A seismic data processing method and device, a storage medium and an electronic device
By generating a training set with multi-frequency and multi-scale features and optimizing the model using a multi-scale enhanced cascaded residual network, the problem of insufficient resolution and denoising performance in seismic data processing was solved, achieving higher resolution and denoising effects, and providing more accurate seismic data for oil and gas exploration.
Patent Information
- Application Number
- CN202510590794.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-08
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies are insufficient in improving resolution and denoising performance in seismic data processing. Traditional methods have limitations, which affect the effectiveness and accuracy of training models.
By acquiring high-frequency and low-frequency seismic profile data, a training set with multi-frequency and multi-scale features is generated. A multi-scale enhanced cascaded residual network is used for training and model optimization. High-fidelity training data is generated using forward modeling methods to avoid the influence of traditional filtering or frequency upscaling. The learning performance is enhanced by combining multi-scale convolutional kernels and cascaded residual structures.
It significantly improves the resolution and denoising performance of seismic data, provides higher fidelity seismic data, can more accurately reflect underground structural information, and improves the accuracy of oil and gas exploration.
Smart Images

Figure CN123066833A_ABST
Abstract
Description
Technical Field
[0001] This disclosure belongs to the field of seismic data processing technology, and particularly relates to a method, apparatus, storage medium and electronic device for processing seismic data. Background Technology
[0002] With the rapid development of deep learning algorithms, their applications in data processing are gradually increasing, including image denoising, super-resolution, and pattern recognition. Deep learning algorithms analyze large amounts of sample data to reveal nonlinear relationships between data points, effectively handling complex, high-dimensional data problems. Although deep learning has demonstrated advantages in multiple fields, its application in seismic data processing is still relatively limited, bringing new research opportunities to the oil and gas geophysics industry.
[0003] In previous algorithms using Convolutional Neural Networks (CNNs) to improve seismic data resolution, the accuracy of network training significantly impacted the denoising performance. When the training samples are not representative or the sample size is too small, the performance of the trained model will degrade. To ensure training accuracy, traditional methods require constructing high-quality training sets. Traditional methods for seismic data resolution enhancement using CNNs primarily rely on two data training strategies: one is to use high-resolution seismic data after filtering and degradation; the other is to use unprocessed raw data after conventional high-resolution processing. While both methods can improve seismic data resolution to some extent, they both have significant limitations. Filtered and degraded data often loses some features of the original signal, while unprocessed data may lack high-frequency information; both situations affect the effectiveness and accuracy of the trained model.
[0004] Improving the resolution of seismic data obtained through model processing and optimizing denoising performance are technical problems that need to be addressed. Summary of the Invention
[0005] Therefore, it is necessary to provide a seismic data processing method, apparatus, storage medium, and electronic device to address the shortcomings of existing technologies in improving the resolution of seismic data and their poor noise reduction performance.
[0006] In a first aspect, embodiments of the present invention provide a method for processing seismic data, the method comprising:
[0007] Acquire seismic profile data of the target area and preprocessed actual seismic data of the target area, wherein the seismic profile data includes high-frequency seismic profile data and low-frequency seismic profile data;
[0008] A training set is generated based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area.
[0009] The training data in the training set is input into a multi-scale augmented cascaded residual network for training and model optimization until the current conditions meet the stopping iteration conditions, at which point the iteration and optimization are stopped, and the optimized training model is obtained.
[0010] The optimized training model is used to process the seismic data to be processed in the target area to obtain processed seismic data.
[0011] Optionally, the stopping iteration condition includes: the model has converged, or...
[0012] The current iteration count has reached the preset iteration count.
[0013] Optionally, before generating the training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area, the method further includes:
[0014] Obtain the first model;
[0015] Obtain the reflection coefficient sequence, high-frequency seismic waveform, and first noise term at high frequency obtained from the interpretation of well logging data;
[0016] The reflection coefficient sequence, the high-frequency seismic waveform, and the first noise term at high frequency are input into the first model for processing, and the high-frequency seismic profile data are output; and
[0017] Obtain the second model;
[0018] Obtain the reflection coefficient sequence, low-frequency seismic waveform, and the second noise term at low frequency;
[0019] The reflection coefficient sequence, the low-frequency seismic waveform, and the second noise term are input into the second model for processing, and the low-frequency seismic profile data are output.
