A shear wave velocity prediction method, device, equipment and readable storage medium

Abstract

The application discloses a shear wave velocity prediction method, device and equipment and a readable storage medium, and is applied to the field of geophysical technology, and comprises the following steps: a shear wave velocity prediction model is constructed in advance based on a KAN network and a sequence displacer; an activation function of the shear wave velocity prediction model is a B-spline base function; logging data of a target area is acquired; and the logging data is input into the shear wave velocity prediction model to obtain a prediction result. The shear wave velocity prediction model based on the KAN network has robustness and precision, and the model is combined with the sequence displacer architecture, so that the model is fused with the KAN network to further perform feature extraction, through a dynamic weight fusion mechanism and a learnable B-spline base function, efficient modeling of the logging data is realized, the model can be used in large oilfields on a large scale, various complex geological conditions can be coped with, shear wave velocity prediction can be accurately performed, and in turn, exploration and development risks can be reduced, and the benefit development of large oilfields can be ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of geophysical technology, in particular to a shear wave velocity prediction method, device, equipment and readable storage medium. BACKGROUND

[0002] Shear wave (shear wave) prediction plays an irreplaceable role in the field of petroleum geology, and its core is to provide key basis for reservoir lithology identification, fracture detection and fluid property discrimination through P-S wave velocity ratio (Vs / Vp) and elastic parameter inversion. At present, the algorithm level in the field of shear wave velocity prediction cannot accurately predict the shear wave velocity under the existing complex geological conditions, and traditional methods such as empirical geophysical model rely on specific physical equations tailored for different types of reservoirs, which is time-consuming and laborious in application.

[0003] Therefore, how to provide a method for accurately predicting shear wave velocity in various complex geological conditions is a technical problem to be solved at present. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a shear wave velocity prediction method, device, equipment and readable storage medium, which solves the problems of poor accuracy and adaptability in the prior art.

[0005] To solve the above technical problems, the present application provides a shear wave velocity prediction method, comprising:

[0006] A shear wave velocity prediction model is constructed in advance based on a KAN network and a sequence permutor; the activation function of the shear wave velocity prediction model is a B-spline basis function;

[0007] Obtain the logging data of the target area;

[0008] Input the logging data into the shear wave velocity prediction model to obtain a prediction result.

[0009] Optionally, inputting the logging data into the shear wave velocity prediction model to obtain a prediction result comprises:

[0010] Input the logging data into the embedding layer of the shear wave velocity prediction model to obtain a mapping result;

[0011] Input the mapping result into the feature interaction layer of the shear wave velocity prediction model to obtain a feature vector; the feature interaction layer is constructed based on the KAN network and combined with the sequence permutor;

[0012] Input the feature vector into the classification layer of the shear wave velocity prediction model to obtain the prediction result.

[0013] Optionally, the logging data is input into an embedding layer of the shear wave velocity prediction model to obtain a mapping result, including:

[0014] A one-dimensional convolution operation is performed on local features of the logging data by a convolution layer of the shear wave velocity prediction model, initial internal input feature variables are extracted, and a dimension of the internal input feature variables of the data is mapped to an internal channel dimension D of the shear wave velocity prediction model to obtain the mapping result ;

[0015] A formula expression of the embedding layer is:

[0016] ;

[0017] wherein, is a parameter of a convolution kernel at a position r and a channel d; R is a kernel size; bd is a bias; and L is a depth of data sampling. is logging data.

[0018] Optionally, the mapping result is input into a feature interaction layer of the shear wave velocity prediction model to obtain a feature vector, including:

[0019] The mapping result is input into a plurality of the feature interaction layers of the shear wave velocity prediction model in sequence to obtain the feature vector; each of the feature interaction layers includes a sequence permuter structure feature extraction module and a fully connected layer.

[0020] A formula expression of the feature interaction layer is:

[0021] ;

[0022] wherein, is an output of a current feature interaction layer. is a feature interaction layer. is an input of the current feature interaction layer; K is a total number of the feature interaction layers; PKAN is the sequence permuter structure feature extraction module; and LP is the fully connected layer.

[0023] Optionally, the sequence permuter structure feature extraction module includes:

[0024] a channel permutation KAN and a depth permutation KAN.

[0025] Correspondingly, a formula expression of the sequence permuter structure feature extraction module PKAN is:

[0026] ;

[0027] ;

[0028] ;

[0029] Correspondingly, the formula expression of the feature interaction layer is :

[0030] = (α +β ) ;

[0031] ;

[0032] wherein, is a channel replacement KAN; is a depth replacement KAN; is an output result of the channel replacement KAN; is an output result of the depth replacement KAN; and are dynamic weights of the channel and the depth respectively; is a normalization function; is an average value.

[0033] Optionally, before the feature vector is input into a classification layer of the S-wave velocity prediction model to obtain the prediction result, the method further comprises the following steps of:

[0034] inputting the feature vector into a full connection layer and a normalization layer of the S-wave velocity prediction model and mapping into a regressor; the formula is: ;

[0035] wherein, ∈R D×1 is a regressor weight; is an output of the regressor; is a normalization layer; is a full connection layer; is a final output of the feature interaction layer.

[0036] Optionally, the logging data comprises:

[0037] density data, photoelectric index data, deep lateral resistivity data, shallow lateral resistivity data, natural gamma data and compensated neutron data.

[0038] The application further provides an S-wave velocity prediction device, comprising:

[0039] a model construction module, configured to construct an S-wave velocity prediction model in advance based on a KAN network and a sequence replacement device; an activation function of the S-wave velocity prediction model is a B-spline base function;

[0040] The well logging data acquisition module is configured to acquire well logging data of a target region.

[0041] The prediction module is configured to input the well logging data into the S-wave velocity prediction model to obtain a prediction result.

[0042] The present application further provides a S-wave velocity prediction device, comprising:

[0043] The memory is configured to store a computer program.

[0044] The processor is configured to implement the S-wave velocity prediction method as described above when executing the computer program.

