracy, resulting in poor universality in different geological environments and difficulty in meeting actual needs.

[0004] In the prior art, many technologies directly use factor analysis to evaluate a certain behavior. For example, Chinese patent CN115860509A discloses a method for evaluating the competitiveness of liquor companies based on factor analysis, and Chinese patent CN116485200A discloses a comprehensive evaluation method for the development effect of low-permeability oil fields based on factor analysis. Although theoretically, the use of factor analysis alone can evaluate these to a certain extent, the inventors found through experiments that the accuracy of the evaluation is relatively low and has weak guiding significance for actual production. Summary of the invention

[0005] To solve at least one of the above problems, the present invention proposes an intelligent evaluation method for post-pressing effect based on factor analysis.

[0006] The technical solution of the present invention is: a method for intelligent evaluation of post-pressing effect based on factor analysis, comprising the following steps: S1. Obtaining production data and parameters affecting production of fractured wells in the same block, and preprocessing them to obtain evaluation parameters; S2. Based on the CatBoost model and cross-validation theory, the evaluation parameters are preliminarily screened to obtain candidate parameters; S3. Based on the factor analysis method, the number of factors involved in the analysis is obtained through the variance contribution rate of each factor and the cumulative variance contribution rate threshold. The comprehensive score calculation formula in the factor analysis is used to obtain the comprehensive scores of different wells. The higher the comprehensive score, the better the post-pressure effect.

[0007] Beneficial effects: The present invention significantly improves the accuracy and efficiency of the post-pressure effect evaluation analysis by combining the CatBoost model to extract the main control factors of yield, cross-validating to determine the optimal number of parameters, and using the factor analysis method to obtain the final post-pressure effect intelligent evaluation results. Compared with the existing technology, it has stronger intelligent analysis capabilities, can accurately identify key factors and reduce the correlation between multiple variables, and provide more comprehensive and stable analysis results. This method not only improves the decision-making support capabilities, but also has strong adaptability, can be applied to different types of data, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 Results graph of the optimal number of parameters determined for cross-validation; Figure 2 It is a comparison chart of the comprehensive factor scores of the present invention and the comprehensive factor scores of direct factor analysis. DETAILED DESCRIPTION

[0009] The specific implementation modes of the present invention will be described clearly and completely below in conjunction with examples and drawings. Obviously, the described examples are only some embodiments of the present invention, rather than all embodiments.

[0010] A method for intelligent evaluation of post-pressing effect based on factor analysis, comprising the following steps: S1. Obtaining production data and parameters affecting production of fractured wells in the same block, and preprocessing them to obtain evaluation parameters; In this step, the parameters affecting production include fracturing parameters, perforation parameters, geological parameters, drainage parameters and logging parameters, wherein the fracturing parameters include total sand volume, total liquid volume, fracturing fluid type, fracturing stage, total cluster number, ball feeding fluid dosage, acid dosage, pre-fluid volume, pre-fluid ratio, sand-carrying fluid dosage, displacement fluid dosage, average sand ratio, quartz sand volume, quartz sand ratio, ceramsite volume, ceramsite ratio, average construction pressure, fracture pressure, average pump stop pressure, and closing pressure; the perforation parameters include average perforation length, average segment length, average segment spacing, perforation concentration, perforation density, and the product of perforation concentration and perforation density; the geological parameters include abundance, gas content, and oil layer thickness; the drainage parameters include: liquid injection intensity, sand addition intensity, construction displacement, and average cluster displacement; the logging parameters include low gamma horizontal section length and low gamma horizontal section length ratio. Since the parameter values ​​are large and numerous, it is difficult to list the original data one by one. In this embodiment, Table 1 lists the statistics of the original data.

[0011] In particular, the parameter "the product of perforation concentration and perforation density" refers to the product of the average value of the perforation concentration and the average value of the perforation density of all the fracturing stages in the same well.

[0012] Table 1 Statistics of original data

[0013] The above parameters can all be obtained from the production process of developed wells, and the parameters obtained by statistically analyzing multiple wells that have been fractured in this block. At the same time, the above parameters are selected as evaluation parameters affecting production based on the analysis of fracture production of multiple oil fields by the inventors: in the actual fracturing development process of oil fields, the above parameters are usually required to be considered when performing fracturing operations. Of course, those skilled in the art can select other parameters that can affect the fracturing effect.

