Continental facies pulveryte lithofacies logging identification method based on SMOTE-XGBoost

Abstract

The invention belongs to the technical field of oil and gas development, and discloses a continental facies pulveryte lithofacies logging identification method based on SMOTE-XGBoost, and the method comprises the steps: carrying out the depth correction of a target layer core in a research area, and carrying out the normalization processing of logging data; the lithofacies are divided based on a lithology-structure-TOC-mineral four-unit method, and a lithofacies-logging database is established; a key well with lithofacies and corresponding logging data is selected as a model training well, training data are preprocessed through an SMOTE method, the problem that model training is insufficient due to the fact that the number of sample categories is unbalanced is solved, and all categories of lithofacies data are made to be the same; and inputting lithofacies and logging data into the XGBoost model for training, and establishing a continental facies pulveryte lithofacies identification model by using default model parameters. According to the method, the nonlinear problem caused by strong heterogeneity of lithofacies can be effectively solved, and the precision of continental facies pulveryte lithofacies identification is obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas development, and in particular to a lithofacies logging identification method for continental fine-grained sedimentary rocks based on SMOTE-XGBoost. Background Art

[0002] Fine-grained sedimentary rock lithofacies is one of the important means to study the enrichment mechanism of continental shale oil. The spatial distribution of lithofacies controls the spatial law of shale oil enrichment, so the spatial distribution state of lithofacies is particularly important. However, it is difficult to meet the needs of regional fine lithofacies distribution research by relying solely on some outcrops, cores, and experimental test analysis methods, and it consumes a lot of manpower, material resources, and financial resources. There is an urgent need for an economical and simple method to quickly identify the spatial distribution of lithofacies. Therefore, logging information with high vertical resolution plays an important role in lithofacies identification. However, factors such as frequent lithology alternation, complex mineral composition, and diverse and complex structures of fine-grained sedimentary rocks cause strong heterogeneity of fine-grained sedimentary rock lithofacies, resulting in the relationship between lithofacies and logging is no longer a simple linear relationship, but a complex nonlinear characteristic. Conventional methods are no longer applicable, which poses a huge challenge to the logging identification of lithofacies.

[0003] When establishing the lithofacies logging identification model for continental fine-grained sedimentary rocks, the number of training sample categories is unbalanced. Because the lithofacies developed in the target layer are mostly shale, while siltstone is mostly interlayered, the imbalance of training samples leads to insufficient training of a few types of lithofacies samples during model training, affecting the accuracy of the model in identifying lithofacies with interlayer development such as siltstone. Summary of the invention

[0004] In response to the problems encountered in the identification of terrestrial fine-grained sedimentary lithofacies, the present invention provides a lithofacies logging identification method for terrestrial fine-grained sedimentary rocks based on SMOTE-XGBoost. The SMOTE method is used to solve the problem of insufficient training of minority class samples caused by the imbalance in the development of lithofacies categories. At the same time, the XGBoost algorithm is combined to process the nonlinear relationship between lithofacies and logging, thereby realizing the identification of terrestrial fine-grained sedimentary rock lithofacies.

[0005] The present invention provides the following technical solutions:

[0006] A method for identifying terrestrial fine-grained sedimentary rock lithofacies by well logging based on SMOTE-XGBoost, comprising the following steps:

[0007] Step 1: Correct the core depth of the target layer in the study area and normalize the logging data;

[0008] Step 2: Divide the lithofacies based on the four-unit method of "lithology-structure-TOC-mineral" and establish a lithofacies-well logging database;

[0009] Step 3: Select a key well with lithofacies and corresponding logging data as the model training well, and pre-process the training data using the SMOTE method to overcome the problem of insufficient model training caused by the imbalance of sample categories, so that the lithofacies data of each category are the same;

[0010] Step 4: Input the lithofacies and logging data into the XGBoost model for training, and use the default model parameters to establish a lithofacies identification model for continental fine-grained sedimentary rocks.

