Logging identification method for Permian sepiolite-bearing formations.

Abstract

To provide an accurate prediction of sepiolite-bearing layer systems and a method for identifying logging of Permian sepiolite-bearing layer systems. [Solution] The method includes step S1 of data preprocessing, step S2 of setting a training set, a validation set, and a test set, step S3 of constructing a logging identification model, step S4 of using the CatBoost algorithm to perform a binary classification judgment on whether talc is present in a sepiolite-bearing layer system, step S5 of using the CatBoost algorithm to perform a multi-value classification judgment on the talc occurrence, step S6 of using the XGBoost algorithm to perform a regression prediction on the talc content of the stratum, step S7 of testing models using the validation set and the test set and selecting the model with the smallest loss function from the remaining models as the final model, and step S8 of using the final model to make predictions on Permian sepiolite-bearing layer systems awaiting identification.

Description

[Technical field]

[0001] The present invention relates to a method for identifying well logs for Permian sepiolite-bearing formations. [Background technology]

[0002] Carbonate rocks have the characteristics of autogenous self-containment and in-source accumulation, and their organic matter and pore growth are closely related to the sepiolite-bearing layer system. In unconventional oil and gas exploration, the sepiolite-bearing layer system has a very strong organic matter adsorption effect in the sepiolite-talc transition process, and is an important hydrocarbon source.

[0003] Well logging prediction can be summarized as a classification problem and a regression problem, and machine learning algorithms have been widely applied in well logging prediction, and good results have been obtained in the fields of lithology prediction, reservoir physical property evaluation and so on. The prediction of sepiolite-containing layer system belongs to the prediction of clay minerals, and its well logging prediction methods are relatively numerous, including the calculation of shale clay mineral content by low resistivity curve characteristics, the analysis of the distribution rule of clay mineral content by neural network model, the calculation of clay mineral distribution in shale containing a large amount of organic matter by natural gamma energy spectrum, the reverse analysis of clay mineral content based on weathering index, the comprehensive calculation of the type and content of strata clay minerals by sonic time difference-resistivity, well logging mineral sensitivity analysis, the combination of X-ray diffraction, and the combination of multiple regression analysis to build a mud shale clay mineral analysis model. At present, there is a lack of research on the well logging response characteristics of clay minerals such as sepiolite and talc, and the current well logging analysis method of clay minerals requires the combination of various means such as energy spectrum logging and XRD. In practical applications, logging data is mainly conventional logging data, and energy spectrum logging data is lacking. Conventional data focuses on interpreting the content of clay minerals, and it is difficult to distinguish the specific types of clay minerals. Therefore, establishing a workflow pattern suitable for logging prediction of meerschaum-bearing formation systems is of great significance to unconventional oil and gas exploration. At the geological level, talc was transformed from meerschaum, so meerschaum must have existed in the past where talc grows, so the presence of meerschaum can be determined by investigating talc. Summary of the Invention [Problem to be solved by the invention]

[0004] The objective of the present invention is to overcome the shortcomings of the prior art, establish an analysis prediction flow for sepiolite-bearing layer systems based on existing drilling data, make the prediction results consistent with the traditional TOC distribution of the first layer of the Maokou Formation, and provide an accurate prediction for sepiolite-bearing layer systems and a logging identification method for Permian sepiolite-bearing layer systems. [Means for solving the problem]

