Shear wave time difference prediction method and device
By combining a hybrid model of CNN and LSTM neural networks to process and group logging data, the problem of missing shear wave logging data was solved, achieving high-precision and efficient shear wave time difference prediction and improving the accuracy of reservoir prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PETROCHINA CO LTD
- Filing Date
- 2022-08-26
- Publication Date
- 2026-07-07
AI Technical Summary
Existing technologies suffer from problems such as missing or poor-quality shear wave logging data, resulting in low prediction accuracy, poor regional applicability, and low computational efficiency.
A machine learning-based approach was adopted, using a neural network constructed by combining CNN and LSTM, along with importance analysis, kurtosis and skewness processing of well logging data, to build a shear wave time difference prediction model, perform data filtering and grouping, and improve prediction accuracy and efficiency.
It enables accurate prediction of shear wave time difference, improves the accuracy of rock physics analysis, lithology identification, reservoir description and fluid identification, and reduces computational complexity and time cost.
Smart Images
Figure CN117669785B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of petroleum exploration and development technology, specifically to a shear wave time difference prediction method, a shear wave time difference prediction device, an electronic device, and a machine-readable storage medium. Background Technology
[0002] Shear wave logging data is one of the important parameters used in rock physical analysis, lithology identification, calculation of rock elasticity parameters, reservoir description, and fluid identification, playing a crucial role in improving reservoir prediction accuracy. Conventional sonic logging can obtain both P-wave and S-wave data, but the quality of the obtained S-wave data is poor or incomplete, insufficient to meet production needs. Dipole sonic logging instruments can obtain better quality S-wave data, but the acquisition cost is high, and data is only collected in key wells or high-risk exploration wells; most wells lack S-wave logging data. Well conditions, logging technology, and cost are the main reasons for the lack of data, making accurate S-wave prediction particularly important.
[0003] Common methods for predicting shear waves include empirical formulas and rock physics models. Empirical formulas calculate shear waves by analyzing the relationship between longitudinal and shear waves and obtaining a fitted linear formula. This method is simple and convenient, and can quickly predict shear waves, but the accuracy of shear wave prediction using empirical formulas is low, and it suffers from poor regional applicability. Rock physics models construct rock skeleton models and fluid parameter models, and calculate shear waves from these models. This method can accurately predict shear waves, but the model requires many accurate parameters, such as rock mineral composition, porosity, and pore structure. These parameters are numerous and difficult to collect, making it difficult to establish an accurate rock physics model, and the computational efficiency is low. In summary, both empirical formulas and rock physics models have certain limitations. Therefore, this application proposes a prediction method based on machine learning. Summary of the Invention
[0004] The purpose of this invention is to provide a method and apparatus for predicting transverse wave time difference, which solves the problems of low prediction accuracy, poor regional applicability, and low computational efficiency of the above-mentioned methods.
[0005] To achieve the above objectives, embodiments of the present invention provide a method for predicting transverse wave time difference, comprising:
[0006] Obtain well logging sample data as the training dataset for the prediction model;
[0007] The training dataset is preprocessed, data is filtered based on importance analysis, and data is grouped based on kurtosis and skewness to obtain the processed training dataset.
[0008] The processed training dataset is then input into a neural network constructed by combining CNN and LSTM to train the transverse wave time difference prediction model.
[0009] Obtain logging data for the predicted shear wave time difference;
[0010] The well logging data is preprocessed and grouped based on kurtosis and skewness to obtain processed well logging data;
[0011] The processed well logging data are used as inputs to the shear wave time difference prediction model to obtain the shear wave time difference.
[0012] Optionally, the CNN neural network and LSTM neural network in the shear wave time difference prediction model are connected through a Dropout layer.
[0013] Optionally, the logging data includes: natural gamma logging data, caliper logging data, spontaneous potential logging data, resistivity logging data, neutron logging data, sonic logging data, and density logging data.
[0014] Optionally, the training dataset is preprocessed, data is filtered based on importance analysis, and data is grouped based on kurtosis and skewness to obtain a processed training dataset, including:
[0015] The training dataset is cleaned, filtered, and normalized to obtain the first training data.
