Methods, equipment, and storage media for selecting optimal parameters of shear wave velocity in oil and gas wells

Abstract

This invention discloses a method, equipment, and storage medium for selecting optimal parameters for shear wave velocity in oil and gas wells. The method includes the following steps: acquiring various logging data; preprocessing the data and dividing it into training and validation sets; systematically eliminating feature parameters in the training set to construct different feature subsets; training a neural network for shear wave velocity prediction to obtain the trained networks corresponding to different feature subsets; obtaining the fitting determination coefficients of different network performances using the validation set; sorting the feature parameters from highest to lowest according to the fitting determination coefficients to determine the influence of different feature subsets on shear wave velocity prediction; performing pairwise correlation analysis on the removed feature parameters corresponding to important feature subsets; when the correlation is higher than a second preset value, selecting the more important parameter from the two correlated parameters; and obtaining the final optimal combination of logging parameters. This invention solves the parameter selection problem in predicting shear wave velocity in oil and gas wells and is suitable as a basis for optimizing logging parameter selection during model training.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of oil and gas well exploration, and particularly relates to a method for selecting optimal parameters of shear wave velocity of an oil and gas well, a device and a storage medium. BACKGROUND

[0002] Shear wave velocity is an important logging parameter in the development of oil and gas wells. In the logging process, the method for obtaining shear wave velocity includes using dipole acoustic logging and laboratory petrophysical experiment measurement. Compared with other logging parameters, the measurement cost of shear wave velocity is higher, so there may be a lack of shear wave velocity information in some actual logging data. Due to the complex nonlinear relationship between different logging parameters, it is usually necessary to use a prediction method to obtain the shear wave velocity.

[0003] In recent years, as a powerful machine learning method, deep learning has matured and been widely applied. By establishing a deep mapping model, deep learning can effectively capture the complex relationship between logging parameters, providing an effective tool for accurate prediction of shear wave velocity. In this process, selecting appropriate feature sets is crucial, but training a neural network model cannot input all logging parameters as feature parameters. Too many feature parameters will cause a substantial increase in computational complexity, model redundancy, dimensionality disaster and other problems.

[0004] Screening feature parameters helps to improve the performance, generalization ability and interpretability of the model, while reducing the computational complexity. The conventional method for selecting feature parameters for training models is to directly calculate the Pearson correlation coefficient or the Spearman rank correlation coefficient of each parameter with the shear wave velocity. However, the Pearson correlation coefficient is sensitive to outliers and is not suitable for nonlinear data, while the Spearman rank correlation coefficient may not be stable for small sample data and is not sensitive to nonlinear relationships.

[0005] Deep learning networks are inherently suitable for mining strong nonlinear relationships between data, so directly using conventional methods to screen feature parameters for training models may not be accurate. Calculating correlation coefficients may not capture the complex interaction effects between parameters, and some parameters may not have strong correlation in multiple regression, but may play an important role in the nonlinear part of the model. SUMMARY

[0006] To solve the technical problem of inaccurate input parameter selection in predicting the shear wave velocity of an oil and gas well in the prior art, the present application provides a method for selecting optimal parameters of shear wave velocity of an oil and gas well, a device and a storage medium.

[0007] The method comprises the following steps:

[0008] S1: Obtain various logging data of the oil and gas well with measured shear wave velocity;

[0009] S2: Preprocess the logging data, and divide the preprocessed data into a training set and a validation set;

[0010] S3: Exclude the feature parameters one by one in the training set, use the remaining feature parameters to construct different feature subsets, train the shear wave velocity prediction neural network using different feature subsets, and obtain the trained network corresponding to different feature subsets;

[0011] S4: Test the trained network corresponding to different feature subsets using the validation set, and obtain the fitting determination coefficients of different network performances;

[0012] S5: According to the fitting determination coefficients of different network performances, the influence degree of different feature subsets on the shear wave velocity prediction is obtained by sorting from high to low;

[0013] S6: The feature parameters corresponding to the feature subsets with fitting determination coefficients lower than a first preset value are collected to obtain a preliminary logging parameter combination; in the preliminary logging parameter combination, parameter two-by-two correlation analysis is performed, and when the correlation is higher than a second preset value, a parameter with higher importance is selected between the two correlation parameters; the parameter with higher importance refers to that the fitting determination coefficient of the corresponding feature subset is lower after removing the parameter;

[0014] S7: The logging parameter combination selected by the parameter is screened to obtain the final preferred logging parameter combination.

[0015] A storage medium stores instructions and data for implementing an optimal parameter selection method for shear wave velocity of an oil and gas well.

[0016] An optimal parameter selection device for shear wave velocity of an oil and gas well comprises a processor and a storage medium; the processor loads and executes the instructions and data in the storage medium to implement an optimal parameter selection method for shear wave velocity of an oil and gas well.

