Transverse wave speed intelligent prediction method and device based on channel attention mechanism
By employing a channel attention mechanism-based approach, a one-dimensional deep convolutional neural network and channel attention mechanism are used to directly extract shear wave velocity features from conventional well logging curves. This solves the problems of complex rock physics modeling methods and low generalization of deep learning models, and achieves efficient and accurate shear wave velocity prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA PETROLEUM & CHEMICAL CORP
- Filing Date
- 2024-10-30
- Publication Date
- 2026-05-05
AI Technical Summary
In existing technologies, rock physics modeling methods for predicting shear wave velocities suffer from problems such as complex calculation processes, numerous parameters, and low prediction accuracy and efficiency. Furthermore, conventional deep learning models struggle to consider the spatial structural characteristics of strata, resulting in poor model generalization.
A channel attention-based approach is adopted. By preprocessing and sensitivity analysis of conventional logging curves, logging curves sensitive to shear wave velocity are selected as feature curves. By combining a one-dimensional deep convolutional neural network and the channel attention mechanism, a shear wave velocity prediction model is constructed, and a high-precision nonlinear mapping relationship between the feature curve and the shear wave velocity is established.
It achieves efficient and accurate shear wave velocity prediction, takes into account the spatial structure information of the stratigraphy, improves the prediction accuracy and the generalization ability of the model, and reduces the computational complexity.
Smart Images
Figure CN121978759A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas geophysics, and more specifically, to an intelligent prediction method and device for shear wave velocity based on a channel attention mechanism. Background Technology
[0002] Shear wave velocity is a key parameter in pre-stack inversion, reservoir prediction, and fluid identification. Measured shear wave velocities of subsurface rock formations can usually be obtained through core sampling and laboratory analysis or dipole shear wave logging. However, due to high costs and other factors, most wells lack measured shear wave data. In practical applications, rock physics modeling is commonly used to predict shear wave velocity. This method utilizes conventional logging curves and, under certain assumptions, describes the pore structure, mineral composition, and elastic modulus of the rock from a microscopic perspective, thereby calculating the shear wave velocity information. However, in practical applications, the accuracy and efficiency of this method for predicting shear wave velocity still face significant challenges. Due to the complexity of sedimentary environments, diagenesis, and subsequent alteration, rock physics models based on certain assumptions struggle to accurately describe the pore structure, fluid characteristics, and their impact on the elastic modulus of the reservoir. Using conventional rock physics models, it is difficult to obtain accurate reservoir shear wave velocities. Furthermore, predicting shear wave velocity using rock physics modeling involves a very complex calculation process, numerous parameters, and low prediction efficiency. The adaptability of rock physics models varies significantly across different reservoirs.
[0003] In recent years, artificial intelligence technologies, represented by deep learning, have developed rapidly, achieving significant progress in areas such as image recognition, speech recognition, and large-scale models. Deep learning technology uses deep neural network models to perform various tasks such as classification, segmentation, and regression fitting. This technology has also been introduced into the field of geophysical exploration and has shown promising application prospects in shear wave velocity prediction. When using deep learning to predict shear wave velocity, it is necessary to collect measured shear wave velocities as label data to train the deep neural network model. However, due to the generally limited availability of measured shear wave velocity data that can be used as label data, coupled with the spatial structural characteristics of reservoirs, such as lateral sedimentary facies variations and vertical lithological interbedded, thin-layered, or multi-stage superpositions, conventional deep neural network prediction of shear wave velocity still suffers from problems such as insufficient accuracy, low model generalization, and weak noise resistance. Summary of the Invention
[0004] The purpose of this invention is to propose an intelligent prediction method and device for shear wave velocity based on channel attention mechanism. It can directly utilize conventional well logging curves to achieve efficient and accurate shear wave velocity prediction. It aims to solve the problems of complex calculation process, many parameters, and insufficient accuracy and efficiency of shear wave velocity prediction in existing technologies using rock physics modeling methods, as well as the problems that conventional deep learning models have difficulty considering the influence of formation spatial structure characteristics on rock shear wave velocity prediction and have low generalization under small sample conditions.