[0014] In the prior art, the relationship between the evaluation parameters that affect production and production is usually used to guide the fracturing development of other wells in the same block. However, those skilled in the art know that the more evaluation parameters there are, the higher the difficulty and the lower the accuracy of the final evaluation; at the same time, since some parameters have little effect on the fracturing effect, it is necessary to screen the number of evaluation parameters to select parameters with a higher correlation with the fracturing effect (that is, production) for evaluation, which can not only greatly reduce the difficulty of evaluation, but also improve the accuracy of evaluation and provide better guidance for actual production.

[0015] Before screening the evaluation parameters, it is necessary to preprocess them because the data range and data integrity of each parameter are different.

[0016] When preprocessing data, considering that there are some qualitative parameters in the evaluation parameters, such as the type of fracturing fluid (such as water-based fracturing fluid, oil-based fracturing fluid, foam fracturing fluid, etc.), positive assignment can be performed according to the fracturing effect of the fracturing fluid type. Therefore, before actual use, we need to assign quantitative values ​​to it in order to convert it into data for subsequent analysis and processing. Among them, the target coding method is often used to assign values ​​according to the strength of the relationship between the parameter and the target variable.

[0017] After the qualitative parameters are quantitatively converted, all evaluation parameters are numerical parameters. However, considering that the collected data may be missing or wrong, we also need to process data outliers and fill in missing values. In this embodiment, the segmented regression interpolation method is used to fit the univariate cubic regression curve with the collected data as the fitting data. When the difference between the true value and the fitting value is greater than a certain threshold, the true value is regarded as an outlier and removed to generate a new missing value. Then, the missing value is replaced by the fitting value.

[0018] After the evaluation parameters are processed and completed, considering the different sizes of each group of data, in order to facilitate subsequent calculations, we also need to normalize these data so that all variables are compressed and mapped to the range of [0,1]. Normalization is a conventional processing method in this field, so its specific operation will not be described in detail here.

[0019] After normalizing the evaluation parameters, in order to facilitate subsequent calculations, it is also necessary to match each set of evaluation parameters with the output to ultimately form a data matrix.

[0020]

[0021] In the formula, X nm represents the mth normalized evaluation parameter corresponding to the nth well, Y n Represents the production of the nth well.

[0022] S2. Based on the CatBoost model and cross-validation theory, the evaluation parameters are preliminarily screened to obtain candidate parameters; In S1, there are 35 evaluation parameters in total. It is known to those skilled in the art that, although the more evaluation parameters there are, the better the final evaluation effect will be, too many evaluation parameters will lead to a significant increase in the amount of calculation and the time of calculation. At the same time, in this field, some technologies use a small number of evaluation parameters based on the specific conditions of a certain block and their actual development experience, which can reduce the amount of calculation and the time of calculation, and can also maintain a certain accuracy of prediction and evaluation. For example, Chinese patent CN117627608A, Chinese patent CN106203847A, etc. However, the source of its evaluation parameters is usually obtained based on experience, which is not only not scientific and objective, but also may have large differences in the application effects of other blocks.

[0023] To this end, in this embodiment, the evaluation parameters are preliminarily screened through the CatBoost model to obtain candidate parameters.

[0024] S21. Based on the cross-validation theory, the hyperparameters of the CatBoost model are selected to obtain the best model, the evaluation parameters are applied to the best model for model training, and the information gain of each evaluation parameter is calculated; In this step, the hyperparameters of the CatBoost model are first optimized using cross-validation theory. The hyperparameters of the CatBoost model include the number of trees, learning rate, maximum depth, etc., which can all be optimized using cross-validation theory. When optimizing, evaluation parameters need to be brought in. When optimizing hyperparameters, the average error is used as the judgment criterion: the hyperparameter group with the smallest average error is usually selected and substituted into the CatBoost model as the best model. Cross-validation theory is a conventional operation in this field, and its specific operation will not be repeated here.

[0025] After obtaining the best model, the data matrix obtained in S1 is brought into the best model for iteration, and new decision trees are continuously added based on the current prediction, so that the prediction error is gradually reduced. The model output F(x) of each round is calculated by the model output F(x) of the previous round. t-1 (x) and the newly added tree h t (x) is updated, the formula is:

[0026] Among them: F t-1 (x) is the model output of the previous round; h t (x) is the The output of the round decision tree; η is the learning rate, which is used to control the influence of each tree on the final prediction result.

[0027] During the construction of each decision tree, CatBoost calculates the information gain of each feature, which is the improvement in prediction when the current feature is split. Since the target variable in this example is output and no intervals are divided, information gain is usually measured based on the variance of the target variable. Information gain measures the reduction in the variance of the target variable after the feature is split. The calculation formula for information gain is as follows:

[0028] Among them: IG(X u ) is the feature X u After splitting the data set, the degree to which the entropy (uncertainty) of the output Y as the target variable is reduced is the information gain; u is the u-th feature in the data matrix X, indicating the feature considered in the decision tree split; Var(Y) is the variance of the target variable Y in the data matrix X, indicating the uncertainty or dispersion in the data set; Indicates that by feature X u The number of subsets after division; |N z| is the number of samples in the subset N; |X| is the total number of samples in the feature matrix X; Var(N z ) is a subset N i The variance of the target variable Y in .

