A rock drillability prediction method based on support vector machine regression model
Through the combination of support vector machine regression model and Gaussian kernel function, the accuracy and speed problems of rock drillability prediction are solved, and the rapid and accurate evaluation of rock drillability of oil and gas wells is achieved, reducing costs and supporting the entire well section predictions.
Patent Information
- Application Number
- CN202410091823.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-23
- Publication Date
- 2024-12-10
- Estimated Expiration
- 2044-01-23
AI Technical Summary
The prior art is difficult to accurately predict rock drillability, resulting in shortening of drill bit life, extended drilling cycle and high cost, and the inability to achieve rapid drillability assessment for the entire well section.
The support vector machine regression model is adopted to obtain the meticulous structure quantization index of rock particles, build and train the model, and use Gaussian kernel function, penalty coefficient and gamma coefficient optimization to achieve rapid prediction of the drillability of rocks in oil and gas wells.
It improves the accuracy and speed of rock drillability prediction, reduces costs, avoids model overfitting, supports rock drillability prediction of deep wells ultra-deep wells, and provides a direction of drilling while drilling prediction for the entire well section.
Smart Images
Figure CN117909933B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of rock mechanics in oil and gas drilling engineering, and in particular relates to a rock drillability prediction method based on a support vector machine regression model. Background Art
[0002] In oil and gas drilling projects, rock drillability has always been the most concerned parameter for drill bit design engineers. At present, the most commonly used methods to obtain rock drillability are laboratory micro-drilling test method and acoustic time difference prediction method. However, the laboratory micro-drilling method has the problems of difficult downhole coring and high drilling cost, and it is extremely difficult to obtain the drillability of the entire well section. The acoustic time difference prediction method first needs to establish a large number of databases to find laws statistically, and secondly, it cannot reveal the mechanism of the influence of acoustic time difference of different lithologies on drillability. However, rock drillability will seriously affect the life of the drill bit, drilling cycle and drilling cost. The extension of the drilling cycle will also cause hydration reaction between the drilling fluid and the surrounding rocks, affecting the stability of the wellbore. Therefore, accurate prediction of rock drillability is of great significance in terms of safe drilling speed and efficiency. There is an urgent need for a rock drillability prediction method for oil and gas wells based on a support vector machine regression model to establish a reliable model for rapid and accurate prediction of rock drillability based on downhole cuttings, and finally realize rapid measurement of rock drillability while drilling. Summary of the invention
[0003] The purpose of the present invention is to provide a rock drillability prediction method based on a support vector machine regression model to solve the problems existing in the above-mentioned prior art.
[0004] To achieve the above object, the present invention provides a rock drillability prediction method based on a support vector machine regression model, comprising:
[0005] Obtain the quantitative index value of the mesostructure of rock particles;
[0006] Build a support vector machine regression model and initialize it;
[0007] After determining the penalty coefficient and gamma coefficient of the support vector machine regression model, the support vector machine regression model is trained by using the mesostructure quantitative index value of the rock particles to obtain a prediction model;
[0008] The mesostructure quantified value of the rock particles in the oil and gas well is obtained and input into the prediction model to predict the drillability grade of the rock in the oil and gas well.
[0009] Optionally, the process of obtaining the mesostructure quantitative index value of rock particles includes: collecting literature data, and obtaining the mesostructure quantitative index based on the literature data analysis; contouring the rock particle image to obtain the absolute positions of several points on the rock particle boundary in the image, and calculating the geometric parameters of the corresponding contour based on the absolute positions of the several points to obtain the mesostructure quantitative index value.
[0010] Optionally, the rock particle image is contoured based on the optical characteristics of the type to which each rock particle belongs.
[0011] Optionally, the mesostructure quantitative indicators include but are not limited to length, area, circumference, roundness, angle factor, aspect ratio, and mesostructure coefficient.
[0012] Optionally, the model is initialized by standardizing and normalizing the input data of the support vector machine regression model.
[0013] Optionally, the process of determining the penalty coefficient and gamma coefficient of the support vector machine regression model includes:
[0014] The parameter range is predefined, the penalty coefficient and the gamma coefficient are determined by an exhaustive method, the model under each pair of coefficient combinations of the penalty coefficient and the gamma coefficient is trained and tested, and the coefficient combination with the best test result is selected as the penalty coefficient and the gamma coefficient of the support vector machine regression model.
[0015] Optionally, a Gaussian kernel function is used as the kernel function of the prediction model.
