A method, device, equipment and storage medium for processing shear wave velocity of logging while drilling

Abstract

The present invention relates to the field of applied geophysical technology, and in particular to a method, device, equipment and storage medium for processing shear wave velocity in logging while drilling. The method comprises: dividing parameter range and step length, determining formation parameter combination, and using it as a data set sample feature; constructing a dispersion data set according to the obtained data set sample label; performing data cleaning on the dispersion data set, and correcting the erroneous samples in the data set; establishing a neural network, training the dispersion data set, and verifying the feasibility of the model; establishing an inversion objective function through a joint inversion method, using a low-frequency band to invert shear wave velocity, and a high-frequency band to invert mud velocity; adding an optimization method to reduce the number of forward calculations in the inversion and improve the inversion efficiency; replacing the current depth point with the next depth point, and returning to execute the optimization method until the shear wave velocity processing of the full depth section is completed. The present invention realizes the correction of rapid dispersion under logging while drilling, improves calculation efficiency, and saves calculation cost.

Description

Technical Field

[0001] The invention relates to the field of applied geophysical technology, and in particular to a method, device, equipment and storage medium for processing shear wave velocity of logging while drilling. Background Art

[0002] As a high-end acoustic logging technology, LWD can provide real-time wellbore stability analysis, formation pore pressure prediction and other services by obtaining the formation longitudinal wave velocity and shear wave velocity. Due to its special dispersion characteristics, the quadrupole sound source is less disturbed by the drill collar in the LWD environment and has been widely used in LWD work. However, the quadrupole wave has a strong dispersion effect, and the presence of the drill collar in the wellbore further affects the dispersion characteristics. In order to obtain the true shear wave velocity information of the formation, the original measurement data needs to be dispersion corrected.

[0003] Chinese patent CN115327640A discloses a method for correcting well seismic dispersion suitable for dense sandstone formations. The patent mainly provides a method for correcting well seismic dispersion in dense sandstone formations. The dispersion correction in the patent is only used for the situation of dense sandstone formations.

[0004] Chinese patent CN115857018A discloses a dipole acoustic logging dispersion correction method and related device based on density clustering. The patent mainly provides a method for correcting well seismic dispersion in dense sandstone formations. The patent discloses a dispersion curve group; the dispersion curve group is processed by density clustering to obtain the effective part of the dispersion curve group; the effective part of the dispersion curve group is fitted to obtain the correction result. However, the invention selects only the effective part of the data for analysis and cannot cover all sample data.

[0005] At present, how to establish a method for processing shear wave velocity of LWD, realize dispersion correction of shear wave velocity of LWD, and comprehensively process all collected data are technical problems that have not yet been solved in this field. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a method, device, equipment and storage medium for processing shear wave velocity of logging while drilling. The technical solution provided by the present invention can construct a dispersion data set by numerically solving the dispersion curve corresponding to the parameter combination, use the machine learning method to train a neural network model that can quickly solve the dispersion curve, and in the inversion objective function established by joint inversion, combine the optimization method to complete the inversion, reduce the calculation cost from the two perspectives of accelerating the forward output dispersion curve and accelerating the inversion to solve the minimum fitting residual, and realize the fast dispersion correction method under logging while drilling. The technical solution is as follows:

[0007] In a first aspect, the present invention provides a method for processing shear wave velocity of logging while drilling, the method comprising the following steps:

[0008] Step S1, according to the sensitivity analysis of different parameters of logging while drilling to quadrupole shear wave dispersion, divide the parameter range and step size, determine the formation parameter combination, and use it as the sample feature of the data set;

[0009] Step S2, numerically solving the dispersion equation of the data set samples to obtain the dispersion curve corresponding to the formation parameter combination as the data set sample label, thereby constructing a dispersion data set;

[0010] Step S3, performing data cleaning on the dispersion data set to correct erroneous samples in the data set;

[0011] Step S4, establishing a neural network, training it on the dispersion data set obtained in step S3, obtaining a neural network model capable of fast dispersion forward modeling, performing prediction error analysis and generalization analysis on the model, and verifying the feasibility of the model;

[0012] Step S5, obtaining the theoretical dispersion curve under the corresponding formation parameters according to the neural network forward model trained in step S4; establishing the inversion objective function through the joint inversion method, using the low-frequency band inversion shear wave velocity and the high-frequency band inversion mud velocity as the inversion object; calculating the fitting residual between the theoretical dispersion curve and the formation dispersion data;

[0013] Step S6, in the process of obtaining the minimum value of the fitting residual in step S5, the interpolation inversion method is replaced by the optimization method, so as to reduce the number of forward modeling calculations in the inversion and improve the inversion efficiency;

[0014] Step S7, replace the current depth point with the next depth point, and return to step S6 to execute the optimization method until the shear wave velocity processing of the full depth segment is completed.

[0015] Furthermore, the step S1 is used to analyze the influence of different formation parameters on the dispersion curve in different frequency bands, that is, sensitivity analysis, and the calculation formula is:

[0016] ,

[0017] Among them, Sensitivity is the parameter sensitivity; is the formation parameter; is the angular frequency; v phase is the theoretical dispersion curve; ∂ is the symbol for partial differential solution.