[0005] The object of the present invention is achieved by the following technical means: A method for identifying a logging well for a Permian sepiolite-bearing layer system includes: Step S1 is data preprocessing, which involves collecting logging data to be predicted, selecting six types of logging data, namely CNL, DEN, GR, RT, RXO, and AC curves, and performing MinMaxScaler normalization on the selected data; Step S2 of setting the well logging data into a training set, a validation set, and a test set in a ratio of 7:1:2; Step S3 of constructing a logging identification model, k and its corresponding talc tag Y k Set D:D={X k ,Y k}, where k=1,2,...,n, n is the total number of well logging data, and X k represents the kth logging curve data, and the logging data X k The feature dataset X consists of It is represented as JPEG0007618127000001.jpg3870, where m is the number of vertical samples. TIFF0007618127000002.tif1125, TIFF0007618127000003.tif43 represents the m-th sampling point of the k-th logging curve data, and the talc tag Y k includes whether talc is present in the stratum, the type of talc occurrence, and the talc content. TIFF0007618127000004.tif2662TIFF0007618127000005.tif54 indicates whether or not talc is present in the layer, TIFF0007618127000006.tif519, 0 represents the absence of talc in the layer, 1 represents the presence of talc in the layer, TIFF0007618127000007.tif54 shows the talc content of the stratum, TIFF0007618127000008.tif54 shows the talc occurrence in the stratum. TIFF0007618127000009.tif525, 0 represents patchy, 1 represents lenticular, 2 represents lamellar, Using machine learning algorithms to predict various tags of talc, Represented as TIFF0007618127000010.tif584, TIFF0007618127000011.tif48 shows the algorithms used to determine the presence or absence of talc, predict the talc content, and predict the occurrence of talc. TIFF0007618127000012.tif420 shows the loss function of the algorithm used to predict the presence or absence of talc, the talc content, and the talc occurrence. Step S4 of inputting the training set into the CatBoost model and performing a binary classification discrimination of whether or not talc is present in the sepiolite-bearing layer system using the CatBoost algorithm; Step S5: inputting the training set data determined to contain talc in step S4 into another CatBoost model and performing a multi-value classification of the talc occurrence using the CatBoost algorithm; Step S6 of inputting the training set data determined to contain talc in S4 into the XGBoost model and performing a regression prediction of the talc content of the stratum using the XGBoost algorithm; inputting the validation set data into the model trained in S4, sequentially performing binary classification of whether talc is present, multi-value classification of talc occurrence, and predicting the talc content, respectively judging whether the trained model has converged in S4, S5, and S6, adjusting the hyperparameters of the model and continuing training if the model has not converged and overfitting has not occurred, terminating training if the model has converged or overfitting has occurred, and inputting the test set data into the model to judge whether the model operates normally, and if so, keeping the model, or if not, excluding the model, and each time the training of each model is completed, calculating the loss function of the model, and selecting the model with the smallest loss function from the remaining models as the final model in step S7; Using the final model, a binary classification of the presence or absence of talc for the Permian sepiolite-bearing layer system awaiting identification, a multi-value classification of the talc occurrence, and a prediction of the talc content are performed, and the effective thickness of the sepiolite-bearing layer system with different occurrences is calculated using the talc content. JPEG0007618127000013.jpg1461n includes step S8, which represents the number of layers in the meerschaum layer system.

[0006] The cross-entropy function is used as the loss function of the CatBoost model in steps S4 and S5, and the cross-entropy calculation method is as follows: TIFF0007618127000014.tif857, where n is the total number of samples, class is the number of types, and y ij represents the tag that the i-th sample contains type j, and p ij represents the predicted value for type j of the i-th sample, The root mean square error (RMSE) is used as the loss function for the XGBoost algorithm in step S6, and its calculation formula is: TIFF0007618127000015.tif932, and f(x i ) is the i-th sample x i represents the predicted value of Y i represents the true value of the i-th sample.

[0007] The beneficial effects of the present invention are as follows: the present invention establishes the analysis prediction flow of sepiolite-bearing layer system based on the existing drilling data, and the prediction result is consistent with the traditional TOC distribution of the first member of the Maokou Formation; it can realize accurate prediction of sepiolite-bearing layer system, which is of great significance to the practical application of unconventional exploration. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] China's meerschaum clay ore has its own deposits, and the future prospects of meerschaum clay mineral resources are very high, with hydrothermal and sedimentary types all having a certain future mineralization potential. Therefore, this study thoroughly compared commonly used machine learning algorithms, and then comprehensively compared the existence of meerschaum-bearing layer systems (binary classification problem), talc occurrence (multiple classification problem) and talc content (regression problem), and established a workflow model applicable to well logging prediction of meerschaum-bearing layer systems. The following further describes the technical proposal of this application.

[0009] Abbreviations and definitions of key terms: Meerschaum is SiO 2 and Mg 2+ Contains a large amount of Al 3+ Meerschaum is a clay mineral that contains small amounts of sepiolite. Modern sepiolite is mainly distributed in the mid-ocean ridges and abyssal plains of the Atlantic and Indian Oceans, and in low-lying continental lake basins of hypersaline lakes. Meerschaum deposits have important industrial value, and most of them occur in terrestrial sedimentary strata of the Neogene. However, the origin, occurrence and significance of sepiolite-talc in oil and gas geology have not yet been studied in depth. For the problem of logging prediction of sepiolite-bearing formations, based on geological data such as rock cores, thin sections, XRD and logging wells, CNL, DEN, GR, RT, RXO and AC logging curves are selected as features based on sensitivity analysis, and the SMOTE algorithm is used to process the balance difference of features. After comparing the performance differences of CatBoost, XGBoost, LightGBM, random forest, ANN, SVM, KNN and Bayesian models in the determination, classification and content prediction of sepiolite-bearing formations, a prediction flow of sepiolite-bearing formations is proposed.