[0016] The technical effects of the present invention are:
[0017] The present invention can predict rock drillability based on the crystal structure and mineralogical characteristics of rock cuttings. The oil and gas well rock drillability prediction method established using the support vector regression model can not only improve the prediction accuracy of the model, but also solve the problem of using downhole rock cuttings to quickly predict rock drillability, providing strong support for the prediction of rock drillability in deep and ultra-deep wells, and providing a direction for the prediction of rock drillability in the entire well section. The present invention reduces the cost of obtaining rock drillability, solves the problem of difficulty in predicting rock drillability in deep and ultra-deep wells by using rock cuttings discharged from drilling fluid, improves the prediction accuracy of rock drillability and avoids the occurrence of model overfitting. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0019] Figure 1It is a flow chart of a rock drillability prediction method based on a support vector machine regression model in an embodiment of the present invention;
[0020] Figure 2 is a flow chart for determining a penalty coefficient and a gamma coefficient in an embodiment of the present invention;
[0021] Figure 3 Schematic diagram of the importance of different mesostructure coefficients to rock drillability in an embodiment of the present invention;
[0022] Figure 4 Schematic diagram of comparison between predicted and actual values of rock drillability in an embodiment of the present invention. DETAILED DESCRIPTION
[0023] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0024] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0025] Embodiment 1
[0026] like Figure 1-4 As shown, this embodiment provides a rock drillability prediction method based on a support vector machine regression model, including:
[0027] The quantitative indicators of the rock's microstructure are obtained, the support vector regression model is initialized, the Gaussian kernel function is determined, the penalty coefficient and the gamma coefficient are determined, the model is trained and tested, and a rock drillability prediction method based on the support vector machine regression model is formed.
[0028] Specifically, obtaining the mesostructure quantitative indicators of rocks includes clarifying the mesostructure quantitative indicators, dividing the contours of mineral particles based on the optical characteristics of minerals, summarizing all contour information of mineral particles, calculating the geometric parameters of the corresponding contours, and obtaining the mesostructure quantitative indicators through programming.
[0029] The optical characteristics of minerals refer to the fact that the images presented by the same mineral under single polarization and orthogonal polarization at different angles are different. Based on this characteristic, the single polarization and orthogonal polarization images of a mineral can be compared to distinguish the actual outline of the mineral. Then, ImageJ software can be used to use "marking points" to surround the mineral particles, and then adjacent "marking points" are connected to form a closed curve to obtain the outline of the particles.
[0030] Specifically, the support vector regression model initialization includes standardizing the mesostructure quantitative indicators and normalizing the data.
[0031] Specifically, determining the Gaussian kernel function involves mapping the input parameters to high dimensions so that it can process nonlinearly separable samples. The use of the Gaussian kernel function can effectively capture the complex relationship between samples and improve the model's fitting and generalization capabilities. The expression of the Gaussian kernel function is as follows:
[0032]
[0033] where κ(x i , x j ) represents the output value of the Gaussian kernel function, x i , x j represents the feature vector of the input sample, ||x i -x j || 2 represents the square of the modulus of the eigenvector difference, and σ represents the bandwidth of the Gaussian kernel function.
[0034] Specifically, determining the penalty coefficient and the gamma coefficient includes changing the penalty coefficient to determine the appropriate relaxation factor, clarifying the lower limit of the model prediction accuracy, debugging the appropriate gamma coefficient, and improving the generalization ability of the kernel function.
[0035] Specifically, model training and testing include importing data into the Jupyter program, dividing 70% of the data into a training set and 30% of the data into a test set, setting the penalty coefficient and Gamma, and obtaining the accuracy of the training set and the test set. The prediction model is obtained based on the training and test results.
[0036] Specifically, the rock drillability prediction method based on the support vector machine regression model includes establishing a support vector machine regression model with the microstructure of particles as an input parameter and the rock drillability grade as an output parameter.
[0037] In this embodiment, a rock drillability prediction method based on a support vector machine regression model mainly includes obtaining quantitative indicators of rock microstructure, initializing a support vector regression model, determining a Gaussian kernel function, determining a penalty coefficient and a gamma coefficient, model training and testing, and forming a rock drillability prediction method based on a support vector machine regression model.
[0038] In this embodiment, it is necessary to determine appropriate mesostructure quantitative indicators through literature research, mainly including length, area, circumference, roundness, angle factor, aspect ratio, mesostructure coefficient, etc., and then use image processing to obtain the absolute position of each marking point on the mineral boundary, calculate the geometric size according to the absolute position of the marking point on the mineral boundary, and finally use MATLAB programming to calculate the value of the mesostructure quantitative indicator according to the calculation model of each mesostructure quantitative indicator; initialize the support vector regression model and normalize the input data; determine the kernel function, penalty coefficient and gamma coefficient, among which the kernel function is a Gaussian kernel function, the penalty coefficient determines the interval size and classification accuracy, and the gamma coefficient affects the Gaussian range of action corresponding to the support vector and thus affects the generalization ability of the kernel function; through the training and testing stages of the model, a method for predicting the drillability of oil and gas well rocks based on the support vector machine regression model is formed.
[0039] Length, area, and circumference refer to classical geometric features, and roundness refers to "the ratio of the area A of an object to the area of a circle with the same circumference P". For example, roundness SF can be calculated using this formula Among them, P is the perimeter, A is the area of the object, and the angle factor refers to AF 1 , can be calculated according to the following two formulas, where x 1 is the number of particles whose longest axis and the direction have an angle of 0°-10°, x 2 is the number of particles whose longest axis and the direction have an angle of 10°-20°, and so on. 9 It is the number of particles whose longest axis and the direction have an angle of 80°-90°.
