Tight sandstone reservoir fluid identification method and device based on machine learning

Abstract

The invention discloses a tight sandstone reservoir fluid identification method and device based on machine learning. The method comprises the steps that label assignment is conducted on logging data of a gas layer, a gas-water same layer, a differential gas layer, a gas-containing water layer, a water layer and a dry layer; an advanced feature LT is constructed, pseudo tag type information of the advanced feature LT is stored, the feature is discretized through one-hot coding, N states are coded through an N-bit state register, each state has an independent register bit, and only one bit in one-hot coding takes effect; analyzing the correlation between the well logging curve data of the selected gas-water identification sensitive characteristics; an XGBoost model is established, and a gradient boosting decision tree is used as a basic learner. Compared with a neural network model for reservoir identification, the method has the advantages that the model can be selected in a more targeted manner, the method can be applicable to small and medium data sets, and the number of samples of various categories is balanced by performing data enhancement on the logging data and supplementing unbalanced logging data samples.

Description

Technical Field

[0001] The present invention belongs to the technical field of petroleum geophysical exploration, and in particular relates to a method and device for identifying fluid in a tight sandstone reservoir based on machine learning. Background Art

[0002] In recent years, machine learning has been widely used in geological modeling and reservoir description and has achieved better results than traditional statistical methods. For example, support vector machine (SVM), decision-making tree (DT), artificial neural network (ANN), etc. can be used for geological parameter estimation, stratigraphic boundary determination and lithology identification. XGBoost is a thriving branch of this field. It can extract complex features from simple data through learning and training of large data sets, and establish high-dimensional mapping relationships through layer-by-layer abstraction. It can be used to solve complex nonlinear geological problems.

[0003] The patent application number is CN 114002744 A, and the name is a patent application for a method for identifying fluids in tight sandstone reservoirs based on deep learning. It provides a method for identifying fluids in tight sandstone reservoirs based on deep learning. First, the features of each point in the layer of the well logging data in the study area are preprocessed and spliced ​​into samples; the samples with different depths in the layer are subjected to depth unification processing; the samples are input into the convolutional neural network to extract high-dimensional nonlinear well logging features; the bidirectional long short-term memory neural network is used to further fuse the feature mapping output by the convolutional neural network, and the multi-scale features of the well logging data are further learned; the tight sandstone reservoir fluid identification model outputs the identification results. Although this patent application can improve the accuracy of reservoir identification, the model method of this patent application is only applicable to the case of large data sets, and the interpretability is relatively poor. It is limited by the problem of unbalanced and insufficient samples in the well logging data set, and the trained model is difficult to meet the accuracy requirements of well logging interpretation and reservoir characterization. Summary of the invention

[0004] In order to overcome the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a method and device for fluid identification in tight sandstone reservoirs based on machine learning, propose a new sample construction method in terms of data enhancement, and improve the algorithm in combination with actual logging experience, which can effectively improve the recognition accuracy.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for identifying fluid in tight sandstone reservoirs based on machine learning comprises the following steps:

[0007] S1: According to the priority order of gas test results, logging results, and logging interpretation, labels are assigned to the six types of logging data, namely, gas layer, gas-water layer, poor gas layer, gas-water layer, water layer, and dry layer; if there is a gas test result, the gas test result is used as the final label value; if there is no gas test result, the logging information is used as the final label value; if there is no gas test result and no logging information, the logging interpretation is used as the final label value;

[0008] S2: Construct a high-level feature LT, save its pseudo-label type information, discretize the feature using one-hot encoding, and use an N-bit state register to encode N states. Each state has an independent register bit, and only one bit in a one-hot encoding is effective.

[0009] S3: Analyze the correlation between the selected well logging curve data of gas and water identification sensitive features;

[0010] S4: The well logging curve data of gas-water identification sensitive features are divided into training set and test set, and the XGBoost model is established, using the gradient boosting decision tree as the basic learner;

[0011] S5: Output the prediction results of tight sandstone reservoir fluid.