[0018] Furthermore, step S2 establishes a dispersion equation according to an equivalent instrument theoretical model of a uniform formation while drilling, and solves dispersion curves corresponding to different parameter combinations based on the dispersion equation as sample labels to produce a data set, wherein the different parameter combinations are sample features; the calculation formula is:

[0019] ,

[0020] Where D is the dispersion matrix using the equivalent instrument theory; k is the wave number; is the angular frequency; F is the relevant parameter of elastic isotropic formation; B is the waveguide in the wellbore; M T is the instrument modulus; is the equivalent instrument radius.

[0021] Furthermore, the step S3 performs data cleaning on all data sets, and the data cleaning includes the following steps:

[0022] Step S3.1: Determine whether the sample is an error sample. If it is an error sample, execute step S3.2; if it is a correct sample, execute step S3.4;

[0023] Step S3.2: Correcting the error samples;

[0024] Step S3.3: Obtain the correct sample after correction, and execute step S3.4 on the sample;

[0025] Step S3.4: pre-training the neural network on the sample;

[0026] Step S3.5: Determine whether there are large fitting residual samples in the samples pre-trained with the neural network in step S3.4. The residual samples execute step S3.2, and the non-residual samples end the correction.

[0027] Furthermore, the step S4 determines the optimal structure of the neural network according to the grid optimization method; trains the established neural network model, and uses mean square error loss to analyze the model training process during the training; and uses mean relative error to analyze the prediction error of the trained model;

[0028] The neural network calculation formula is:

[0029] ,

[0030] Among them, v pred Predicted dispersion curves for the network model; is the nonlinear activation function; w i 、b i , i=1, 2, 3, respectively representing the weight and bias between each connection layer; is the input sample feature value;

[0031] The mean square error loss is MeanSquaredErrorLoss, referred to as MSE, and the calculation formula is:

[0032] ,

[0033] Among them, v theory is the dispersion curve label of numerical simulation; v pred is the dispersion curve predicted by the network model; i, j represents the jth point on the i-th group of dispersion curves; N is the number of data sets;

[0034] The mean relative error is MeanRelativeError, referred to as MRE, and the calculation formula is:

[0035] ,

[0036] Among them, v pred is the dispersion curve predicted by the network model; v theory is the dispersion curve label of numerical simulation; N is the number of frequency calculations; Ω is the frequency calculation range; is the angular frequency.

[0037] Furthermore, step S5 is based on the shear wave velocity and mud speed Based on the sensitivity characteristics in different frequency bands, the inversion objective function E is established:

[0038] ,

[0039] in, For low frequency band, For high frequency band; and They are the formation dispersion data of low frequency band and high frequency band respectively; is the weight factor; is the angular frequency; is the symbol for the derivative in the integral.

[0040] Furthermore, step S6 is the shear wave velocity and mud speed The joint inversion method of the invention reduces the number of forward modeling times n required to be calculated in the inversion by an optimization method. The calculation formula of the method is:

[0041] ,

[0042] ,

[0043] ,

[0044] in, for location; for speed; is the gradient; The correction amount brought by the current gradient to the speed; is the learning rate; is the momentum coefficient.

[0045] In a second aspect, the present invention provides a shear wave velocity processing device for logging while drilling, the device comprising:

[0046] The sensitivity analysis module divides the parameter range and step size according to the sensitivity analysis of different parameters of LWD to quadrupole shear wave dispersion, determines the formation parameter combination, and uses it as the sample feature of the data set;

[0047] The data set construction module solves the dispersion equation numerically output by the sensitivity analysis module, obtains the dispersion curve corresponding to the formation parameter combination as the data set sample label, and thus constructs the dispersion data set;

[0048] The data correction module cleans the dispersion data set output by the data set construction module and corrects the erroneous samples in the data set;

[0049] The feasibility verification module builds a neural network and trains it on the dispersion data set obtained by the data correction module to obtain a neural network model that can quickly perform dispersion forward modeling. The model is then subjected to prediction error analysis and generalization analysis to verify the feasibility of the model.

[0050] The inversion module obtains the theoretical dispersion curve under the corresponding formation parameters according to the neural network forward model trained by the feasibility verification module; establishes the inversion objective function through the joint inversion method, adopts the low-frequency band inversion of shear wave velocity and the high-frequency band inversion of mud velocity as the inversion object; calculates the fitting residual between the theoretical dispersion curve and the formation dispersion data;

[0051] The optimization module uses the optimization method to replace the interpolation inversion method in the process of obtaining the minimum value of the fitting residual in the inversion module, thereby reducing the number of forward calculations in the inversion and improving the inversion efficiency;

[0052] The loop iteration module replaces the current depth point with the next depth point, and returns to the optimization module to execute the optimization method until the shear wave velocity processing of the full depth segment is completed.

[0053] In a third aspect, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program;

[0054] The computer program is stored in the memory;

[0055] The computer program comprises instructions, which, when executed by the processor, cause the electronic device to perform the method described in any one of the first aspects.

[0056] In a fourth aspect, the computer-readable storage medium includes a stored program, and when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods described in the first aspect.