[0020] Optionally, before inputting the training data from the training set into the multi-scale augmented cascaded residual network for training and model optimization, the method further includes:
[0021] Obtain the multi-scale enhanced cascaded residual network and obtain the cascaded residual structure of the multi-scale enhanced cascaded residual network, wherein the multi-scale enhanced cascaded residual network includes scale convolution kernels;
[0022] The multi-scale convolution kernel captures detailed information from seismic profile data.
[0023] The cascaded residual structure enhances the learning performance of the model to be optimized.
[0024] Optionally, the step of processing the seismic data to be processed in the target area based on the optimized training model to obtain processed seismic data includes:
[0025] Acquire the seismic data to be processed in the target area;
[0026] The seismic data to be processed in the target area is denoised to obtain denoised seismic data.
[0027] The noise-reduced seismic data is input into the optimized training model for further processing to obtain the processed seismic data.
[0028] Optionally, before generating the training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area, the method further includes:
[0029] Obtain actual seismic data for the target area;
[0030] The actual earthquake data is preprocessed according to a preset method to obtain the preprocessed actual earthquake data.
[0031] Optionally, the preset method includes: noise reduction method and data standardization processing.
[0032] Optionally, after obtaining the optimized trained model, the method further includes:
[0033] A test set is generated based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area.
[0034] The performance of the multi-scale enhanced cascaded residual network is evaluated based on the test data in the test set, and the evaluation results are obtained.
[0035] Optionally, the evaluation results include at least: evaluation results related to resolution improvement effect and evaluation results related to noise reduction performance.
[0036] Optionally, the evaluation metrics used to assess the performance of the multi-scale enhanced cascaded residual network based on the test data in the test set include at least: signal-to-noise ratio, structural similarity, and visual effect.
[0037] Secondly, embodiments of the present invention provide a seismic data processing apparatus, the apparatus comprising:
[0038] The acquisition module is used to acquire seismic profile data of the target area and preprocessed actual seismic data of the target area. The seismic profile data includes high-frequency seismic profile data and low-frequency seismic profile data.
[0039] The generation module is used to generate a training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area;
[0040] The training and optimization module is used to input the training data in the training set into the multi-scale augmented cascaded residual network for training and model optimization until the current conditions meet the stopping iteration conditions, at which point the iteration and optimization are stopped, and the optimized training model is obtained.
[0041] The processing module is used to process the seismic data to be processed in the target area based on the optimized training model, so as to obtain the processed seismic data.
[0042] Optionally, the stopping iteration condition includes: the model has converged, or...
[0043] The current iteration count has reached the preset iteration count.
[0044] Optionally, the acquisition module is further configured to: acquire a first model before generating a training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area; and acquire the reflection coefficient sequence, high-frequency seismic waveform and first noise term at high frequency obtained from the interpretation of well logging data;
[0045] The processing module is also used for:
[0046] The reflection coefficient sequence, the high-frequency seismic waveform, and the first noise term at high frequency are input into the first model for processing, and the high-frequency seismic profile data are output; and
[0047] The acquisition module is also used to: acquire a second model before generating a training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area; and acquire the reflection coefficient sequence, low-frequency seismic waveform and the second noise term at low frequency.
[0048] The processing module is also used for:
[0049] The reflection coefficient sequence, the low-frequency seismic waveform, and the second noise term are input into the second model for processing, and the low-frequency seismic profile data are output.
[0050] Optionally, the acquisition module is further configured to: acquire the multi-scale augmented cascaded residual network and acquire the cascaded residual structure of the multi-scale augmented cascaded residual network before inputting the training data in the training set into the multi-scale augmented cascaded residual network for training and model optimization, wherein the multi-scale augmented cascaded residual network includes scale convolution kernels.