[0005] To achieve the above objectives, in a first aspect, the present invention proposes an intelligent prediction method for shear wave velocity based on a channel attention mechanism, comprising:
[0006] Conventional logging curves were preprocessed and subjected to shear wave velocity sensitivity analysis, and logging curves sensitive to shear wave velocity were selected as characteristic curves.
[0007] The characteristic curves are standardized.
[0008] A training set with spatial structure information is constructed based on the standardized feature curves.
[0009] A shear wave velocity prediction model is constructed based on a one-dimensional deep convolutional neural network and a channel attention mechanism. The shear wave velocity prediction model is trained using the training set to establish a high-precision nonlinear mapping relationship between the feature curve and the shear wave velocity.
[0010] The trained shear wave velocity prediction model was evaluated using a test set until the prediction accuracy was confirmed to meet the requirements.
[0011] The trained shear wave velocity prediction model is used to predict the shear wave velocity in wells lacking measured shear wave velocities, and the corresponding shear wave velocity curves are obtained.
[0012] Optionally, preprocessing and shear wave velocity sensitivity analysis can be performed on conventional logging curves, including:
[0013] Collect well data with measured shear wave velocities;
[0014] Quality control analysis was conducted on the logging curves, and preprocessing, including editing and correction, was performed on problematic curves.
[0015] Sensitivity analysis of conventional logging curves was conducted using cross-plot analysis and Pearson correlation coefficient, and logging curves sensitive to shear wave velocity were selected as characteristic curves.
[0016] Optionally, the characteristic curve is standardized, including:
[0017] The characteristic curve is standardized using the Z-score standardization method. The standardization calculation formula is as follows:
[0018]
[0019] In the formula, x represents the data to be standardized. mean x is the average value. std Let x be the variance. norm This is the standardized data.
[0020] Optionally, the input data dimension of each sample in the training set is M×N, where M is the step size, i.e. the number of the sample point and its adjacent sampling points in the vertical direction, and N is the number of feature curves; the label data corresponding to the sample is the measured shear wave velocity of the sample.
[0021] Optionally, the label data setting modes for the samples include:
[0022] Top-level mode: The label data is the shear wave velocity corresponding to the first sampling point in the input sampling points of the current sample;
[0023] Low-level mode: The label data is the transverse wave velocity corresponding to the last sampling point in the input sampling points of the current sample;
[0024] In intermediate mode, the label data is the transverse wave velocity corresponding to the intermediate sampling point among the input sampling points of the current sample.
[0025] Optionally, the shear wave velocity prediction model includes multiple attention mechanism modules connected in sequence and two fully connected layers; wherein each attention mechanism module includes a convolutional layer, a batch normalization layer, a ReLU function nonlinear transformation, a fully connected layer, and a Softmax function nonlinear transformation; the number of convolutional kernels in the multiple attention mechanism modules gradually increases.
[0026] Optionally, during training, the attention mechanism module is used to extract feature data related to shear wave velocity from the training set. The entire feature extraction process includes convolution operations, batch normalization, nonlinear mapping, channel attention weight calculation, and feature weight assignment, as shown below:
[0027] z (l) =Conv1(W i (l) ,a (l-1) )+b i (2)
[0028] p (l) =BN(z) (l) (3)
[0029] q (l) =f(p (l) (4)
[0030] r (l)=f(W j (l) ,q (l) )+b j (5)
[0031]
[0032] a (l) =M(q) (l) ,s (l) (7)
[0033] In the formula, a (l-1) a is the output of the previous layer. (l) For the output of the current layer, Conv1 indicates that a one-dimensional convolution operation is performed, and the input data will be padded during the operation. W i (l) Let b be the weight matrix of the i-th convolutional kernel in the current layer. i Let p be the i-th bias term; BN is the batch normalization function, p (l) For the batch-standardized results, f is used as the activation function for nonlinear transformation, with ReLU as the activation function; W j (l) Let b be the weight matrix of the first fully connected layer. j This is the bias term for the j-th neuron; b is the weight matrix of the second fully connected layer. k is the bias term for the k-th neuron; g is the activation function that performs a non-linear transformation, using Softmax as the activation function to learn weight coefficients for different features; M is the matrix multiplication operation, which assigns corresponding weight coefficients to each feature and outputs the result of the current attention mechanism module.