[0057] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0058] 1. Based on the theory of quadrupole acoustic wave propagation in logging while drilling, the present invention adopts a machine learning method to obtain a dispersion forward model through neural network training. The model can quickly calculate the theoretical dispersion curve; in the inversion, the joint inversion method can effectively solve the problem of multiple solutions, and the optimization method can reduce the number of forward models and further reduce the calculation cost.

[0059] 2. The data cleaning method provided by the present invention can realize the correction of erroneous samples. During the operation process, it can quickly determine whether the theoretical dispersion curve data obtained is correct, and correct the erroneous data. The correct samples and the corrected samples are pre-trained through a neural network, and the corrected correct sample data is output, thereby realizing full coverage operation of the sample data.

[0060] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application so that it can be implemented in accordance with the contents of the specification, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following is a detailed description of the preferred embodiments of the present application in conjunction with the accompanying drawings as follows.

[0061] Based on the detailed description of the specific embodiments of the present application in combination with the accompanying drawings below, those skilled in the art will become more aware of the above and other objects, advantages and features of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings without creative work. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual scale.

[0063] Figure 1 A work flow chart for a shear wave velocity processing method, device, equipment and storage medium for logging while drilling;

[0064] Figure 2 Data cleaning flow chart for the dispersion data set;

[0065] Figure 3 This is the parameter sensitivity analysis diagram of the quadrupole acoustic field dispersion while drilling. Figure 3 (a) is the formation parameter sensitivity analysis diagram of the fast formation. Figure 3 Middle (b) is the formation parameter sensitivity analysis diagram of the slow formation;

[0066] Figure 4 Flowchart for training neural network models;

[0067] Figure 5 This is the error statistics of the neural network model in the test set;

[0068] Figure 6 This is a comparison chart of the application effects of the numerical forward method and the model calculation method. Figure 6 (a) is the fitting residual diagram of the numerical forward modeling method. Figure 6 (b) is the fitted residual graph calculated by the neural network model. Figure 6 (c) is the fitting curve after dispersion correction and the coincidence diagram of formation dispersion data;

[0069] Figure 7 Schematic diagram of the solution path for the optimization method. Figure 7 (a) is a three-dimensional descent process diagram. Figure 7 (b) is the two-dimensional projection diagram of the three-dimensional descent process;

[0070] Figure 8 This is the result of fast formation shear wave velocity processing in LWD. DETAILED DESCRIPTION

[0071] To make the purpose, technical scheme and advantages of the embodiment of the present application clearer, the technical scheme in the embodiment of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiment of the present application. Obviously, the described embodiment is a part of the embodiment of the present application, rather than all of the embodiments. In the following description, specific details such as specific configuration and components are provided only to help fully understand the embodiments of the present application. Therefore, it should be clear to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. In addition, for clarity and brevity, the description of known functions and structures is omitted in the embodiment.

[0072] It should be understood that the references to "one embodiment" or "this embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present application. Therefore, the references to "one embodiment" or "this embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0073] In addition, the present application may repeat reference numerals and / or letters in different examples. This repetition is for the purpose of simplicity and clarity, and does not in itself indicate the relationship between the various embodiments and / or settings discussed.

[0074] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist at the same time. The term " / and" in this article describes another type of association object relationship, indicating that there can be two relationships. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the previous and next associated objects are in an "or" relationship.

[0075] The term "at least one" in this article is merely a description of the association relationship of associated objects, indicating that there may be three relationships. For example, at least one of A and B can mean: A exists alone, A and B exist at the same time, and B exists alone.

[0076] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusions. Example 1

[0077] This embodiment provides a method for processing shear wave velocity of logging while drilling. Figure 1 As shown, the method comprises the following steps:

[0078] Step S1: According to the sensitivity analysis of different parameters under logging while drilling to quadrupole shear wave dispersion, the parameter range and step size are divided, the formation parameter combination is determined, and it is used as the sample feature of the data set; this step is used to analyze the influence of different formation parameters on the dispersion curve in different frequency bands, that is, the sensitivity analysis, and the calculation formula is:

[0079] ,

[0080] Among them, Sensitivity is the parameter sensitivity; is the formation parameter; is the angular frequency; v phase is the theoretical dispersion curve; ∂ is the symbol for partial differential solution.

[0081] Step S2, by numerically solving the dispersion equation of the data set sample, the dispersion curve corresponding to the formation parameter combination is obtained as the data set sample label, thereby constructing a dispersion data set; the dispersion equation is established according to the equivalent instrument theoretical model of the uniform formation while drilling, and the dispersion curves corresponding to different parameter combinations are solved based on the dispersion equation as sample labels to produce a data set, and the different parameter combinations are sample features; the calculation formula is:

[0082] ,

[0083] Where D is the dispersion matrix using the equivalent instrument theory; k is the wave number; is the angular frequency; F is the relevant parameter of elastic isotropic formation; B is the waveguide in the wellbore; M T is the instrument modulus; is the equivalent instrument radius.

[0084] Step S3: Clean the dispersion data set and correct the erroneous samples in the data set.

