Comprehensive method for analyzing compressibility of shale reservoir based on machine learning

Abstract

The invention discloses a comprehensive method for analyzing compressibility of a shale reservoir based on machine learning, and relates to the technical field of oil and gas field development, the comprehensive method for analyzing compressibility of the shale reservoir based on machine learning mainly comprises the following steps: obtaining a training data set according to core experiment data and corresponding logging data; constructing a three-parameter prediction model by using a random forest regression model; training the three-parameter prediction model by using the training data set to obtain a trained three-parameter prediction model, and predicting the to-be-predicted logging data to obtain three parameters; and calculating a compressibility coefficient based on the three parameters so as to obtain compressibility classification evaluation of the shale reservoir in the research area. By implementing the comprehensive method for analyzing the compressibility of the shale reservoir based on machine learning provided by the invention, the compressibility of the shale reservoir can be efficiently, accurately and universally evaluated.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas field development, and more specifically, to a comprehensive method for analyzing the compressibility of shale reservoirs based on machine learning. Background Art

[0002] With the continuous growth of world oil and gas demand and the continuous decline of conventional oil and gas production, unconventional oil and gas with great resource potential has gradually become a strategic replacement resource in the oil field. Shale oil, as one of the important components of unconventional oil and gas, has the characteristics of poor reservoir properties, strong density, high clay mineral content, and great difficulty in development. These factors make shale oil difficult to exploit, with low recovery rate and little benefit. Therefore, it is particularly important to take corresponding transformation measures for shale reservoirs to achieve effective development of shale oil, and the evaluation of shale reservoir compressibility is an important technical component before shale reservoir transformation. Compressibility refers to the ability of the reservoir to be effectively fractured to form a complex fracture network. There are many studies on the evaluation of the compressibility of shale oil reservoirs, and various factors are considered. Among them, the brittleness index, Young's modulus, and Poisson's ratio are more common and easier to obtain parameters in compressibility research. The higher the brittleness index and Young's modulus, the lower the Poisson's ratio, and the stronger the compressibility.

[0003] The determination of compressibility parameters is the basis for compressibility evaluation. It is too time-consuming and expensive to obtain the corresponding parameters through experiments. At present, the use of machine learning and logging data provides new ideas for predicting compressibility evaluation parameters. However, most of them are single-parameter predictions, involving multiple parameters and the need to establish multiple different models, which is relatively cumbersome. Although the existing BP multi-output neural network can predict multiple parameters, the premise is that the independent variables (logging parameters) corresponding to each parameter to be predicted are the same from the same core sample. For the brittleness index, Young's modulus, and Poisson's ratio, the corresponding experiments are X-ray diffraction analysis technology (X-Ray Diffraction, XRD) experiments and rock triaxial stress experiments. In actual exploration, core sampling is mostly inconsistent, and traditional BP multi-output neural network models are difficult to adapt. Although deep learning can solve this problem, the model establishment process is cumbersome and the training time cost is high. Summary of the invention

[0004] The purpose of the present invention is to provide a comprehensive method for analyzing the compressibility of shale reservoirs based on machine learning, which can efficiently, accurately and universally evaluate the compressibility of shale reservoirs and a computer program product.

[0005] The present invention provides a comprehensive method for analyzing the compressibility of shale reservoirs based on machine learning, comprising the following steps: S1: acquiring core test data and corresponding well logging data; obtaining a training data set according to the core test data and the corresponding well logging data; S2: constructing a three-parameter prediction model using a random forest regression model; S3: training the three-parameter prediction model using the training data set to obtain a trained three-parameter prediction model; S4: predicting the well logging data to be predicted using the trained three-parameter prediction model to obtain a brittleness index prediction value, a Young's modulus prediction value and a Poisson's ratio prediction value; S5: calculating a compressibility coefficient according to the brittleness index prediction value, the Young's modulus prediction value and the Poisson's ratio prediction value; and obtaining a classification evaluation of the compressibility of the shale reservoir in the study area according to the compressibility coefficient.