[0040]
[0041] AF 1 =ANGLE FACTOR / 5 The aspect ratio refers to the ratio of the longest axis to the shortest axis of the particle.
[0042] The microstructure coefficient TC refers to Where TC is the microstructure coefficient; AW is the mineral filling coefficient, N 0 、N 1 are the number of particles with aspect ratio (Feret's definition) lower or higher than the preset value (usually 2), SF 0 The arithmetic mean of the discriminated shape factors of all particles, AR 1 The arithmetic mean of the discriminated aspect ratios of all particles, AF 1 Angular factor of the particle.
[0043] like Figure 2 As shown, Figure 2 This is the process of determining the penalty coefficient and the gamma coefficient.
[0044] In this embodiment, the range of the penalty coefficient and the gamma coefficient is first predefined, and then the penalty coefficient and the gamma coefficient are determined by an exhaustive method. Then, the performance of the coefficient combination is cross-validated. The training set and test set scores of each pair of penalty coefficient and gamma coefficient combination are calculated by using the mean square error (MSE) to judge the quality of each pair of penalty coefficient and gamma coefficient combination. The smaller the MSE, the better the score. At this time, the penalty coefficient and gamma coefficient are the preferred combination, and the preferred parameter combination is used as the penalty coefficient and gamma coefficient stage of the model.
[0045]
[0046] Where N is the number of samples, y p，i is the predicted drillability, y t，i It’s the actual drillability.
[0047] like Figure 3 As shown, Figure 3 is the importance of different mesostructure coefficients on rock drillability. The importance is calculated by using the Lagrange multiplier a i It can be seen that the larger the Lagrange multiplier, the higher the corresponding importance.
[0048] In this embodiment, the angle factor, the microstructure coefficient, the aspect ratio and the roundness have the greatest influence on the rock drillability, while the area, the particle homogeneity and the compaction degree have the least influence on the rock drillability.
[0049] The Lagrangian function of the original problem is as follows:
[0050] Y(w,b,ζ,α,β)=Y 1 +Y 2 -Y 3 -Y 4
[0051]
[0052]
[0053]
[0054]
[0055] where Y 1 is the mean norm of the model, Y 2 is the penalty factor, Y 3 Constraint function between model output and actual label, Y 4is the constraint function of the slack variable, w is the weight vector, b is the intercept, ζ is the slack variable vector, α i and β i is the Lagrange multiplier vector.
[0056] Solve the Lagrangian function to maximize it, solve the dual problem, calculate the partial derivatives of the weight vector, intercept, and slack variable and set them equal to zero, and obtain the Lagrangian multiplier. The larger the Lagrangian multiplier, the higher the importance. When a i and β i When the effects on output parameters are inconsistent, use a i As a basis for judgment.
[0057] like Figure 4 As shown, Figure 4 Comparison results of predicted and actual values of rock drillability.
[0058] In this embodiment, the predicted drillability and the measured drillability show a good correlation, which further indicates that the oil and gas well rock drillability prediction method based on the support vector machine regression model has a good promotion prospect.
[0059] The above is only a preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A rock drillability prediction method based on a support vector machine regression model, characterized in that: The following steps are involved: Obtain the quantitative index value of the mesostructure of rock particles; The process of obtaining the mesostructure quantitative index value of rock particles includes: collecting literature data, and obtaining the mesostructure quantitative index according to the literature data analysis; performing contour division on the rock particle image, obtaining the absolute positions of several points on the rock particle boundary in the image, and calculating the geometric parameters of the corresponding contour based on the absolute positions of several points, and obtaining the mesostructure quantitative index value; The rock grain image is contoured based on the optical characteristics of the type to which each rock grain belongs; The mesostructure quantitative indicators include, but are not limited to, length, area, perimeter, roundness, angle factor, aspect ratio, and mesostructure coefficient; Build a support vector machine regression model and initialize it; After determining the penalty coefficient and gamma coefficient of the support vector machine regression model, the support vector machine regression model is trained by using the mesostructure quantitative index value of the rock particles to obtain a prediction model; The process of determining the penalty coefficient and gamma coefficient of the support vector machine regression model includes: Predefine the parameter range, use the exhaustive method to determine the penalty coefficient and the gamma coefficient, train and test the model under each pair of coefficient combinations of the penalty coefficient and the gamma coefficient, and select the coefficient combination with the best test result as the penalty coefficient and the gamma coefficient of the support vector machine regression model; The mesostructure quantified value of the rock particles in the oil and gas well is obtained and input into the prediction model to predict the drillability grade of the rock in the oil and gas well.
2. The rock drillability prediction method based on the support vector machine regression model according to claim 1 is characterized in that: The model is initialized by standardizing and normalizing the input data of the support vector machine regression model.
3. The rock drillability prediction method based on support vector machine regression model according to claim 1 is characterized in that: A Gaussian kernel function is used as the kernel function of the prediction model.