A well logging algorithm identification method for Permian sepiolite-bearing strata
By applying machine learning algorithms on drilling data, an analysis and prediction process for sepiolite-containing layer systems was established, which solved the shortcomings of prediction of sepiolite-containing layer systems in the existing technology, and achieved accurate prediction and TOC distribution prediction consistent with traditional methods, which was of great significance to unconventional oil and gas exploration.
Patent Information
- Application Number
- CN202311162217.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-11
- Publication Date
- 2025-04-18
- Estimated Expiration
- 2043-09-11
AI Technical Summary
The prior art is difficult to accurately predict sepiolite-containing layer systems, especially in the absence of energy spectrum logging data, and conventional logging data is difficult to distinguish specific clay mineral types.
Through machine learning algorithms based on existing drilling data, an analysis and prediction process for sepiolite-containing layer systems is established, and the existence, production and content of talc is predicted using CatBoost, XGBoost and other models to achieve accurate prediction of sepiolite-containing layer systems.
The TOC distribution prediction consistent with the traditional method is achieved, and the sepiolite-containing layer system can be accurately predicted, which is of great significance to unconventional oil and gas exploration.
Smart Images
Figure BDA0004440852310000021 
Figure BDA0004440852310000024 
Figure BDA0004440852310000031
Abstract
Description
Technical Field
[0001] The invention relates to a well logging algorithm identification method for Permian sepiolite-containing strata. Background Art
[0002] Carbonate rocks have the characteristics of self-generation and self-storage, and the accumulation of hydrocarbons in the source, and the development of organic matter and pores is closely related to the sepiolite-containing strata. In unconventional oil and gas exploration, the sepiolite-containing strata have a strong organic matter adsorption effect during the sepiolite-talc transformation process and are an important hydrocarbon source.
[0003] Well logging prediction can be summarized as classification and regression problems. Machine learning algorithms have been widely used in well logging prediction and have achieved good results in the fields of lithology prediction and reservoir physical property evaluation. The prediction of sepiolite-bearing strata belongs to clay mineral prediction. There are many logging prediction methods, including calculating the clay mineral content of shale based on low resistivity curve characteristics, analyzing the distribution law of clay mineral content by neural network model, calculating the distribution of clay minerals in organic-rich shales by natural gamma ray spectrum, inverting clay mineral content based on alteration index, calculating the type and content of formation clay minerals by acoustic time difference-resistivity, analyzing the sensitivity of well logging minerals, and establishing a mudstone clay mineral analysis model by combining X-diffraction and multivariate regression analysis. At present, there is a lack of research on the logging response characteristics of clay minerals such as sepiolite and talc, and the current clay mineral logging analysis method needs to combine multiple means such as spectral logging and XRD. In practical applications, conventional logging data is mainly used, and spectral logging data is lacking. Conventional data tend to interpret clay mineral content and it is difficult to distinguish specific clay mineral types. Therefore, establishing a workflow model suitable for logging prediction of sepiolite-containing strata is of great significance for unconventional oil and gas exploration. At the geological level, talc evolved from sepiolite, so sepiolite must have existed in places where talc developed. Therefore, the existence of sepiolite can be determined by testing talc. Summary of the invention
[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a logging algorithm identification method for the Permian sepiolite-containing strata, which establishes an analysis and prediction process for sepiolite-containing strata based on existing drilling data, and the prediction result can be consistent with the traditional Maoyi Formation TOC distribution, and can achieve accurate prediction of the sepiolite-containing strata.
