Intelligent seismic data reconstruction method and system based on frequency wavenumber domain
By adopting an intelligent method based on the frequency wave number domain in seismic data reconstruction, using convolutional neural networks and specific loss functions, the problem of low quality of seismic data reconstruction in the existing technology is solved, and higher quality data reconstruction is achieved, especially when processing large-area missing data.
Patent Information
- Application Number
- CN202311421376.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-30
- Publication Date
- 2025-05-02
AI Technical Summary
When existing seismic data reconstruction methods deal with incomplete or irregular seismic data, there are problems such as large calculation amount, large errors, and limited reconstruction effect. Especially when large-scale data is missing, deep learning methods face problems such as training difficulties, poor generalization ability, and loss of detailed features.
An intelligent seismic data reconstruction method based on the frequency wave number domain is adopted. By making the frequency wave number domain training data set, a frequency wave number domain convolution neural network is constructed, and feature extraction parts such as the Unet network, residual module, convolution attention module, etc. are used to combine the weighted loss function of the root mean square error function and the structural similarity error function, and Adam adaptive learning rate optimization algorithm is used to conduct network training and data reconstruction.
The reconstruction quality of earthquake data is improved, and the high-frequency information and local detailed characteristics can be reconstructed while retaining the main information, which enhances the network's ability to learn different frequency information of earthquake data, and improves the data integrity and signal-to-noise ratio.
Smart Images
Figure CN119917783A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of seismic data processing, and in particular relates to an intelligent seismic data reconstruction method and system based on frequency wavenumber domain. Background Art
[0002] In the process of seismic exploration and acquisition, the acquired data is often incomplete or irregular due to ground infrastructure limitations, budget and time constraints, data processing and transmission failures, etc., which seriously affects the fidelity, resolution and signal-to-noise ratio of subsequent data processing, thereby reducing the reliability of the final data interpretation. Therefore, it is necessary to reconstruct the incomplete or irregular seismic data to improve the integrity of the seismic data. High-quality reconstruction of missing seismic data plays a very important role in improving the accuracy of subsequent seismic data processing and interpretation.
[0003] At present, the main methods for seismic data reconstruction include methods based on wave equations, methods based on predictive filtering, methods based on mathematical transformation, and methods based on compressed sensing. Among them, the method based on wave equations uses the physical properties of seismic waves propagating in the strata to reconstruct the seismic wave field. This algorithm has a large amount of calculation and will be accompanied by the occurrence of false frequency phenomena; the method based on predictive filtering uses the spectrum of the spatial prediction filter to represent the coefficients of the linear equation in the frequency space domain, but this algorithm has large errors; the method based on mathematical transformation uses mathematical transformation to process seismic signals and reconstructs them in the transform domain. The key to this type of algorithm is to select an appropriate transformation algorithm to best sparse seismic signals. This type of algorithm has good reconstruction effects and high computational efficiency, but there is still room for improvement; the algorithm based on compressed sensing is a data reconstruction method supported by compressed sensing theory. It can reconstruct complete data that meets the accuracy requirements using missing signals, but it needs to meet two prerequisites: one is sparsity and the other is irrelevance.
[0004] In recent years, deep learning methods have shown strong advantages in many fields, such as image recognition, image classification, speech recognition, etc. Deep learning methods can learn the features between large amounts of data and fit complex functional relationships through a large number of training parameters. They have good versatility. Image reconstruction methods based on deep learning can reconstruct high-resolution images and restore more texture details. Subsequently, residual networks, attention mechanisms, and generative adversarial networks were introduced into image reconstruction, and the reconstructed images were more realistic.
[0005] Deep learning does not require prior information, such as sparsity and low rank, so it has advantages in seismic data reconstruction. However, deep learning also has its own limitations, such as the need for a large number of training samples, time-consuming training, and limited generalization ability of different models. The commonly used seismic data reconstruction method based on adversarial generative networks requires a lot of time in the training phase, and the detailed features of complex seismic data are not reconstructed in place. In particular, for the case of a large amount of missing continuous data, such as large obstacles, deep learning methods face great challenges. Large areas of missing data lead to difficulties in training, poor generalization ability, and loss of detailed features. Summary of the invention
[0006] The purpose of the present invention is to solve the above-mentioned problems existing in the prior art and to provide an intelligent seismic data reconstruction method and system based on the frequency wavenumber domain, thereby improving the reconstruction quality of seismic data.
[0007] One of the purposes of the present invention is to provide an intelligent seismic data reconstruction method based on frequency wavenumber domain.
