Sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree

Through microscopic image processing and GBDT decision tree technology, the problem of insufficient accuracy of sandstone abrasive prediction is solved, and the precise division and abrasive prediction of rock mineral particle structure are achieved, which improves drilling efficiency and cost savings.

CN117975452BActive Publication Date: 2024-12-03SOUTHWEST PETROLEUM UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410133165.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-31
Publication Date
2024-12-03
Estimated Expiration
2044-01-31

AI Technical Summary

Technical Problem

The prior art is difficult to accurately predict the abrasiveness of sandstones, especially in consideration of the mineral particle structure inside the rock, which leads to difficulty in centering downhole, high drilling costs and difficulty in obtaining the abrasiveness level value of the entire well section.

Method used

Using a method based on microscopic image processing and GBDT decision tree, the rock sample images are acquired, image processing and crystal boundary division are performed, the geometric parameters and rock maturity of the crystal are extracted, and the sandstone abrasive prediction model is constructed, and the GBDT decision tree algorithm is used to establish Mineral particle structure and composition predict abrasiveness.

Benefits of technology

It realizes the precise division of rock mineral particle profiles, improves the prediction accuracy of sandstone abrasiveness, avoids overfitting problems, and provides a new method for obtaining sandstone abrasiveness in the whole well section, which can quickly calculate and visualize the results and support drill bit life predict.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117975452B_ABST
    Figure CN117975452B_ABST
Patent Text Reader

Abstract

The present invention discloses a sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree, comprising: obtaining a rock sample image to be tested; performing image processing on the image, dividing the crystal boundary of the processed image, extracting the geometric parameters of the crystal, and determining the rock maturity of the processed image; performing a rock abrasiveness test on the image to obtain the rock abrasiveness of the image; performing correlation analysis and screening of characteristic parameters of the rock maturity, the geometric parameters and the rock abrasiveness, and constructing a sandstone abrasiveness prediction model based on microscopic image processing and GBDT decision tree. The present invention achieves the purpose of accurately predicting the abrasiveness of sandstone in the entire well section using downhole rock cuttings.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of oil and gas drilling engineering, and in particular relates to a sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree. Background Art

[0002] In oil and gas drilling projects, sandstone abrasiveness has always been the parameter that drill bit design engineers pay the most attention to. At present, the main methods for obtaining sandstone abrasiveness are CERCHAR grinding method, bond ball milling method, logging parameter calculation method, and rock mechanical property method. Among them, the most commonly used is CERCHAR grinding method, but there are also problems such as difficulty in coring downhole, high drilling cost, and the obtained rock cuttings are mostly irregular in shape, which makes it difficult to conduct effective experiments, and it is extremely difficult to obtain the abrasiveness level of the entire well section. Sandstone abrasiveness affects the life of the drill bit, drilling cycle and drilling cost. Accurately predicting sandstone abrasiveness is of great significance in improving drilling efficiency, saving drilling costs and reducing drilling cycles. In the prior art, a relationship model is proposed based on the internal friction angle and the acoustic wave time difference in the direction perpendicular to the bedding plane, and the drilling angle to determine the internal friction angle of the shale in different drilling directions, that is, the abrasiveness of the direction. This method is first limited by the correlation between the rock physical properties and rock abrasiveness at the macro level, and secondly, it does not consider the influence of the structure of the mineral particles inside the rock on the abrasiveness. The prior art also proposes a method of using the logging data of the rock sample to calculate the abrasiveness parameters of the rock sample, and using density logging data, resistivity logging data, neutron porosity logging data, sonic logging data and natural gamma logging data to directly evaluate the rock abrasiveness parameters of the conglomerate formation, but does not consider the influence of the structure of the mineral particles inside the rock on the abrasiveness. In the "Automatic Segmentation and Extraction Method of Rock Cuttings from Sand Sample Images", a new sand sample image segmentation and extraction method is proposed, which can realize the intelligent segmentation and extraction of rock cuttings. However, it is still impossible to segment and extract particles of sandstone with complex structural characteristics and cementation characteristics, and it is impossible to segment the boundaries of mineral particles in the rock microstructure image. Therefore, it is urgent to propose a sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree. By using the GBDT decision tree method to establish a model for predicting the abrasiveness of sandstone mineral particle structure and mineral composition, the purpose of accurately predicting the abrasiveness of sandstone in the entire well section by using downhole cuttings is achieved.