[0005] The object of the present invention is achieved through the following technical scheme: A well logging identification method for the Permian sepiolite-containing strata, comprising the following steps:
[0006] S1. Data preprocessing: Collect the well logging data to be predicted, select six types of well logging data, CNL, DEN, GR, RT, RXO, and AC curves, and perform MinMaxScaler normalization on the selected data;
[0007] S2, set the logging data as training set, validation set, and test set in a ratio of 7:1:2;
[0008] S3. Establishing a well logging recognition model: transform the well logging data X k and its corresponding talc label Y k Construct set D: D = {X k ,Y k}, k = 1, 2, ..., n, n is the total number of logging data; where X k Represents the kth well logging curve data, which is composed of the well logging data X k The constructed feature data set X is expressed as:
[0009]
[0010] m represents the number of vertical sampling, Indicates the mth sampling point of the kth logging curve data;
[0011] Talc Label Y k It includes whether there is talc in the formation, the type of talc occurrence and the talc content, expressed as:
[0012]
[0013] Indicates whether there is talc in the stratum, 0 means that there is no talc in the formation, and 1 means that there is talc in the formation; Indicates the talc content of the formation; Indicates the occurrence of talc in the formation. 0 means macular, 1 means lenticular, and 2 means lamellar;
[0014] The machine learning algorithm is used to predict various labels of talc, which can be expressed as:
[0015]
[0016] in, Indicates whether talc exists, the algorithm used to predict the talc content and the talc occurrence. Indicates the loss function of the algorithm used to predict the presence of talc, talc content, and talc occurrence;
[0017] S4, inputting the training set data into the CatBoost model, and using the CatBoost algorithm to make a binary classification judgment on whether talc exists in the sepiolite-containing strata;
[0018] S5, inputting the training set data judged as having talc in S4 into another CatBoost model, and using the CatBoost algorithm to perform multi-classification judgment on the occurrence of talc;
[0019] S6, inputting the training set data determined by S4 as having talc into the XGBoost model, and using the XGBoost algorithm to perform regression prediction on the talc content of the formation;
[0020] S7, input the validation set data into the model trained in S4, and perform binary classification judgment on the presence of talc, multi-classification judgment on the occurrence of talc, and prediction of talc content in turn, and respectively judge whether the models trained in S4, S5, and S6 have converged. If the model has not converged and there is no overfitting, adjust the model hyperparameters and continue training; if the model converges or overfitting occurs, end the training; then input the test set data into the model to judge whether the model is running normally, if so, retain the model, otherwise discard the model; after each model training is completed, calculate the model's loss function, and select the model with the smallest loss function from the retained models as the final model;
[0021] S8. The final model is used to make a binary judgment on the presence of talc, a multi-classification judgment on the occurrence of talc, and a prediction of the talc content for the Permian sepiolite-bearing strata to be identified. The talc content is used to calculate the effective thickness of sepiolite-bearing strata with different occurrences:
[0022]
[0023] n represents the number of layers in the sepiolite system.
[0024] The loss function of the CatBoost model in steps S4 and S5 adopts the cross entropy function, and the cross entropy calculation method is:
[0025]
[0026] Among them, n represents the total number of samples, class represents the number of categories, and y ij Indicates that the i-th sample contains the label of category j, p ij Represents the predicted value of the i-th sample for category j;
[0027] The loss function of the XGBoost algorithm in step S6 uses the root mean square error RMSE, which is calculated as follows:
[0028]
[0029] f(x i ) represents the i-th sample x i The predicted value, Y i Represents the true value of the i-th sample.
[0030] The beneficial effects of the present invention are as follows: the present invention establishes a process for analyzing and predicting the sepiolite-containing strata based on existing drilling data, and the predicted results can be consistent with the traditional Mao-1 TOC distribution. It can achieve accurate prediction of the sepiolite-containing strata and has far-reaching significance for the practical application of unconventional exploration. DETAILED DESCRIPTION
[0031] my country's sepiolite clay deposits have their own unique deposits, and the prospect of sepiolite clay mineral resources is extremely impressive. Whether it is hydrothermal or sedimentary, it has certain mineralization prospects. Therefore, this study, based on a comprehensive comparison of commonly used machine learning algorithms, comprehensively compares the existence of sepiolite-containing strata (binary classification problem), talc occurrence (multi-classification problem) and talc content (regression problem), and establishes a workflow model suitable for logging prediction of sepiolite-containing strata. The technical solution of the present invention is further described below.