[0008] A second object of the present invention is to provide an intelligent seismic data reconstruction system based on frequency-wavenumber domain.
[0009] A third object of the present invention is to provide a computer-readable storage medium.
[0010] A fourth object of the present invention is to provide a computer device.
[0011] The present invention is achieved through the following technical solutions:
[0012] The first aspect of the present invention provides an intelligent seismic data reconstruction method based on frequency wavenumber domain, which specifically comprises the following steps:
[0013] Step 1, prepare a frequency wavenumber domain training data set;
[0014] Step 2, construct a frequency wavenumber domain convolutional neural network;
[0015] Step 3, using the training data set as input, training the frequency-wavenumber domain convolutional neural network to obtain a trained frequency-wavenumber domain convolutional neural network;
[0016] Step 4: input the seismic data to be reconstructed into the trained frequency-wavenumber domain convolutional neural network to obtain the reconstructed seismic data.
[0017] A further improvement of the present invention is:
[0018] In step 1, a frequency wavenumber domain training data set is prepared, and the specific operations include:
[0019] Step 1.1, obtain the seismic trace data x0 without missing in the time domain;
[0020] Step 1.2: Obtain the sampling matrix of the missing seismic trace data according to the missing seismic trace data. The missing seismic trace data can be expressed as:
[0021] y=Rx
[0022] Among them, R is the sampling matrix, x represents the intact seismic trace data, and y represents the missing seismic trace data;
[0023] Step 1.3, sampling the non-missing seismic trace data according to the sampling matrix R to obtain new missing seismic trace data x1;
[0024] Step 1.4, perform FK transformation on x0 and x1 respectively to obtain Γ(x0) and Γ(x1) in the frequency wavenumber domain;
[0025] Step 1.5: Create seismic data slices of uniform size as training data sets.
[0026] A further improvement of the present invention is:
[0027] In step 2, a frequency-wavenumber domain convolutional neural network is constructed, and the specific operations include:
[0028] Step 2.1, data preprocessing: Γ(x0) and Γ(x1) in the frequency wavenumber domain obtained in step 1 are used as the model label and model input respectively;
[0029] Step 2.2, feature extraction: The feature extraction part consists of Unet network, residual module, and convolutional attention module;
[0030] Step 2.3, loss function setting: select the weighted sum of the root mean square error function and the structural similarity error function as the new loss function;
[0031] Step 2.4, Optimization algorithm: Use Adam adaptive learning rate method as the optimization algorithm.
[0032] A further improvement of the present invention is:
[0033] The root mean square error function calculation formula is as follows:
[0034]
[0035] Among them, RMSE is the root mean square error; y i is the true value, that is, the true value of the i-th data point in the data set; x i is the reconstruction value, that is, the reconstruction value of the i-th data point; n represents the number of data points in the data set.
[0036] A further improvement of the present invention is:
[0037] The structural similarity function calculation formula is as follows:
[0038]
[0039] Among them, SSIM represents the structural similarity value of x and y, y is the true value, that is, the true value in the data set; x is the reconstructed value, μ x , μ y represent the average values of x and y, σ x , σ y represent the standard deviation of x and y, σ xy represents the covariance of x and y, C1 and C2 are constants used to prevent the denominator from being zero.
[0040] A further improvement of the present invention is:
[0041] The loss function formula is as follows:
[0042] loss(x,y)=k1*RMSE+k2*(1-SSIM)
[0043] Among them, loss(x,y) represents the loss function, k1 and k2 represent the weight coefficients of root mean square error and structural similarity, respectively.
[0044] A further improvement of the present invention is:
[0045] Step 4: input the seismic data to be reconstructed into the trained frequency-wavenumber domain convolutional neural network to obtain the reconstructed seismic data. The specific operations include:
[0046] The seismic data to be reconstructed is generated into seismic data slices of specified size according to the method of making training data set in step 1, and is input into the trained frequency-wavenumber domain convolutional neural network to obtain reconstructed frequency domain data, which is then spliced into the original size and reconstructed time domain data is obtained through FK inverse transformation.
[0047] A second aspect of the present invention provides an intelligent seismic data reconstruction system based on frequency wavenumber domain, comprising:
[0048] A data set making unit, used for making a frequency wave number domain training data set;
[0049] A network construction unit, used to construct a frequency-wavenumber domain convolutional neural network;
[0050] A network training unit is used to train the frequency-wavenumber domain convolutional neural network using a training data set as input to obtain a trained frequency-wavenumber domain convolutional neural network;
[0051] The data reconstruction unit is used to input the seismic data to be reconstructed into the trained frequency-wavenumber domain convolutional neural network to obtain the reconstructed seismic data.