[0004] To achieve the above object, the present invention provides a sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree, comprising:

[0005] Acquire an image of the rock sample to be tested;

[0006] Performing image processing on the image, dividing the crystal boundary of the processed image, extracting the geometric parameters of the crystal, and determining the rock maturity of the processed image;

[0007] Conducting a rock abrasiveness test on the image to obtain the rock abrasiveness of the image;

[0008] The rock maturity, the geometric parameters and the rock abrasiveness are subjected to correlation analysis and screening of characteristic parameters, and a sandstone abrasiveness prediction model based on microscopic image processing and GBDT decision tree is constructed.

[0009] Optionally, the method for conducting a rock abrasiveness test on the image includes: conducting an abrasiveness test on the rock sample image using an abrasiveness test method recommended by the International Society for Rock Mechanics, testing each sample a preset number of times, and taking the average value to obtain the rock abrasiveness of the image.

[0010] Optionally, the image is processed, and the method for dividing the crystal boundaries of the processed image includes: creating a topological map using the grayscale value of the image based on the classical segmentation method of the watershed algorithm of the image, and the watershed of the topological map represents the segmented area; generating a topological map by simulating a diffusion process, and the diffusion process uses a true value mask drawn by manual annotation; training a neural network to predict the horizontal and vertical gradients of the topological map, and a binary map indicating whether a given pixel is inside or outside the area of ​​interest; in image processing, the neural network predicts horizontal and vertical gradients to form a vector field; tracking the vector field through a gradient tracking process, routing all pixels of a given cell to the center; restoring individual mineral particles and their precise shapes by grouping together pixels that converge to the center; and further refining the shape of each mineral particle through pixels located outside the particle predicted by a convolutional neural network.

[0011] Optionally, the method for determining the rock maturity of the processed image includes: judging the processed image, if the clay impurity content in the processed image exceeds 5%, the particle sorting is less than 15% and the roundness is less than 0.5, it is an immature structure; if the clay impurity content in the processed image is less than 5%, the particle sorting is between 15% and 35% and the roundness is between 0.5 and 0.8, it is a sub-mature structure; if the processed image contains a small amount or no clay impurities, the particle sorting is higher than 35%, and the roundness is greater than 0.8, it is a mature structure.

[0012] Optionally, the geometric parameters of the crystal include shape factor, roundness, compaction, area, diameter and circumference.

[0013] Optionally, the method for performing correlation analysis and screening of characteristic parameters of the rock maturity, the geometric parameters and the rock abrasiveness includes: sorting the correlation between the geometric parameters and mineral components and the abrasiveness, screening the geometric parameters and mineral components with high correlation with the abrasiveness as input parameters for establishing a abrasiveness prediction model, obtaining the geometric parameters and abrasiveness data as the total data set, performing correlation analysis on the extracted geometric parameters and mineral components and the rock abrasiveness, and obtaining the correlation between the geometric parameters and mineral components and the rock abrasiveness.

[0014] Optionally, the method for obtaining the correlation between the extracted geometric parameters and mineral components and the rock abrasiveness is:

[0015]

[0016] Where y is w i Spearman correlation with sandstone abrasiveness; w i is the difference in the rank value of the i-th data; n is the total number of samples.

[0017] Optionally, a method for constructing a sandstone abrasiveness prediction model based on microscopic image processing and GBDT decision tree is:

[0018] The initial objective function of the model is:

[0019]

[0020] Among them, O(m,f0(b)) is the initial objective function of the model; N is the number of samples, m i is the true label of the i-th sample, f0(b i ) is the predicted value of the initial model for the i-th sample;

[0021] The final objective function of the model is:

[0022]

[0023] Among them, O(m,F(b)) is the final objective function of the model; F(b) is the final prediction of the model; m i is the true label of the i-th sample; fj is the j-th decision tree; F(bi) is the predicted value of the i-th sample; L(fj) is the regularization term of the decision tree;

[0024] The input parameters of acoustic wave velocity and Leeb hardness were substituted into the final objective function to obtain the results. The results were compared with the abrasiveness to obtain the accuracy of the prediction model, and a sandstone abrasiveness prediction model based on microscopic image processing and GBDT decision tree was constructed.