[0032] Abbreviations and Definitions of Key Terms: Sepiolite is a SiO2 and Mg-rich 2+ , poor Al 3+ Clay minerals, modern sepiolite is mainly distributed in the mid-ocean ridges and deep-sea plains of the Atlantic and Indian Oceans and the low terrigenous basin centers of hypersaline lakes. Sepiolite deposits have important industrial value and are mostly produced in Neogene continental sedimentary strata. However, the genesis, occurrence and oil and gas geological significance of sepiolite-talc are not yet in-depth. Aiming at the problem of well logging prediction of sepiolite-containing strata, based on geological data such as cores, thin sections, XRD, and logging, CNL, DEN, GR, RT, RXO, and AC logging curves were selected as features based on sensitivity analysis. After using the SMOTE algorithm to process the difference in feature balance, the performance differences of CatBoost, XGBoost, LightGBM, random forest, ANN, SVM, KNN and Bayesian models in the determination, classification and content prediction of sepiolite-containing strata were compared, and a prediction process for sepiolite-containing strata was proposed.
[0033] A well logging identification method for the Permian sepiolite-containing strata of the present invention comprises the following steps:
[0034] S1. Data preprocessing: Collect the well logging data to be predicted, select six types of well logging data, CNL, DEN, GR, RT, RXO, and AC curves, and perform MinMaxScaler normalization on the selected data;
[0035] S2, set the logging data as training set, validation set, and test set in a ratio of 7:1:2;
[0036] S3. Establishing a well logging recognition model: transform the well logging data X k and its corresponding talc label Y kConstruct set D: D = {X k ,Y k}, k = 1, 2, ..., n, n is the total number of logging data; where X k Represents the kth well logging curve data, which is composed of the well logging data X k The constructed feature data set X is expressed as:
[0037]
[0038] m represents the number of vertical sampling, Indicates the mth sampling point of the kth logging curve data;
[0039] Talc Label Y k It includes whether there is talc in the formation, the type of talc occurrence and the talc content, expressed as:
[0040]
[0041] Indicates whether there is talc in the stratum. 0 means that there is no talc in the formation, and 1 means that there is talc in the formation; Indicates the talc content of the formation; Indicates the occurrence of talc in the formation. 0 means macular, 1 means lenticular, and 2 means lamellar;
[0042] The machine learning algorithm is used to predict various labels of talc, which can be expressed as:
[0043]
[0044] in, Indicates whether talc exists, the algorithm used to predict the talc content and the talc occurrence. express It is a function whose input is an m-dimensional real number and whose output is a one-dimensional real number; The loss function of the algorithm used to predict whether talc exists, talc content, and talc occurrence is represented by the following cross entropy loss function or root mean square error RMSE; the above formula indicates that among the three models of whether talc exists, talc content prediction, and talc occurrence judgment, the model with the smallest loss function is selected.
[0045] S4, inputting the training set data into the CatBoost model, and using the CatBoost algorithm to make a binary classification judgment on whether talc exists in the sepiolite-containing strata;
[0046] S5, inputting the training set data judged as having talc in S4 into another CatBoost model, and using the CatBoost algorithm to perform multi-classification judgment on the occurrence of talc;
[0047] The loss function of the CatBoost model in steps S4 and S5 adopts the cross entropy function, and the cross entropy calculation method is:
[0048]
[0049] Among them, n represents the total number of samples, class represents the number of categories, and y ij Indicates that the i-th sample contains the label of category j, p ij Represents the predicted value of the i-th sample for category j;
[0050] S6. Input the training set data that S4 determined to contain talc into the XGBoost model, and use the XGBoost algorithm to perform regression prediction on the talc content of the formation; the loss function of the XGBoost algorithm uses the root mean square error RMSE, and its calculation formula is as follows:
[0051]
[0052] f(x i ) represents the i-th sample x i The predicted value, Y i Represents the true value of the i-th sample.