[0006] Furthermore, the core experimental data include brittleness index data, Young's modulus data and Poisson's ratio data; the corresponding logging data include natural gamma data, resistance data, acoustic wave time difference data, density data and neutron data.

[0007] Furthermore, the training data set includes a first data set, a second data set and a third data set; the first data set includes brittleness index data and corresponding logging data; the second data set includes Young's modulus data and corresponding logging data; the third data set includes Poisson's ratio data and corresponding logging data.

[0008] Furthermore, the above brittleness index data are obtained by X-ray diffraction analysis experiments; Young's modulus data and Poisson's ratio data are obtained by rock triaxial stress experiments.

[0009] Furthermore, the above three-parameter prediction model includes a first prediction model, a second prediction model and a third prediction model.

[0010] Furthermore, the above-mentioned step S3 specifically includes: taking the brittleness index data in the first data set, the Young's modulus data in the second data set and the Poisson's ratio data in the third data set as dependent variables, and taking the corresponding logging data as independent variables, training and predicting the three-parameter prediction model to obtain the predicted brittleness index, predicted Young's modulus and predicted Poisson's ratio; obtaining the first determination coefficient based on the brittleness index data and the predicted brittleness index in the first data set; obtaining the second determination coefficient based on the Young's modulus data and the predicted Young's modulus in the second data set; obtaining the third determination coefficient based on the Poisson's ratio data and the predicted Poisson's ratio in the third data set; when any one of the first determination coefficient, the second determination coefficient and the third determination coefficient is less than the preset determination coefficient threshold, continuing training and prediction; when the first determination coefficient, the second determination coefficient and the third determination coefficient are all not less than the preset determination coefficient threshold, stopping training to obtain a trained three-parameter prediction model.

[0011] Furthermore, the above preset determination coefficient threshold is 0.8.

[0012] Furthermore, the above step S5 specifically includes: obtaining the compressibility coefficient according to the predicted value of the brittleness index, the predicted value of the Young's modulus and the predicted value of the Poisson's ratio, such as the formula: , in, is the compressibility coefficient, is the predicted value of brittleness index, is the predicted value of Young’s modulus, is the predicted value of Poisson’s ratio; The compressibility coefficient is normalized to obtain the compressibility classification evaluation of shale reservoirs in the study area.

[0013] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned comprehensive method for analyzing the compressibility of shale reservoirs based on machine learning.

[0014] The implementation of the comprehensive method for analyzing shale reservoir compressibility based on machine learning and the computer program product provided by the present invention has the following beneficial effects: The present invention uses the measured core data (brittleness index, Young's modulus, Poisson's ratio) as dependent variables Y, Y', Y'', and the corresponding logging data X, X', X'' as independent variables, and divides the training set and the test set. By introducing the random forest algorithm, a variety of different input parameters and a variety of different output parameter prediction models are established, which can handle complex data structures with multiple inputs and multiple outputs, avoiding the trouble of separately establishing multiple models in traditional methods, and the problem that different experimental data correspond to different logging parameters, and realizing simultaneous prediction and output of multiple parameters. In practical applications, in terms of ensuring the accuracy of model prediction, the efficiency is greatly improved compared with conventional machine learning prediction methods. Through the trained model, multi-parameter prediction from a single well to the study area can be achieved, that is, by integrating the brittleness index, Young's modulus, and Poisson's ratio, three parameters commonly used to evaluate reservoir compressibility, a comprehensive prediction can be made, which can more comprehensively evaluate the compressibility; On the basis of parameter completion, the compressibility coefficient is calculated using the relationship between the three parameters and compressibility, and three types of reservoirs (high compressibility reservoirs, medium compressibility reservoirs, and low compressibility reservoirs) are divided according to the data characteristics of the study area. When evaluating low-porosity and low-permeability reservoirs, one of the most effective methods is to use the idea of ​​grading and classification to conduct a comprehensive evaluation of the reservoir. Dividing the reservoir and effectively predicting it can save a lot of manpower, material resources, and financial resources and greatly improve exploration efficiency.