[0051] The device further includes:
[0052] The capture module is used to capture detailed information of seismic profile data through the multi-scale convolution kernel;
[0053] An enhancement module is used to enhance the learning performance of the model to be optimized through the cascaded residual structure.
[0054] Optionally, the processing module is specifically used for:
[0055] Acquire the seismic data to be processed in the target area;
[0056] The seismic data to be processed in the target area is denoised to obtain denoised seismic data.
[0057] The noise-reduced seismic data is input into the optimized training model for further processing to obtain the processed seismic data.
[0058] Optionally, the acquisition module is further configured to: acquire actual seismic data of the target area before generating a training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area;
[0059] The processing device further includes:
[0060] The preprocessing module is used to preprocess the actual seismic data according to a preset method to obtain the preprocessed actual seismic data.
[0061] Optionally, the preset method includes: noise reduction method and data standardization processing.
[0062] Optionally, the generation module is further configured to: after obtaining the optimized training model, generate a test set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area;
[0063] The processing device further includes:
[0064] The evaluation module is used to evaluate the performance of the multi-scale enhanced cascaded residual network based on the test data in the test set, and obtain the evaluation results.
[0065] Optionally, the evaluation results include at least: evaluation results related to resolution improvement effect and evaluation results related to noise reduction performance.
[0066] Optionally, the evaluation metrics used to assess the performance of the multi-scale enhanced cascaded residual network based on the test data in the test set include at least: signal-to-noise ratio, structural similarity, and visual effect.
[0067] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of the first aspect.
[0068] Fourthly, an electronic device is provided, including a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method of the first aspect.
[0069] In this embodiment of the invention, seismic profile data of the target area and preprocessed actual seismic data of the target area are acquired. The seismic profile data includes high-frequency seismic profile data and low-frequency seismic profile data. A training set is generated based on the seismic profile data and the preprocessed actual seismic data of the target area. The training data in the training set is input into a multi-scale enhanced cascaded residual network for training and model optimization until the current conditions meet the stopping iteration condition, at which point iteration and optimization cease, resulting in an optimized training model. Finally, the optimized training model is used to process the seismic data to be processed in the target area, resulting in processed seismic data. The processing method provided by this embodiment of the invention, by introducing high-frequency and low-frequency seismic profile data to generate a training set and by using a multi-scale enhanced cascaded residual network to more comprehensively learn the characteristics of the seismic data, not only significantly improves the resolution of the seismic data processed by the optimized training model but also optimizes the denoising performance. Attached Figure Description
[0070] Exemplary embodiments of the present invention can be more fully understood by referring to the accompanying drawings. The drawings are provided to further illustrate the embodiments of the present invention and form part of the specification. They are used together with the embodiments of the present invention to explain the invention and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.
[0071] Figure 1 A flowchart illustrating a method for processing seismic data according to an exemplary embodiment of the present invention;
[0072] Figure 2 This is a flowchart of a high-resolution processing method for multi-scale network seismic data based on forward modeling training.
[0073] Figure 3 To generate seismic profiles of different frequencies for forward modeling of well logging data;
[0074] Figure 4 Before high-resolution processing ( Figure 4 In Figure 4 a) and after processing ( Figure 4 In Figure 4 b) Seismic profile;
[0075] Figure 5 Before high-resolution processing ( Figure 5 In Figure 5 a) and after processing ( Figure 5 In Figure 5 b) Seismic profile;
[0076] Figure 6 Before high-resolution processing ( Figure 6 In Figure 6 a) and after processing ( Figure 6 In Figure 6 b) Spectrum analysis diagram;
[0077] Figure 7 A schematic diagram of the structure of a seismic data processing apparatus 700 provided according to an exemplary embodiment of the present invention. Detailed Implementation
[0078] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
[0079] It should be noted that, unless otherwise stated, the technical or scientific terms used in this invention should have the ordinary meaning as understood by one of ordinary skill in the art.
[0080] Furthermore, the terms "first" and "second," etc., are used to distinguish different objects, not to describe a specific order. Additionally, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to those processes, methods, products, or devices.