[0011] Furthermore, in step 1, we first collect the well logging curves of the core wells, realize the splicing and standardization of the well logging curves, and normalize the well logging curves to eliminate the influence of well logging data of different dimensions, and normalize the curve values ​​to the interval [0,1]. The core depth may be inconsistent with the well logging depth, and the core depth is corrected by observing the well logging characteristics corresponding to the core. The normalization formula is as follows:

[0012]

[0013] Where X norm is the normalized data, X i is the original data, X max , X min are the maximum and minimum values ​​of the logging data respectively.

[0014] Furthermore, step 2 proposes a four-unit lithofacies division scheme of "lithology-structure-TOC-mineral". Lithology is divided into three types: shale-siltstone-carbonate rock, and each axis is divided by 50%, which is further divided into three types: shale, siltstone and limestone (cloud) rock; primary structure is divided into three types: massive, laminar and layered; mineral composition is divided into three types: clay-felsic-carbonate, and each axis is divided by 50%, which is further divided into three types: clay, felsic and lime; fine-grained sediments are divided into three types according to the TOC value: high organic matter content (TOC value greater than 2%), medium organic matter content (TOC value 1% to 2%) and low organic matter content (TOC value less than 1%). The fine-grained sediments in the study area are divided into six types: A high-organic mud-grade laminated clay shale, B medium-organic mud-grade laminated felsic shale, C low-organic silt-containing laminated felsic shale, D layered siltstone, and E layered shell-clastic limestone.

[0015] Furthermore, in step 3, 70% of the data set is divided into a training set and 30% is divided into a test set to test the model validation effect, and then SMOTE processing is performed. SMOTE is an oversampling technology. i , first calculate its K-nearest neighbors of the same category KN_x i In SMOTE, K is usually set to 5. From KN_x iA sample X2 is randomly selected from the sample X1 and the difference between the corresponding attribute j between the samples X1 and X2 is diff j =x 2j -x 1j The specific formula is as follows:

[0016] f 1j =x 1j +diff j ×rand[0,1]=x 1j +(x 2j -x 1j )×rand[0,1](2)

[0017] Where f 1j are the newly generated minority class samples.

[0018] Furthermore, in step 4, XGBoost is a Boosting ensemble learning method that uses CART as the base learner. The specific formula is as follows:

[0019]

[0020] In the formula is the predicted value, x i is the i-th sample, f k is an independent function in the space, and F is the function space composed of all CART.

[0021] The objective function formula is as follows:

[0022]

[0023] Where L is the loss function, is the error between the predicted value and the true value, and Ω is the regularization function. The regularization function is as follows:

[0024]

[0025] Where T is the number of leaf nodes in each tree, w is the weight of each leaf, λ is the 2 regularization coefficient, and γ is the split threshold. The formula is as follows:

[0026]

[0027]

[0028]

[0029]

[0030]

[0031] XGBoost is calculated by a greedy algorithm. Assuming that the node before the column is j, its contribution formula to the objective function is as follows:

[0032]

[0033] The contribution formula of the two child nodes to the objective function is as follows:

[0034]

[0035] At this time, the objective function can be changed as follows:

[0036]

[0037] Finally, the information gain formula obtained by each split of the objective function is as follows:

[0038]

[0039] Where GLGR is the information gain statistic of the first-order gradient of the left and right leaves, and HLHR is the information gain statistic of the second-order gradient of the left and right leaves.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The present invention uses a combination of core observation, thin section identification, organic carbon, X-diffraction and other means to perform lithofacies division of fine-grained sedimentary rocks, combines logging data with high vertical resolution to establish a lithofacies-logging database, and establishes a SMOTE-XGBoost model to effectively deal with the nonlinear problems caused by strong lithofacies heterogeneity, thereby significantly improving the accuracy of lithofacies identification of terrestrial fine-grained sedimentary rocks. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Attached Figure 1 This is a flow chart of the lithofacies logging identification method for continental fine-grained sedimentary rocks based on SMOTE-XGBoost;

[0043] Attached Figure 2 This is a lithofacies division diagram of continental fine-grained sedimentary rocks based on the principle of "lithology-structure-TOC-mineral" division;

[0044] Attached Figure 3 This is the effect diagram of single well lithofacies prediction;

[0045] Attached Figure 4 Analysis chart of the prediction results for the test set. DETAILED DESCRIPTION

[0046] In order to more clearly explain the purpose, method and steps of this application, Figure 1-4 Provide an expanded description.