[0016] Optionally, the training dataset is preprocessed, filtered based on importance analysis, and grouped based on kurtosis and skewness to obtain a processed training dataset, further including:
[0017] Data with a correlation coefficient greater than a first preset coefficient value that is selected from the first training data are used as the second training data;
[0018] Calculate the correlation coefficient between the two different types of data in the second training data respectively;
[0019] If there are two different types of data with a correlation coefficient greater than the second preset coefficient value, then select one type of data from the two different types of data, and use the other different types of data in the second training data with a correlation coefficient less than or equal to the second preset coefficient value as the third training data.
[0020] Optionally, the training dataset is preprocessed, filtered based on importance analysis, and grouped based on kurtosis and skewness to obtain a processed training dataset, further including:
[0021] Based on the preset kurtosis coefficient and preset skewness coefficient, the third training data is divided into at least two sets of logging data, which are used as the processed training data.
[0022] Optionally, the correlation coefficient is calculated using the Pearson correlation coefficient formula.
[0023] This invention also provides a transverse wave time difference prediction device, comprising:
[0024] The training data acquisition module is used to acquire well logging sample data as the training dataset for the prediction model.
[0025] The first data processing module is used to preprocess the training dataset, filter data based on importance analysis, and group data based on kurtosis and skewness to obtain the processed training dataset.
[0026] The model training module is used to input the processed training dataset into a neural network built by combining CNN and LSTM to train the model and obtain the transverse wave time difference prediction model.
[0027] The input data acquisition module is used to acquire well logging data of the shear wave time difference to be predicted;
[0028] The second data processing module is used to preprocess the logging data, group the logging data based on kurtosis and skewness, and obtain the processed logging data.
[0029] The result output module is used to take the processed well logging data as input to the shear wave time difference prediction model to obtain the shear wave time difference.
[0030] This invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described transverse wave time difference prediction method.
[0031] On the other hand, the present invention provides a machine-readable storage medium storing instructions for causing a machine to perform the above-described transverse wave time difference prediction method.
[0032] This technical solution combines a neural network built with CNN and LSTM to construct a shear wave time difference prediction model. The well logging data to be predicted is preprocessed and grouped based on kurtosis and skewness before being input into the shear wave time difference prediction model to obtain the shear wave time difference. The calculation is simple, highly practical, and can accurately predict shear wave time difference. It can provide necessary parameters for rock physics analysis, lithology identification, calculation of rock elastic mechanical parameters, reservoir description, and fluid identification.
[0033] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description
[0034] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:
[0035] Figure 1 This is a flowchart illustrating the shear wave time difference prediction method provided by the present invention;
[0036] Figure 2 This is a schematic diagram of the structure of the shear wave time difference prediction model provided by the present invention;
[0037] Figure 3 This is a schematic diagram showing the positions of different peaks provided by the present invention;
[0038] Figure 4 This is a schematic diagram of positions with different degrees of skewness provided by the present invention;
[0039] Figure 5 This is a schematic diagram of the transverse wave time difference prediction device provided by the present invention;
[0040] Figure 6 This is a schematic diagram comparing the transverse wave time difference obtained by this solution with the transverse wave time difference in the prior art.
[0041] Explanation of reference numerals in the attached figures
[0042] 10 - Training data acquisition module; 20 - First data processing module;
[0043] 30 - Model training module; 40 - Input parameter acquisition module;
[0044] 50 - Second data processing module; 60 - Result output module. Detailed Implementation
[0045] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.
[0046] In the embodiments of the present invention, unless otherwise stated, directional terms such as "up," "down," "left," and "right" generally refer to the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use.
[0047] The terms “first,” “second,” “third,” etc., are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.
[0048] Furthermore, terms such as "approximately" and "basically" are used to indicate that the relevant content does not require absolute precision, but rather allows for a certain degree of deviation. For example, "approximately equal" does not simply mean absolute equality; because absolute "equality" is difficult to achieve in actual production and operation, a certain degree of deviation is generally present. Therefore, in addition to absolute equality, "approximately equal to" also includes the aforementioned situation where a certain degree of deviation exists. Using this as an example, in other cases, unless otherwise specified, terms such as "approximately" and "basically" have similar meanings as described above. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0049] Figure 1 This is a flowchart illustrating the shear wave time difference prediction method provided by the present invention; Figure 2 This is a schematic diagram of the structure of the shear wave time difference prediction model provided by the present invention; Figure 3 This is a schematic diagram showing the positions of different peaks provided by the present invention; Figure 4 This is a schematic diagram of positions with different degrees of skewness provided by the present invention; Figure 5 This is a schematic diagram of the transverse wave time difference prediction device provided by the present invention; Figure 6 This is a schematic diagram comparing the transverse wave time difference obtained by this solution with the transverse wave time difference in the prior art.