[0025] Technical effects of the present invention:

[0026] (1) It provides a method for accurately dividing the contours of rock mineral particles, solving the problem of difficult division of mineral particles in rocks and providing a theoretical basis for artificial intelligence to extract rock mesostructure;

[0027] (2) The prediction accuracy of sandstone abrasiveness is improved, and the accuracy of the prediction model can be further improved as the number of samples increases;

[0028] (3) After inputting data, the results can be quickly calculated and displayed visually;

[0029] (4) The use of the GBDT decision tree algorithm ensures that the model based on the regression of mineral grain structure and mineralogical characteristics with sandstone abrasiveness will not have the problem of overfitting during the training process;

[0030] (5) A new method for obtaining the abrasiveness of sandstone in the entire well section is provided, which can provide a scientific basis for predicting the life of the drill bit. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] 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:

[0032] Figure 1 It is a schematic diagram of the process of predicting sandstone abrasiveness based on microscopic image processing and GBDT decision tree according to an embodiment of the present invention;

[0033] Figure 2 It is a schematic diagram of the training process of the mineral particles of sandstone based on the watershed algorithm in an embodiment of the present invention;

[0034] Figure 3 A schematic diagram of the accuracy of dividing mineral grain boundaries using a watershed algorithm based on image marking in an embodiment of the present invention;

[0035] Figure 4 A schematic diagram of the learning process of a single GBDT decision tree algorithm according to an embodiment of the present invention;

[0036] Figure 5 Schematic diagram of correlation analysis results of characteristic parameters in an embodiment of the present invention;

[0037] Figure 6 Schematic diagram of the prediction accuracy of abrasiveness of different parameters according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] 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.

[0039] 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.

[0040] like Figure 1 As shown, this embodiment provides a sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree, including:

[0041] Obtain input parameters 101, conduct abrasiveness test to obtain rock abrasiveness 102, image processing 103, divide mineral grain boundaries based on image labeling watershed algorithm 104, determine rock maturity 105, extract particle geometric parameters 106, correlation analysis and screening of characteristic parameters 107, establish sandstone abrasiveness prediction model based on microscopic image processing and GBDT decision tree 108.

[0042] In this embodiment, 101, 102, and 103 are the preparation stage for extracting input parameters, 104, 105, and 106 are the stage for obtaining input parameters, 107 is the stage for analyzing and screening parameters, and 108 is the stage for predicting abrasiveness based on mineral particle structure and mineralogical characteristics.

[0043] The input parameters 101 include the abrasiveness, maturity, and geometric parameters of mineral particles of the rock.

[0044] The rock abrasiveness test is carried out to obtain the rock abrasiveness 102, which includes using the abrasiveness test method recommended by the International Society of Rock Mechanics to test the rock samples, testing each sample five times, and finally taking the average value as the actual abrasiveness.

[0045] Image processing 103 includes making thin slices and taking microscopic images, which provide a basis for the classification of maturity and the intelligent extraction of particle contours.

[0046] The watershed algorithm based on image marking divides the boundaries of mineral particles 104, including grayscale transformation, identification of the boundaries of mineral particles, clarification of the ownership of mineral particles, watershed transformation and image segmentation, to obtain a model dedicated to the automatic recognition and segmentation of the contours of rock mineral particles. The processed image is judged. If the clay impurity content in the processed image exceeds 5%, the particle sorting is less than 15%, and the roundness is less than 0.5, it is an immature structure; if the clay impurity content in the processed image is less than 5%, the particle sorting is between 15% and 35%, and the roundness is between 0.5 and 0.8, it is a sub-mature structure; if the processed image contains a small amount or no clay impurity, the particle sorting is higher than 35%, and the roundness is greater than 0.8, it is a mature structure.

[0047] Determining the rock maturity 105 includes grading all rock samples into three levels of structural maturity: immature, semi-mature, and mature. The specific grading standards are as follows: immature - the clay matrix content in the rock sample exceeds 5%, the particle sorting is less than 15%, and the roundness is less than 0.5; semi-mature - the clay matrix content in the rock sample is less than 5%, the particle sorting is between 15% and 30%, and the roundness is between 0.5 and 0.8; mature - the rock sample contains a small amount or no clay matrix, the particle sorting is higher than 35%, and the roundness is greater than 0.8. The maturity is structurally characterized and sequentially coded, and immature, semi-mature, and mature are coded as 0, 1, and 2 respectively.

[0048] Extracting particle geometry parameters 106 includes calculating the shape factor, roundness, compaction, area, diameter, circumference, etc. of the mineral particles.

[0049] The correlation analysis and screening 107 of characteristic parameters include sorting the correlation between geometric parameters and mineral components and abrasiveness, screening geometric parameters and mineral components with high correlation with abrasiveness as input parameters for establishing a grindability prediction model, and obtaining geometric parameters and grindability data as a total data set S. Correlation analysis is performed on the extracted geometric parameters and mineral components obtained by XRD experiment and the grindability data obtained by experiment to obtain the correlation between the extracted geometric parameters and mineral components and grindability. The calculation expression of Spearman correlation is:

[0050]

[0051] Where y is the parameter w i Spearman correlation with sandstone abrasiveness grade; w i is the difference in the rank value of the i-th data; n is the total number of samples.