[0010] The method for identifying a logging well for a Permian sepiolite-bearing layer system according to the present invention comprises the steps of: Step S1 is data preprocessing, which involves collecting logging data to be predicted, selecting six types of logging data, namely CNL, DEN, GR, RT, RXO, and AC curves, and performing MinMaxScaler normalization on the selected data; Step S2 of setting the well logging data into a training set, a validation set, and a test set in a ratio of 7:1:2; Step S3 of constructing a logging identification model, k and its corresponding talc tag Y k Set D:D={X k ,Y k}, where k=1,2,...,n, n is the total number of well logging data, and X k represents the kth logging curve data, and the logging data X k The feature dataset X consists of It is represented as JPEG0007618127000016.jpg3870, where m represents the number of vertical samples. TIFF0007618127000017.tif1125, TIFF0007618127000018.tif43 represents the m-th sampling point of the k-th logging curve data, Talc Tag Y k includes whether talc is present in the stratum, the type of talc occurrence, and the talc content. TIFF0007618127000019.tif2662TIFF0007618127000020.tif54 indicates whether or not talc is present in the layer, TIFF0007618127000021.tif519, 0 represents the absence of talc in the stratum, 1 represents the presence of talc in the stratum, TIFF0007618127000022.tif54 shows the talc content of the stratum, TIFF0007618127000023.tif54 shows the talc occurrence in the stratum. TIFF0007618127000024.tif525, 0 represents patchy, 1 represents lenticular, 2 represents lamellar, Using machine learning algorithms to predict various tags of talc, TIFF0007618127000025.tif584 Among them, TIFF0007618127000026.tif48 shows the algorithms used to determine the presence or absence of talc, predict the talc content, and predict the occurrence of talc. TIFF0007618127000027.tif418 is TIFF0007618127000028.tif48 represents a function whose input is m-dimensional real numbers and whose output is 1-dimensional real numbers. TIFF0007618127000029.tif420 represents the loss function of the algorithm used to determine the presence or absence of talc, predict the talc content, and predict the occurrence of talc, i.e., the cross-entropy loss function or root mean square error (RMSE) below. The above equation represents the selection of the model with the smallest loss function for each of the three models for determining the presence or absence of talc, predicting the talc content, and determining the occurrence of talc.

[0011] Step S4 of inputting the training set into the CatBoost model and performing a binary classification discrimination of whether or not talc is present in the sepiolite-bearing layer system using the CatBoost algorithm; Step S5: inputting the training set data determined to contain talc in step S4 into another CatBoost model and performing a multi-value classification of the talc occurrence using the CatBoost algorithm; The cross-entropy function is used as the loss function of the CatBoost model in steps S4 and S5, and the cross-entropy calculation method is as follows: TIFF0007618127000030.tif857, where n is the total number of samples, class is the number of types, and y ij represents the tag that the i-th sample contains type j, and p ij represents the predicted value for type j of the i-th sample, Step S6, in which the training set data determined to contain talc in S4 is input to the XGBoost model, and a regression prediction is made for the talc content of the stratum using the XGBoost algorithm, and the root mean square error (RMSE) is used as the loss function of the XGBoost algorithm, and the calculation formula is: TIFF0007618127000031.tif932, and f(x i ) is the i-th sample x i represents the predicted value of Y i step S6, where represents the true value of the i-th sample; inputting the validation set data into the model trained in S4, sequentially performing binary classification of whether talc is present, multi-value classification of talc occurrence, and predicting the talc content, respectively judging whether the trained model has converged in S4, S5, and S6, adjusting the hyperparameters of the model and continuing training if the model has not converged and overfitting has not occurred, terminating training if the model has converged or overfitting has occurred, and inputting the test set data into the model to judge whether the model operates normally, and if so, keeping the model, or if not, excluding the model, and each time the training of each model is completed, calculating the loss function of the model, and selecting the model with the smallest loss function from the remaining models as the final model in step S7; Using the final model, a binary classification of the presence or absence of talc for the Permian sepiolite-bearing layer system awaiting identification, a multi-value classification of the talc occurrence, and a prediction of the talc content are performed, and the effective thickness of the sepiolite-bearing layer system with different occurrences is calculated using the talc content. JPEG0007618127000032.jpg1461n includes step S8, which represents the number of layers in the meerschaum layer system.