[0012] Optionally, in step S1, the screened gas-water identification sensitive characteristic data include photoelectric index curve PE, natural potential curve SP, aperture logging curve CAL, natural gamma curve GR, uranium curve U, thorium curve TH, potassium curve K, acoustic time difference curve AC, compensated neutron curve CNL, compensated density curve DEN, shallow lateral resistivity curve RLLS and deep lateral resistivity curve RLLD.

[0013] Optionally, in step S1, one data point is taken for every 0.125 m of logging data, and the middle 20% of the data is taken for the well section with less than 10 points per layer; the middle 40% of the data is taken for the well section with more than 10 points per layer and less than 20 points per layer; and the middle 60% of the data is taken for the well section with more than 20 points per layer.

[0014] Optionally, before step S1, reservoir and gas-water characteristic analysis and logging response research are carried out to screen gas-water identification sensitive characteristic data.

[0015] Optionally, before step S1, the input logging data is preprocessed using a z-score standardization method.

[0016] Optionally, in step S3, the Pearson correlation coefficient is used to analyze the linear correlation between the features.

[0017] Optionally, in step S3, the Kendall rank correlation coefficient and the Spearman rank correlation coefficient of the Copula function are used to quantitatively describe the correlation between the logging curves.

[0018] A device for a tight sandstone reservoir fluid identification method based on machine learning, comprising:

[0019] The first processing module is used to assign labels to the six types of well logging data, namely, gas layer, gas-water layer, poor gas layer, gas-water layer, water layer and dry layer, according to the priority order of gas test results, logging results and logging interpretation; if there is a gas test result, the gas test result is used as the final label value; if there is no gas test result, the logging information is used as the final label value; if there is no gas test result and no logging information, the logging interpretation is used as the final label value;

[0020] The second processing module is used to construct a high-level feature LT, save its pseudo-label type information, discretize the feature using one-hot encoding, and encode N states using an N-bit state register. Each state has an independent register bit, and only one bit in a one-hot encoding is effective.

[0021] The third processing module is used to analyze the correlation between the selected well logging curve data of the gas-water identification sensitive features;

[0022] The fourth processing module is used to divide the well logging curve data of gas-water identification sensitive features into a training set and a test set, establish an XGBoost model, and use a gradient boosting decision tree as a basic learner;

[0023] The fifth processing module is used to output the prediction result of the tight sandstone reservoir fluid.

[0024] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of a method for identifying fluids in tight sandstone reservoirs based on machine learning when executing the computer program.

[0025] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for identifying fluid in a tight sandstone reservoir based on machine learning.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] Compared with the use of neural network models to identify reservoirs, the present invention can select models more specifically and can be applicable to the situation of small and medium-sized data sets. At the same time, the application uses an integrated model to reduce time and space costs and obtain prediction results more quickly. The calculation method of the present application performs data enhancement on well logging data, supplements unbalanced well logging data samples, and balances the number of samples of each category. Due to the use of a tree model, there is no need to perform strict missing value processing, data type conversion, data scaling normalization, and complex data processing work such as initialization learning rate parameter adjustment on the data. And the xgboost model used in the present application has a ranking of feature importance, which can be directly analyzed according to the results.

[0028] Furthermore, the present application adopts the Kendall rank correlation coefficient and the Spearman rank correlation coefficient of the Copula function to quantitatively describe the degree of correlation between logging curves, which can better describe the nonlinear, non-normal, non-monotonic or heteroscedastic correlation between logging curves, thereby providing more information and basis for the analysis and processing of logging data.

[0029] Furthermore, the present application samples data by taking the middle 20% of the data from the well sections with less than 10 points per layer; taking the middle 40% of the data from the well sections with more than 10 points per layer and less than 20% of the data; and taking the middle 60% of the data from the well sections with more than 20 points per layer. This can avoid data redundancy and overfitting caused by too dense sampling, thereby obtaining an accurate and generalizable model. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings described herein are for illustration purposes only and are not intended to limit the scope of the present disclosure in any way.

[0031] In the attached picture:

[0032] Figure 1 This is a flow chart of the overall calculation idea of ​​the present invention;

[0033] Figure 2 This is the XGBoost fluid identification flow chart of the present invention. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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 should fall within the scope of protection of the present invention.