[0081] This invention provides a method and apparatus for processing seismic data, a computer-readable medium, and an electronic device, which will be described below with reference to the accompanying drawings.
[0082] Please refer to Figure 1 It illustrates flowcharts of seismic data processing methods provided by some embodiments of the present invention, such as... Figure 1 As shown, the method for processing seismic data may include the following steps:
[0083] Step S101: Obtain seismic profile data of the target area and preprocessed actual seismic data of the target area. The seismic profile data includes high-frequency seismic profile data and low-frequency seismic profile data.
[0084] In practical application scenarios, the process of generating high-frequency seismic profile data and low-frequency seismic profile data is as follows: High-frequency seismic profile data can be generated by the following formula (1).
[0085] Shigh(t)=R(t)*Whigh(t)+Nhigh(t) Formula (1);
[0086] In the above formula (1), Shigh(t) is the high-frequency seismic profile, R(t) is the reflection coefficient sequence obtained from the well logging data, which represents the reflection characteristics of the formation interface; Whigh(t) is the high-frequency seismic waveform, and Nhigh(t) is the first noise term at high frequency, which is used to simulate the noise component in the actual seismic data.
[0087] Low-frequency seismic profile data can be generated using the following formula (2).
[0088] Slow(t)=R(t)*Wlow(t)+Nlow(t) formula (2);
[0089] In the above formula (2), Slow(t) is the low-frequency seismic profile, R(t) is the reflection coefficient sequence obtained from the well logging data, which represents the reflection characteristics of the formation interface; Wlow(t) is the low-frequency seismic waveform, and Nlow(t) is the second noise term at low frequency, which is used to simulate the noise component in the actual seismic data.
[0090] In the processing method provided in this embodiment of the invention, the forward modeling method uses well logging data to connect well lines and generates the aforementioned high-frequency and low-frequency seismic profile data through convolution operations. Based on the geological model and well logging data, forward modeling technology is used to generate seismic profiles of different frequencies (high-frequency and low-frequency).
[0091] In practical applications, actual seismic data can be obtained, as well as actual seismic exploration data of the target area. These data can reflect the true reflection information of the underground structure of the target area.
[0092] In one example, before generating a training set based on seismic profile data of the target area and preprocessed actual seismic data of the target area, the seismic data processing method provided in this embodiment of the invention may further include the following steps:
[0093] Obtain actual seismic data for the target area;
[0094] The actual earthquake data is preprocessed according to a preset method to obtain preprocessed actual earthquake data.
[0095] In one example, the preset methods include: noise reduction and data standardization.
[0096] Step S102: Generate a training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area.
[0097] In one example, before generating a training set based on seismic profile data of the target area and preprocessed actual seismic data of the target area, the seismic data processing method provided in this embodiment of the invention may further include the following steps:
[0098] Obtain the first model;
[0099] Obtain the reflection coefficient sequence, high-frequency seismic waveform, and first noise term at high frequency obtained from the interpretation of well logging data;
[0100] The reflection coefficient sequence, high-frequency seismic waveform, and the first noise term at high frequency are input into the first model for processing, and high-frequency seismic profile data are output; and
[0101] Obtain the second model;
[0102] Obtain the reflection coefficient sequence, low-frequency seismic waveform, and the second noise term at low frequency;
[0103] The reflection coefficient sequence, low-frequency seismic waveform, and second noise term are input into the second model for processing, and low-frequency seismic profile data are output.
[0104] It should be noted that in the processing method provided in this embodiment of the invention, the first model is a high-frequency model. In one application scenario, the algorithm corresponding to this first model is specifically described as follows:
[0105] Shigh(t)=R(t)*Whigh(t)+Nhigh(t) Formula (1);
[0106] In the above formula (1), Shigh(t) is the high-frequency seismic profile, R(t) is the reflection coefficient sequence obtained from the well logging data, which represents the reflection characteristics of the formation interface; Whigh(t) is the high-frequency seismic waveform, and Nhigh(t) is the first noise term at high frequency, which is used to simulate the noise component in the actual seismic data.