[0052] Establishing a sandstone abrasiveness prediction model 108 based on microscopic image processing and GBDT decision tree includes randomly dividing the total data set S to obtain a training set S1 and a test set S2, wherein the training set S1 accounts for 83.3% of the total data set S and the test set S2 accounts for 16.7% of the total data set S; training the training set S1 using the GBDT decision tree method to obtain a trained GBDT regression model; using the geometric parameters and mineral composition of the test set S2 to bring into the established regression model to obtain the predicted abrasiveness; by comparing and analyzing the abrasiveness of the test set S2 with the predicted value, the prediction accuracy of the regression model can be obtained to evaluate the quality of the regression model; based on the downhole rock cuttings, the mineral grain structure and mineralogical characteristics are obtained, and the mineral grain structure and mineralogical characteristics are brought into the established GBDT regression model to obtain the abrasiveness, and finally a model for predicting abrasiveness based on the mineral grain structure and mineralogical characteristics of sandstone is obtained.

[0053] The initial objective function of the model is:

[0054]

[0055] Among them, O(m,f0(b)) is the initial objective function of the model; N is the number of samples, m i is the true label of the i-th sample, f0(b i ) is the predicted value of the initial model for the i-th sample.

[0056] The final objective function of the model is:

[0057]

[0058] Where O(m,F(b)) is the final objective function of the model; F(b) is the final prediction of the model; m i is the true label of the i-th sample; fj is the j-th decision tree; F(bi) is the predicted value of the i-th sample; L(fj) is the regularization term of the decision tree.

[0059] like Figure 2 As shown, Figure 2 The training process of the mineral particles of sandstone in the embodiment is based on the watershed algorithm. In this embodiment, the watershed algorithm based on image labeling is used. The training process is as follows Figure 2 As shown in the figure, an initial model is trained first, and this initial model is applied to the user's data. The result will be manually corrected by the user, and then retrained based on the corrected result to improve the accuracy of the algorithm for mineral segmentation. The mineral particles of the rock are segmented automatically. With the increase of model training, the model's intelligent recognition and segmentation accuracy of the edges of mineral particles in rock thin section images are getting higher and higher. The successful application of this algorithm saves a lot of time for extracting rock microstructures.

[0060] like Figure 3 As shown, Figure 3 This is the accuracy of dividing the mineral grain boundaries using the watershed algorithm based on image marking in the embodiment.

[0061] Among them, A is the area of ​​the mineral particle; p is the circumference of the mineral particle; Rnd is the roundness; L is the Feret diameter.

[0062] In this embodiment, the precision 301 of intelligently segmenting particles shows that the watershed algorithm can achieve the precision requirement of manually segmenting mineral particles.

[0063] like Figure 4 As shown, Figure 4 This is the learning process of a single GBDT decision tree algorithm.

[0064] In this embodiment, all tree structures are listed 401, objective functions of all tree structures are calculated 402, the optimal tree structure is found 403, the prediction value of each leaf node is calculated 404, the available features of each leaf node are listed 405, the optimal splitting point is obtained 406, and the decision tree is constructed 407.

[0065] The initial objective function of all tree structures is:

[0066]

[0067] in, is the initial objective function of all tree structures; M is the loss function, which is used to measure how well the model fits the training data; N is called the regularization term, which is used to measure the complexity of the learned model.

[0068] The final objective function of all tree structures is:

[0069]

[0070] Among them, O is the final objective function of all tree structures; T is the number of leaf nodes; define the set Ij = {i|q(xi) = j}; define q(x) means that sample x is on a leaf node; b is a constant.

[0071] The formula for calculating the predicted value of each root node is:

[0072]

[0073] in is the root node prediction formula; define the set Ij = {i|q(xi) = j}; define q(x) means that sample x is on a leaf node; is a constant.

[0074] like Figure 5 As shown, Figure 5 It is the correlation analysis result of the characteristic parameters in the embodiment.

[0075] In this embodiment, Spearman correlation analysis is performed on A, P, L, AR, SF, Rnd, Eds, Cpa, t, g, TC, M, and CAI, and parameters with high correlation with abrasiveness are selected as input parameters in the data set to improve the efficiency of model training and model prediction based on the GBDT decision tree prediction method. Among them, A-area; P-perimeter; L-Feret diameter; AR-minimum aspect ratio; SF-shape factor; Rnd-roundness; Eds-edge smoothness; Cpa-compactness; t-particle size uniformity; g-interlocking index; TC-mesostructure coefficient; M-maturity; CAI-abrasiveness.

[0076] like Figure 6 As shown, Figure 6 Test set accuracy for predicting sandstone abrasiveness for example.