[0034] Optionally, during training, the mean squared error function is used to calculate the error between the model's predictions and the true labels, with the objective function being:
[0035]
[0036] In the formula, E is the objective function, and y i ′ represents the predicted shear wave velocity, y i This represents the measured shear wave velocity;
[0037] The gradient descent algorithm is used to backpropagate the residual values calculated from the objective function to update the network parameters and train the model.
[0038] Secondly, this invention proposes an intelligent prediction device for shear wave velocity based on a channel attention mechanism, comprising:
[0039] The preprocessing and sensitivity analysis module is used to preprocess conventional logging curves and perform shear wave velocity sensitivity analysis, and selects logging curves sensitive to shear wave velocity as characteristic curves.
[0040] A standardization processing module is used to standardize the feature curves;
[0041] The training set construction module is used to construct a training set with spatial structure information based on the standardized feature curves.
[0042] The testing module is used to perform quality control evaluation on the trained shear wave velocity prediction model using a test set until the prediction accuracy is confirmed to meet the requirements.
[0043] The prediction module is used to predict the shear wave velocity of wells lacking measured shear wave velocities using the trained shear wave velocity prediction model, and obtain the corresponding shear wave velocity curve.
[0044] Thirdly, the present invention provides an electronic device, the electronic device comprising:
[0045] At least one processor; and,
[0046] A memory communicatively connected to the at least one processor; wherein,
[0047] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the intelligent prediction method for transverse wave velocity based on the channel attention mechanism described in the first aspect.
[0048] The beneficial effects of this invention are as follows:
[0049] (1) Compared with the prior art, the present invention is more efficient in predicting shear wave velocity. It does not require a complex rock physics modeling process. It adopts an end-to-end approach and directly uses characteristic curves to achieve rapid prediction of shear wave velocity.
[0050] (2) This invention takes into account the sedimentary characteristics of the strata and adds the spatial structure information of the strata to the constructed training set. The input data of each sample contains the characteristic curve information of adjacent sampling points within a certain step size range, which can effectively improve the accuracy and reliability of shear wave velocity prediction.
[0051] (3) The present invention constructs an intelligent transverse wave velocity prediction model based on a one-dimensional convolutional neural network and an attention mechanism. The model consists of multiple attention mechanism modules. The model can automatically extract structured features related to transverse wave velocity directly from the input feature curves and can automatically assign different weights to each feature, so that the model can pay more attention to important features related to transverse wave velocity. The model has high prediction accuracy, strong generalization ability and good noise resistance.
[0052] The system of the present invention has other features and advantages that will be apparent from or will be set forth in detail in the accompanying drawings and following detailed description, which together serve to explain the particular principles of the invention. Attached Figure Description
[0053] The above and other objects, features and advantages of the present invention will become more apparent from the accompanying drawings, in which like reference numerals generally denote like parts.
[0054] Figure 1 This is a schematic diagram of an intelligent prediction process for transverse wave velocity based on a channel attention mechanism, according to an embodiment of the present invention.
[0055] Figure 2 This is a characteristic curve diagram of the sensitivity to shear wave velocity in one embodiment of the present invention.
[0056] Figure 3 This is a schematic diagram illustrating the establishment of a training set containing spatial structure information in one embodiment of the present invention.
[0057] Figure 4 This is a schematic diagram of a transverse wave velocity prediction network model based on a channel attention mechanism in one embodiment of the present invention.
[0058] Figure 5 This is a schematic diagram of the channel attention mechanism module unit structure in one embodiment of the present invention.
[0059] Figure 6 This is a comparison curve of the application effect of actual data shear wave velocity prediction in one embodiment of the present invention. Detailed Implementation
[0060] Shear wave velocity (SVR) is a key parameter for detailed reservoir characterization. Current methods for predicting SVR using rock physics modeling are typically based on certain assumptions, making it difficult to accurately describe the actual rock structure and elastic modulus. This results in insufficient prediction accuracy, and these methods are computationally complex with numerous parameters, leading to low prediction efficiency. In recent years, deep learning technology has made progress in reservoir SVR prediction, demonstrating promising application results. However, due to the limited availability of measured SVR velocities as label data, and the fact that actual reservoirs often exhibit spatial structural features such as interbedded lithology, sedimentary facies variations, and multiple superimposed phases, conventional deep neural networks suffer from weak generalization and noise resistance, resulting in insufficient prediction accuracy and reliability. Therefore, they still face significant challenges in practical applications.