[0045] The present application further provides a computer readable storage medium, wherein the computer readable storage medium stores computer executable instructions, and the computer executable instructions are loaded and executed by a processor to implement the S-wave velocity prediction method as described above.

[0046] It can be seen that the present application constructs a S-wave velocity prediction model in advance based on a KAN network and a sequence permutor; the activation function of the S-wave velocity prediction model is a B-spline basis function; well logging data of a target region is acquired; and the well logging data is input into the S-wave velocity prediction model to obtain a prediction result. The S-wave velocity prediction model based on the KAN network has robustness and precision, and the model is combined with the sequence permutor architecture, so that it is fused with the KAN network to perform feature extraction, and through a dynamic weight fusion mechanism and a learnable B-spline basis function, efficient modeling of the well logging data is realized. The model can be used in large oilfields on a large scale to cope with various complex geological conditions, accurately perform S-wave velocity prediction, and thus reduce exploration and development risks and ensure the benefit development of large oilfields.

[0047] In addition, the present application further provides a S-wave velocity prediction device, equipment and computer readable storage medium, which also have the above beneficial effects. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of the provided drawings.

[0049] Figure 1 A flowchart of a S-wave velocity prediction method provided by the present application embodiment;

[0050] Figure 2 A flowchart of a S-wave velocity prediction model provided by the present application embodiment;

[0051] Figure 3 A clastic rock single well different depth learning S-wave velocity prediction result comparison chart provided for an embodiment of the present application;

[0052] Figure 4 A carbonate rock single well different depth learning S-wave velocity prediction result comparison chart provided for an embodiment of the present application;

[0053] Figure 5 A volcanic rock single well different depth learning S-wave velocity prediction result comparison chart provided for an embodiment of the present application;

[0054] Figure 6 A structure schematic diagram of a S-wave velocity prediction device provided for an embodiment of the present application;

[0055] Figure 7 A structure schematic diagram of a S-wave velocity prediction device provided for an embodiment of the present application. DETAILED DESCRIPTION

[0056] To make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0057] S-wave (shear wave) prediction plays an irreplaceable role in the field of petroleum geology, and its core is to provide key basis for reservoir lithology identification, fracture detection, fluid property discrimination and sweet spot prediction through P-S wave velocity ratio (Vs / Vp) and elastic parameter inversion. First, S-wave velocity is more sensitive to rock skeleton, which can effectively distinguish complex lithology, such as dolomite and mudstone in carbonate rock, or sandstone and mudstone ratio in clastic rock. Second, S-wave splitting phenomenon (fast and slow S-wave) is an important tool for detecting fractures. Fast wave (S1) propagates along the fracture surface, and slow wave (S2) is perpendicular to the fracture surface. The amplitude difference between the two can quantify the fracture density and direction, providing basis for horizontal well trajectory design and significantly improving drilling success rate. In addition, the comprehensive analysis of S-wave and P-wave amplitude ratio can determine whether the "bright spot" is a gas layer. For example, when P-wave amplitude increases and S-wave does not change significantly, it indicates a gas-bearing reservoir. This method is particularly important in natural gas exploration. At the same time, S-wave velocity prediction combined with porosity, brittleness index and other parameters can optimize the fracturing scheme and improve the development efficiency of tight reservoirs.

[0058] However, the current shear wave velocity prediction is faced with the following two problems: firstly, the measured shear wave data is scarce and of uneven quality, especially in deep or complex formations, the cost of shear wave logging is high and the technical difficulty is great, leading to the prediction relying on empirical formula (such as Greenberg-Castagna model, that is, a shear wave velocity prediction method established by applying statistical method according to rock sample longitudinal wave velocity and shear wave velocity measurement data and logging data) or theoretical model (such as Xu-White model), but the applicability of these models is limited in areas with strong heterogeneity and complex mineral composition (such as carbonate rock containing argillaceous or anhydrite), and needs to be corrected in combination with regional geological characteristics. Secondly, the problem of multiple solutions is prominent, especially in the fracture development area or the mixed phase of fluid, the shear wave response is easily disturbed by the pore structure, pressure change and fluid saturation, for example, the dynamic change of reservoir elastic parameters after CO2 injection needs to rely on the joint modeling of Hertz-Mindlin (a contact model for describing particle contact) and Gassmann equation (the equation establishes the relationship between rock bulk modulus, porosity, pore fluid bulk modulus and rock skeleton bulk modulus), but the uncertainty of key parameters such as coordination number increases the prediction difficulty.

[0059] It can be seen that the inherent complexity of the current oil and gas reservoir makes it difficult to establish a high-precision comprehensive model for each field. At present, the algorithm level in the field of shear wave velocity prediction cannot accurately predict the shear wave velocity under the existing complex geological conditions, and the traditional method of empirical geophysical model depends on specific physical equations tailored for different types of reservoirs, which is time-consuming and laborious.

[0060] Therefore, the present application proposes a shear wave velocity prediction model based on KAN (Kolmogorov-Arnold Networks, a new type of neural network architecture) network and sequence permutation period, which is called SuperKAN model. The model combines the architecture of sequence permutation and can be used in large oilfields on a large scale to cope with various complex geological conditions and accurately predict shear wave velocity, thereby reducing the risk of exploration and development and ensuring the benefit development of large oilfields.

[0061] For details, please refer to Figure 1 , Figure 1 A flowchart of a shear wave velocity prediction method provided by the embodiment of the present application. The method can include:

[0062] S101: Constructing a shear wave velocity prediction model based on KAN network and sequence permutation in advance; the activation function of the shear wave velocity prediction model is B-spline basis function.

[0063] The execution subject of the embodiment is a terminal. The embodiment does not limit the type of terminal as long as it can complete the operation of the shear wave velocity prediction method.

[0064] The core function of the sequence replacer is the structured reorganization and explainable processing of sequence data, which essentially realizes the modular reorganization of input sequences by dynamically adjusting the order of neuron connections, thereby revealing the dependency between variables. The KAN network is an adaptive activation network designed based on the Kolmogorov-Arnold theorem, which realizes the nonlinear mapping and feature expression of input data through a learnable B-spline basis function and a dynamic grid adjustment mechanism. Define the input data X ∈ N ×Cin×Din , N is the number of input samples, C in is the first feature dimension of the input data X, and D in is the second feature dimension of the input data X. The final output of the KAN network is determined by the basis function expansion result and linear transformation:

[0065] .