[0107] In the processing method provided in this embodiment of the invention, the second model is a low-frequency model. In one application scenario, the algorithm corresponding to this second model is specifically described as follows:
[0108] Slow(t)=R(t)*Wlow(t)+Nlow(t) formula (2);
[0109] In the above formula (2), Slow(t) is the low-frequency seismic profile, R(t) is the reflection coefficient sequence obtained from the well logging data, which represents the reflection characteristics of the formation interface; Wlow(t) is the low-frequency seismic waveform, and Nlow(t) is the second noise term at low frequency, which is used to simulate the noise component in the actual seismic data.
[0110] In the processing method provided in the embodiments of the present invention, the forward modeling method uses well logging data to connect well lines and generates the above-mentioned high-frequency seismic profile and low-frequency seismic profile through convolution operation.
[0111] In the processing method provided in this embodiment of the invention, the training set is a high-fidelity training set with multi-frequency and multi-scale features; in this way, it can simulate the real scenario of earthquake data processing and provide rich learning materials for deep learning models.
[0112] Step S103: Input the training data in the training set into the multi-scale augmented cascaded residual network for training and model optimization until the current conditions meet the stopping iteration condition, that is, reach 500,000 iterations, stop the iteration and optimization, and obtain the optimized training model.
[0113] It should be noted that the 500,000 iterations mentioned above are merely an example, and no specific limit is made on the exact number of iterations.
[0114] In the processing method provided in this embodiment of the invention, seismic profile data generated by forward modeling of well logging data is used as training data, avoiding the influence of frequency upscaling or filtering algorithms, and more realistically reflecting the subsurface seismic response, thus achieving higher data fidelity. Furthermore, the seismic profile obtained by forward modeling of well logging data can more realistically reflect the frequency characteristics and signal distribution of the subsurface seismic response, and is more consistent with the characteristics of actual seismic data, thereby providing training data with higher fidelity. Moreover, since the training data is directly derived from the forward modeling of well logging data, the processing method provided in this embodiment of the invention is not affected by traditional filtering degradation or frequency upscaling, avoiding deviations and information loss that may be introduced by preprocessing steps. In addition, through the above-mentioned forward modeling of well logging data, the introduction of noise can be effectively controlled, ensuring that the signals in the training data retain their original characteristics as much as possible, thereby reducing the impact of noise interference during the training process.
[0115] In practical applications, preprocessing steps may include noise reduction and data standardization. The preprocessing steps described above are merely examples and can be adjusted according to the needs of different application scenarios. No specific limitations are imposed on the preprocessing steps here.
[0116] In one example, before inputting the training data from the training set into the multi-scale augmented cascaded residual network for training and model optimization, the seismic data processing method provided in this embodiment of the invention may further include:
[0117] Obtain a multi-scale enhanced cascaded residual network and obtain the cascaded residual structure of the multi-scale enhanced cascaded residual network, which includes scale convolution kernels;
[0118] Capture detailed information from seismic profile data using multi-scale convolution kernels;
[0119] Enhance the learning performance of the model to be optimized by using cascaded residual structures.
[0120] It should be noted that MECRN (Multiscale Enhanced Cascade Residual Network) captures multiscale information from seismic data through convolutional kernels of different sizes and enhances the signal recognition and recovery capabilities of the model under optimization through a dual-path enhanced residual learning mechanism. The dual-path structure design helps the model under optimization to simultaneously learn the direct features and residual information of the input data, thereby better recovering high-frequency weak signals. The multiscale convolutional kernels of the multiscale enhanced cascade residual network can capture detailed information from seismic data; furthermore, the cascaded residual structure of the network enhances the learning performance of the model under optimization. Moreover, the design of the multiscale enhanced cascade residual network takes into account the characteristics of seismic data, such as signal continuity and hierarchical structure.
[0121] By using the aforementioned multi-scale enhanced cascaded residual network, the characteristics of seismic data can be learned more comprehensively. This not only significantly improves the resolution of seismic data processed by the optimized training model, but also optimizes the denoising performance.