[0017] The beneficial effects provided by the present application are:

[0018] 1. For a limited number of logging parameters, the workload of training the model is not large.

[0019] 2. The collinearity problem between features is considered, and the mutual influence between features can be captured.

[0020] 3. The influence of the feature parameters on the performance of the model is quantified, and the importance ranking of the feature parameters is provided. Compared with the correlation analysis, this method is more suitable for selecting the optimized logging parameters for model training. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a schematic diagram of the method flow of the present application;

[0022] Figure 2 is a schematic diagram of conventional logging data of oil and gas wells as a training set in the embodiment of the present application;

[0023] Figure 3 is a schematic diagram of conventional logging data of oil and gas wells as a verification set in the embodiment of the present application;

[0024] Figure 4 is a high-contribution logging parameter correlation heat map provided in the embodiment of the present application

[0025] Figure 5 is a schematic diagram of the working of the hardware device of the present application. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical scheme and advantages of the present application clearer, the embodiments of the present application will be further described below with reference to the drawings.

[0027] Please refer to Figure 1 , Figure 1 is a schematic diagram of the method flow of the present application.

[0028] The present application provides a method for selecting the best parameter of the shear wave velocity of an oil and gas well, comprising the following steps:

[0029] S1: obtaining various logging data of an oil and gas well whose shear wave velocity has been measured;

[0030] It should be noted that in step S1, the various logging parameters include all logging parameters used to measure the shear wave velocity.

[0031] S2: pre-processing the logging data and dividing the pre-processed data into a training set and a verification set;

[0032] It should be noted that since the measured data may have measurement abnormalities or missing conditions, the missing value filling and abnormal value replacement operations are first performed. These operations will significantly reduce the influence of unconventional values on subsequent network training. Finally, normalization processing is performed.

[0033] For filling of missing items, the average value of the data before and after the missing item can be used for filling, or other existing filling methods can also be used.

[0034] For replacement of abnormal values, after the abnormal items are determined by means of data mean, variance, etc., they are replaced by the data mean.

[0035] Normalization effectively eliminates dimensional differences between different logging parameters, allowing the model to converge faster. Gradient descent is a common algorithm used in most models. Data with dimensional differences can find the optimal solution more quickly, and it can also improve the performance and generalization ability of subsequent models.

[0036] It should be noted that both the training set and the validation set include a feature part and a label part; the feature part consists of different logging parameters, and the label part consists of shear wave velocity.

[0037] S3: Eliminate feature parameters one by one in the training set, and use the remaining feature parameters to construct different feature subsets; use the different feature subsets to train the transverse wave velocity prediction neural network to obtain the trained network corresponding to the different feature subsets;

[0038] It should be noted that the shear wave velocity prediction neural network employs a combination model of 1D-CNN and LSTM. The core components of the network include fully connected layers, one-dimensional convolutional layers, and Long Short-Term Memory (LSTM) layers. Initial input data passes through a fully connected layer containing 100 nodes for initial feature extraction and linear transformation. Next, the data passes through two one-dimensional convolutional layers, each containing 64 nodes with a kernel size of 2, to capture local features of the input data. Subsequently, the data is passed to two LSTM layers, each with 40 nodes, to process time-series data, capture temporal dependencies, and long-term memory. To connect the output of the LSTM layers to the fully connected layers, a Flatten layer is used to transform the high-dimensional data into a one-dimensional form. Finally, the data is output through a fully connected layer with one node.

[0039] The process of constructing the feature subset in step S3 is as follows:

[0040] S31. Obtain the feature set T = [t1, t2, ..., t] from the entire training set. n ];

[0041] S32. Excluding the first feature parameter from the feature set, we obtain the feature subset T1 = [t2, ..., t n ];

[0042] S33. Excluding the second feature parameter from the feature set, we obtain the feature subset T2 = [t1, t3, ..., t n ];

[0043] S34. Similarly, exclude the nth feature parameter from the feature set to obtain the feature subset T. n =[t1,t2,…,t n-1 ].

[0044] Through the above process, the initial feature set is split one by one, and each time one of the logging parameters is removed, thereby generating a feature subset equal in number to the types of feature parameters. This feature set splitting method for the S-wave velocity prediction is based on the limited types of logging parameters, and thus has little workload.

[0045] After selecting the model, the performance index is selected as MSE, because MSE has sensitivity to all errors, which helps to comprehensively understand the prediction accuracy of the model in model evaluation.

[0046] In step S3, when the network is trained, the same feature subset is used for multiple repeated training, and multiple fitting determination coefficients are obtained. After averaging the multiple fitting determination coefficients, the fitting determination coefficient of the feature subset is obtained.