[0050] like Figure 1 The present embodiment provides a method for predicting transverse wave time difference, including:
[0051] Step 101: Obtain well logging sample data as the training dataset for the prediction model;
[0052] Step 102: Preprocess the training dataset, filter the data based on importance analysis, and group the data based on kurtosis and skewness to obtain the processed training dataset;
[0053] Step 103: Input the processed training dataset into the neural network built by the combination of CNN and LSTM to train it and obtain the transverse wave time difference prediction model.
[0054] Step 104: Obtain well logging data for the predicted shear wave time difference;
[0055] Step 105: Preprocess the logging data and group the logging data based on kurtosis and skewness to obtain processed logging data;
[0056] Step 106: Use the processed logging data as input to the shear wave time difference prediction model to obtain the shear wave time difference.
[0057] Specifically, in step 101, the well logging sample data needs to be processed, including: preprocessing, data filtering based on importance analysis, and data grouping based on kurtosis and skewness to ensure that the data format remains consistent, facilitating machine learning and enabling accurate identification of the shear wave time difference prediction model data, thus achieving accurate prediction of the shear wave time difference. In step 105, the well logging data is preprocessed and grouped based on kurtosis and skewness to obtain at least two sets of processed well logging data. Using these at least two sets of processed well logging data as inputs to the shear wave time difference prediction model can yield more accurate shear wave time differences and reduce the computational load in the prediction process, thereby improving efficiency. Furthermore, in this embodiment, the method steps for preprocessing the well logging data to be predicted for shear wave time difference and grouping the well logging data based on kurtosis and skewness are similar to those for preprocessing the training dataset and grouping the data based on kurtosis and skewness, and will not be repeated here.
[0058] Furthermore, the CNN neural network and LSTM neural network in the shear wave time difference prediction model are connected through a Dropout layer.
[0059] Specifically, such as Figure 2 As shown, in this embodiment, the transverse wave time difference prediction model is trained using a neural network constructed from a hybrid CNN and LSTM. The CNN and LSTM neural networks are connected via a Dropout layer, a structure used to reduce overfitting in the neural network. A CNN, or Convolutional Neural Network, is a type of neural network, specifically a feedforward neural network. Its weight-sharing structure makes it more similar to biological neural networks, reducing the complexity of the network model and the number of weights. The CNN model structure includes three layers: convolution, pooling, and fully connected layers. Its artificial neurons can respond to a portion of the surrounding units within their coverage area, thus considering local features of the data. Convolutional layers perform convolution on the input data to reduce the number of parameters and connections, thereby significantly reducing the number of model iterations and iteration time. Pooling layers, also known as downsampling layers, are commonly used components in convolutional neural networks. They mainly reduce the dimensionality of the data, remove redundant information, compress features, simplify network complexity, and facilitate neural network learning. Fully connected layers usually appear in the last few layers and are used to perform a weighted sum of the features designed earlier. Their function is to map the distributed local features extracted by the previous convolutions to the sample label space.
[0060] LSTM (Long Short-Term Memory) neural networks are a type of recurrent neural network specifically designed to address the long-term dependency problem inherent in general neural networks. They are suitable for processing and predicting important events with very long intervals and delays in time series. An LSTM mainly consists of cell states, a forget gate, an input gate, and an output gate. The cell state manages the flow of information stored in each cell. The forget gate decides whether to delete some information, primarily processing information passed from the previous time step and information input at the current time step. The input gate detects whether there is input and decides whether to input the data into the cell state memory. The output gate outputs the result based on the cell state, which includes information from the current time step and previous time steps.
[0061] Furthermore, the logging data includes: natural gamma logging data, caliper logging data, spontaneous potential logging data, resistivity logging data, neutron logging data, sonic logging data, and density logging data.