[0052] According to a third aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores at least one computer-executable program, and when the at least one program is executed by the computer, the computer executes the steps in the intelligent seismic data reconstruction method based on the frequency wavenumber domain as described above.
[0053] A fourth aspect of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the intelligent seismic data reconstruction method based on the frequency wavenumber domain as described above.
[0054] Compared with the prior art, the present invention has the following beneficial effects:
[0055] The present invention transforms the seismic data in the time-space domain into the frequency-wavenumber domain as network input, extracts seismic image features in the frequency-wavenumber domain, and enhances the network's ability to learn different frequency information of seismic data. While retaining the main information, high-frequency information and local detail features can also be better reconstructed, thereby improving the reconstruction quality of seismic data and having important significance for high-precision seismic data reconstruction and processing. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 It is a flow chart of an intelligent seismic data reconstruction method based on frequency wavenumber domain in an embodiment of the present invention;
[0057] Figure 2 It is the complete data in the time and space domain in the embodiment of the present invention;
[0058] Figure 3 In the embodiment of the present invention b Figure 2 The corresponding frequency-wavenumber domain data;
[0059] Figure 4 It is the under-sampled data in the time and space domain in the example of the present invention;
[0060] Figure 5 In the embodiment of the present invention Figure 4 The corresponding frequency-wavenumber domain data. DETAILED DESCRIPTION
[0061] The present invention is further described in detail below in conjunction with the accompanying drawings:
[0062] [Example 1]
[0063] The embodiment of the present invention provides an intelligent seismic data reconstruction method based on frequency wavenumber domain, such as Figure 1 As shown, the method comprises the following steps:
[0064] Step 1, create a frequency wavenumber domain training data set and a test data set;
[0065] Specific operations include:
[0066] Step 1.1, obtain the seismic trace data x0 without missing in the time domain;
[0067] Step 1.2: Obtain the sampling matrix of the missing seismic trace data according to the missing seismic trace data. The missing seismic trace data can be expressed as:
[0068] y=Rx
[0069] Among them, R is the sampling matrix, x represents the intact seismic trace data, and y represents the missing seismic trace data;
[0070] Step 1.3, sampling the non-missing seismic trace data according to the sampling matrix R to obtain new missing seismic trace data x1;
[0071] Step 1.4, perform FK transformation on x0 and x1 respectively to obtain Γ(x0) and Γ(x1) in the frequency wave number domain; FK transformation adopts the existing technology and will not be described in detail here;
[0072] Step 1.5: Create seismic data slices of uniform size as training data sets.
[0073] The method for preparing the test data set is the same as that for the training data set, and will not be repeated here.
[0074] Step 2, construct a frequency wavenumber domain convolutional neural network;
[0075] Specific operations include:
[0076] Step 2.1, data preprocessing: Γ(x0) and Γ(x1) in the frequency wavenumber domain obtained in step 1 are used as the model label and model input respectively;
[0077] Step 2.2, feature extraction: The feature extraction part consists of Unet network, residual module, and convolutional attention module;
[0078] The Unet network is widely used in seismic data feature extraction, denoising and reconstruction. It achieves feature fusion of different scales and maintains more texture detail features through the symmetrical structure of encoding and decoding and multiple upsampling and downsampling.
[0079] The residual module can reduce the gradient vanishing problem that occurs during the training of multi-layer convolutional neural networks, and at the same time facilitates the back propagation of gradients, speeds up training efficiency, and improves network robustness;
[0080] The convolutional attention module is mainly used to enhance the feature extraction capability and focus on local detail information.
[0081] Step 2.3, loss function setting: select the weighted sum of the root mean square error function and the structural similarity error function as the new loss function;
[0082] The root mean square error function calculation formula is as follows:
[0083]
[0084] Among them, RMSE is the root mean square error; y i is the true value, that is, the true value of the i-th data point in the data set; x i is the reconstruction value, that is, the reconstruction value of the i-th data point; n represents the number of data points in the data set.
[0085] The structural similarity function calculation formula is as follows:
[0086]
[0087] Among them, SSIM represents the structural similarity value of x and y, y is the true value, that is, the true value in the data set; x is the reconstructed value, μ x , μ y represent the average values of x and y, σ x , σ y represent the standard deviation of x and y, σ xy represents the covariance of x and y, C1 and C2 are constants used to prevent the denominator from being zero.
[0088] The loss function formula is as follows:
[0089] loss(x,y)=k1*RMSE+k2*(1-SSIM)
[0090] Among them, loss(x,y) represents the loss function, k1 and k2 represent the weight coefficients of root mean square error and structural similarity, respectively.