[0035] It should be understood by those skilled in the art that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0036] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0037] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0038] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0040] It should be noted that in the claims, any reference signs placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The present application may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third etc. does not indicate any order. These words may be interpreted as names.

[0041] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0042] The present invention is described in detail below with reference to the accompanying drawings.

[0043] like Figure 1 As shown, a method for identifying fluid in a tight sandstone reservoir based on machine learning of the present invention comprises the following steps:

[0044] A logging database for the study area was constructed, and the logging data was preprocessed to select appropriate features and determine the feature data. In order to meet the fit between model recognition and actual logging experience, each layer of logging data to be input into the model was adjusted to an interval that fits the actual experience; a data point was taken every 0.125m in the logging data, and the thickness of each layer was different, resulting in a different number of points in each layer. The interval for each layer was set as follows: the middle 20% of the data was taken for each layer with less than 10 points; the middle 40% of the data was taken for each layer with more than 10 and less than 20 points; the middle 60% of the data was taken for each layer with more than 20 points.

[0045] During the logging process, the logging tool will be lowered and withdrawn to perform continuous measurements in the wellbore. The logging tool will record the physical parameter values ​​corresponding to each depth (logging depth) to obtain raw data. When using machine learning for fluid identification in tight sandstone reservoirs, sampling that is too dense or too sparse may have some consequences and challenges on the results. Too much dense sampling may lead to data redundancy and overfitting. If the spacing between data points is very small, the model may rely too much on these points and fail to generalize, resulting in poor performance on new data. Excessive sampling will increase the amount of data and computational complexity, which may lead to an increase in model training time and computing resources. Sampling that is too sparse may lead to information loss and model inaccuracy. If the spacing between data points is too large, the model may not be able to capture subtle changes and features, making it impossible to accurately identify fluids. Lack of sufficient sampling may lead to insufficient model training and failure to capture the overall characteristics of the data, thus affecting the performance and generalization ability of the model. In summary, the data can be sampled by taking the middle 20% of the data from the sections with less than 10 points per layer; taking the middle 40% of the data from the sections with more than 10 points per layer and less than 20 points per layer; and taking the middle 60% of the data from the sections with more than 20 points per layer. This method can obtain an accurate and generalizable model.

[0046] Based on the comprehensive collection of relevant basic data by professionals in this field, reservoir and gas-water characteristic analysis and logging response research were carried out to screen the features sensitive to gas-water identification. Finally, the 12 features shown in Table 1 below were selected.

[0047] Table 1

[0048]

[0049]

[0050] Due to the complexity and changeability of the actual underground geological conditions, there are often large magnitude differences between underground formation logging data. In order to reduce the reconstruction error caused by the magnitude difference of the data, the z-score standardization method is used to preprocess the input logging data; its specific expression is:

[0051]

[0052] Among them, x * is the standardized data, x i is the original input data, x i ′ is the mean value of the input data, and s is the standard deviation of the input data.

[0053] In the well logging database of the study area, the number of well logging data samples in the six categories of gas layer, gas-water layer, poor gas layer, gas-water layer, water layer and dry layer is not balanced. Unbalanced samples will cause the training model to focus on the category with a large number of samples, and make wrong judgments on the category with a small number of samples, which will lead to the lack of generalization ability of the model. First, the well logging data is enhanced, and a method of pseudo-labeling all samples in the well logging database is proposed. The pseudo-label assignment rule is as follows:

[0054] According to the priority order of gas test results, logging results and logging interpretation, labels are assigned to six types of logging data, namely gas layer, gas-water layer, poor gas layer, gas-water layer, water layer and dry layer; if there is a gas test result, the gas test result is used as the final label value; if there is no gas test result, the logging information is used as the final label value; if there is no gas test result and no logging information, the logging interpretation is used as the final label value.

[0055] Construct a high-level feature LT, save its pseudo-label type information, discretize the feature with one-hot encoding, and use N-bit state registers to encode N states. Each state has an independent register bit, and only one bit in a one-hot encoding is effective. The new samples obtained through the above steps supplement the unbalanced samples to balance the number of samples in each category.