[0047] It should be noted that due to the instability of neural network training, the same feature subset is used for three repeated training in the present application, and the three models are used for S-wave velocity prediction, and the determination coefficient R-squared of each prediction is calculated. The R-squared calculation formula is The coefficient is a general formula for calculating the determination coefficient of the neural network. The average of the three determination coefficients is taken to obtain a comprehensive model performance evaluation, which is equivalent to the fitting result obtained by training the model in the absence of one parameter.

[0048] S4, using the validation set to test the trained network corresponding to different feature subsets, to obtain the fitting determination coefficients of different network performances;

[0049] It should be noted that the closer the fitting result is to 1, the less effective the parameter is in predicting the S-wave velocity with multiple logging parameters, and the more inclined to be discarded.

[0050] S5: According to the fitting determination coefficients of different network performances, from high to low, the influence degree of different feature subsets on S-wave velocity prediction is obtained;

[0051] By arranging the obtained fitting results from small to large, it is equivalent to arranging the importance of logging parameters for S-wave velocity prediction from large to small, which helps to objectively identify and quantify the relative importance of logging parameters in the model.

[0052] S6: The feature parameters corresponding to the feature subsets whose fitting determination coefficients are lower than the first preset value are excluded to obtain a preliminary logging parameter combination. In the preliminary logging parameter combination, a two-by-two correlation analysis is performed on the parameters, and when the correlation is higher than the second preset value, a parameter with higher importance is selected between the two correlated parameters; the parameter with higher importance refers to the fitting determination coefficient of the corresponding feature subset after removing the parameter is lower;

[0053] In simple terms, the meaning of this step is as follows: for example, four feature subsets b, c, d and a, c, d and a, b, d and a, b, c are constructed, and the network prediction parameter f is directly trained using the four feature subsets respectively, so as to obtain four corresponding fitting determination coefficients, of course, each model is repeatedly trained three times to take the average value, and finally the fitting effect of removing the parameter a, the fitting effect of removing the parameter b, the fitting effect of removing the parameter c and the fitting effect of removing the parameter d are obtained. Then the worse the fitting effect is, the more important the parameter is, which is equivalent to the lower the fitting determination coefficient is. Then the important parameters are selected for correlation analysis.

[0054] S7: After parameter selection, the selected logging parameter combination is used as the final selected logging parameter combination.

[0055] As an embodiment, for the obtained fitting effect ranking, 75% is set as a preset value. When the prediction result of the model is less than the preset value, it indicates that the missing parameter is not ideal for the prediction effect of the shear wave velocity. On the contrary, if the model performs well without this parameter, it indicates that the parameter can be discarded when constructing the feature set.

[0056] The corresponding parameters with fitting effects lower than the preset threshold of 75% are screened out, and then the heat map correlation analysis is performed on these parameters. If the analysis shows that the correlation coefficient between two parameters exceeds 80%, the two parameters are strongly correlated data, and simultaneously, the training of the network as a feature set may increase the model operation amount and will not improve the model. Therefore, one of the parameters with smaller influence degree can be discarded. After the arrangement and screening of the logging parameters, the optimized logging parameter combination for training the neural network model is obtained.

[0057] As an embodiment, the present application selects two wells with measured shear wave velocities in different work areas, fills and replaces the missing values and abnormal values in the logging data, selects Figure 2 Well A as the training set, Figure 3 Well B as the verification set, since the present application focuses on the screening of logging parameters, the parameters of the neural network have been debugged before testing, and the network structure and the number of training will remain constant in subsequent training. The different feature subsets after splitting are input into the network in turn and repeatedly trained three times, the R-squared of the prediction results of the three times and the measured values are averaged, and the results in Table 1 are obtained. Wherein Vp represents the longitudinal wave velocity calculated by the acoustic logging value, CNL represents the neutron logging value, RHOB represents the density logging value, GR represents the natural gamma logging value, SP represents the natural electric potential logging value, DEPTH represents the logging depth, RT represents the resistivity logging value, and Vs represents the shear wave velocity.

[0058] Table 1 experimental results

[0059]

[0060] Then, the logging parameters with the determination coefficient less than 0.75 are subjected to heat map correlation analysis, so that Figure 4 , and Figure 4 , the relatively good logging parameter selection can be made. It can be clearly seen from Table 1 that the prediction of the longitudinal wave velocity Vp to the transverse wave velocity Vs is particularly important, and the effect of removing the resistivity parameter RT is much better than that of adding the resistivity parameter RT, which further illustrates the importance of selecting the logging parameter combination, and it can be seen from Figure 4 that there is no strong correlation between the parameters with high contribution to the prediction of the transverse wave velocity, so the priority ranking of the selected logging parameter combination is from left to right as shown in Table 2.