[0122] Using the aforementioned forward modeling method, a series of high-fidelity high-frequency and low-frequency seismic profile data were generated. These data were used to train deep learning networks, particularly MECRN. MECRN has a dual-path cascaded residual network structure, extracting shallow information from DAS records through a dual-path mechanism. MECRN improves the resolution and denoising performance of seismic data. The training data for the high-frequency model is mainly used to improve the network's ability to identify and recover high-frequency information in seismic data, which is particularly important for recovering weak high-frequency signals. The training data for the low-frequency model helps the network better understand the low-frequency characteristics of seismic data, thereby optimizing the overall denoising performance of the seismic data. Since the training data includes not only high-frequency model training data but also low-frequency model training data, the processing method provided in this embodiment of the invention can effectively improve the resolution and denoising performance of seismic data, providing more accurate subsurface structure information for seismic exploration.
[0123] By employing the frequency division training method described above, MECRN can more comprehensively learn the characteristics of seismic data, significantly improving the resolution of seismic data and optimizing denoising performance. The application of this processing method provides higher-quality seismic data for oil and gas exploration and has significant practical application value.
[0124] During the model training process described above, backpropagation and gradient descent were used to optimize the network parameters, with the goal of minimizing the difference between the model output and the actual seismic data labels. The choice of loss function directly affects model performance; commonly used loss functions include mean squared error and cross-entropy.
[0125] In one example, the stopping condition for iteration includes: the model has converged.
[0126] In another example, the stopping iteration condition includes: the current iteration count has reached the preset iteration count.
[0127] In one example, after obtaining the optimized training model, the seismic data processing method provided in this embodiment of the invention may further include the following steps:
[0128] A test set is generated based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area.
[0129] The performance of multi-scale enhanced cascaded residual networks is evaluated based on test data from the test set, and the evaluation results are obtained.
[0130] In one example, the evaluation results include at least: evaluation results related to the resolution improvement effect and evaluation results related to the noise reduction performance.
[0131] In one example, the evaluation metrics used to assess the performance of a multi-scale augmented cascaded residual network based on test data in a test set include at least: signal-to-noise ratio, structural similarity, and visual effect.
[0132] In the processing method provided in this embodiment of the invention, test set evaluation: The performance of MECRN is evaluated using a test dataset independent of the training set, including resolution enhancement and denoising capabilities. Evaluation metrics may include signal-to-noise ratio, structural similarity, and visual effects, etc.
[0133] Parameter tuning: Based on the test results, the model is further optimized by adjusting factors such as network structure parameters and learning rate. Possible optimization strategies include cross-validation and hyperparameter search.
[0134] Iterative optimization: Performance evaluation and parameter tuning is an iterative process that continues until the model performance reaches the expected target or can no longer be significantly improved.
[0135] Step S104: Process the seismic data to be processed in the target area based on the optimized training model to obtain processed seismic data.
[0136] In one example, the seismic data to be processed in the target area is processed based on the optimized trained model to obtain processed seismic data, including the following steps:
[0137] Acquire seismic data to be processed for the target area;
[0138] The seismic data to be processed in the target area is denoised to obtain denoised seismic data.
[0139] The denoised seismic data is input into the optimized training model for further processing to obtain the processed seismic data.
[0140] It should be noted that by performing noise reduction processing on the seismic data to be processed in the target area, noise interference can be effectively reduced.
[0141] like Figure 2 The diagram shown is a flowchart of a high-resolution processing method for multi-scale network seismic data trained based on forward modeling data.
[0142] As Figure 2As shown, the above processing flow includes the following: ① Constructing a training set. A real noise database is established, and 800 synthetic data points and 200 forward modeling data points based on real noise are constructed as the training set. ② Training the network. The original data is input into the deep learning network to learn signal features, noise features, and the differences between signal and noise. ③ Optimizing the network model. The model is iteratively optimized by controlling the loss function and adjusting the convolution kernel parameters to obtain a high-quality model. During model optimization, the model convergence is determined based on the loss function, or the number of iterations has reached the preset number of iterations, to determine whether to stop iteration until the learned model (optimized model) is obtained. ④ Obtaining the trained model and then entering the testing phase, i.e., testing new data to achieve denoising of low signal-to-noise ratio data. In the testing phase, the new test data is input into the learning model for data processing, and the processing results are obtained and output.