[0085] Step S4, establish a neural network, train it on the dispersion data set obtained in step S3, obtain a neural network model capable of fast dispersion forward modeling, perform prediction error analysis and generalization analysis on the model, and verify the feasibility of the model; determine the optimal structure of the neural network according to the grid optimization method; train the established neural network model, use the mean square error loss to analyze the model training process during training, and when the mean square error loss is greater than the threshold, back propagate the error, optimize the weights, biases and hyperparameters, and then re-perform the next round of training until the mean square error loss is less than the set threshold; use the average relative error of the trained model to analyze the prediction error;

[0086] The neural network calculation formula is:

[0087] ,

[0088] Among them, v pred Predicted dispersion curves for the network model; is the nonlinear activation function; w i 、b i (i=1,2,3) represent the weight and bias between each connection layer; is the input sample feature value;

[0089] The mean square error loss is MeanSquaredErrorLoss, referred to as MSE, and the calculation formula is:

[0090] ,

[0091] Among them, v theory is the dispersion curve label of numerical simulation; v pred is the dispersion curve predicted by the network model; i, j represents the jth point on the i-th group of dispersion curves; N is the number of data sets;

[0092] The mean relative error is MeanRelativeError, referred to as MRE, and the calculation formula is:

[0093] ,

[0094] Among them, v pred is the dispersion curve predicted by the network model; v theory is the dispersion curve label of numerical simulation; N is the number of frequency calculations; Ω is the frequency calculation range; is the angular frequency.

[0095] Step S5: Establish an inversion objective function through a joint inversion method, use the low-frequency band to invert the shear wave velocity, and the high-frequency band to invert the mud velocity; this step is based on the shear wave velocity and mud speed Based on the sensitivity characteristics in different frequency bands, the inversion objective function E is established:

[0096] ,

[0097] in, For low frequency band, For high frequency band; and They are the formation dispersion data of low frequency band and high frequency band respectively; is the weight factor; is the angular frequency; d is the derivative symbol in the integral.

[0098] Step S6: In the process of obtaining the minimum value of the fitting residual in step S5, the optimization method is used to replace the interpolation inversion method to reduce the number of forward calculations in the inversion and improve the inversion efficiency. This step is for the shear wave velocity and mud speed The joint inversion method of the invention reduces the number of forward modeling times n required to be calculated in the inversion by an optimization method. The calculation formula of the method is:

[0099] ,

[0100] ,

[0101] ,

[0102] in, for location; for speed; is the gradient; The correction amount brought by the current gradient to the speed; is the learning rate; is the momentum coefficient.

[0103] Step S7, replace the current depth point with the next depth point, and return to step S6 to execute the optimization method until the shear wave velocity processing of the full depth segment is completed.

[0104] The technical effect achieved by this embodiment is as follows: the present invention provides a method for processing shear wave velocity of logging while drilling. On the basis of the theory of quadrupole acoustic wave propagation in logging while drilling, a dispersion forward model is obtained by neural network training using a machine learning method. The model can quickly calculate the theoretical dispersion curve. In inversion, the joint inversion method can effectively solve the problem of multiple solutions, and the optimization method can reduce the number of forward models and reduce the computing cost. Example 2

[0105] Based on the dispersion data set obtained in step S2 of Example 1, this embodiment provides a data cleaning method for shear wave velocity of logging while drilling. For a schematic diagram, see Figure 2 As shown, the data cleaning includes the following steps:

[0106] Step S3.1: Determine whether the sample is an error sample. If it is an error sample, execute step S3.2; if it is a correct sample, execute step S3.4;

[0107] Step S3.2: Correcting the error samples;

[0108] Step S3.3: Obtain the correct sample after correction, and execute step S3.4 on the sample;

[0109] Step S3.4: pre-training the neural network on the sample;

[0110] Step S3.5: Determine whether there are large fitting residual samples in the samples pre-trained with the neural network in step S3.4. The residual samples execute step S3.2, and the non-residual samples end the correction.

[0111] The technical effect achieved by this embodiment is: the data cleaning method provided by this embodiment can realize the correction of erroneous samples, and in the operation process it can quickly determine whether the obtained theoretical dispersion curve data is correct, and correct the erroneous data, and pre-train the correct samples and the corrected samples through the neural network, and output the corrected correct sample data, thereby realizing full coverage operation of the sample data. Example 3

[0112] Based on Examples 1 and 2, this embodiment provides an implementation process of a method for processing shear wave velocity of logging while drilling, and the specific implementation process is as follows:

[0113] Step S1: Take the formation parameters in the data set as sample features, and the dispersion curve corresponding to the parameter combination as the sample label. Before making the data set, it is necessary to perform sensitivity analysis on the 8 input formation parameters, and define the appropriate step size and parameter range according to the parameter sensitivity. The sensitivity calculation expression is as follows:

[0114] ,

[0115] Among them, Sensitivity is the parameter sensitivity; is the formation parameter; is the angular frequency; v phase is the theoretical dispersion curve; ∂ is the symbol for partial differential solution.