[0015] The present invention can adapt to complex coring experimental conditions and the problem that different experimental data correspond to different logging parameters, make full use of new and old data, avoid waste of resources, and is of great significance for re-examination of old wells and secondary exploration. The selection of parameters in the present invention should be based on actual needs in exploration and can be flexibly adjusted, providing reliable support for sweet spot evaluation in a variety of complex geological environments and having wide adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 is a flow chart of a comprehensive method for analyzing shale reservoir compressibility based on machine learning provided by the present invention; Figure 2 is a prediction error graph of the three-parameter model provided by the present invention; Figure 3 This is a diagram for comprehensive prediction of multiple parameters of a single well and classification evaluation of the compressibility of shale reservoirs provided by the present invention. DETAILED DESCRIPTION

[0017] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0018] Figure 1 A schematic diagram of a comprehensive method for analyzing the compressibility of shale reservoirs based on machine learning in this embodiment is shown. In this embodiment, the comprehensive method for analyzing the compressibility of shale reservoirs based on machine learning includes the following steps: S1: Obtain core test data and corresponding well logging data; obtain a training data set based on the core test data and the corresponding well logging data; For an exemplary embodiment, the core test data includes brittleness index data, Young's modulus data and Poisson's ratio data; the corresponding logging data includes natural gamma data, resistance data, acoustic wave time difference data, density data and neutron data; For an exemplary embodiment, the training data set includes a first data set, a second data set and a third data set; the first data set includes brittleness index data and corresponding well logging data; the second data set includes Young's modulus data and corresponding well logging data; the third data set includes Poisson's ratio data and corresponding well logging data; For an exemplary embodiment, the brittleness index data is obtained by X-ray diffraction analysis experiment; the Young's modulus data and Poisson's ratio data are obtained by rock triaxial stress experiment; Specifically, in step S1, core test data and logging data corresponding to each quality parameter are collected; wherein the core test data include three quality parameters: brittleness index, in %, obtained by X-ray diffraction analysis technology (XRD) experiment; Young's modulus, in GPa, obtained by rock triaxial stress experiment; Poisson's ratio, obtained by rock triaxial stress experiment; logging data include: natural gamma, referred to as GR, in API; electrical resistance, referred to as RT, in Ω·m; acoustic time difference, referred to as AC, in us / m; density, referred to as DEN, in g / cm 3 ; Neutron, referred to as CNL, unit is %; Sort the above data and check whether the data type parameters are missing. If a parameter is missing, remove all data of the same batch of the sample; if there is no missing, retain it, get the training data set, and proceed to the next step; S2: Use the random forest regression model to build a three-parameter prediction model; For an exemplary embodiment, the three-parameter prediction model includes a first prediction model, a second prediction model, and a third prediction model; Specifically, in step S2, a multi-model parallel construction method is used based on Python programming software to realize multi-target prediction of random forest. Specifically, an independent random forest model is constructed for each target variable. The three models have different independent variables and corresponding dependent variables, and their hyperparameters are adjusted according to the characteristics of each target. S3: training the three-parameter prediction model using the training data set to obtain a trained three-parameter prediction model; For an exemplary embodiment, step S3 specifically includes: taking the brittleness index data in the first data set, the Young's modulus data in the second data set and the Poisson's ratio data in the third data set as dependent variables, and taking the corresponding logging data as independent variables, training and predicting the three-parameter prediction model to obtain the predicted brittleness index, predicted Young's modulus and predicted Poisson's ratio; obtaining the first determination coefficient according to the brittleness index data and the predicted brittleness index in the first data set; obtaining the second determination coefficient according to the Young's modulus data and the predicted Young's modulus in the second data set; obtaining the third determination coefficient according to the Poisson's ratio data and the predicted Poisson's ratio in the third data set; when any one of the first determination coefficient, the second determination coefficient and the third determination coefficient is less than a preset determination coefficient threshold, continuing training and prediction; when the first determination coefficient, the second determination coefficient and the third determination coefficient are not less than the preset determination coefficient threshold, stopping training to obtain a trained three-parameter prediction model; For an exemplary embodiment, the preset determination coefficient threshold is 0.8; Specifically, in step S3, the three-parameter prediction model is trained based on the random forest algorithm, with the logging data as the independent variable and the experimental data as the dependent variable. The training process includes: data import, data partitioning, data normalization, model initialization, setting hyperparameters, model training and prediction; training the model multiple times and calculating the determination coefficient R 2 , until the prediction accuracy reaches a certain threshold, and a trained three-parameter prediction model is obtained; It should be noted that the trained three-parameter prediction model includes a trained first prediction model, a trained second prediction model and a trained third prediction model; Optionally, the above step S3 further includes: drawing a scatter point fitting graph of the training set and the test set predicted dependent variables and the true value; S4: using the trained three-parameter prediction model to predict the logging data to be predicted, and obtaining the predicted values ​​of brittleness index, Young's modulus and Poisson's ratio; S5: According to the predicted values ​​of brittleness index, Young's modulus and Poisson's ratio, the compressibility coefficient is obtained; according to the compressibility coefficient, the compressibility classification evaluation of shale reservoirs in the study area is obtained; For an exemplary embodiment, step S5 specifically includes: obtaining a compressibility coefficient according to a brittleness index prediction value, a Young's modulus prediction value, and a Poisson's ratio prediction value, such as the formula: , in, is the compressibility coefficient, is the predicted value of brittleness index, is the predicted value of Young’s modulus, is the predicted value of Poisson’s ratio; The compressibility coefficient was normalized to obtain the compressibility classification evaluation of shale reservoirs in the study area; It should be noted that because the brittleness index and Young's modulus are proportional to the compressibility of shale reservoirs, and Poisson's ratio is inversely proportional to the compressibility, the compressibility coefficient is calculated as (brittleness index*Young's modulus / Poisson's ratio); Table 1: Shale reservoir compressibility classification evaluation table