[0029] S22, sorting the evaluation parameters based on information gain; In S21, the information gain value of each evaluation parameter is calculated, and the information gain value is used as an indicator to measure the importance of the parameter. Then we sort them according to their size to facilitate subsequent selection. In this embodiment, the final sorting is shown in Table 2.

[0030] Table 2 The importance ranking of the best model parameters

[0031] S23. Combined with the cross-validation theory, determine the number N1 of better parameters, and then use the top N1 evaluation parameters in S22 as candidate parameters.

[0032] After the number of optimal parameters is determined through cross-validation theory, the top N1 evaluation parameters are selected as candidate parameters to facilitate subsequent factor analysis. Figure 1 As shown in Figure 2, the optimal number of parameters is determined to be 25 through cross-validation theory. Therefore, the top 25 evaluation parameters in Table 2 are selected as candidate parameters.

[0033] S3. Based on the factor analysis method, the number of factors involved in the analysis is obtained through the variance contribution rate of each factor and the cumulative variance contribution rate threshold. The comprehensive score calculation formula in the factor analysis is used to obtain the comprehensive scores of different wells, and the post-pressure effect is evaluated based on the comprehensive score.

[0034] The factor analysis method used in this embodiment mainly includes the following steps: S31, performing correlation test on candidate parameters; In this step, the main purpose of conducting correlation tests on the initially screened parameters is to ensure that the data are suitable for factor analysis and to verify whether there are structural associations between variables that are suitable for extracting common factors. Factor correlation tests mainly include correlation matrix tests, KMO tests, and Bartlett sphericity tests.

[0035] In this example, the Pearson correlation coefficients in the constructed correlation matrix are mostly greater than 0.3, indicating that there is a significant correlation between the variables and the data can be used for factor analysis; at the same time, the calculated value of the KMO test is 0.6841, which is greater than or equal to 0.6, indicating that the data is suitable for factor analysis; the p-value in the Bartlett sphericity test is 0.4715, which is less than 0.5, also indicating that the data is suitable for factor analysis. Passing the three tests is a prerequisite for factor analysis. In this example, all three tests are passed, indicating that the data is very suitable for analysis by factor analysis.

[0036] S32, perform factor extraction and factor rotation; (1) Factor extraction

[0037] In factor analysis, the most commonly used method is principal component analysis. When selecting the number of factors to be extracted, the following criteria are usually used: ① Kaiser criterion: select factors with eigenvalues ​​greater than 1; ② Variance contribution rate: select factors that can explain more than 90% of the cumulative variance; At the same time, in the actual analysis process, if too few factors are screened, the explanatory power of the original data will be poor. Therefore, we usually need to use variance contribution rate as the main screening criterion.

[0038] According to the above two criteria, in this example, the cumulative variance explanation rate of the first 10 factors is greater than 90%, so the number of comprehensive determination factors is 10, and the total variance explanation table is shown in Table 3.

[0039] Table 3 Total variance explained

[0040] (2) Factor rotation The goal of factor analysis is to convert the original multiple variables into a few factors through linear combination. The purpose of rotation is to make each factor have a higher load with as few original variables as possible, so that it is easier to interpret. As for the load, it is obtained through the factor extraction process. Each factor load represents the linear relationship or correlation between the original variable and the extracted factor. The method of obtaining it is common knowledge in this field. During the rotation process, the factor loading matrix will change, but the overall explanatory structure and fit (such as variance explanation) of the factor analysis model remain unchanged. Common rotation methods are: ① Orthogonal rotation: assuming that the factors are independent of each other; ② Oblique rotation: allowing the factors to be correlated.

[0041] ① Orthogonal rotation: The goal of orthogonal rotation is to make the factors after rotation remain independent of each other, that is, there is no correlation between the factors after rotation. Therefore, the rotation matrix Q of orthogonal rotation must be an orthogonal matrix, that is

[0042] in, is the transposed matrix of matrix Q; I is the identity matrix.

[0043] The factor loading matrix after rotation for:

[0044] Among them, Λ is the original factor loading matrix; Q is the orthogonal rotation matrix.