[0116] The influence of drill collar instruments on acoustic wave propagation cannot be ignored, so the dispersion equation of the uniform formation model while drilling needs to be derived before sensitivity analysis. Due to the increase in the boundary conditions and instrument parameters of the model, the dispersion equation takes a long time to solve, which is not conducive to the subsequent calculation of large-scale dispersion data sets. It is necessary to simplify it using the equivalent instrument theory. This method equates the hollow drill collar to a solid elastic rod. The simplified dispersion equation is as follows:

[0117] ,

[0118] The T matrix is ​​a 6×6 complex matrix, in which the 6 rows represent the influence of the 6 stress vectors, and the 1st, 3rd, and 5th columns of the 6 columns represent the inward propagating P waves, SH waves, and SV waves, and the 2nd, 4th, and 6th columns represent the diffuse P waves, SH waves, and SV waves. 11 and T 41 Made the following transformations:

[0119] ,

[0120] ,

[0121] Among them, n=1, 2 represent dipole and quadrupole sound fields respectively; is the wellbore radius; is the radial wave number of the borehole mud longitudinal wave; , are the modified Bessel functions of the first and second kinds, respectively; is the emission coefficient and The ratio of the instrument modulus added by the equivalent instrument theory and equivalent instrument radius Two parameter variables.

[0122] According to the derived dispersion equation, the sensitivity analysis diagrams of fast formation and slow formation parameters are obtained, such as Figure 3 As shown, Figure 3 (a) is the formation parameter sensitivity analysis of the fast formation equivalent instrument model. When the frequency approaches low frequency, the shear wave velocity has a high sensitivity, but as the frequency increases, the sensitivity of the mud velocity increases, and it plays a dominant role in the dispersion curve, while the sensitivity of the instrument modulus is very low and does not change with the increase of frequency. Different from the fast formation, Figure 3 (b) is a formation parameter sensitivity analysis diagram of shear wave velocity in slow formations. It can be seen from the figure that the shear wave velocity in slow formations has a high sensitivity in the entire frequency range, and the sensitivities of the remaining numbers are at a low value. The instrument modulus sensitivity is consistent with the trend in fast formations and is basically not affected by frequency factors.

[0123] Shear wave velocity, mud velocity, equivalent instrument radius and borehole radius are formation parameters that have a greater impact on the dispersion curve. In order to better describe the impact of the parameters on the dispersion curve, a relatively larger range and smaller step size will be set for them, which is conducive to fine characterization of sample characteristics. The instrument modulus is basically not affected by frequency in both fast and slow formations, and the dispersion curve has a low sensitivity to the instrument modulus, so it is not used as a sample characteristic parameter. Table 1 shows the parameter range and step size settings in the LWD quadrupole wave dispersion data set. A total of 861,840 formation parameter combinations were formed through permutations and combinations.

[0124] Table 1 Sample feature combinations of LWD quadrupole wave dispersion dataset

[0125] ,

[0126] Step S2: After determining the formation parameter combination, the corresponding dispersion curve can be obtained by solving the dispersion equation. By using the dispersion equation derived in step S1 using the equivalent instrument theory, the calculation cost of the solution is reduced, and the solution process can be expressed as:

[0127] ,

[0128] Where D is the dispersion matrix using the equivalent instrument theory; k is the wave number; is the angular frequency; F is the relevant parameter of elastic isotropic formation; B is the waveguide in the wellbore; M T is the instrument modulus; is the equivalent instrument radius.

[0129] Step S3: Due to the multi-solution nature of the dispersion equation, the solution strategy needs to be continuously adjusted during the solution process to ensure the correctness of the calculation, and the wrong labels in the data set need to be screened and corrected.

[0130] like Figure 2 As shown in the data cleaning flow chart of the dispersion data set, the data cleaning process screens out most of the error samples by setting the error sample evaluation index, and corrects the error samples. The corrected samples and the remaining samples are added to the neural network pre-training. Since most of the samples involved in the pre-training are correct, there are certain physical constraints between the input and output ends of the neural network, and a large fitting error will appear at the position of the error sample. Therefore, the fitting effect analysis of the trained model can screen out the remaining error samples. After a cycle, all error samples can be gradually corrected to correct samples.

[0131] Step S4: Use a feedforward neural network to perform model training. The neural network pre-training is as follows: Figure 4 As shown in Figure 1, the training content and process of the neural network model are shown. The structure of the model and the choice of algorithm during training directly affect the model performance. Figure 4 The neural network structure shown in the figure inputs formation parameters through 7 neurons in the input layer during training; the design of multiple hidden layers and nonlinear activation functions is used to increase the complexity of the model, making it better suited for regression fitting; the output layer outputs dispersion curve prediction points that are equidistant at 0.2kHz. The model can be expressed as a function of 7 independent variables:

[0132] ,

[0133] Among them, v pred is the dispersion curve predicted by the network model; ReLU is the nonlinear activation function; w i 、b i (i=1,2,3) represent the weight and bias between each connection layer respectively; x is the input sample feature value.

[0134] The model uses two layers of nonlinear activation functions and one layer of linear activation function, where the nonlinear activation function is The activation function, its mathematical expression is:

[0135] ,

[0136] The training set and validation set are divided into 95:5 ratios, and the model is optimized using the Resilient Propagation Algorithm (Rprop). This algorithm controls the change of the learning rate during training to prevent the learning rate from being too large or too small during training. The mathematical expression of the algorithm principle is as follows:

[0137] ,

[0138] in, is the gradient value obtained by parameter calculation; t is the number of training iterations; β is the learning rate; S is the final selected gradient size.