[0066] Where Y is the output target of the current layer; is the learnable weight; B ∈ N×Din×(T×P) is the basis function tensor, T is the spline order, and P is the number of segments; W spline is the spline coefficient.

[0067] The adaptive activation calculation process in the KAN network first calculates the basis function tensor through the initial grid tensor, then obtains the spline coefficient as the network weight, and then updates G in the back propagation process, while updating the basis function tensor, to achieve the adaptive effect. For details, please refer to the following process:

[0068] (1) Perform basis expansion through the B-spline basis function to obtain the basis function tensor. The construction of the basis function tensor is based on the uniform distribution of the preset grid range (G=[a,b]), and then it is expanded recursively:

[0069] .

[0070] Where G ∈ Din×(K×S) is the initial grid tensor.

[0071] (2) KAN fits the spline coefficient W spline by least squares method:

[0072] ;

[0073] Where W is the coefficient vector defined on each grid interval, which is used to combine the B-spline basis function to form the final spline function.

[0074] (3) To realize the self-adaptation to the input data X distribution, the KAN network dynamically adjusts the grid position G in the training process adaptive In the back propagation process, the grid is expanded at both ends to support recursive calculation:

[0075] ;

[0076] .

[0077] wherein, is a mixing coefficient; G adaptive is a dynamically adjusted grid position; X sorted is the result of sorting the input data from large to small; I T is an index vector; is the new position of the dynamically adjusted grid.

[0078] It can be seen that the KAN network realizes more fine and transparent feature expression of the internal law of data through the dynamically learnable activation function and the hierarchical function combination mechanism. The KAN network parameterizes each connection edge as a differentiable spline function, allowing the model to adaptively adjust the nonlinear form of each feature interaction in the training process, thereby directly capturing the basic mathematical relationship between the input and the output in a shallow network with fewer parameters. This hierarchical feature construction makes the feature learning process change from the traditional neural network's fuzzy mapping to a traceable and intervenable symbolic reasoning path.

[0079] Further, on the basis of the KAN network, a feature interaction layer is established by fusing a sequence permuter, and then a high-efficiency modeling of the logging data can be realized through a dynamic weight fusion mechanism and a learnable B-spline basis function to obtain a shear wave velocity prediction model, which includes an embedding layer, a feature interaction layer and a classification layer. For details, please refer to Figure 2 , Figure 2 is a flowchart of a shear wave velocity prediction model provided by an embodiment of the present application.

[0080] S102: Obtain the logging data of a target area.

[0081] In well logging prediction, data selection and collection are the core to ensure the accuracy and reliability of the model. Reasonable screening of logging parameters can enhance the geological pertinence of the model and avoid redundant interference. And high-quality logging data should be collected to avoid the influence of equipment and human factors, so that the logging data can truly reflect the characteristics of the formation and reduce noise and errors. Systematic data collection needs to cover multi-dimensional information of target horizons to ensure the representativeness and integrity of samples, thereby improving the evaluation accuracy of the model. In this embodiment, six logging data related to the physical properties and lithological characteristics of rocks are selected, including density (DEN), photoelectric index (PE), deep lateral resistivity (LLD), shallow lateral resistivity (LLS), natural gamma (GR) and compensated neutron (CNL).

[0082] S103: input the logging data into the shear wave velocity prediction model to obtain a prediction result.

[0083] The shear wave velocity prediction model of this step is the shear wave velocity prediction model constructed in step S101, which is composed of three parts of embedding layer, feature interaction layer and classification layer. The model combines the sequence displacer architecture, which is combined with the KAN network to extract features, and realizes efficient modeling of logging data through a dynamic weight fusion mechanism and a learnable B-spline basis function.

[0084] Further, the above inputting the logging data into the shear wave velocity prediction model to obtain a prediction result can specifically include:

[0085] Step 11: input the logging data into the embedding layer of the shear wave velocity prediction model to obtain a mapping result.

[0086] It should be noted that the embedding layer in this embodiment is a one-dimensional convolution layer. Through the convolution layer of the shear wave velocity prediction model, one-dimensional convolution operation is performed on the local features of the logging data ∈ N×C×L to extract the initial internal input feature variable X 0 ∈ N×D×L and map the dimension of the internal input feature variable of the data to the internal channel dimension D of the shear wave velocity prediction model to obtain a mapping result ; the formula expression of the embedding layer is:

[0087] ;

[0088] wherein, R represents the parameters of the convolution kernel at position r and channel d; R is the kernel size; bd is the bias, specifically the bias term of the d-th channel of the output feature map, and the bias bd is a learnable parameter; L is the depth of data sampling. This is well logging data.

[0089] Step 12: Input the mapping result into the feature interaction layer of the shear wave velocity prediction model to obtain the feature vector.

[0090] It should be noted that the feature interaction layer in this embodiment is constructed based on a KAN network and combined with a sequence permutation unit. This embodiment utilizes stacked K-layer feature interaction layers to achieve multi-scale feature interaction. Specifically, the mapping result is input into multiple consecutive feature interaction layers of the shear wave velocity prediction model to obtain feature vectors; each feature interaction layer includes a sequence permutation unit structure feature extraction module and a fully connected layer; the formula expression for the feature interaction layer is:

[0091] ;

[0092] in, This is the output of the current feature interaction layer; For feature interaction layer; is the input to the current feature interaction layer; K is the total number of feature interaction layers; PKAN is the sequence permuter structure feature extraction module; LP is the fully connected layer.

[0093] like Figure 2 As shown, the sequence permuter structural feature extraction module in this embodiment consists of two parallel feature extraction branches: channel permutation (KAN) and depth permutation (KAN). The sequence permuter is a dynamic feature interaction module designed for lithological data. Through the fusion of the channel-space dual-branch architecture with the KAN basis function, it achieves adaptive modeling of the nonlinear coupling relationship between logging parameters.