[0062] Furthermore, the training dataset is preprocessed, data is filtered based on importance analysis, and data is grouped based on kurtosis and skewness to obtain the processed training dataset, including:
[0063] The training dataset is cleaned, filtered, and normalized to obtain the first training data.
[0064] Specifically, the training dataset includes historical well logging sample data. Data cleaning removes outliers from the well logging curves. Outliers may be caused by the logging environment or human error. Logging environment causes include wellbore enlargement or large well deviation, special reservoirs, instrument performance constraints, and instrument malfunctions. These outliers can severely affect neural network model training; common methods include deleting and replacing outlier data. Data filtering smooths the data to remove noise and abrupt changes. Normalization involves subtracting the minimum value from the current value and dividing by the difference between the maximum and minimum values. Its purpose is to limit the data to a certain range, eliminate the adverse effects of outlier samples, and improve the model's convergence speed and accuracy.
[0065] Furthermore, the training dataset is preprocessed, data is filtered based on importance analysis, and data is grouped based on kurtosis and skewness to obtain the processed training dataset, which also includes:
[0066] Data with a correlation coefficient greater than a first preset coefficient value that is selected from the first training data are used as the second training data;
[0067] Calculate the correlation coefficient between the two different types of data in the second training data respectively;
[0068] If there are two different types of data with a correlation coefficient greater than the second preset coefficient value, then select one type of data from the two different types of data, and use the other different types of data in the second training data with a correlation coefficient less than or equal to the second preset coefficient value as the third training data.
[0069] Specifically, the correlation coefficient characterizes the importance of data and the degree of association between different data. Generally, the larger the correlation coefficient, the stronger the association between the data. In neural networks, the quality of the output largely depends on the input data. Providing too much data to the machine learning model can lead to reduced prediction accuracy, prolonged training time, and an increased possibility of overfitting. Therefore, selecting appropriate input data is essential. In this embodiment, by calculating the correlation coefficient between the input data and the prediction result (i.e., the shear wave time difference) of the shear wave time difference prediction model, the input data with the highest correlation to the prediction result can be accurately determined. This identifies the most important input data for the prediction result (i.e., the shear wave time difference) of the shear wave time difference prediction model, thereby achieving data filtering, reducing the amount of invalid input data, and thus reducing the computational load and time of the model. At the same time, it can improve prediction efficiency and ensure that the prediction result of the shear wave time difference prediction model is more accurate.
[0070] In this embodiment, by calculating the correlation coefficient between the input data and the prediction result (i.e., the shear wave time difference) of the shear wave time difference prediction model, data with a correlation coefficient greater than the first preset coefficient value are selected as the second training data. However, there may be some data with high similarity in the second training data. Using the data with high similarity as input data at the same time will cause the repeated use of variables and data redundancy. Therefore, for different types of data in the second training data, the correlation coefficient of any two different types of data is calculated. If the correlation coefficient of two different types of data is greater than the second preset coefficient value, then for these two different types of data with a correlation coefficient greater than the second preset coefficient value, any one type of data is selected from the two different types of data and merged with the remaining different types of data in the second training data with a correlation coefficient less than or equal to the second preset coefficient value as the third training data.
[0071] For example: After filtering by correlation coefficient, the second training data is obtained. The second training data consists of five different types of data (X1, X2, X3, X4, and X5). By calculating the correlation coefficient of any two data sets among the five sets, X1 and X2, X1 and X3, X1 and X4, X1 and X5, X2 and X3, X2 and X4, X2 and X5, X3 and X4, X3 and X5, and X4 and X5, only the correlation coefficient of data X2 and X3 is greater than the second preset coefficient value. Therefore, any one type of data from data X2 and X3 (e.g., choosing data of type X3) can be selected together with the remaining data in the second training data (i.e., data X1, X4, and X5 whose correlation coefficient is less than or equal to the second preset coefficient value) as the third training data. Thus, the third training data is (X1, X3, X4, and X5).