[0139] Since the input is 7 parameters and the data dimension is relatively high, this method of setting the upper and lower limits of the gradient change can effectively prevent the model from falling into a local minimum.

[0140] The loss function is of great significance in guiding the update direction of model parameters and reflecting the performance of the model. The loss function used in training is the mean square error loss (MSE), and its mathematical expression is:

[0141] ,

[0142] Where i and j represent the jth point on the i-th dispersion curve; N is the number of data sets; v pred is the dispersion curve predicted by the network model; v theory is the dispersion curve label of the numerical simulation. During training, when the loss function is greater than the threshold, the error is back-propagated, and the weights, biases, and hyperparameters are optimized through the elastic back-propagation algorithm, and the next round of training is repeated until the loss function is less than the set threshold, and the dispersion calculation model is obtained.

[0143] In order to evaluate the model training effect, the dispersion calculation model is predicted in the test set, and the mean relative error (MeanRelativeError, MRE) is used to measure the prediction effect. Its mathematical expression is:

[0144] ,

[0145] Among them, v pred is the dispersion curve predicted by the network model; v theory is the dispersion curve label of numerical simulation; N is the number of frequency calculations; Ω is the frequency calculation range; is the angular frequency.

[0146] Figure 5 This is the error statistics of the trained neural network model in the test set, where more than 99% of the sample errors are less than 0.4%, indicating that the overall convergence effect of the model is good and the generalization ability is strong. In most of the error cases, the prediction curve basically coincides with the theoretical dispersion curve, which is negligible compared with the measured dispersion error caused by the instrument measurement. The reason for the large error in some samples is that the input formation parameter combination is at the boundary of the data set, while this type of parameter combination rarely appears in actual logging, which does not affect the actual use of the model.

[0147] It has been verified that the prediction accuracy of the dispersion forward neural network model trained by the above algorithm is relatively high overall, and the model has strong generalization ability within the range of formation parameter values. The computational cost of solving the dispersion curve is greatly reduced in the calculation, and it can be used to replace the numerical solution process.

[0148] Step S5: Since the shear wave velocity and the mud velocity have high sensitivity in fast formations, the shear wave velocity and the mud velocity are selected as the inversion objects. To this end, the low frequency band and the high frequency band are connected by a weight factor using a joint inversion method, and the inversion objective function is established as follows:

[0149] ,

[0150] in, For low frequency band, For high frequency band; and They are the formation dispersion data of low frequency band and high frequency band respectively; is the weight factor; is the angular frequency; d is the derivative symbol in the integral.

[0151] This function calculates the fitting residual between the theoretical dispersion curve and the formation dispersion data. In the traditional method, the theoretical dispersion curve is obtained by numerically solving the dispersion equation. This step can quickly obtain the theoretical dispersion curve under the corresponding formation parameters according to the neural network forward model trained in step S4. Figure 6 (a) and (b) are the fitted residuals obtained by the numerical forward modeling method and the neural network model respectively. It can be seen that the distribution of the fitted residuals of the two methods is basically the same; Figure 6 (c) shows that the fitting curve after dispersion correction is in good agreement with the formation dispersion data. The two groups of comparisons together show that the application of the neural network model method in inversion is effective and feasible.

[0152] Step S6: Since the inversion objective function is continuous and differentiable, in the process of obtaining the minimum value of the fitting residual in step S5, the optimization method is used to replace the interpolation inversion method to further improve the inversion efficiency.

[0153] The optimization method used in the present invention is the momentum gradient descent method, which essentially uses gradient information and learning rate to determine the descent amplitude and direction, and then adjusts the step size and direction through the momentum parameter according to the gradient change direction to reduce oscillations during the convergence process, thereby improving the efficiency and stability of the optimization algorithm. Figure 7 In (a), this method can be understood as a three-dimensional descent process, during which the descent speed is gradually slowed down by the friction force; Figure 7In (b), the three-dimensional descent is projected onto two dimensions. When seeking the minimum value of the fitting residual, there is no need to calculate all the points on the two-dimensional surface. Instead, the minimum position of the curve can be found by directionally calculating the values ​​of several points according to the gradient changes, which effectively improves the computational efficiency of the inversion solution.

[0154] Step S7, repeat the contents of step S6 at each depth point until the shear wave velocity processing of all depth segments is completed. Figure 8 This is the result of fast shear wave velocity processing using neural network and optimization method. The theoretical dispersion curve after dispersion correction is highly consistent with the formation dispersion data, and the inversion value of shear wave velocity is basically consistent with the formation value. This fast shear wave velocity processing method, device, equipment and storage medium based on supervised machine learning model has the characteristics of efficient calculation and stable inversion, and provides a new processing strategy for shear wave dispersion processing in drilling environment.