[0094] Correspondingly, the formula expression for the sequence permuter structural feature extraction module PKAN is:

[0095] ;

[0096] ;

[0097] ;

[0098] Correspondingly, the formula expression of the feature interaction layer for:

[0099] = (α) + β ) ;

[0100] ;

[0101] wherein, is a channel-wise replacement KAN; is a depth-wise replacement KAN; is an output result of the channel-wise replacement KAN; is an output result of the depth-wise replacement KAN; and are dynamic weights for the channel and the depth, respectively; is a normalization function; is an average value.

[0102] It can be seen that, for the input data of each layer, the parallel passing through the channel branch and the spatial branch respectively obtains and . In order to enable the depth branch to obtain the depth characteristics of the logging data, the features are first transposed and reconstructed, and after feature extraction by the KAN layer, the original dimension of the input data is restored. Subsequently, the dynamic weights and of the two branches are calculated; finally, the sequence replacer multiplies the structures of the two branches with the channel-depth weights respectively, calculates the dynamic feature values, and adds the original data to obtain the output . The sequence replacer realizes feature extraction of the logging data in the channel and the depth directions, and realizes adaptive allocation of the channel-depth by combining the feature contribution degree.

[0103] Step 13: input the feature vector into the classification layer of the shear wave velocity prediction model to obtain a prediction result.

[0104] In the embodiment, the classification layer is used to output the prediction result. Further, before the feature vector is input into the classification layer of the shear wave velocity prediction model to obtain the prediction result, the above can further include:

[0105] input the feature vector into the fully connected layer and the normalization layer of the shear wave velocity prediction model, and map to the regressor; the formula is: ;

[0106] wherein, ∈R D×1 is a regressor weight; is an output of the regressor; is a normalization layer; is a fully connected layer; is the final output of the feature interaction layer.

[0107] The application embodiment provides a shear wave velocity prediction method, a shear wave velocity prediction model is constructed in advance based on a KAN network and a sequence displacer; an activation function of the shear wave velocity prediction model is a B-spline basis function; logging data of a target area is acquired; and the logging data is input into the shear wave velocity prediction model to obtain a prediction result. The shear wave velocity prediction model based on the KAN network has robustness and accuracy, and the model is combined with the sequence displacer architecture, so that the KAN network is fused to perform feature extraction, and through a dynamic weight fusion mechanism and a learnable B-spline basis function, efficient modeling of the logging data is realized. The model can be used in large oilfields on a large scale to cope with various complex geological conditions, accurately predict shear wave velocity, and thus reduce exploration and development risks and ensure the benefit development of large oilfields. Moreover, the shear wave velocity prediction model established by the fusion of the sequence displacer and the KAN network combines the channel-space separation interaction mechanism of the sequence displacer and the nonlinear expression ability and interpretability of the KAN network, and enhances the feature extraction effect of the shear wave velocity prediction model on the logging data.

[0108] In order to verify the feasibility of the method, volcanic rock reservoirs and clastic rock reservoirs are selected as research objects, and the two structural blocks have complex geological structures, which can well verify the feasibility of the model. The data set records six logging data related to the density (DEN), photoelectric index (PE), deep lateral resistivity (LLD), shallow lateral resistivity (LLS), natural gamma (GR) and compensated neutron (CNL) of the physical properties and lithological characteristics of the rock. Four evaluation criteria of Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and correlation coefficient R2 are used for analysis. The specific introduction is as follows:

[0109] MAPE is a statistical index for measuring the error between predicted values and actual values, which is the average value of percentage error, and is commonly used to evaluate the prediction accuracy of regression models. Its characteristics are to express the error in percentage form, and to treat positive and negative errors equally, reflecting the average deviation of predicted values from actual values. The formula of MAPE is as follows:

[0110] .

[0111] Wherein, s is the total number of samples, yi is the true value of the i th sample. is the predicted value of the i th sample.

[0112] MAE is also a regression analysis indicator that measures the average difference between predicted and actual values. It reflects the accuracy of model prediction by calculating the average of the absolute errors of all sample points, and the smaller the value, the better the prediction effect. The MAE formula is as follows:

[0113] .

[0114] RMSE is a commonly used error measurement indicator in regression analysis, used to measure the average deviation between predicted and actual values. It is obtained by taking the square root of the mean square error (MSE), and essentially reflects the standard deviation level difference between predicted and actual values. The smaller the value, the higher the prediction accuracy. The formula is as follows:

[0115] .

[0116] R2 (R-Squared) is a core indicator in regression analysis for measuring the fitting effect of the model. It reflects the similarity between the measured and predicted results, and the value range in practical application is [0, 1]. The closer R2 is to 1, the better the model fitting effect, and the higher the precision of the established model. The formula of R2 is as follows:

[0117] .

[0118] where SST (Total Sum of Squares) is the sum of squared errors between the actual value and the mean. The formula of SST is as follows:

[0119] .

[0120] where, is the mean of the actual value. SSE (Sum of Squared Errors) is the sum of squared errors between the actual value and the predicted value. The formula of SSE is as follows:

[0121] .

[0122] SSR (Sum of Squared Regression) is the sum of squared errors between the predicted value and the mean. The formula of SSR is as follows:

[0123] .