[0091] Step 2.4, optimization algorithm: Use Adam adaptive learning rate method as the optimization algorithm, which can dynamically adjust the learning rate, has fast convergence speed and good reconstruction effect.
[0092] Step 3, training the frequency wavenumber domain convolutional neural network;
[0093] Specific operations include:
[0094] The frequency-wavenumber domain convolutional neural network constructed in step (2) is trained using the training data set produced in step (1) to obtain a trained frequency-wavenumber domain convolutional neural network.
[0095] Step 4, inputting the seismic data to be reconstructed into the trained frequency-wavenumber domain convolutional neural network to obtain the reconstructed seismic data;
[0096] Specific operations include:
[0097] The seismic data to be reconstructed is generated into seismic data slices of specified size according to the method of making training data set in step 1, and is input into the trained frequency-wavenumber domain convolutional neural network to obtain reconstructed frequency domain data, which is then spliced into the original size and reconstructed time domain data is obtained through FK inverse transformation.
[0098] [Example 2]
[0099] In order to make the above-mentioned objects, features and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and specific implementation methods.
[0100] like Figure 1 As shown, Figure 1 The figure is a flow chart of the steps of the intelligent seismic data reconstruction method based on the frequency wavenumber domain.
[0101] Step 001: Obtain complete seismic data and undersampled data in the time and space domains, and select seismic data with different characteristics as much as possible, such as Figure 2 and Figure 4 shown.
[0102] Step 002: According to the FK transformation, the time-space domain undersampled seismic data and the complete seismic data are transformed into frequency-wavenumber domain data respectively as training data and label data for deep learning, such as Figure 3 and Figure 5 shown.
[0103] Step 003: Construct a convolutional neural network, which is mainly composed of Unet network, residual module and convolutional attention module. Unet network is widely used in seismic data feature extraction, denoising and reconstruction. Through the symmetrical structure of encoding and decoding and multiple upsampling and downsampling, it can achieve feature fusion of different scales and maintain more texture detail features. The residual module can reduce the gradient vanishing problem of multi-layer convolutional neural network during training, and at the same time help the back propagation of gradient, speed up training efficiency and improve network robustness. The convolutional attention module is mainly used to enhance feature extraction capabilities and focus on local detail information.
[0104] Step 004: Loss function setting, select the weighted sum of the root mean square error function and the structural similarity error function as the new loss function.
[0105] Step 005: Determine the optimization algorithm and use the Adam adaptive learning rate method as the optimization algorithm, which can dynamically adjust the learning rate, has fast convergence speed and good reconstruction effect.
[0106] Step 006: Perform network training. After preparing the training set and network structure, perform network training. Continuously optimize network parameters through loss function and optimization algorithm until the set number of iterations is reached, and finally obtain a trained network.
[0107] Step 007: Perform network application, generate a matrix of specified size for the seismic data to be reconstructed according to the method of step 001, input the trained network, output the reconstructed data, and then splice it into the original size, and obtain the reconstructed time domain data through FK inverse transformation.
[0108] [Example 3]
[0109] The embodiment of the present invention provides an intelligent seismic data reconstruction system based on frequency wavenumber domain, comprising:
[0110] A data set making unit, used for making a frequency wave number domain training data set;
[0111] A network construction unit, used to construct a frequency-wavenumber domain convolutional neural network;
[0112] A network training unit is used to train the frequency-wavenumber domain convolutional neural network using a training data set as input to obtain a trained frequency-wavenumber domain convolutional neural network;
[0113] The data reconstruction unit is used to input the seismic data to be reconstructed into the trained frequency-wavenumber domain convolutional neural network to obtain the reconstructed seismic data.
[0114] [Example 4]
[0115] An embodiment of the present invention provides a computer-readable storage medium, which stores at least one computer-executable program. When the at least one program is executed by the computer, the computer executes the steps in the intelligent seismic data reconstruction method based on the frequency wavenumber domain as described in Example 1.
[0116] [Example 5]
[0117] An embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the intelligent seismic data reconstruction method based on the frequency wavenumber domain as described in Example 1.
[0118] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0119] The above technical solution is only one implementation mode of the present invention. For those skilled in the art, it is easy to make various types of improvements or modifications based on the principles disclosed in the present invention, and it is not limited to the technical solution described in the above specific embodiments of the present invention. Therefore, the above description is only preferred and does not have a restrictive meaning.