[0155] The technical effect achieved by this embodiment is: the present invention provides a specific content of the processing of shear wave velocity of logging while drilling. On the basis of the theory of quadrupole acoustic wave propagation of logging while drilling, a dispersion forward model is obtained by neural network training using a machine learning method. The model can quickly calculate the theoretical dispersion curve; in the inversion, the joint inversion method can effectively solve the problem of multiple solutions, and the optimization method can reduce the number of forward models and further reduce the calculation cost. Compared with the traditional method, the present invention greatly reduces the calculation cost in the processing of shear wave velocity of logging while drilling, which is conducive to the application of this technology in the logging field and realizes the efficient correction of the original measured dispersion data. Example 4

[0156] Based on embodiments 1-3, this embodiment provides a device for executing a method for processing shear wave velocity of logging while drilling, the device comprising: a sensitivity analysis module, a data set construction module, a data correction module, a feasibility verification module, an inversion module, an optimization module, and a loop iteration module;

[0157] The sensitivity analysis module divides the parameter range and step size according to the sensitivity analysis of different parameters of LWD to quadrupole shear wave dispersion, determines the formation parameter combination, and uses it as the sample feature of the data set;

[0158] The data set construction module solves the dispersion equation numerically output by the sensitivity analysis module, obtains the dispersion curve corresponding to the formation parameter combination as the data set sample label, and thus constructs the dispersion data set;

[0159] The data correction module cleans the dispersion data set output by the data set construction module and corrects the erroneous samples in the data set;

[0160] The feasibility verification module builds a neural network and trains it on the dispersion data set obtained by the data correction module to obtain a neural network model that can quickly perform dispersion forward modeling. The model is then subjected to prediction error analysis and generalization analysis to verify the feasibility of the model.

[0161] The inversion module obtains the theoretical dispersion curve under the corresponding formation parameters according to the neural network forward model trained by the feasibility verification module; establishes the inversion objective function through the joint inversion method, adopts the low-frequency band inversion of shear wave velocity and the high-frequency band inversion of mud velocity as the inversion object; calculates the fitting residual between the theoretical dispersion curve and the formation dispersion data;

[0162] The optimization module uses the optimization method to replace the interpolation inversion method in the process of obtaining the minimum value of the fitting residual in the inversion module, thereby reducing the number of forward calculations in the inversion and improving the inversion efficiency;

[0163] The loop iteration module replaces the current depth point with the next depth point, and returns to the optimization module to execute the optimization method until the shear wave velocity processing of the full depth segment is completed.

[0164] The technical effect achieved by this embodiment is: this embodiment provides a device for executing the shear wave velocity processing method of logging while drilling, which can directly analyze the collected shear wave dispersion data to achieve data processing and correction. Example 5

[0165] Based on Examples 1-4, this example provides an electronic device, and the specific examples of this application do not limit the specific implementation of the computing device.

[0166] The electronic device includes: a processor, a memory, and a computer program; the computer program is stored in the memory; the computer program includes instructions, and the instructions can be specifically used to execute the LWD shear wave velocity processing method in any of the above-mentioned method embodiments. The specific implementation of the instructions can refer to the corresponding steps in the embodiment of the LWD shear wave velocity processing method and the corresponding description in the device, which will not be repeated here. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described equipment can refer to the corresponding process description in the aforementioned method embodiment, which will not be repeated here.

[0167] The algorithm or display provided here is not inherently related to any specific computer, virtual system or other equipment. Various general systems can also be used together with the teaching based on this. According to the above description, it is obvious to construct the structure required for this type of system. In addition, the present application embodiment is not directed to any specific programming language yet. It should be understood that various programming languages ​​can be utilized to realize the content of the present application described here, and the above description of specific languages ​​is to disclose the best mode of implementation of the present application.

[0168] The technical effect achieved by this embodiment is: an electronic device provided by this embodiment realizes the processing and storage of the shear wave velocity processing method of logging while drilling.

[0169] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for processing shear wave velocity of logging while drilling, characterized in that: The method comprises the following steps: Step S1, according to the sensitivity analysis of different parameters of logging while drilling to quadrupole shear wave dispersion, divide the parameter range and step size, determine the formation parameter combination, and use it as the sample feature of the data set; Step S2, numerically solving the dispersion equation of the data set samples to obtain the dispersion curve corresponding to the formation parameter combination as the data set sample label, thereby constructing a dispersion data set; Step S3, performing data cleaning on the dispersion data set to correct erroneous samples in the data set; Step S4, establishing a neural network, training it on the dispersion data set obtained in step S3, obtaining a neural network model capable of fast dispersion forward modeling, performing prediction error analysis and generalization analysis on the model, and verifying the feasibility of the model; Step S5, obtaining the theoretical dispersion curve under the corresponding formation parameters according to the neural network forward model trained in step S4; establishing the inversion objective function through the joint inversion method, using the low-frequency band inversion shear wave velocity and the high-frequency band inversion mud velocity as the inversion object; calculating the fitting residual between the theoretical dispersion curve and the formation dispersion data; Step S6, in the process of obtaining the minimum value of the fitting residual in step S5, the interpolation inversion method is replaced by the optimization method, so as to reduce the number of forward modeling calculations in the inversion and improve the inversion efficiency; Step S7, replacing the current depth point with the next depth point, and returning to step S6 to execute the optimization method until the shear wave velocity processing of the full depth segment is completed; The step S4 determines the optimal structure of the neural network according to the grid optimization method; trains the established neural network model, and uses the mean square error loss to analyze the model training process during the training; and uses the average relative error to analyze the prediction error of the trained model; The neural network calculation formula is: ， Among them, v pred Predicted dispersion curves for the network model; is a nonlinear activation function; w i 、b i Represent the weights and biases between each connection layer; is the input sample feature value; The mean square error loss calculation formula is: ， Among them, v theory is the dispersion curve label of numerical simulation; v pred is the dispersion curve predicted by the network model; i, j represents the jth point on the i-th group of dispersion curves; N is the number of data sets; The average relative error calculation formula is: ， Among them, v pred is the dispersion curve predicted by the network model; v theory is the dispersion curve label of numerical simulation; N1 is the number of frequency calculations; Ω is the frequency calculation range; is the angular frequency; The step S5 is based on the shear wave velocity and mud speed The sensitivity characteristics in different frequency bands will Incorporating the inversion function of shear wave velocity improves the dispersion fitting effect and establishes the inversion objective function E: ， in, For low frequency band, For high frequency band; and They are the formation dispersion data of low frequency band and high frequency band respectively; is the weight factor; is the angular frequency; d is the derivative symbol in the integral.