[0124] To verify the effectiveness of the model in the task of predicting shear wave velocity, eleven mainstream comparative models (CNN (Convolutional Neural Network), TCN (Temporal Convolution Network), LSTM (Long Short-Term Memory Network), GRU (Gated Recurrent Unit), N-BEATS (Neural Network-Based Time Series Prediction Model), ViT (Visual Model Based on Transformer Architecture), Informer (A Transformer Model Designed for Long Sequence Time Series Prediction), MLP-Mixer (Multi-Layer Perceptron), TSMixer (Model Based on Multi-Layer Perceptron), PatchMixer (A New Model Based on CNN for Long Time Series Prediction), and MSD-Mixer (Multi-Granularity Patch Division Time Series Model)) were selected for systematic comparative experiments, and the prediction performance of each model was comprehensively evaluated from multiple dimensions such as MAPE, MAE, RMSE, and R2. As shown in Table 1, SuPerKAN (the shear wave velocity prediction model of the present application) achieved the best results in all four indicators. Specifically, the MAPE of SuPerKAN was only 1.88%, which was about 9% lower than the second-ranked TSMixer (2.07%), indicating that it had the smallest relative error on different amplitude data. In terms of MAE, SuPerKAN performed significantly better than the second-best TSMixer, with a reduction of 5.02. In terms of RMSE, SuPerKAN was the best, reaching 66.29, indicating that it had the strongest suppression effect on extreme deviations. The R2 of SuPerKAN reached 0.9470, meaning that the model could explain about 94.7% of the observed variance, which was higher than the second-best Informer (0.9418) and TSMixer (0.9404), and its fitting accuracy was the most reliable when predicting shear wave velocity. In contrast, the R2 of traditional CNN, TCN, LSTM, and GRU models was mostly in the range of 0.89-0.92, reflecting the insufficient ability of single convolution or recurrent structure to capture complex geological time series features. The overall architecture of MLP-Mixer, TSMixer, PatchMixer, and MSD-Mixer was better than that of traditional networks, among which Informer, TSMixer, and MSD-Mixer introduced time attention mechanism and multi-scale information processing, and the R2 was improved to 0.91-0.94. SuPerKAN effectively balanced global dependence and local details through the synergistic effect of multi-scale complementary feature fusion and self-attention mechanism, and had stronger modeling capability for common non-stationary changes (such as sudden enrichment and periodic fluctuations) in geological sequences.In summary, among the twelve compared deep learning architectures, SuPerKAN achieved the optimal effect of shear wave velocity prediction with the lowest MAPE, MAE, RMSE and the highest R2, which has significant application potential and promotion value, and also shows the superiority of SuPerKAN in the algorithm aspect.

[0125] Table 1 shear wave velocity (Vs) prediction results of the clastic rock data set using various deep learning methods (the best result is shown in bold)

[0126]

[0127] Figure 3 A clastic rock single well different depth learning shear wave velocity prediction result comparison chart is provided for the embodiment of the application. In the single well verification, Figure 3The blue line in the middle is the actual result, and the red line represents the logging curve result of the shear wave velocity prediction by multiple deep learning methods. The left column contains key logging curves such as density (DEN), photoelectric index (PE), deep lateral resistivity (LLD), and shallow lateral resistivity (LLS), which reflect the physical properties and lithological characteristics of clastic rocks and are important input parameters for the model. In the right columns, the blue curve represents the actual Vs value, and the red curve represents the predicted value of the corresponding model. By comparing the fitting degree of the red and blue curves, the prediction ability of each model for Vs can be intuitively evaluated. The clastic rock data has strong heterogeneity, and Vs is sensitive to lithology, porosity, and fluid properties. Deep learning models can capture the complex nonlinear relationships between logging data to predict Vs. Different deep learning models have different structures and different advantages and disadvantages in predicting Vs of clastic rocks. By comparing the fitting degree of the red and blue curves in the figure, the adaptability of each model to the characteristics of clastic rock data (heterogeneity, sequence dependence, and multi-parameter correlation) can be clearly analyzed. The fitting effect of SuperKAN is generally good. From the depth trend, the prediction curve can better follow the fluctuations of the actual value, especially in some key depth sections (such as around 2300 meters), the curve shape and the fluctuation characteristics of the actual value are highly consistent, indicating that SuperKAN can effectively capture the variation of Vs. Compared with adjacent models (such as MSD-Mixer, PatchMixer, etc.), the deviation of the blue curve of SuperKAN from the red actual value curve is smaller, and it closely fits in most areas, showing that its data feature capturing ability is stronger when predicting Vs, and it can better adapt to the complexity of clastic rock data, providing a more reliable result for Vs prediction. The red and blue curves of CNN fit well in local fluctuations (3265-3275m), indicating that it can effectively extract local features, but may lack the ability to capture long sequence depth trends due to local processing characteristics. The TCN curve fits well in the depth sequence change, indicating that it has strong processing ability for the sequence characteristics of clastic rocks with gradual changes in sedimentary environment, and can capture the evolution rule of Vs with depth. Compared with LSTM and GRU, the GRU curve is smoother, indicating that it reduces the computational burden while remembering long-term information, and has both efficiency and effectiveness in processing clastic rock data. LSTM has high fitting degree in areas with sharp depth changes (such as 3295-3315m), indicating that LSTM can adapt to the complex sequence dependence characteristics of clastic rocks. The overall curve of MSD-Mixer fluctuates well with the actual value, indicating that it has strong modeling ability for channel dependence. However, the model is unstable in some areas (3345-3365m).

[0128] The strong heterogeneity and complex pore structure of carbonate reservoirs result in the prediction of S-wave velocity Vs being influenced by multiple factors such as porosity, fractures, and mineral composition. Table 2 shows the prediction performance of Vs by various deep learning methods. The MAPE of SuPerKAN reaches 0.88%, which is significantly better than other methods, indicating that it has the strongest control over relative error. Most of the remaining methods are around 1.0%-1.5%, and the recurrent neural networks such as LSTM and GRU are slightly higher. The MAE of SuPerKAN is 41.48, which is the best. Among the remaining methods, the MAE of TSMixer is also excellent, which is 51.91, and the error of traditional GRU and CNN is more than 65. The RMSE of SuPerKAN is the smallest, which is 30.12, reduced by 6.61 compared with the second-best MSD-Mixer, reflecting that it has the best inhibition effect on larger prediction errors. The R2 of SuPerKAN reaches 0.9811, which is the best. SuPerKAN can accurately depict local features such as pore structure with the help of the strong nonlinear mapping ability of KAN. SuPermutator strengthens the capture of global information across reservoirs through adaptive sequence rearrangement. The complementary features of the two are synergistically integrated on the vertical-horizontal multi-scale, effectively suppressing the adverse effects of well logging noise on Vs prediction and significantly improving the robustness of the model. Traditional time series networks (CNN, LSTM, GRU) focus on local convolution or short-term recursive dependence, making it difficult to capture long-distance associations across scales and layers. The error indicators are generally high, indicating that their ability to describe the response of non-homogeneous carbonate rocks is limited. MSD-Mixer strengthens global dependence modeling through the design of multi-scale receptive field and double-channel mixing, which is only second to SuPerKAN in RMSE and R2, and is suitable for processing well logging data with long sequences and high noise.