[0072] For example: After filtering by correlation coefficient, the second training data is obtained. The second training data consists of five different types of data (X1, X2, X3, X4, and X5). By calculating the correlation coefficient of any two data sets among the five sets, X1 and X2, X1 and X3, X1 and X4, X1 and X5, X2 and X3, X2 and X4, X2 and X5, X3 and X4, X3 and X5, and X4 and X5, the correlation coefficients of data X2 and X3, as well as data X4 and X5, are all greater than the second preset coefficient value. Then, one set of data is randomly selected from data X2 and X3 (e.g., data of type X2), and one set of data is randomly selected from X4 and X5 (e.g., data of type X5). Together with the remaining data in the second training data (i.e., data with a correlation coefficient less than or equal to the second preset coefficient value: X1), they are used as the third training data. Therefore, the third training data is (X1, X2, and X5).
[0073] Furthermore, the training dataset is preprocessed, data is filtered based on importance analysis, and data is grouped based on kurtosis and skewness to obtain the processed training dataset, which also includes:
[0074] Based on preset kurtosis coefficient and preset skewness coefficient, the third training data is divided into at least two sets of logging data, which are used as the processed training dataset.
[0075] Specifically, in this embodiment, well logging data is grouped by characterizing the sharpness of the P-wave travel curve peaks and the degree of asymmetry in data distribution. Each group of data is then used as input to the model, which improves the model's prediction accuracy. For example... Figure 3 and 4As shown, kurtosis, also known as kurtosis coefficient, is a characteristic number that characterizes the peak value of a probability density distribution curve at the mean. In other words, kurtosis is a statistic that describes the steepness of the distribution of all values in a population; that is, it reflects the sharpness of the peak. Skewness, also known as skewness coefficient, is a measure of the direction and degree of skewness in the distribution of statistical data; it is a numerical characteristic of the degree of asymmetry in the distribution of statistical data.
[0076] Furthermore, the correlation coefficient is calculated using the Pearson correlation coefficient formula.
[0077] Specifically, the Pearson correlation coefficient, also known as the Pearson product-moment correlation coefficient, is widely used to measure the degree of correlation between two variables X and Y. Its value ranges between -1 and 1. The formula for calculating the Pearson correlation coefficient is:
[0078]
[0079] The formula shows that the Pearson correlation coefficient is the covariance of X and Y divided by the standard deviation of X multiplied by the standard deviation of Y.
[0080] like Figure 5 As shown, this embodiment also provides a shear wave time difference prediction device, including:
[0081] The training data acquisition module 10 is used to acquire well logging sample data as the training dataset for the prediction model.
[0082] The first data processing module 20 is used to preprocess the training dataset, filter data based on importance analysis, and group data based on kurtosis and skewness to obtain the processed training dataset.
[0083] The model training module 30 is used to input the processed training dataset into the neural network built by CNN and LSTM respectively for training, so as to obtain the transverse wave time difference prediction model.
[0084] The input data acquisition module 40 is used to acquire well logging data of the shear wave time difference to be predicted;
[0085] The second data processing module 50 is used to preprocess the logging data, group the logging data based on kurtosis and skewness, and obtain the processed logging data.
[0086] The result output module 60 is used to take the processed logging data as input to the shear wave time difference prediction model to obtain the shear wave time difference.
[0087] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described transverse wave time difference prediction method.
[0088] This embodiment also provides a machine-readable storage medium storing instructions for causing a machine to perform the above-described transverse wave time difference prediction method.
[0089] Example 1
[0090] A training dataset is obtained. The well logging sample dataset is preprocessed, data is filtered based on importance analysis, and data is grouped based on kurtosis and skewness to obtain a preprocessed training dataset. A new neural network structure (a hybrid CNN and LSTM network) is formed by concatenating a CNN and an LSTM network through a Dropout layer. The processed training dataset is then input into the hybrid CNN and LSTM network respectively to train the shear wave time difference prediction model, as detailed below:
[0091] Well logging data undergoes data cleaning, filtering, and normalization to obtain the first set of well logging data, ensuring an accurate and consistent format for machine learning. Commonly found numbers in well logging data, such as "-9999," are removed during data cleaning. Data filtering utilizes median filtering to remove spikes and glitches. Normalization limits the data to values between 0 and 1, improving model convergence speed and accuracy.