[0061] The purpose of this invention is to overcome the aforementioned technical shortcomings and provide an intelligent prediction method and device for shear wave velocity based on a channel attention mechanism. This invention performs shear wave velocity sensitivity analysis based on the preprocessing of conventional logging curves, selecting logging curves with good correlation to shear wave velocity as characteristic curves. By standardizing the characteristic curves, the influence of numerical unit differences between different types of characteristic curves on the model prediction results is reduced. Based on this, a training set with spatial structure information is created. A high-precision shear wave depth prediction model is constructed based on a one-dimensional deep convolutional neural network and a channel attention mechanism. The model is trained through convergence to establish a high-precision nonlinear mapping relationship between the characteristic curves and shear wave velocity. The trained model is tested and evaluated using a test set. Finally, the trained prediction model is used to predict wells lacking measured shear wave velocities, obtaining the corresponding shear wave velocity curves. This invention establishes an intelligent prediction method for shear wave velocity applicable to complex reservoirs. This method can consider formation spatial structure information, establish a shear wave velocity prediction model based on a one-dimensional deep convolutional neural network and a channel attention mechanism, and can quickly achieve high-precision shear wave velocity prediction directly using conventional logging curves under small sample conditions.
[0062] The invention will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
[0063] Example 1
[0064] This embodiment provides an intelligent prediction method for shear wave velocity based on a channel attention mechanism. The specific process of this method is as follows: Figure 1 As shown, it includes the following steps:
[0065] S1: Preprocess the conventional logging curves and perform shear wave velocity sensitivity analysis, and select the logging curves that are sensitive to shear wave velocity as the characteristic curves;
[0066] This step involves preprocessing conventional logging curves and performing shear wave velocity sensitivity analysis, specifically including:
[0067] Collect well data with measured shear wave velocities;
[0068] Quality control analysis was conducted on the logging curves, and preprocessing, including editing and correction, was performed on problematic curves.
[0069] Sensitivity analysis of conventional logging curves was conducted using cross-plot analysis and Pearson correlation coefficient, and logging curves sensitive to shear wave velocity were selected as characteristic curves.
[0070] In practice, well data with measured shear wave velocities are collected. However, due to various factors such as measuring instruments, methods, and environment, the original conventional logging curves often exhibit problems such as outliers, missing curves, and mismatches in inter-well consistency. Therefore, quality control analysis of the logging curves is necessary, along with preprocessing such as editing and correction of problematic curves to improve the reliability of the logging data. There are many types of conventional logging curves, each with different correlations to shear wave velocity. To reduce information redundancy in the training data and improve the training efficiency of the prediction model, this method uses cross-plot analysis and Pearson correlation coefficient to conduct sensitivity analysis on conventional logging curves, selecting characteristic curves that are more sensitive to shear wave velocity to create the input data for the training set.
[0071] S2: Standardize the characteristic curve;
[0072] Specifically, this step uses the Z-score standardization method to standardize the feature curves in order to reduce the impact of unit differences in the feature curve values on the model prediction results. The standardization calculation process is shown below:
[0073]
[0074] In the formula, x represents the data to be standardized. mean x is the average value. std Let x be the variance. norm This is the standardized data.
[0075] S3: Construct a training set with spatial structure information based on the standardized feature curves;
[0076] Preferably, the input data dimension of each sample in the training set constructed in this step is M×N, where M is the step size, i.e. the number of the sample point and its adjacent sampling points in the vertical direction, and N is the number of feature curves; the label data corresponding to the sample is the measured shear wave velocity of the sample.
[0077] The label data setting modes for the samples include:
[0078] Top-level mode: The label data is the shear wave velocity corresponding to the first sampling point in the input sampling points of the current sample;
[0079] Low-level mode: The label data is the transverse wave velocity corresponding to the last sampling point in the input sampling points of the current sample;
[0080] In intermediate mode, the label data is the transverse wave velocity corresponding to the intermediate sampling point among the input sampling points of the current sample.