[0061] Table 2 Optimal parameter combination

[0062] Vp CNL RHOB GR SP DEPTH RT

[0063] Please refer to Figure 5 , Figure 5 is a hardware device working schematic diagram of an embodiment of the present application, and the hardware device specifically comprises: an oil and gas well transverse wave velocity optimal parameter selection device 401, a processor 402 and a storage medium 403.

[0064] The oil and gas well transverse wave velocity optimal parameter selection device 401: the oil and gas well transverse wave velocity optimal parameter selection device 401 realizes the oil and gas well transverse wave velocity optimal parameter selection method.

[0065] The processor 402: the processor 402 loads and executes the instructions and data in the storage medium 403 to realize the oil and gas well transverse wave velocity optimal parameter selection method.

[0066] The storage medium 403: the storage medium 403 stores instructions and data; and the storage medium 403 is used to realize the oil and gas well transverse wave velocity optimal parameter selection method.

[0067] The present application has the following beneficial effects:

[0068] 1. For a limited number of logging parameters, the workload is not large when training the model.

[0069] 2. The collinearity problem between features is considered, and the mutual influence between features can be captured.

[0070] 3. The influence of the feature parameters on the model performance is quantified, and the importance ranking of the feature parameters is provided. Compared with the correlation analysis, this method is more suitable for the optimization of the logging parameters for model training.

[0071] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for optimizing parameter selection of shear wave velocity in an oil and gas well, characterized in that: The method comprises the following steps: S1: obtaining various logging data of an oil and gas well with measured shear wave velocity; In step S1, the various logging data includes all logging parameters for measuring shear wave velocity, including: P-wave velocity Vp, neutron logging value CNL, density logging value RHOB, natural gamma logging value GR, natural potential logging value SP, logging depth DEPTH, and resistivity logging value RT; S2: preprocessing the logging data and dividing the preprocessed data into a training set and a validation set; S3: excluding logging parameters one by one in the training set, using the remaining logging parameters to construct different feature subsets, training a shear wave velocity prediction neural network using different feature subsets, and obtaining a trained network corresponding to different feature subsets; S4: testing the trained network corresponding to different feature subsets using the validation set to obtain fitting determination coefficients of different network performances; S5: sorting the fitting determination coefficients of different network performances from high to low to obtain the influence degree of different feature subsets on shear wave velocity prediction; S6: collecting feature parameters corresponding to feature subsets with fitting determination coefficients lower than a first preset value to obtain a preliminary logging parameter combination; performing parameter correlation analysis on the preliminary logging parameter combination, and selecting a parameter with higher importance between two correlated parameters when the correlation is higher than a second preset value; the parameter with higher importance refers to a parameter whose fitting determination coefficient corresponding to its feature subset is lower after removing the parameter; S7: selecting the logging parameter combination obtained after parameter selection as the final preferred logging parameter combination.

2. The method of optimizing parameters of shear wave velocity in an oil and gas well of claim 1, wherein: The preprocessing includes missing value filling, abnormal value replacement, and normalization processing.

3. The method of claim 1, wherein: The training set and the validation set each include a feature part and a label part; the feature part is different logging parameters, and the label part is shear wave velocity.

4. The method of optimizing parameters of shear wave velocity in an oil and gas well of claim 3, wherein: The process of constructing feature subsets in step S3 is as follows: S31, acquire feature part in whole training set T [ t 1, t 2,…, t n ]; S32、from the feature part, exclude the first feature parameter to obtain a feature subset T 1=[ t 2,…, t n ]; S33, exclude the second feature parameter from the feature part to obtain a feature subset T 2=[ t 1, t 3,…, t n ]; S34, and so on, exclude the i-th feature parameter from the feature part, to obtain a feature subset n T n [ t 1, t 2,…, t n-1 ]​ 5. The method of optimizing parameters for shear wave velocity in an oil and gas well of claim 1, wherein: The shear wave velocity prediction neural network adopts a combined model of 1D-CNN and LSTM.

6. The method of optimizing parameters for shear wave velocity in an oil and gas well of claim 1, wherein: In step S3, when training the network, multiple fitting determination coefficients are obtained by repeatedly training the same feature subset multiple times, and the average value of the multiple fitting determination coefficients is taken as the fitting determination coefficient of the feature subset.

7. A storage medium characterized by: The storage medium stores instructions and data for implementing the oil and gas well shear wave velocity optimal parameter selection method of any one of claims 1-6.

8. An apparatus for optimizing parameter selection for shear wave velocity in an oil and gas well, the apparatus comprising: It comprises: A processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the oil and gas well shear wave velocity optimal parameter selection method of any one of claims 1-6.

Images