[0129] Table 2 Vs prediction results of various deep learning methods on carbonate rock data set (the best result is in bold)

[0130]

[0131] Figure 4A carbonate rock single well different depth learning shear wave velocity prediction result comparison chart is provided for the embodiments of the application. SuPerKAN shows the best fitting accuracy and robustness in the Vs prediction of carbonate rocks, and the research interval is about 3720-3700 m in depth. The prediction curve of the SuPerKAN model in multiple complex layers is close to the measured data, showing higher fitting accuracy and change response ability, especially in the sections with dense fractures or strong reservoir heterogeneity, which can effectively capture the small nonlinear changes of shear wave velocity. In the section with obvious wave velocity mutation (such as 3450-3550 m), most models (such as CNN, LSTM, N-BEATS, etc.) have fitting lag or prediction deviation in the steep change interval, indicating that they have certain limitations in dealing with lithology mutation and strong fracture influence in carbonate rocks. Some models based on attention mechanism or mixed structure (such as ViT, MLP-Mixer, TSMixer, etc.) perform well in relatively stable strata, but still have overfitting or insufficient prediction in the rapid change zone; comprehensive comparison shows that the prediction curve of SuPerKAN is smoother and has stronger trend tracking ability, and maintains good consistency and detail recovery ability at the velocity mutation point, highlighting the advantages of SuPerKAN in feature selection, adaptive structure modeling and high-dimensional data collaborative modeling.

[0132] Volcanic rocks usually have complex structures, such as air pores, fracture development, and large changes in lithology, which will affect Vs, and are currently a difficult problem in the exploration and development of major oilfields. Vs is also important in volcanic rocks because different lithology and structure will cause velocity differences. The results are shown in Table 3, and the SuPerKAN model achieves the best performance in all evaluation indicators, with MAPE of 1.33%, MAE of 41.57, RMSE of 67.44, and R2 of 0.9748, indicating that the model has significant advantages in describing the nonlinear changes and spatial heterogeneity of Vs. In addition, mixed structure models such as TSMixer, MSD-Mixer, and MLP-Mixer also perform well, indicating that this method has strong ability in modeling complex time series dependence and structural features in seismic sequences. In contrast, traditional convolutional models (such as CNN) have lower prediction accuracy, reflecting their shortcomings in long-range dependence modeling.

[0133] Table 3 Comparison of shear wave velocity (Vs) prediction results using various deep learning methods on the volcanic rock data set (the best result is shown in bold)

[0134]

[0135] Different deep learning models perform differently in predicting Vs in different depth ranges (about 3970-4080 m) of volcanic rocks. Some models (such as those with high curve fitting degree) can more accurately capture the characteristics of volcanic rock data, indicating that deep learning technology has application potential for Vs prediction in complex volcanic reservoirs. Figure 5 The volcanic rock single-well different depth learning shear wave velocity prediction result comparison chart provided by the embodiment of the present application is shown in Figure 5 It can be seen from Figure 5 that the prediction curve of SuperKAN closely matches the actual curve in some depth sections, especially in areas where volcanic lithology changes regularly or pore structure has relatively stable influence, and can better capture the trend of Vs change, indicating that it has strong learning and fitting ability for the nonlinear relationship commonly found in volcanic rock data, and can effectively mine the complex correlation between well parameters (such as CNL, DEN, Vp, etc.) and Vs, which shows that the model has certain advantages in processing complex geological data such as volcanic rocks. In the strong nonlinear distribution area of volcanic rocks (such as near the depth of 3980-4000 m), the prediction values of most traditional models (such as CNN, LSTM, GRU, etc.) deviate significantly, and it is difficult to accurately fit the actual wave velocity trend, which shows that the modeling ability of the model in response to complex geological structures is insufficient. The Transformer structure (such as Informer, ViT) and MLP-Mixer type model performs well in some continuous sections (3960-3980 m), but still has obvious lag or overfitting phenomenon in the position where the wave velocity changes dramatically. The prediction curve of the SuPerKAN model has the highest fitting degree with the measured data, especially in multiple velocity mutation interfaces and nonlinear disturbance intervals, and the prediction result can accurately capture the jump change of shear wave velocity, indicating that it has excellent feature extraction and expression ability in processing volcanic heterogeneity structure, fracture system and formation interface characteristics.

[0136] In the prediction of the above three different complex lithology shear waves, it can be directly seen that no matter what the geological conditions are, what the lithology is, SuPerKAN is superior to the other deep learning models in terms of adaptability in shear wave velocity prediction, and the accuracy is also superior to the other models. Therefore, the SuPerKAN model is more secure and reliable than the other models. In the task of predicting shear wave velocity in various types of rock reservoirs in oilfields, the SuPerKAN model shows excellent results. It can maintain robustness for unbalanced data sets, can cope with complex geological environments, accurately predict shear wave Vs of various lithologies, and reduce exploration and development risks.

[0137] The shear wave velocity prediction device provided by the embodiment of the present application is described below. The shear wave velocity prediction device described below can be correspondingly referred to the shear wave velocity prediction method described above.

[0138] For details, please refer to Figure 6 ,Figure 6 A structure diagram of a shear wave velocity prediction device provided by an embodiment of the present application can include:

[0139] A model construction module 100 is configured to construct a shear wave velocity prediction model in advance based on a KAN network and a sequence permutor; an activation function of the shear wave velocity prediction model is a B-spline basis function;

[0140] A logging data acquisition module 200 is configured to acquire logging data of a target region;

[0141] A prediction module 300 is configured to input the logging data into the shear wave velocity prediction model to obtain a prediction result.