[0081] Specifically, sedimentation creates spatial structural features in stratigraphic distribution. A sample within a stratigraphic layer generally exhibits similar distribution characteristics or trends to other samples within a certain distance above and below it. The curve information of these adjacent samples significantly and effectively influences the shear wave velocity of the current sample. Therefore, predicting the shear wave velocity of a sample within a stratigraphic layer requires incorporating not only the characteristic curve information of that sample but also the curve information of adjacent sample points. This greatly helps the model obtain accurate and reliable prediction results. This method constructs a training set with spatial structural information. The input data dimension for each sample in this training set is M×N, where M is the step size, i.e., the number of the sample point and its vertically adjacent sampling points (the value of which can be flexibly set according to the actual stratigraphic conditions), and N is the number of characteristic curves. The label data corresponding to the sample is the measured shear wave velocity of that sample. There are three modes for setting the corresponding label data: the top-level mode, where the label data is the shear wave velocity corresponding to the first sampling point in the input sampling points of the current sample; the bottom-level mode, where the label data is the shear wave velocity corresponding to the last sampling point in the input sampling points of the current sample; and the middle mode, where the label data is the shear wave velocity corresponding to the middle sampling point in the input sampling points of the current sample.
[0082] S4: Construct a shear wave velocity prediction model based on a one-dimensional deep convolutional neural network and channel attention mechanism, and train the shear wave velocity prediction model using the training set to establish a high-precision nonlinear mapping relationship between the feature curve and the shear wave velocity.
[0083] Specifically, this step constructs a transverse wave velocity prediction network model based on a one-dimensional deep neural network and a channel attention mechanism. Compared to conventional fully connected neural networks, convolutional neural networks perform a series of convolution operations on the input data through convolution kernels, enabling hierarchical and automatic feature extraction. They possess powerful nonlinear expressive capabilities, can learn structured information from the input data, and require fewer trainable parameters. The channel attention mechanism automatically learns different weight coefficients for each feature extracted by the network, thus giving different features differentiated importance. This allows the network model to focus more on important features, effectively improving the model's training efficiency and prediction accuracy. This method utilizes a fully connected neural network, a one-dimensional deep convolutional neural network, and a channel attention mechanism to construct attention mechanism modules. Each attention mechanism module sequentially includes a convolutional layer, a batch normalization layer, a ReLU function nonlinear transformation, a fully connected layer, and a Softmax function nonlinear transformation. Finally, multiple attention mechanism modules and two fully connected layers are stacked to form the transverse wave velocity prediction model; the number of convolution kernels in the attention mechanism modules gradually increases within the model.
[0084] During training, the attention mechanism module is used to extract feature data related to shear wave velocity from the training set. The entire feature extraction process includes convolution operations, batch normalization, nonlinear mapping, channel attention weight calculation, and feature weight assignment, which can be represented as follows:
[0085] z (l) =Conv1(W i (l) ,a (l-1) )+b i (2)
[0086] p (l) =BN(z) (l) (3)
[0087] q (l) =f(p (l) (4)
[0088] r (l) =f(W j (l) ,q (l) )+b j (5)
[0089]
[0090] a (l) =M(q) (l) ,s (l) (7)
[0091] In the formula, a (l-1) a is the output of the previous layer. (l) For the output of the current layer, Conv1 indicates that a one-dimensional convolution operation is performed, and the input data will be padded during the operation. W i (l) Let b be the weight matrix of the i-th convolutional kernel in the current layer. i Let p be the i-th bias term; BN is the batch normalization function, p (l) For the batch-standardized results, f is used as the activation function for nonlinear transformation, with ReLU as the activation function; Wj j (l) Let b be the weight matrix of the first fully connected layer. j This is the bias term for the j-th neuron; b is the weight matrix of the second fully connected layer. k is the bias term for the k-th neuron; g is the activation function that performs a non-linear transformation, using Softmax as the activation function to learn weight coefficients for different features; M is the matrix multiplication operation, which assigns corresponding weight coefficients to each feature and outputs the result of the current attention mechanism module.