[0045] A common orthogonal rotation method is Varimax rotation, whose goal is to maximize the variance of factor loadings so that the loading of each factor is as large as possible on some variables and as small as possible on other variables. The objective function of Varimax rotation is:

[0046] Among them, λ ij is the element of the original factor loading matrix, which represents the loading of the i-th factor on the j-th variable. By optimizing this objective function, the rotation matrix Q is adjusted; is to determine the number of factors to be finally calculated; M is the number of candidate parameters finally determined.

[0047] ② Oblique rotation: The difference between oblique rotation and orthogonal rotation is that oblique rotation allows correlation between factors. Therefore, the rotation matrix Q in the oblique rotation is no longer an orthogonal matrix, but allows correlation between factors. This means that the factor loading matrix after rotation may be correlated.

[0048] In the case of skew rotation, the rotation matrix Q is no longer an orthogonal matrix, so it satisfies the following conditions:

[0049] The factor loading matrix after rotation is still:

[0050] The goal of the oblique rotation is to make the structure of the factor loading matrix more concise while allowing correlation between factors. A common oblique rotation method is the Oblimin rotation, which is optimized based on the following objective function:

[0051] in, is the correlation between factors, that is, the covariance between factors; α is the parameter that controls the correlation between factors. Rotation helps us reduce the correlation between factors, making each factor more independent, thereby enhancing the explanatory power of the model; is to determine the number of factors to be finally calculated; M is the number of candidate parameters finally determined.

[0052] In this example, the cumulative variance explanation rate of the first 10 factors in the rotated factor variance table is 84.625%. Extracting 10 factors can better meet the needs of factor analysis.

[0053] S33. Calculate the score of each factor based on the variance explanation rate (1) Load factor calculation

[0054] The rotated factors need to be interpreted based on their loads. Each factor represents a potential structure in the data. The specific characteristics reflected by each factor can be judged by combining the size of the factor load. In this example, the load coefficients of factor 1 with the total liquid volume and the amount of quartz sand are 0.95 and 0.93, respectively, indicating that factor 1 has a high load with the total liquid volume and the amount of quartz sand. The remaining factors can also be represented by the rotated factor load coefficients. The rotated factor load coefficient table is shown in Table 4.

[0055] Table 4 Load factor coefficient table after rotation

[0056] (2) Calculation of comprehensive score

[0057] By extracting the number of factors in step S31 and calculating the proportion of the variance explanation rate of each factor in the cumulative variance explanation rate, the score of each factor can be determined. The variance explanation rate ratio can be obtained from Table 3 or from the variance explanation rate ratio formula. The variance explanation rate ratio formula is as follows:

[0058] Among them, λ i is the variance contribution rate of the i-th factor; is the cumulative variance contribution rate of v factors.

[0059] At the same time, the component matrix table can reflect the contribution of each original parameter to different factors, that is, each factor can be expressed by different parameters. The relationship between each factor and parameter can be clarified through the component matrix table, and then the linear expression of each factor can be obtained. The calculation formula of the principal component of the factor is as follows:

[0060] Among them, R i Calculate the principal component value of the i-th factor; a ij is the factor loading of the i-th factor on the j-th variable; X j is the standardized value of the j-th variable.

[0061] The specific values ​​in the above factor calculation formula need to be obtained from the component matrix table, as shown in Figure 4.

[0062] The comprehensive score calculation formula is obtained based on the proportion of the variance explanation rate of each factor after rotation to the cumulative variance explanation rate. The comprehensive score calculation formula of different wells reflects the fracturing construction conditions of different wells. The comprehensive score calculation formula is as follows:

[0063] in, To determine the number of factors for the final calculation; For the The proportion of variance explained by each factor; is the number of holes in the nth well. The principal component calculation value of the factor; S n is the comprehensive score of the nth well.

[0064] The comprehensive score is regarded as a comprehensive evaluation index reflecting the post-fracturing effect, and is compared with the production of FSI production profile logging parameters to verify whether the method can effectively reflect the production and effect after fracturing, and then to test whether it can be compared with the comprehensive score of efficient factor analysis of oil and gas resources to verify that the factor analysis of the preferred parameters provides fracturing parameter optimization and helps sustainable exploitation of resources compared with the factor analysis of non-preferred parameters; at the same time, the results of the present invention are also compared with the direct effect.

[0065] In this example, the results of the embodiment of the present invention are compared with the comprehensive scores of direct factor analysis and also compared with the FSI output, as shown in Table 5 and Figure 2 shown.