[0047] The examples described below are only preferred embodiments of the present invention. It should be noted that the following description is only for the purpose of explaining the present invention and is not intended to limit the present invention in any form. Any simple modification, equivalent changes and modifications made to the embodiments according to the technical essence of the present invention are within the scope of the technical solution of the present invention. In the following examples, the materials, reagents, etc. used were obtained from commercial sources unless otherwise specified.

[0048] Specific related process reference Figure 1 The present invention proposes a lithofacies logging identification method for continental fine-grained sedimentary rocks based on SMOTE-XGBoost. The application example is the H-1 well of the continental fine-grained sedimentary rock formation of the Qingshankou Formation in the southern Songliao Basin. The lithofacies are divided by combining core, thin section, TOC, and X-diffraction data, and the corresponding relationship between lithofacies and logging data is established for lithofacies identification. The specific steps are as follows:

[0049] Step 1: Observe the core of the target layer of Well H-1 carefully, observe the response of lithology on the well logging, and perform core depth correction. First, splice the well logging curves on the well logging data, and then perform normalization to normalize the well logging data to the [0-1] interval.

[0050] Step 2: Take fine samples of the target layer core in the range of 0.3-3m for indoor experimental analysis, observe the bedding structure of each lithofacies on thin sections, analyze the geochemical characteristics of the lithofacies on TOC, and analyze the mineral characteristics by X-diffraction. Comprehensively analyze the above characteristics to establish a four-unit lithofacies division scheme of "lithology-structure-TOC-mineral".

[0051] Step 3: Select the corresponding logging data from the divided lithofacies, which are four curves: natural gamma (GR), deep lateral resistivity (RLLD), acoustic time difference (AC), and density (DEN). The sampling interval is 0.125m and the depth range is 2043.75-2052m. The SMOTE method is used to calculate the data with equal quantities of 6 types of lithofacies from the samples with unbalanced development, so as to prepare for the subsequent model establishment.

[0052] Step 4: 70% of the H-1 well data set is the training set and 30% is the test set, which are input into the XGBoost model. The specific parameters are learning rate 0.1, number of base learners 50, maximum tree depth 3, and others are default. Iteration is continued until the error is minimized. The trained terrestrial fine-grained sedimentary rock lithofacies recognition model has a prediction accuracy of up to 80.46% in the test set.

[0053] The SMOTE-XGBoost-based lithofacies logging identification method for terrestrial fine-grained sedimentary rocks invented in this invention establishes a correlation between the lithofacies divided by limited and expensive core and testing costs and the corresponding conventional logging data, saving a lot of manpower, material and financial resources. The SMOTE method is used to deal with the imbalance problem between lithofacies samples, overcoming the problem of insufficient model training on a few types of lithofacies samples. At the same time, the XGBoost method in ensemble learning is used, and multiple base classifiers are used to participate in the training at the same time, which handles the nonlinear relationship between complex lithofacies logging, comprehensively improves the accuracy of model identification of lithofacies, and provides a convenient, accurate and efficient technical foundation for exploring the distribution status of lithofacies.

[0054] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not limitations on the implementation methods of the present invention. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the implementation methods here. All obvious changes or modifications derived from the technical solution of the present invention are still within the protection scope of the present invention.

Claims

1. A method for identifying terrestrial fine-grained sedimentary rock lithofacies by well logging based on SMOTE-XGBoost, characterized in that: The following steps are involved: Step 1: Correct the core depth of the target layer in the study area and normalize the logging data; Step 2: Divide the lithofacies based on the "lithology-structure-TOC-mineral" four-unit method and establish a lithofacies-well logging database; Step 3: Select a key well with lithofacies and corresponding logging data as the model training well, and pre-process the training data using the SMOTE method to overcome the problem of insufficient model training caused by the imbalance of sample categories, so that the lithofacies data of each category are the same; Step 4: Input the lithofacies and logging data into the XGBoost model for training, and use the default model parameters to establish a lithofacies identification model for continental fine-grained sedimentary rocks.