[0056] The various logging curves in the logging data are not independent of each other, and there is a certain degree of interdependence. In order to analyze the correlation between the features and find out the internal connection, the Pearson correlation coefficient is used to analyze the linear correlation between the features. The calculation formula is:

[0057]

[0058] In the formula, mi and ni are the data of two different variables, t is the number of data, and (m i ,n i ),i=1,2,3…,t.

[0059] There is not only linear correlation between logging data, but also a large nonlinear relationship. The Kendall rank correlation coefficient τ and Spearman rank correlation coefficient ρ of the Copula function are used to quantitatively describe the correlation between logging curves, and compared with the Pearson correlation coefficient, which can better describe the nonlinear, non-normal, non-monotonic or heteroscedastic correlation between logging curves, thereby providing more information and basis for the analysis and processing of logging data. By comparing the correlation coefficients of various methods, it is found that there is a nonlinear relationship between fluid characteristics. If there is a nonlinear relationship between two variables, then using the Pearson correlation coefficient to describe the correlation between them may be biased. In this case, using the Spearman rank correlation coefficient or the Kendall rank correlation coefficient can better describe the monotonic relationship between them. When the conclusion that there is a nonlinear relationship between fluid characteristics is obtained, the integrated model xgboost can be used for identification and subsequent feature analysis, because xgboost performs well in processing nonlinear data.

[0060] The new category feature constructed in step S2 can be regarded as the classification result of a classifier. Instead of using this result directly, the feature used to identify this result is used instead. Figure 2 As shown in the figure, the well logging curve data of gas-water identification sensitive features are divided into training set and test set, and the XGBoost model is established, using the gradient boosting decision tree as the basic learner. XGBoost is used for fluid identification; the important parameters of XGBoost are configured as follows: gamma is 0.1; max_depth is 8; subsample is 0.7; colsample_bytree is 0.5; min_child_weight is 3;

[0061] The steps of XGBoost fluid identification are as follows:

[0062] (1) Input Features: The input of the model is a set of feature vectors, which represent the feature values ​​of the observed samples.

[0063] (2) Data preprocessing: XGBoost preprocesses and transforms input features, including encoding categorical features and processing missing values.

[0064] (3) Correlation analysis: Pearson’s correlation coefficient is used to analyze the linear correlation between features.

[0065] (4) Dataset partitioning: Divide the logging data into training set and test set.

[0066] (5) Building the XGBoost Classifier: XGBoost uses the Gradient Boosting Decision Tree as the basic learner. The decision tree makes predictions by learning the relationship between sample features and labels.

[0067] (6) Output: The output of the model is the predicted categories, which are gas layer, gas-water layer, poor gas layer, gas-water layer, water layer and dry layer.

[0068] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. It should be understood that the described embodiments are only part of the embodiments of the present invention, not all of the embodiments, and therefore should not be regarded as limiting the scope of protection. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technical personnel in this field without creative work are within the scope of protection of the present invention.

[0069] Example 1

[0070] As shown in Table 2 below, take the identification of No. 38, Li 25 Well as an example.

[0071] Step 1: Build a well logging database for the study area and preprocess the well logging data; select appropriate features, determine feature data, and select PE, SP, CAL, GR, U, TH, K, AC, CNL, DEN, RLLS, and RLLD as features; obtain 84 well logging data;

[0072] Step 2: Due to the problem of sample imbalance, data enhancement is performed on the unbalanced samples to balance the number of samples; in the original data, the ratio of gas layer, gas-water layer, poor gas layer, gas-water layer, water layer, and dry layer is 1.2:0.6:0.5:1:0.8:1.4, which shows a significant imbalance. After supplementation, the ratio is close to 1:1:1:1:1:1, so that the samples are balanced.

[0073] Step 3: Analyze the obtained feature data and the correlation between the features; obtain their linear and nonlinear correlation.

[0074] Step 4: Based on the feature analysis results, the XGBoost model was used to model the tight sandstone reservoir logging data and output the identification results. The identification accuracy of each point in each layer reached 80.4%, and the identification accuracy of each layer reached 84.3%. Li 25 Well No. 38 was also identified as a gas layer by the model, which was verified.