[0053] S7, input the validation set data into the model trained in S4, and perform binary classification judgment on the presence of talc, multi-classification judgment on the occurrence of talc, and prediction of talc content in turn, and respectively judge whether the models trained in S4, S5, and S6 have converged. If the model has not converged and there is no overfitting, adjust the model hyperparameters and continue training; if the model converges or overfitting occurs, end the training; then input the test set data into the model to judge whether the model is running normally, if so, retain the model, otherwise discard the model; after each model training is completed, calculate the model's loss function, and select the model with the smallest loss function from the retained models as the final model;
[0054] S8. The final model is used to make a binary judgment on the presence of talc, a multi-classification judgment on the occurrence of talc, and a prediction of the talc content for the Permian sepiolite-bearing strata to be identified. The talc content is used to calculate the effective thickness of sepiolite-bearing strata with different occurrences:
[0055]
[0056] n represents the number of layers in the sepiolite system.
[0057] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.
Claims
1. A well logging identification method for Permian sepiolite-containing strata, characterized in that: The following steps are involved: S1. Data preprocessing: Collect the well logging data to be predicted, select six types of well logging data, CNL, DEN, GR, RT, RXO, and AC curves, and perform MinMaxScaler normalization on the selected data; S2, set the logging data as training set, validation set, and test set in a ratio of 7:1:2; S3. Establishing a well logging recognition model: transform the well logging data X k and its corresponding talc label Y k Construct set D: D={X k ,Y k }, k = 1, 2, ..., n, n is the total number of logging data; where X k Represents the kth well logging curve data, which is composed of the well logging data X k The constructed feature data set X is expressed as: m represents the number of vertical sampling, Indicates the mth sampling point of the kth logging curve data; Talc Label Y k It includes whether there is talc in the formation, the type of talc occurrence and the talc content, expressed as: Y1 T Indicates whether there is talc in the stratum, Y1 T ∈[0, 1], 0 means there is no talc in the formation, 1 means there is talc in the formation; Indicates the talc content of the formation; Indicates the occurrence of talc in the formation. 0 means macular, 1 means lenticular, and 2 means lamellar; The machine learning algorithm is used to predict various labels of talc, which can be expressed as: Among them, F i j (x) indicates whether talc exists, the algorithm used to predict the talc content and the talc occurrence, Indicates the loss function of the algorithm used to predict the presence of talc, talc content, and talc occurrence; S4, inputting the training set data into the CatBoost model, and using the CatBoost algorithm to make a binary classification judgment on whether talc exists in the sepiolite-containing strata; S5, inputting the training set data judged as having talc in S4 into another CatBoost model, and using the CatBoost algorithm to perform multi-classification judgment on the occurrence of talc; S6, inputting the training set data determined by S4 as having talc into the XGBoost model, and using the XGBoost algorithm to perform regression prediction on the talc content of the formation; S7, input the validation set data into the model trained in S4, and perform binary classification judgment on the presence of talc, multi-classification judgment on the occurrence of talc, and prediction of talc content in turn, and respectively judge whether the models trained in S4, S5, and S6 have converged. If the model has not converged and there is no overfitting, adjust the model hyperparameters and continue training; if the model converges or overfitting occurs, end the training; then input the test set data into the model to judge whether the model is running normally, if so, retain the model, otherwise discard the model; after each model training is completed, calculate the model's loss function, and select the model with the smallest loss function from the retained models as the final model; S8. The final model is used to make a binary judgment on the presence of talc, a multi-classification judgment on the occurrence of talc, and a prediction of the talc content for the Permian sepiolite-bearing strata to be identified. The talc content is used to calculate the effective thickness of sepiolite-bearing strata with different occurrences: n represents the number of layers in the sepiolite system.
2. A well logging identification method for Permian sepiolite-containing strata according to claim 1, characterized in that: The loss function of the CatBoost model in steps S4 and S5 adopts the cross entropy function, and the cross entropy calculation method is: Among them, n represents the total number of samples, class represents the number of categories, and y ij Indicates that the i-th sample contains the label of category j, p ij Represents the predicted value of the i-th sample for category j; The loss function of the XGBoost algorithm in step S6 uses the root mean square error RMSE, which is calculated as follows: f(x i ) represents the i-th sample x i The predicted value, Y i Represents the true value of the i-th sample.