[0092] Meanwhile, during training, the mean squared error function is used to calculate the error between the model's prediction results and the true labels. The objective function can be expressed by the following formula:
[0093]
[0094] In the formula, E is the objective function, and y i ′ represents the predicted shear wave velocity, y i The measured shear wave velocity is used. The gradient descent algorithm is employed to backpropagate the residual values calculated from the objective function to update the network parameters, thereby training the model.
[0095] S5: Use a test set to perform quality control evaluation on the trained shear wave velocity prediction model until the prediction accuracy is confirmed to meet the requirements.
[0096] This step uses a test set to perform quality control evaluation on the trained model. If the model's prediction accuracy does not meet the requirements, the training set parameters or model structure are adjusted, and the model is retrained until the model's prediction accuracy meets the requirements.
[0097] S6: Use the trained shear wave velocity prediction model to predict the shear wave velocity of wells lacking measured shear wave velocities, and obtain the corresponding shear wave velocity curve.
[0098] This method takes into account formation spatial structure information and establishes a shear wave velocity prediction model based on a one-dimensional deep convolutional neural network and channel attention mechanism. It can quickly achieve high-precision shear wave velocity prediction by directly using conventional logging curves under small sample conditions.
[0099] Example 2
[0100] This embodiment provides an intelligent prediction method for shear wave velocity based on a channel attention mechanism. The method flow is as follows: Figure 1 The method specifically includes:
[0101] (1) Predictive processing of conventional logging curves and selection of sensitive characteristic curves. Well data with measured shear wave velocities are collected from the study area, generally including conventional logging curves for P-wave velocity, density, gamma, compensated neutron, resistivity, well diameter, and spontaneous potential. First, the conventional logging curves are preprocessed, including outlier editing and multi-well consistency correction. Then, sensitive characteristic curve selection is carried out. Through cross-plot analysis of conventional logging curves and shear wave velocity curves and calculation of Pearson correlation coefficients, the correlation between each logging curve and the shear wave velocity curve is analyzed. Simultaneously, logging curves with high correlation are selected as characteristic curves for creating a training set. For example... Figure 2As shown, this embodiment preferentially selects longitudinal wave velocity (VP), density (DEN), gamma (GR), compensated neutron (CNL), and resistivity (RS) as characteristic curves, and will use these sensitive characteristic curves to create a training set. It should be noted that the sensitive characteristic curves may differ in different work areas, and the selection of characteristic curves is also affected by the completeness of the collected original logging data. The specific selection should be based on the results of actual cross-plot analysis or correlation coefficient calculation.
[0102] (2) Characteristic curve standardization. The selected characteristic curves and their corresponding measured shear wave velocities are Z-score standardized to reduce the impact of differences in numerical units between different types of curves on model prediction, thereby improving the training efficiency and prediction accuracy of the model.
[0103] (3) Create a training set containing information on the spatial structure of the formation. Each sample in the training set has an input data dimension of M×N, where M is the spatial step size and N is the number of feature curves. The spatial step size can be set according to features such as the depth sampling interval of the well logging data and the actual formation thickness. For example... Figure 3 As shown, in this embodiment, the intermediate mode is used to create the training set. The spatial step size M of each sample is 25, that is, 25 adjacent sampling points are selected. The number of feature curves N is 5. The label data of the sample is the transverse wave velocity corresponding to the intermediate sampling point.
[0104] (4) Construct an intelligent prediction model for shear wave velocity based on a channel attention mechanism. For example... Figure 4 As shown, the shear wave velocity prediction model network structure used in this embodiment includes three channel attention mechanism modules and two fully connected layers. The network structure of each channel attention mechanism module is as follows: Figure 5 As shown in the diagram, the convolutional kernels in the three attention mechanism modules increase sequentially to 32, 64, and 128, respectively. The multidimensional features output by the last attention mechanism module are flattened before being input into the subsequent fully connected layers. The first fully connected layer contains 1024 neurons and is used to extract transverse wave information from all features output by the attention mechanism modules. The second fully connected layer has only one neuron and is used to output the predicted transverse wave velocity. It is important to note that the intelligent transverse wave velocity prediction method proposed in this invention uses a flexibly adjustable network structure. During use, the number of network layers can be flexibly adjusted according to the size of the training set in the actual work area and the prediction effect of the test data to obtain better prediction results.