[0019] The above compressibility coefficients are normalized, and the compressibility classification evaluation of shale reservoirs in the study area is completed according to Table 1.

[0020] In an exemplary embodiment, the comprehensive method for analyzing the compressibility of shale reservoirs based on machine learning is implemented in the following manner. In this embodiment, the comprehensive method for analyzing the compressibility of shale reservoirs based on machine learning includes the following steps: Obtain core test data and corresponding well logging data; obtain a training data set based on the core test data and the corresponding well logging data; The three-parameter prediction model was established using Python programming software, including: (1) Import necessary libraries, including pandas, numpy, and related modules in sklearn; (2) Import data: read the Excel file from the specified path and store it in the data variable; (3) Define independent and dependent variables: Divide the data into three parts, each of which contains a set of independent variables (X) and a corresponding dependent variable (Y); (4) Use the train_test_split function to divide each group of independent variables and dependent variables into a training set and a test set, with the test set accounting for 20%, and then normalize them; (5) Initialize three RandomForestRegressor random forest regression models and set different hyperparameters, including the number of trees, maximum depth, minimum number of sample leaves, and minimum number of sample splits; Training model: Use the training data set to train each model in the three-parameter prediction model; Make predictions: Make predictions on the training set and the test set respectively to obtain the prediction results; Calculate and display R 2 Value: Calculate the R of each model on the training set and test set 2 value; Save prediction results: Create two DataFrames to store the true values ​​and predicted values ​​of the training set and test set respectively; It should be noted that in this embodiment, since the experimental data of Young's modulus and Poisson's ratio are too scarce to meet the basic quantity of model establishment, the Young's modulus and Poisson's ratio in well logging are used instead of experimental values ​​to establish the model, such as Figure 2 The following is a graph showing the prediction error of the three-parameter model; Figure 2 The results show that the model is more effective.

[0021] Based on the above prediction results, the classification and evaluation of the compressibility of shale reservoirs in the study area are obtained; Figure 3 Shown is a diagram of comprehensive prediction of multiple parameters for a single well and classification and evaluation of the compressibility of shale reservoirs.