[0092] The correlation between different types of data and shear wave velocity in the first logging data is obtained, usually calculated using the Pearson correlation coefficient method. The curves most correlated with shear wave transit time (DTS) (correlation coefficient greater than a first preset coefficient value, which is set to 0.15 in this embodiment) are, in order: DTC, CNL, DEN, GR, and RI. DTC, CNL, DEN, GR, and RI are then used as the second logging data. The comparison of the correlation coefficients between each data point and shear wave transit time (DTS) is shown in Table 1 below.
[0093] Table 1 Comparison of Correlation Coefficients
[0094] CNL 1 0.99 0.81 0.81 0.52 0.17 0.11 -0.18 DEN 0.99 1 0.79 0.79 0.48 0.17 0.11 -0.12 DTC 0.81 0.79 1 1 0.5 0.15 0.1 -0.42 DTS 0.81 0.79 1 1 0.5 0.15 0.1 -0.42 GR 0.52 0.48 0.5 0.5 1 0.19 0.17 -0.52 RI 0.17 0.17 0.15 0.15 0.19 1 0.98 -0.066 RT 0.11 0.11 0.1 0.1 0.17 0.98 1 -0.066 SP -0.18 -0.12 -0.42 -0.42 -0.52 -0.066 -0.066 1 CNL DEN DTC DTS GR RI RT SP
[0095] The first column lists the names of the eight input curves, from top to bottom: CNL (compensated neutron), DEN (bulk density), DTC (longitudinal wave transit time), DTS (transverse wave transit time), GR (natural gamma), RI (shallow resistivity), RT (deep resistivity), and SP (natural potential). The ninth row lists the names of the eight input curves, from left to right: CNL (compensated neutron), DEN (bulk density), DTC (longitudinal wave transit time), DTS (transverse wave transit time), GR (natural gamma), RI (shallow resistivity), RT (deep resistivity), and SP (natural potential). The numbers represent the Pearson correlation coefficients between the input curves; a larger value indicates a higher correlation, and a smaller value indicates a lower correlation.
[0096] If two highly correlated variables appear simultaneously in the input data, it will lead to variable duplication and data redundancy. Therefore, the correlation coefficients among the five data points DTC, CNL, DEN, GR, and RI are calculated. For example, if the correlation coefficients of two data points, CNL and DEN, are greater than the second preset coefficient value, using CNL and DEN as input variables in the model is equivalent to using the variable "porosity (both CNL and DEN are curves for evaluating porosity)" twice, which easily leads to data redundancy and increases computation time. Therefore, after comprehensive consideration, this embodiment selects DTC, DEN, GR, and RI as the third logging data.
[0097] The third logging data were grouped using kurtosis and skewness as indicators. Wells with kurtosis and skewness greater than 1 in their P-wave transit time were divided into the first group, and wells with kurtosis and skewness less than 1 in their P-wave transit time were divided into the second group, as shown in Table 2 below:
[0098] Table 2. Grouping data for kurtosis and skewness
[0099]
[0100] Two sets of well logging data were used as the processed training datasets and then input into a neural network constructed using a hybrid CNN and LSTM to train the model, thus obtaining the shear wave time difference prediction model.
[0101] A21-A24 are four new wells with shear wave transit time to be predicted. Well logging data preprocessing and data grouping based on kurtosis and skewness were used to obtain processed well logging data, which were then used as input variables for the shear wave transit time prediction model. The preprocessing and data grouping based on kurtosis and skewness steps are similar to the methods used when processing the training dataset, and will not be repeated here. The results predicted by this invention, along with regional empirical formulas and rock physics modeling methods, are as follows: Figure 6 As shown in Table 3, the prediction accuracy of different methods is compared:
[0102] Table 3 Comparison of Prediction Accuracy of Different Methods
[0103] hashtag Intelligent prediction method Empirical Formula Method Rock physical modeling A21 93.83％ 90.91％ 92.46％ A22 94.43％ 91.25％ 93.95％ A23 94.51％ 91.13％ 91.14％ A24 95.49％ 92.46％ 90.71％
[0104] Figure 6 The first channel displays natural gamma (GR), spontaneous potential (SP), and wellbore diameter (CALI). Natural gamma and spontaneous potential characterize lithological variations, while wellbore diameter indicates wellbore quality. The second channel is the depth channel, representing the distance from the wellhead to the measured well section (i.e., the target layer). The third channel displays three porosity curves, including longitudinal wave transit time (DTC), bulk density (DEN), and compensated neutron (CNL) curves, typically used for porosity calculations; here, they are used to predict shear wave transit time. The fourth channel displays resistivity curves, including deep resistivity (RT), shallow resistivity (RI), and low resistivity (RXO), typically used for identifying oil, gas, and water layers and calculating saturation; here, they are used to predict shear wave transit time. The fifth channel is the shear wave comparison channel, including shear wave transit time (DTS) and intelligent prediction method, used to compare the shear wave transit time obtained by the intelligent prediction method with the actual measured shear wave transit time. The sixth track is for shear wave comparison, including shear wave transit time (DTS) and empirical formula method, used to compare the shear wave transit time obtained by empirical formula method with the actual measured shear wave transit time. The seventh track is for shear wave comparison, including shear wave transit time (DTS) and rock physics modeling method, used to compare the shear wave transit time obtained by rock physics modeling method with the actual measured shear wave transit time.