[0012] Those skilled in the art should understand that the embodiments described herein are intended to help the reader understand the principles of the present invention, and should understand that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can create various other specific modifications and combinations that do not deviate from the essence of the present invention according to the technical suggestions disclosed in the present invention, and these modifications and combinations are within the scope of protection of the present invention.

Claims

1. A method for identifying a logging well for a Permian sepiolite-bearing formation, comprising: Step S1 is a data pre-processing step in which logging data to be predicted is collected, six types of logging data, namely CNL, DEN, GR, RT, RXO, and AC curves, are selected from the collected data, and the selected data is subjected to MinMaxScaler normalization processing; Step S2 of setting the well logging data into a training set, a validation set, and a test set in a ratio of 7:1:2; Step S3 of constructing a logging identification model, ｋ and the corresponding talc tag Y ｋ Set D: D = {X ｋ , Y ｋ }, k=1, 2, . . . , n, n is the total number of well logging data, where X ｋ represents the k-th logging curve data, and the logging data X ｋ The feature dataset X is composed of where m represents the number of samples in the vertical direction. 、 represents the m-th sampling point of the k-th logging curve data; Talc Tag Y ｋ includes whether talc is present in the stratum, the type of talc occurrence, and the talc content. indicates whether talc is present in the layer, , 0 represents the absence of talc in the stratum, 1 represents the presence of talc in the stratum, represents the talc content of the formation, indicates the talc occurrence in the stratum, , 0 represents patchy, 1 represents lenticular, and 2 represents lamellar; Using machine learning algorithms to predict various tags of talc, is expressed as represents the algorithm used to determine the presence or absence of talc, predict the talc content, and predict the occurrence of talc, step S3, where S represents the loss function of the algorithm used to predict the presence or absence of talc, the talc content, and the talc occurrence; Step S4: inputting the training set into the CatBoost model and performing a binary classification judgment on whether or not talc is present in the sepiolite-bearing layer system using the CatBoost algorithm; Step S5: inputting the training set data determined to contain talc in step S4 into another CatBoost model and performing a multi-value classification of the talc occurrence using the CatBoost algorithm; Step S6 of inputting the training set data determined to contain talc in step S4 into the XGBoost model and performing a regression prediction of the talc content of the stratum using the XGBoost algorithm; inputting the validation set data into the model trained in S4, sequentially performing binary classification judgment of whether talc is present or not, multi-value classification judgment of talc occurrence, and prediction of talc content, respectively judging whether the trained models in S4, S5, and S6 have converged, respectively, adjusting the hyperparameters of the model and continuing training if the model has not converged and overfitting has not occurred, terminating training if the model has converged or overfitting has occurred, and inputting the test set data into the model to judge whether the model operates normally, and if so, keeping the model, otherwise excluding the model, and each time the training of each model is completed, calculating the loss function of the model, and selecting the model with the smallest loss function from the remaining models as the final model in step S7; Using the final model, a binary classification of the presence or absence of talc for the Permian sepiolite-bearing layer system awaiting identification, a multi-value classification of the talc occurrence, and a prediction of the talc content are performed, and the effective thickness of the sepiolite-bearing layer system with different occurrences is calculated using the talc content. Step S8, where n represents the number of layers in the sepiolite layer system; A method for identifying a logging well for a Permian sepiolite-bearing layer system, comprising:

2. A cross-entropy function is used as the loss function of the CatBoost model in steps S4 and S5, and the cross-entropy calculation method is as follows: where n represents the total number of samples, class represents the number of types, and y ｉｊ represents the tag in which the i-th sample contains type j, and p ｉｊ represents the predicted value of the i-th sample for type j, The root mean square error (RMSE) is used as the loss function of the XGBoost algorithm in step S6, and the calculation formula is and f(x ｉ ) is the i-th sample x ｉ represents the predicted value of ｉ represents the true value of the i-th sample.

2. The method for identifying a logging well for a Permian sepiolite-bearing formation according to claim 1.

Images