[0142] Based on the above embodiment, the prediction module 300 can include:

[0143] An embedding unit is configured to input the logging data into an embedding layer of the shear wave velocity prediction model to obtain a mapping result;

[0144] A feature interaction unit is configured to input the mapping result into a feature interaction layer of the shear wave velocity prediction model to obtain a feature vector; the feature interaction layer is constructed based on the KAN network and in combination with the sequence permutor;

[0145] A classification unit is configured to input the feature vector into a classification layer of the shear wave velocity prediction model to obtain the prediction result.

[0146] Based on the above embodiment, the embedding unit can include:

[0147] An embedding subunit is configured to perform one-dimensional convolution operation on local features of the logging data through a convolution layer of the shear wave velocity prediction model to extract initial internal input feature variables, and map a dimension of the internal input feature variables of the data to a dimension D of internal channels of the shear wave velocity prediction model to obtain the mapping result.

[0148] A formula expression of the embedding layer is:

[0149]

[0150] wherein, is a parameter of a convolution kernel at a position r and a channel d; R is a kernel size; bd is a bias; and L is a depth of data sampling. is logging data.

[0151] Based on the above embodiment, the feature interaction unit can include:

[0152] ​​The feature interaction subunit is configured to input the mapping result into a plurality of feature interaction layers of the shear wave velocity prediction model in sequence to obtain the feature vector. Each feature interaction layer comprises a sequence permutor structure feature extraction module and a fully connected layer.

[0153] The formula expression of the feature interaction layer is:

[0154] ;

[0155] wherein, is the output of the current feature interaction layer; is the feature interaction layer; is the input of the current feature interaction layer; K is the total number of feature interaction layers; PKAN is the sequence permutor structure feature extraction module; and LP is the fully connected layer.

[0156] Based on the above embodiment, the sequence permutor structure feature extraction module can comprise:

[0157] channel permutation KAN and depth permutation KAN;

[0158] The formula expression of the sequence permutor structure feature extraction module PKAN is:

[0159] ;

[0160] ;

[0161] ;

[0162] The formula expression of the feature interaction layer is:

[0163] = (α +β ) ;

[0164] ;

[0165] wherein, is the channel permutation KAN; is the depth permutation KAN; is the output result of the channel permutation KAN; is the output result of the depth permutation KAN; and are dynamic weights of the channel and the depth, respectively; is a normalization function; is the average value.

[0166] ​Based on the above embodiment, the S-wave velocity prediction device can further comprise:

[0167] a regression module configured to input the feature vector into a full connection layer and a normalization layer of the S-wave velocity prediction model, and map into a regressor, and the formula is:

[0168] wherein, ∈R D×1 is a regressor weight; is an output of the regressor; is a normalization layer; is a full connection layer; is the final output of the feature interaction layer.

[0169] Based on the above embodiment, the logging data acquisition module can comprise:

[0170] a data acquisition unit configured to acquire density data, photoelectric index data, deep lateral resistivity data, shallow lateral resistivity data, natural gamma data, and compensated neutron data.

[0171] It should be noted that the order of the modules and units in the above S-wave velocity prediction device can be changed without affecting the logic.

[0172] The S-wave velocity prediction device provided by the embodiment of the application comprises a model construction module 100, a logging data acquisition module 200, and a prediction module 300.

[0173] The S-wave velocity prediction device provided by the embodiment of the application will be described below.

[0174] Please refer to Figure 7 , Figure 7 FIG. 1 is a structural schematic diagram of an S-wave velocity prediction device provided by an embodiment of the application.

[0175] ​a memory 10 for storing a computer program;

[0176] a processor 20 for executing the computer program to implement the shear wave velocity prediction method described above.

[0177] The memory 10, the processor 20 and the communication interface 31 can communicate with each other through a communication bus 32.

[0178] In the embodiment of the present application, the memory 10 stores one or more programs, which can include program codes including computer operation instructions. In the embodiment of the present application, the memory 10 can store programs for implementing the following functions:

[0179] The shear wave velocity prediction model is constructed based on the KAN network and the sequence transposer in advance, and the activation function of the shear wave velocity prediction model is a B-spline basis function;

[0180] Obtain the logging data of the target area;

[0181] Input the logging data into the shear wave velocity prediction model to obtain the prediction result.

[0182] In a possible implementation, the memory 10 can include a program storage area and a data storage area. The program storage area can store an operating system and at least one application required by a function, etc. The data storage area can store data created in the use process.

[0183] In addition, the memory 10 can include a read-only memory and a random access memory, and provide instructions and data for the processor. Part of the memory can also include an NVRAM. The memory stores an operating system and operation instructions, executable modules or data structures, or subsets or extended sets thereof, wherein the operation instructions can include various operation instructions for implementing various operations. The operating system can include various system programs for implementing various basic tasks and processing hardware-based tasks.

[0184] The processor 20 can be a central processing unit (CPU), an application specific integrated circuit, a digital signal processor, a field programmable gate array or other programmable logic device. The processor 20 can be a microprocessor or any conventional processor. The processor 20 can call the program stored in the memory 10.

[0185] The communication interface 31 can be an interface of a communication module, used for connecting with other devices or systems.

[0186] Of course, it should be noted that, Figure 7The shown structure does not constitute a limitation on the transverse wave velocity prediction device in the embodiments of the present application, and in actual applications, the transverse wave velocity prediction device can include more or fewer components than those shown, or some components can be combined. Figure 7

[0187] The computer readable storage medium provided by the embodiments of the present application is described as follows, and the computer readable storage medium described below can be referred to in correspondence with the transverse wave velocity prediction method described above.

[0188] The present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the transverse wave velocity prediction method described above.