[0105] (5) Use test data to perform quality control evaluation on the trained model. Create a certain scale of test data, which is not used in model training but is used to evaluate the trained model. If the prediction accuracy meets the requirements, the model training is complete. If the prediction accuracy does not meet the requirements, the training set parameters or model structure need to be adjusted, and training is performed again until the accuracy requirements are met. In this embodiment, the prediction results for blind wells are as follows: Figure 6 As shown, the predicted shear wave velocity matches the true value well throughout the entire depth range, with a root mean square error of 10³ and an average relative error of 3.5%, indicating that the prediction method and device provided by this invention can obtain high-precision shear wave velocity prediction results.
[0106] (6) Use the trained model to predict shear wave velocity from actual data. Apply the trained model to other wells without measured shear wave velocity and obtain the predicted shear wave velocity results.
[0107] Example 3
[0108] This embodiment provides an intelligent prediction device for shear wave velocity based on a channel attention mechanism, including:
[0109] The preprocessing and sensitivity analysis module is used to preprocess conventional logging curves and perform shear wave velocity sensitivity analysis, and selects logging curves sensitive to shear wave velocity as characteristic curves.
[0110] A standardization processing module is used to standardize the feature curves;
[0111] The training set construction module is used to construct a training set with spatial structure information based on the standardized feature curves.
[0112] The testing module is used to perform quality control evaluation on the trained shear wave velocity prediction model using a test set until the prediction accuracy is confirmed to meet the requirements.
[0113] The prediction module is used to predict the shear wave velocity of wells lacking measured shear wave velocities using the trained shear wave velocity prediction model, and obtain the corresponding shear wave velocity curve.
[0114] For the specific functions of each module in this embodiment, please refer to Embodiment 1 or 2 above, which will not be repeated here.
[0115] Example 4
[0116] This embodiment provides an electronic device, the electronic device comprising:
[0117] At least one processor; and,
[0118] A memory communicatively connected to the at least one processor; wherein,
[0119] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the intelligent prediction method for transverse wave velocity based on the channel attention mechanism described in Embodiment 1 or 2 above.
[0120] An electronic device according to embodiments of the present disclosure includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.
[0121] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory.
[0122] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.
[0123] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.
[0124] Example 5
[0125] This embodiment provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to execute the intelligent prediction method for transverse wave velocity based on the channel attention mechanism described in Embodiment 1 or 2 above.
[0126] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When these non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the methods described in the foregoing embodiments of the present disclosure are performed.
[0127] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).
[0128] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.
Claims
1. An intelligent prediction method for shear wave velocity based on a channel attention mechanism, characterized in that, include: Conventional logging curves were preprocessed and subjected to shear wave velocity sensitivity analysis, and logging curves sensitive to shear wave velocity were selected as characteristic curves. The characteristic curves are standardized. A training set with spatial structure information is constructed based on the standardized feature curves. A shear wave velocity prediction model is constructed based on a one-dimensional deep convolutional neural network and a channel attention mechanism. The shear wave velocity prediction model is trained using the training set to establish a high-precision nonlinear mapping relationship between the feature curve and the shear wave velocity. The trained shear wave velocity prediction model was evaluated using a test set until the prediction accuracy was confirmed to meet the requirements. The trained shear wave velocity prediction model is used to predict the shear wave velocity in wells lacking measured shear wave velocities, and the corresponding shear wave velocity curves are obtained.
2. The method according to claim 1, characterized in that, Preprocessing and shear wave velocity sensitivity analysis of conventional well logging curves, including: Collect well data with measured shear wave velocities; Quality control analysis was conducted on the logging curves, and preprocessing, including editing and correction, was performed on problematic curves. Sensitivity analysis of conventional logging curves was conducted using cross-plot analysis and Pearson correlation coefficient, and logging curves sensitive to shear wave velocity were selected as characteristic curves.