Claims
1. An intelligent seismic data reconstruction method based on frequency wavenumber domain, characterized in that: The specific steps include: Step 1, prepare a frequency wavenumber domain training data set; Step 2, construct a frequency wavenumber domain convolutional neural network; Step 3, using the training data set as input, training the frequency-wavenumber domain convolutional neural network to obtain a trained frequency-wavenumber domain convolutional neural network; Step 4: input the seismic data to be reconstructed into the trained frequency-wavenumber domain convolutional neural network to obtain the reconstructed seismic data.
2. The intelligent seismic data reconstruction method based on frequency wavenumber domain according to claim 1 is characterized in that: In step 1, a frequency wavenumber domain training data set is prepared, and the specific operations include: Step 1.1, obtain the seismic trace data x0 without missing in the time domain; Step 1.2: Obtain the sampling matrix of the missing seismic trace data according to the missing seismic trace data. The missing seismic trace data can be expressed as: y=Rx Among them, R is the sampling matrix, x represents the intact seismic trace data, and y represents the missing seismic trace data; Step 1.3, sampling the non-missing seismic trace data according to the sampling matrix R to obtain new missing seismic trace data x1; Step 1.4, perform FK transformation on x0 and x1 respectively to obtain Γ(x0) and Γ(x1) in the frequency wavenumber domain; Step 1.5: Create seismic data slices of uniform size as training data sets.
3. The intelligent seismic data reconstruction method based on frequency wavenumber domain according to claim 2 is characterized in that: In step 2, a frequency-wavenumber domain convolutional neural network is constructed, and the specific operations include: Step 2.1, data preprocessing: Γ(x0) and Γ(x1) in the frequency wavenumber domain obtained in step 1 are used as the model label and model input respectively; Step 2.2, feature extraction: The feature extraction part consists of Unet network, residual module, and convolutional attention module; Step 2.3, loss function setting: select the weighted sum of the root mean square error function and the structural similarity error function as the new loss function; Step 2.4, Optimization algorithm: Use Adam adaptive learning rate method as the optimization algorithm.
4. The intelligent seismic data reconstruction method based on frequency wavenumber domain according to claim 3 is characterized in that: The root mean square error function calculation formula is as follows: Among them, RMSE is the root mean square error; y i is the true value, that is, the true value of the i-th data point in the data set; x i is the reconstruction value, that is, the reconstruction value of the i-th data point; n represents the number of data points in the data set.
5. The intelligent seismic data reconstruction method based on frequency wavenumber domain according to claim 4 is characterized in that: The structural similarity function calculation formula is as follows: Among them, SSIM represents the structural similarity value of x and y, y is the true value, that is, the true value in the data set; x is the reconstructed value, μ x , μ y represent the average values of x and y, σ x , σ y represent the standard deviation of x and y, σ xy represents the covariance of x and y, C1 and C2 are constants used to prevent the denominator from being zero.
6. The intelligent seismic data reconstruction method based on frequency wavenumber domain according to claim 5 is characterized in that: The loss function formula is as follows: loss(x, y)=k1*RMSE+k2*(1-SSIM) Among them, loss(x, y) represents the loss function, k1 and k2 represent the weight coefficients of root mean square error and structural similarity, respectively.
7. The intelligent seismic data reconstruction method based on frequency wavenumber domain according to claim 1 is characterized in that: Step 4: input the seismic data to be reconstructed into the trained frequency-wavenumber domain convolutional neural network to obtain the reconstructed seismic data. The specific operations include: The seismic data to be reconstructed is generated into seismic data slices of specified size according to the method of making training data set in step 1, and is input into the trained frequency-wavenumber domain convolutional neural network to obtain reconstructed frequency domain data, which is then spliced into the original size and reconstructed time domain data is obtained through FK inverse transformation.
8. An intelligent seismic data reconstruction system based on frequency wavenumber domain, characterized in that: include: A data set making unit, used for making a frequency wave number domain training data set; A network construction unit, used to construct a frequency-wavenumber domain convolutional neural network; A network training unit is used to train the frequency-wavenumber domain convolutional neural network using a training data set as input to obtain a trained frequency-wavenumber domain convolutional neural network; The data reconstruction unit is used to input the seismic data to be reconstructed into the trained frequency-wavenumber domain convolutional neural network to obtain the reconstructed seismic data.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer-executable program, and when the at least one program is executed by the computer, the computer executes the steps in the intelligent seismic data reconstruction method based on the frequency-wavenumber domain as described in any one of claims 1-7.
10. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the intelligent seismic data reconstruction method based on the frequency wavenumber domain as described in any one of claims 1 to 7.