[0075] Table 2

[0076]

[0077]

[0078] Unless otherwise specified, the working modes or control modes involved in the above embodiments are all conventional working modes or control modes in the art.

[0079] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in the field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for identifying fluid in tight sandstone reservoirs based on machine learning, characterized in that: The following steps are involved: S1: According to the priority order of gas test results, logging results, and logging interpretation, labels are assigned to the six types of logging data, namely, gas layer, gas-water layer, poor gas layer, gas-water layer, water layer, and dry layer; if there is a gas test result, the gas test result is used as the final label value; if there is no gas test result, the logging information is used as the final label value; if there is no gas test result and no logging information, the logging interpretation is used as the final label value; S2: Construct a high-level feature LT, save its pseudo-label type information, discretize the feature using one-hot encoding, and use an N-bit state register to encode N states. Each state has an independent register bit, and only one bit in a one-hot encoding is effective. S3: Analyze the correlation between the selected well logging curve data of gas and water identification sensitive features; S4: The well logging curve data of gas-water identification sensitive features are divided into training set and test set, and the XGBoost model is established, using the gradient boosting decision tree as the basic learner; S5: Output the prediction results of tight sandstone reservoir fluid.

2. The method for identifying fluid in tight sandstone reservoirs based on machine learning according to claim 1, characterized in that: In step S1, the screened gas-water identification sensitive characteristic data include photoelectric index curve PE, natural potential curve SP, aperture logging curve CAL, natural gamma curve GR, uranium curve U, thorium curve TH, potassium curve K, acoustic time difference curve AC, compensated neutron curve CNL, compensated density curve DEN, shallow lateral resistivity curve RLLS and deep lateral resistivity curve RLLD.

3. The method for identifying fluid in tight sandstone reservoirs based on machine learning according to claim 1, characterized in that: In step S1, a data point is taken every 0.125m of the logging data. For the well section with less than 10 points per layer, the middle 20% of the data is taken; for the well section with more than 10 points per layer and less than 20 points per layer, the middle 40% of the data is taken; and for the well section with more than 20 points per layer, the middle 60% of the data is taken.

4. The method for identifying fluid in tight sandstone reservoirs based on machine learning according to claim 1, characterized in that: Before step S1, reservoir and gas-water characteristic analysis and logging response research are carried out to screen sensitive characteristic data for gas-water identification.

5. The method for identifying fluid in tight sandstone reservoirs based on machine learning according to claim 1, characterized in that: Before step S1, the input logging data is preprocessed using the z-score standardization method.

6. The method for identifying fluid in tight sandstone reservoirs based on machine learning according to claim 1, characterized in that: In step S3, the linear correlation between the features is analyzed using the Pearson correlation coefficient.

7. The method for identifying fluid in tight sandstone reservoirs based on machine learning according to claim 1, characterized in that: In step S3, the Kendall rank correlation coefficient and the Spearman rank correlation coefficient of the Copula function are used to quantitatively describe the correlation between the logging curves.

8. A device for a tight sandstone reservoir fluid identification method based on machine learning, characterized in that: include: The first processing module is used to assign labels to the six types of well logging data, namely, gas layer, gas-water layer, poor gas layer, gas-water layer, water layer and dry layer, according to the priority order of gas test results, logging results and logging interpretation; if there is a gas test result, the gas test result is used as the final label value; if there is no gas test result, the logging information is used as the final label value; if there is no gas test result and no logging information, the logging interpretation is used as the final label value; The second processing module is used to construct a high-level feature LT, save its pseudo-label type information, discretize the feature using one-hot encoding, and encode N states using an N-bit state register. Each state has an independent register bit, and only one bit in a one-hot encoding is effective. The third processing module is used to analyze the correlation between the selected well logging curve data of the gas-water identification sensitive features; The fourth processing module is used to divide the well logging curve data of gas-water identification sensitive features into a training set and a test set, establish an XGBoost model, and use a gradient boosting decision tree as a basic learner; The fifth processing module is used to output the prediction result of the tight sandstone reservoir fluid.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of a method for identifying fluids in tight sandstone reservoirs based on machine learning as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of a method for identifying fluid in a tight sandstone reservoir based on machine learning as described in any one of claims 1 to 7.

Images