[0066] Table 5 Statistical table of comprehensive scores of the embodiment of the present invention and comprehensive scores of direct factor analysis

[0067] The results show that the factor analysis method for parameter optimization can better reflect the change trend of the actual production, and further verify the effectiveness and reliability of the selected parameters. Therefore, the invention constructed by the present invention can effectively guide the construction site to carry out fracturing operations and sustainable development, and realize the intelligent evaluation of the post-fracturing effect.

[0068] In this embodiment, CatBoost is used to extract the main controlling factors of production. In essence, it captures the nonlinear characteristic relationship between factors through a tree model based on gradient boosting, and uses feature importance analysis to identify key variables. Since the production of oil and gas resources is affected by multiple complex parameters such as fracturing, perforation, and logging, the CatBoost model has good high-dimensional feature processing capabilities, can maintain robustness in the case of non-balanced and missing data, and determines the importance value of the main controlling factors through information gain.

[0069] Furthermore, through cross-validation (determine the number of main control factors), the optimal feature combination that can maximize the information expression ability and does not introduce redundancy is determined from the perspective of generalization ability. This can avoid overfitting or feature redundancy and improve the stability of the model.

[0070] On this basis, factor analysis is applied to reduce the dimensionality of the main controlling factors, mainly using the collinearity characteristics between variables to extract potential common factors to reduce dimensions, remove noise, and retain comprehensive variables with large amounts of information. The factor scores can be used as compressed new features to input into the intelligent evaluation system, so that the model can maintain high explanatory power and prediction accuracy while reducing complexity. Ultimately, this method can more accurately characterize the key influencing factors of oil and gas resource production and improve the accuracy and stability of intelligent evaluation.

[0071] The above description is only a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of the technical solution of the present invention.

Claims

1. A method for intelligent evaluation of post-pressing effect based on factor analysis, characterized in that: The following steps are involved: S1. Obtaining production data and parameters affecting production of fractured wells in the same block, and preprocessing them to obtain evaluation parameters; S2. Based on the CatBoost model and cross-validation theory, the evaluation parameters are preliminarily screened to obtain candidate parameters; S3. Based on the factor analysis method, the number of factors involved in the analysis is obtained through the variance contribution rate of each factor and the cumulative variance contribution rate threshold. The comprehensive score calculation formula in the factor analysis is used to obtain the comprehensive scores of different wells, and the post-pressure effect is evaluated based on the comprehensive score.

2. The method according to claim 1, characterized in that The parameters affecting production include fracturing parameters, perforation parameters, geological parameters, drainage parameters and logging parameters.

3. The method according to claim 2, characterized in that The fracturing parameters include total sand volume, total liquid volume, fracturing fluid type, fracturing stage, total cluster number, ball feeding fluid dosage, acid liquid dosage, pre-fluid volume, pre-fluid ratio, sand-carrying fluid dosage, displacement fluid dosage, average sand ratio, quartz sand volume, quartz sand ratio, expanded clay volume, expanded clay ratio, average construction pressure, fracture pressure, average pump stop pressure, and closing pressure; the perforation parameters include average perforation length, average segment length, average segment spacing, perforation concentration, perforation density, and the product of perforation concentration and perforation density; the geological parameters include abundance, gas content, and oil layer thickness; the drainage parameters include: liquid injection intensity, sand addition intensity, construction displacement, and average cluster displacement; the logging parameters include low gamma horizontal section length and low gamma horizontal section length ratio.

4. The method according to claim 1, characterized in that In S1, the following methods are used for preprocessing: qualitative data are converted into quantitative data, outliers and missing values ​​are processed, data are standardized, and a data matrix is ​​established.

5. The method according to claim 1, characterized in that: S2 includes the following sub-steps: S21. Based on the cross-validation theory, the hyperparameters of the CatBoost model are selected to obtain the best model, the evaluation parameters are applied to the best model for model training, and the information gain of each evaluation parameter is calculated; S22, sorting the evaluation parameters based on information gain; S23. Combined with the cross-validation theory, determine the number N1 of better parameters, and then use the top N1 evaluation parameters in S22 as candidate parameters.

6. The method according to claim 1, characterized in that S3 includes the following sub-steps: S31, performing correlation test on candidate parameters; S32, extract factors based on principal component analysis and obtain principal component calculation values, and then perform factor rotation; S33. Calculate the score of each factor based on the variance explanation rate; S34. Combine the principal component calculation value of each factor with the score to calculate the comprehensive score of the factor. The higher the comprehensive score, the better the post-compression effect.

7. The method according to claim 6, characterized in that In S31, the correlation test method is at least one of a correlation matrix test, a KMO test, and a Bartlett sphericity test.

8. The method according to claim 6, characterized in that In S32, the factor rotation includes orthogonal rotation and oblique rotation.

Images