[0105] pass Figure 6 A comparison with Table 3 shows that the shear wave time difference predicted by the method of this application has the advantages of high accuracy, small error and strong generalization ability compared with the shear wave time difference obtained by regional empirical formula and rock physics modeling method.
[0106] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention.
[0107] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not describe the various possible combinations separately.
[0108] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.
[0109] Furthermore, various different implementations of the present invention can be combined arbitrarily, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed in the present invention.
Claims
1. A method for predicting transverse wave time difference, characterized in that, include: Obtain well logging sample data as the training dataset for the prediction model; The training dataset is preprocessed, filtered based on importance analysis, and grouped based on kurtosis and skewness to obtain the processed training dataset, including: The training dataset is cleaned, filtered, and normalized to obtain the first training data; Data with a correlation coefficient greater than a first preset coefficient value that is selected from the first training data are used as the second training data; Calculate the correlation coefficient between the two different types of data in the second training data respectively; If there are two different types of data with a correlation coefficient greater than the second preset coefficient value, then select one type of data from the two different types of data and use the other different types of data in the second training data with a correlation coefficient less than or equal to the second preset coefficient value as the third training data. The processed training dataset is then input into a neural network constructed by combining CNN and LSTM to train the transverse wave time difference prediction model. Obtain logging data for the predicted shear wave time difference; The well logging data is preprocessed and grouped based on kurtosis and skewness to obtain processed well logging data; The processed well logging data are used as inputs to the shear wave time difference prediction model to obtain the shear wave time difference.
2. The method according to claim 1, characterized in that, The CNN neural network and LSTM neural network in the shear wave time difference prediction model are connected through a Dropout layer.
3. The method according to claim 1, characterized in that, The logging data includes: natural gamma logging data, caliper logging data, spontaneous potential logging data, resistivity logging data, neutron logging data, sonic logging data, and density logging data.
4. The method according to claim 1, characterized in that, The training dataset is preprocessed, filtered based on importance analysis, and grouped based on kurtosis and skewness to obtain the processed training dataset, which also includes: Based on preset kurtosis coefficient and preset skewness coefficient, the third training data is divided into at least two sets of logging data, which are used as the processed training dataset.
5. The method according to claim 1, characterized in that, The correlation coefficient was calculated using the Pearson correlation coefficient formula.
6. A transverse wave time difference prediction device, used to implement the transverse wave time difference prediction method according to claim 1, characterized in that, include: The training data acquisition module is used to acquire well logging sample data as the training dataset for the prediction model. The first data processing module is used to preprocess the training dataset, filter data based on importance analysis, and group data based on kurtosis and skewness to obtain the processed training dataset. The model training module is used to input the processed training dataset into a neural network built by combining CNN and LSTM to train the model and obtain the transverse wave time difference prediction model. The input data acquisition module is used to acquire well logging data of the shear wave time difference to be predicted; The second data processing module is used to preprocess the logging data, group the logging data based on kurtosis and skewness, and obtain the processed logging data. The result output module is used to take the processed well logging data as input to the shear wave time difference prediction model to obtain the shear wave time difference.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the transverse wave time difference prediction method according to any one of claims 1-5.
8. A machine-readable storage medium storing instructions for causing a machine to perform the transverse wave time difference prediction method according to any one of claims 1-5.