3. The method according to claim 1, characterized in that, The characteristic curve is standardized, including: The characteristic curve is standardized using the Z-score standardization method. The standardization calculation formula is as follows: In the formula, x represents the data to be standardized. mean x is the average value. std Let x be the variance. norm For standardization The data after that.
4. The method according to claim 1, characterized in that, The input data dimension of each sample in the training set is M×N, where M is the step size, i.e. the number of the sample point and its adjacent sampling points in the vertical direction, and N is the number of feature curves; the label data corresponding to the sample is the measured shear wave velocity of the sample.
5. The method according to claim 4, characterized in that, The label data setting modes for the samples include: Top-level mode: The label data is the shear wave velocity corresponding to the first sampling point in the input sampling points of the current sample; Low-level mode: The label data is the transverse wave velocity corresponding to the last sampling point in the input sampling points of the current sample; In intermediate mode, the label data is the transverse wave velocity corresponding to the intermediate sampling point among the input sampling points of the current sample.
6. The method according to claim 1, characterized in that, The shear wave velocity prediction model includes multiple attention mechanism modules connected in sequence and two fully connected layers; each attention mechanism module includes a convolutional layer, a batch normalization layer, a ReLU function nonlinear transformation, a fully connected layer, and a Softmax function nonlinear transformation; the number of convolutional kernels in the multiple attention mechanism modules gradually increases.
7. The method according to claim 6, characterized in that, During training, the attention mechanism module is used to extract feature data related to shear wave velocity from the training set. The entire feature extraction process includes convolution operations, batch normalization, nonlinear mapping, channel attention weight calculation, and feature weight assignment, which is represented as follows: z (l) =Conv1(W i (l) ,a (l-1) )+b i (2) p (l) =BN(with (l) ) (3) q (l) =f(p (l) ) (4) r (l) =f(W j (l) ,q (l) )+b j (5) a (l) =M(q (l) ,s (l) ) (7) In the formula, a (l-1) a is the output of the previous layer. (l) For the output of the current layer, Conv1 indicates that a one-dimensional convolution operation is performed, and the input data will be padded during the operation. W i (l) Let b be the weight matrix of the i-th convolutional kernel in the current layer. i Let p be the i-th bias term; BN is the batch normalization function, p (l) For the batch-standardized results, f is used as the activation function for nonlinear transformation, with ReLU as the activation function; W j (l) Let b be the weight matrix of the first fully connected layer. j This is the bias term for the j-th neuron; b is the weight matrix of the second fully connected layer. k is the bias term for the k-th neuron; g is the activation function that performs a non-linear transformation, using Softmax as the activation function to learn weight coefficients for different features; M is the matrix multiplication operation, which assigns corresponding weight coefficients to each feature and outputs the result of the current attention mechanism module.
8. The method according to claim 7, characterized in that, During training, the mean squared error function is used to calculate the error between the model's predictions and the true labels. The objective function is: In the formula, E is the objective function, and y i ′ represents the predicted shear wave velocity, y i This represents the measured shear wave velocity; The gradient descent algorithm is used to backpropagate the residual values calculated from the objective function to update the network parameters and train the model.
9. A smart predictive device for transverse wave velocity based on a channel attention mechanism, characterized in that, include: The preprocessing and sensitivity analysis module is used to preprocess conventional logging curves and perform shear wave velocity sensitivity analysis, and selects logging curves sensitive to shear wave velocity as characteristic curves. A standardization processing module is used to standardize the feature curves; The training set construction module is used to construct a training set with spatial structure information based on the standardized feature curves. The prediction model building module is used to build a shear wave velocity prediction model based on a one-dimensional deep convolutional neural network and a channel attention mechanism. The shear wave velocity prediction model is trained using the training set to establish a high-precision nonlinear mapping relationship between the feature curve and the shear wave velocity. The testing module is used to perform quality control evaluation on the trained shear wave velocity prediction model using a test set until the prediction accuracy is confirmed to meet the requirements. The prediction module is used to predict the shear wave velocity of wells lacking measured shear wave velocities using the trained shear wave velocity prediction model, and obtain the corresponding shear wave velocity curve.
10. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the intelligent prediction method for transverse wave velocity based on the channel attention mechanism as described in any one of claims 1-8.