[0022] This embodiment provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned comprehensive method for analyzing the compressibility of shale reservoirs based on machine learning.

[0023] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A comprehensive method for analyzing the compressibility of shale reservoirs based on machine learning, characterized in that: The following steps are involved: S1: Acquire core test data and corresponding well logging data; obtain a training data set according to the core test data and the corresponding well logging data; S2: Use the random forest regression model to build a three-parameter prediction model; S3: Using the training data set to train the three-parameter prediction model to obtain a trained three-parameter prediction model; S4: using the trained three-parameter prediction model to predict the well logging data to be predicted, to obtain a brittleness index prediction value, a Young's modulus prediction value, and a Poisson's ratio prediction value; S5: Calculate the compressibility coefficient according to the predicted value of the brittleness index, the predicted value of the Young's modulus and the predicted value of the Poisson's ratio; and obtain a classification evaluation of the compressibility of the shale reservoir in the study area according to the compressibility coefficient.

2. The comprehensive method for analyzing shale reservoir compressibility based on machine learning according to claim 1, characterized in that: The core experimental data include brittleness index data, Young's modulus data and Poisson's ratio data; the corresponding logging data include natural gamma data, resistance data, acoustic wave time difference data, density data and neutron data.

3. The comprehensive method for analyzing shale reservoir compressibility based on machine learning according to claim 1, characterized in that: The training data set includes a first data set, a second data set and a third data set; the first data set includes brittleness index data and corresponding well logging data; the second data set includes Young's modulus data and corresponding well logging data; the third data set includes Poisson's ratio data and corresponding well logging data.

4. The comprehensive method for analyzing shale reservoir compressibility based on machine learning according to any one of claims 2 and 3, characterized in that: The brittleness index data are obtained by X-ray diffraction analysis experiments; the Young's modulus data and the Poisson's ratio data are obtained by rock triaxial stress experiments.

5. The comprehensive method for analyzing shale reservoir compressibility based on machine learning according to claim 1, characterized in that: The three-parameter prediction model includes a first prediction model, a second prediction model and a third prediction model.

6. The comprehensive method for analyzing shale reservoir compressibility based on machine learning according to claim 3, characterized in that: Step S3 specifically includes: taking the brittleness index data in the first data set, the Young's modulus data in the second data set and the Poisson's ratio data in the third data set as dependent variables, and taking the corresponding logging data as independent variables, training and predicting the three-parameter prediction model to obtain the predicted brittleness index, predicted Young's modulus and predicted Poisson's ratio; obtaining a first determination coefficient based on the brittleness index data in the first data set and the predicted brittleness index; obtaining a second determination coefficient based on the Young's modulus data in the second data set and the predicted Young's modulus; obtaining a third determination coefficient based on the Poisson's ratio data in the third data set and the predicted Poisson's ratio; when any one of the first determination coefficient, the second determination coefficient and the third determination coefficient is less than a preset determination coefficient threshold, continuing training and prediction; when the first determination coefficient, the second determination coefficient and the third determination coefficient are all not less than the preset determination coefficient threshold, stopping training to obtain a trained three-parameter prediction model.

7. The comprehensive method for analyzing shale reservoir compressibility based on machine learning according to claim 6, characterized in that: The preset determination coefficient threshold is 0.

8.

8. The comprehensive method for analyzing shale reservoir compressibility based on machine learning according to claim 1, characterized in that: Step S5 specifically includes: obtaining a compressibility coefficient according to the predicted value of the brittleness index, the predicted value of the Young's modulus and the predicted value of the Poisson's ratio, such as the formula: ， in, is the compressibility coefficient, is the predicted value of brittleness index, is the predicted value of Young’s modulus, is the predicted value of Poisson’s ratio; The compressibility coefficient is normalized to obtain the compressibility classification evaluation of the shale reservoir in the study area.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the comprehensive method for analyzing the compressibility of shale reservoirs based on machine learning as described in any one of claims 1 to 8 are implemented.

Images