[0077] In this example, three related input parameters, namely rock maturity, mesostructure parameter TC value and equivalent quartz content EQC, are introduced. According to the corresponding added input parameters, the GBDT decision tree is used again to train the abrasiveness model, and the abrasiveness model accuracy under different input parameters is obtained. It can be found that the accuracy of the abrasiveness model obtained after the GBDT decision tree is trained increases with the increase of input parameters, which further proves that the sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree has a good application effect.

[0078] The above are only preferred specific implementations 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 sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree, characterized in that: include: Acquire an image of the rock sample to be tested; Performing image processing on the image, dividing the crystal boundary of the processed image, extracting the geometric parameters of the crystal, and determining the rock maturity of the processed image; Conducting a rock abrasiveness test on the image to obtain the rock abrasiveness of the image; The rock maturity, the geometric parameters and the rock abrasiveness are subjected to correlation analysis and screening of characteristic parameters, and a sandstone abrasiveness prediction model based on microscopic image processing and GBDT decision tree is constructed; The method for performing correlation analysis and screening of characteristic parameters of the rock maturity, the geometric parameters and the rock abrasiveness comprises: sorting the correlations between the geometric parameters and the mineral components and the abrasiveness, screening the geometric parameters and the mineral components with high correlations with the abrasiveness as input parameters for establishing an abrasiveness prediction model, using the obtained geometric parameters and abrasiveness data as a total data set, performing correlation analysis on the extracted geometric parameters and the mineral components and the rock abrasiveness, and obtaining the correlations between the geometric parameters and the mineral components and the rock abrasiveness; The method for obtaining the correlation between the geometric parameters and the mineral composition and the abrasiveness of the rock is: Where y is w i Spearman correlation with sandstone abrasiveness; w i is the difference in the rank value of the i-th data; n is the total number of samples; The method for constructing a sandstone abrasiveness prediction model based on microscopic image processing and GBDT decision tree is: The initial objective function of the model is: Among them, O(m,f0(b)) is the initial objective function of the model; N is the number of samples, m i is the true label of the i-th sample, f0(b i ) is the predicted value of the initial model for the i-th sample; The final objective function of the model is: Among them, O(m,F(b)) is the final objective function of the model; F(b) is the final prediction of the model; m i is the true label of the i-th sample; fj is the j-th decision tree; F(bi) is the predicted value of the i-th sample; L(fj) is the regularization term of the decision tree; The input parameters of acoustic wave velocity and Leeb hardness were substituted into the final objective function to obtain the results. The results were compared with the abrasiveness to obtain the accuracy of the prediction model, and a sandstone abrasiveness prediction model based on microscopic image processing and GBDT decision tree was constructed.

2. The sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree according to claim 1 is characterized in that: The method for conducting a rock abrasiveness test on the image includes: conducting an abrasiveness test on the rock sample image using an abrasiveness test method recommended by the International Society for Rock Mechanics, testing each sample a preset number of times, and obtaining the rock abrasiveness of the image after taking the average value.

3. The sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree according to claim 1 is characterized in that: The image is processed, and the method for dividing the crystal boundary of the processed image includes: creating a topological map using the gray value of the image based on the classical segmentation method of the watershed algorithm of the image, and the watershed of the topological map represents the segmented area; generating a topological map by simulating a diffusion process, and the diffusion process uses a true value mask drawn by manual annotation; training a neural network to predict the horizontal and vertical gradients of the topological map, and a binary map indicating whether a given pixel is inside or outside the region of interest; in image processing, the neural network predicts the horizontal and vertical gradients to form a vector field; tracking the vector field through a gradient tracking process, routing all pixels of a given cell to the center; restoring individual mineral particles and their precise shapes by grouping together pixels that converge to the center; and further refining the shape of each mineral particle through pixels outside the particle predicted by a convolutional neural network.

4. The sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree according to claim 1 is characterized in that: The method for determining rock maturity from a processed image includes: judging the processed image, if the clay impurity content in the processed image exceeds 5%, the particle sorting is less than 15%, and the roundness is less than 0.5, it is an immature structure; if the clay impurity content in the processed image is less than 5%, the particle sorting is between 15% and 35%, and the roundness is between 0.5 and 0.8, it is a sub-mature structure; if the processed image contains a small amount or no clay impurity, the particle sorting is higher than 35%, and the roundness is greater than 0.8, it is a mature structure.

5. The sandstone abrasiveness prediction method based on microscopic image processing and GBDT decision tree according to claim 1, characterized in that: The geometric parameters of the crystal include shape factor, roundness, compaction, area, diameter and circumference.