2. The method for identifying terrestrial fine-grained sedimentary rock lithofacies logging based on SMOTE-XGBoost according to claim 1, characterized in that: In step 1, we first collect a relatively complete number of logging curves of core wells, realize logging curve splicing and standardization correction, and at the same time, in order to eliminate the influence of logging data of different dimensions, realize logging curve normalization, normalize the curve value in the interval [0,1]. The core depth may be inconsistent with the logging depth, and the core depth is corrected by observing the corresponding logging characteristics of the core.

3. The method for identifying terrestrial fine-grained sedimentary rock lithofacies logging based on SMOTE-XGBoost according to claim 2, characterized in that: The normalization formula is as follows: Where X norm is the normalized data, X i is the original data, X max , X min are the maximum and minimum values ​​of the logging data respectively.

4. The method for identifying terrestrial fine-grained sedimentary rock lithofacies logging based on SMOTE-XGBoost according to claim 1, characterized in that: Step 2 proposes a four-unit lithofacies division scheme of "lithology-structure-TOC-mineral": lithology is divided into three types: shale-siltstone-carbonate rock, and each axis is divided by 50%, further subdivided into three types: shale, siltstone, limestone or dolomite; primary structure is divided into three types: massive, laminar and layered; mineral composition is divided into three types: clay-felsic-carbonate, and each axis is divided by 50%, further subdivided into three types: clay, felsic and lime.

5. The method for identifying terrestrial fine-grained sedimentary rock lithofacies logging based on SMOTE-XGBoost according to claim 4, characterized in that: In "lithology-structure-TOC-mineral", fine-grained sediments are divided into three types according to the TOC value: high organic matter content, TOC value greater than 2%; medium organic matter content, TOC value 1% to 2%; low organic matter content, TOC value less than 1%.

6. The method for identifying terrestrial fine-grained sedimentary rock lithofacies logging based on SMOTE-XGBoost according to claim 4, characterized in that: The fine-grained sediments in the study area are divided into six types: A high-organic mud-grade laminated clay shale, B medium-organic mud-grade laminated felsic shale, C low-organic silt-containing laminated felsic shale, D layered siltstone, and E layered shell-clastic limestone.

7. The method for identifying terrestrial fine-grained sedimentary rock lithofacies logging based on SMOTE-XGBoost according to claim 1, characterized in that: In step 3, 70% of the data set is divided into a training set and 30% is divided into a test set to test the model validation effect, and then SMOTE processing is performed. SMOTE is an oversampling technology. i , first calculate its K-nearest neighbors of the same category KN_x i In SMOTE, K is usually set to 5. i A sample X2 is randomly selected from the sample X1 and the difference between the corresponding attribute j between the samples X1 and X2 is diff j =x 2j -x 1j , the specific formula is as follows: f 1j =x 1j +diff j ×rand[0,1]=x 1j +(x 2j -x 1j )×rand[0,1](2) Where f 1j are the newly generated minority class samples.

8. The method for identifying terrestrial fine-grained sedimentary rock lithofacies logging based on SMOTE-XGBoost according to claim 1, characterized in that: In step 4, XGBoost is a boosting integrated learning method that uses CART as the base learner. The specific formula is as follows: In the formula is the predicted value, x i is the i-th sample, f k is an independent function in the space, and F is the function space composed of all CARTs; The objective function formula is as follows: Where L is the loss function, is the error between the predicted value and the true value, and Ω is the regularization function, where the regularization function is as follows: Where T is the number of leaf nodes in each tree, w is the weight of each leaf, λ is the 2 regularization coefficient, and γ is the split threshold. The formula is as follows:

9. The method for identifying terrestrial fine-grained sedimentary rock lithofacies logging based on SMOTE-XGBoost according to claim 8, characterized in that: XGBoost is calculated by a greedy algorithm. Assuming that the node before the column is j, its contribution formula to the objective function is as follows: The contribution formula of the two child nodes to the objective function is as follows: At this time, the objective function can be changed as follows: Finally, the information gain formula obtained by each split of the objective function is as follows: Where GLGR is the information gain statistic of the first-order gradient of the left and right leaves, and HLHR is the information gain statistic of the second-order gradient of the left and right leaves.

Images