2. A method for processing shear wave velocity of logging while drilling according to claim 1, characterized in that: The sensitivity analysis in step S1 specifically analyzes the influence of different formation parameters on the dispersion curve in different frequency bands, and the calculation formula is: ， Among them, Sensitivity is the parameter sensitivity; is the formation parameter; is the angular frequency; v phase is the theoretical dispersion curve; ∂ is the symbol for partial differential solution.

3. A method for processing shear wave velocity of logging while drilling according to claim 2, characterized in that: The step S2 establishes a dispersion equation according to an equivalent instrument theoretical model of a uniform formation while drilling, and solves dispersion curves corresponding to different parameter combinations based on the dispersion equation as sample labels to produce a data set, wherein the different parameter combinations are sample features; the data set calculation formula is: ， Where D is the dispersion matrix using the equivalent instrument theory; k is the wave number; is the angular frequency; F is the relevant parameter of elastic isotropic formation; B is the waveguide in the wellbore; M T is the instrument modulus; is the equivalent instrument radius.

4. A method for processing shear wave velocity of logging while drilling according to claim 3, characterized in that: The step S3 performs data cleaning on all data sets, and the data cleaning includes the following steps: Step S3.1: Determine whether the sample is an error sample. If it is an error sample, execute step S3.2; if it is a correct sample, execute step S3.4; Step S3.2: classify the error samples according to the reasons for the iteration errors, and update different initial iteration parameters for different categories to iteratively correct the error samples; Step S3.3: Obtain the correct sample after correction, and execute step S3.4 on the sample; Step S3.4: pre-training the neural network on the sample; Step S3.5: Determine whether there are large fitting residual samples in the samples pre-trained with the neural network in step S3.

4. The residual samples execute step S3.2, and the non-residual samples end the correction.

5. A method for processing shear wave velocity of logging while drilling according to claim 4, characterized in that: The step S6 is the shear wave velocity and mud speed The joint inversion method reduces the number of forward modeling times n required to be calculated in the inversion by an optimization method, and the calculation formula is: ， ， ， in, The correction amount brought by the current gradient to the speed; is the learning rate; is the gradient; for speed; is the momentum coefficient; For location.

6. A shear wave velocity processing device for logging while drilling for executing the method according to any one of claims 1 to 5, characterized in that: The device comprises: The sensitivity analysis module divides the parameter range and step size according to the sensitivity analysis of different parameters of LWD to quadrupole shear wave dispersion, determines the formation parameter combination, and uses it as the sample feature of the data set; The data set construction module solves the dispersion equation numerically output by the sensitivity analysis module, obtains the dispersion curve corresponding to the formation parameter combination as the data set sample label, and thus constructs the dispersion data set; The data correction module cleans the dispersion data set output by the data set construction module and corrects the erroneous samples in the data set; The feasibility verification module builds a neural network and trains it on the dispersion data set obtained by the data correction module to obtain a neural network model that can quickly perform dispersion forward modeling. The model is then subjected to prediction error analysis and generalization analysis to verify the feasibility of the model. The inversion module obtains the theoretical dispersion curve under the corresponding formation parameters according to the neural network forward model trained by the feasibility verification module; establishes the inversion objective function through the joint inversion method, adopts the low-frequency band inversion of shear wave velocity and the high-frequency band inversion of mud velocity as the inversion object; calculates the fitting residual between the theoretical dispersion curve and the formation dispersion data; The optimization module uses the optimization method to replace the interpolation inversion method in the process of obtaining the minimum value of the fitting residual in the inversion module, thereby reducing the number of forward calculations in the inversion and improving the inversion efficiency; The loop iteration module replaces the current depth point with the next depth point, and returns to the optimization module to execute the optimization method until the shear wave velocity processing of the full depth segment is completed.

7. An electronic device, characterized in that: The electronic device comprises: a processor, a memory, and a computer program; The computer program is stored in the memory; The computer program includes instructions, and when the instructions are executed by the processor, the electronic device executes the method for processing shear wave velocity of logging while drilling as claimed in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for processing shear wave velocity of logging while drilling according to any one of claims 1 to 5.

Images