[0189] The computer readable storage medium can include a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0190] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts of each embodiment can be referred to each other. For the device disclosed by the embodiments, since it corresponds to the method disclosed by the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0191] The skilled person can further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware, computer software or a combination of both. In order to clearly show the interchangeability of hardware and software, the composition and steps of each example have been described in general in the above description. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0192] ​Finally, it needs to be mentioned that, in this document, relationships such as first and second, and the like, are intended to distinguish one entity or action from another, without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0193] The transverse wave velocity prediction method, device, equipment and computer readable storage medium provided by the present application are described in detail above, the principle and implementation manner of the present application are described by applying specific examples in this document, the above example description is only for helping to understand the method of the present application and its core idea; meanwhile, for the general technical personnel in the art, according to the idea of the present application, the specific implementation manner and application range will have changes, and the above is summarized, the content of the specification should not be understood as the limitation of the present application.

Claims

1. A shear wave velocity prediction method characterized by, The method comprises the following steps: A shear wave velocity prediction model is constructed in advance based on a KAN network and a sequence permutor; an activation function of the shear wave velocity prediction model is a B-spline base function; Logging data of a target area is obtained; The logging data is input into the shear wave velocity prediction model to obtain a prediction result; The logging data is input into the shear wave velocity prediction model to obtain a prediction result, comprising: The logging data is input into an embedding layer of the shear wave velocity prediction model to obtain a mapping result; The mapping result is input into a feature interaction layer of the shear wave velocity prediction model to obtain a feature vector; the feature interaction layer is constructed based on a KAN network and in combination with a sequence permutor; The feature vector is input into a classification layer of the shear wave velocity prediction model to obtain the prediction result; The mapping result is input into a feature interaction layer of the shear wave velocity prediction model to obtain a feature vector, comprising: The mapping result is input into a plurality of the feature interaction layers of the shear wave velocity prediction model to obtain the feature vector; each of the feature interaction layers comprises a sequence permutor structure feature extraction module and a fully connected layer; A formula expression of the feature interaction layer is: ； wherein, is the output of the current feature interaction layer; is the feature interaction layer; is the input of the current feature interaction layer; K is the total number of feature interaction layers; PKAN is the sequence permuter structure feature extraction module; LP is the fully connected layer; The sequence permutor structure feature extraction module comprises: Channel permutation KAN and depth permutation KAN; A formula expression of the sequence permutor structure feature extraction module PKAN is: ； ； ； Correspondingly, the formula expression of the feature interaction layer is : = （α +β ） ； ； wherein, is a channel-wise KAN; is a depth-wise KAN; is an output of the channel-wise KAN; is an output of the depth-wise KAN; and are dynamic weights for the channel and depth, respectively; is a normalization function; is an average.

2. The shear wave velocity prediction method of claim 1, wherein, The logging data is input into an embedding layer of the shear wave velocity prediction model to obtain a mapping result, comprising: The local features of the logging data are subjected to one-dimensional convolution operation by a convolution layer of the shear wave velocity prediction model, initial internal input feature variables are extracted, and the dimension of the internal input feature variables of the data is mapped to the internal channel dimension D of the shear wave velocity prediction model to obtain the mapping result ; A formula expression of the embedding layer is: ； wherein, are parameters of the convolution kernel at position r, channel d; R is the kernel size; b d is the bias; L is the depth of the data sample; is the well log data.

3. The shear wave velocity prediction method of claim 1, wherein, Before the feature vector is input into a classification layer of the shear wave velocity prediction model to obtain the prediction result, the method further comprises: The feature vector is input into a full connection layer and a normalization layer of the shear wave velocity prediction model and mapped into a regressor; a formula is as follows: ; wherein, ∈ R D×1 is the regressor weight; is the output of the regressor; is the normalization layer; is the fully connected layer; is the final output of the feature interaction layer.

4. The shear wave velocity prediction method of claim 1, wherein, The logging data comprises: Density data, photoelectric index data, deep lateral resistivity data, shallow lateral resistivity data, natural gamma data and compensated neutron data.

5. A shear wave velocity prediction device using the shear wave velocity prediction method according to claim 1, characterized by The method comprises the following steps: A model construction module is configured to construct a shear wave velocity prediction model in advance based on a KAN network and a sequence permutor; an activation function of the shear wave velocity prediction model is a B-spline base function; A logging data acquisition module is configured to obtain logging data of a target area; A prediction module is configured to input the logging data into the shear wave velocity prediction model to obtain a prediction result; The prediction module comprises: An embedding unit is configured to input the logging data into an embedding layer of the shear wave velocity prediction model to obtain a mapping result; A feature interaction unit is configured to input the mapping result into a feature interaction layer of the shear wave velocity prediction model to obtain a feature vector; the feature interaction layer is constructed based on a KAN network and in combination with a sequence permutor; A classification unit is configured to input the feature vector into a classification layer of the shear wave velocity prediction model to obtain the prediction result; The feature interaction unit comprises: A feature interaction subunit is configured to input the mapping result into a plurality of the feature interaction layers of the shear wave velocity prediction model to obtain the feature vector; each of the feature interaction layers comprises a sequence permutor structure feature extraction module and a fully connected layer; A formula expression of the feature interaction layer is: ； wherein, is the output of the current feature interaction layer; is the feature interaction layer; is the input of the current feature interaction layer; K is the total number of feature interaction layers; PKAN is the sequence permuter structure feature extraction module; LP is the fully connected layer; The sequence permutor structure feature extraction module comprises: Channel permutation KAN and depth permutation KAN; Correspondingly, the formula expression of the sequence permutor structure feature extraction module PKAN is: ； ； ； Correspondingly, the formula expression of the feature interaction layer is : = （α +β ） ； ； wherein, is a channel-wise KAN; is a depth-wise KAN; is an output of the channel-wise KAN; is an output of the depth-wise KAN; and are dynamic weights for the channel and depth, respectively; is a normalization function; is an average.

6. A shear wave velocity prediction device characterized by comprising: Comprise: A memory for storing a computer program; A processor for executing the computer program to implement the shear wave velocity prediction method of any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions, and the computer executable instructions are loaded and executed by the processor to implement the shear wave velocity prediction method of any one of claims 1 to 4.

Images