[0143] Taking surface seismic data from the loess hilly area in region A as an example, the processing method provided in this embodiment of the invention, through training and processing using a multi-scale enhanced cascaded residual network, improves the data resolution of region A from 32Hz to 36Hz. Comparative analysis before and after processing shows that the processing method provided in this embodiment of the invention can significantly recover high-frequency weak signals and improve the resolution and denoising performance of seismic data.
[0144] like Figure 3 As shown, this is a forward modeling simulation of well logging data to generate seismic profiles at different frequencies. Figure 3 As shown, Figure 3 middle Figure 3 The frequency of a is 50Hz. Figure 3 In Figure 3 The frequency of b is 15Hz. Figure 3 In Figure 3 a represents the seismic profile modeled by the 50Hz Ricker wavelet forward model. Figure 3 In Figure 3 b is the seismic profile obtained from the forward modeling of the 15Hz Ricker wavelet. Figure 3 In Figure 3 a, Figure 3 In Figure 3 b represents the training sets of forward modeling high-frequency seismic profile data and forward modeling low-frequency seismic profile data, respectively (the horizontal axis represents line number and trace number, and the vertical axis represents time, in ms).
[0145] Figure 4 Before high-resolution processing ( Figure 4 In Figure 4 a) and after processing ( Figure 4 In Figure 4 b) Seismic profile; where the horizontal axis represents line and trace number, and the vertical axis represents time in milliseconds. Figure 4It displays the seismic profile in a variable density manner.
[0146] Figure 5 Before high-resolution processing ( Figure 5 In Figure 5 a) and after processing ( Figure 5 In Figure 5 b) Seismic profile; Figure 5 a is the seismic profile before high-resolution operation. Figure 5 b is the seismic profile after high-resolution processing. After high-resolution processing, the dominant seismic frequency is increased by 4Hz, and the thin-layer display is clearer (the horizontal axis is the line number, and the vertical axis is time, in ms). Figure 5 It displays local seismic profiles as waveforms.
[0147] like Figure 6 As shown, Figure 6 In Figure 6 'a' represents the seismic profile spectrum before high-resolution operation, with a dominant frequency of 32Hz. Figure 6 In Figure 6 b represents the seismic profile spectrum after high-resolution processing, with a dominant frequency of 36Hz. After high-resolution processing, the seismic dominant frequency is increased by 4Hz, resulting in clearer thin-layer displays (the horizontal axis represents line and trace numbers, and the vertical axis represents time, in milliseconds).
[0148] The processing method provided by the embodiments of the present invention can improve the data resolution of the target area from 32Hz to 36Hz.
[0149] The seismic data processing method provided in this embodiment of the invention, by introducing high-frequency and low-frequency seismic profile data to generate a training set, and by using a multi-scale enhanced cascaded residual network to more comprehensively learn the characteristics of the seismic data, not only significantly improves the resolution of the seismic data processed by the optimized training model, but also optimizes the denoising performance. Furthermore, the processing method provided in this embodiment of the invention can effectively improve the resolution and denoising performance of seismic data, providing more accurate subsurface structure information for seismic exploration. Further, the processing method provided in this embodiment of the invention has the characteristic of high fidelity: the seismic profiles obtained from forward modeling can more realistically reflect the frequency characteristics and signal distribution of subsurface seismic responses, and are more consistent with the characteristics of actual seismic data, thus providing training data with higher fidelity. In addition, the processing method provided in this embodiment of the invention is unaffected by traditional processing: since the training data comes directly from the forward modeling of well logging data, it is not affected by traditional filtering degradation or frequency upscaling processing, avoiding the bias and information loss that may be introduced by preprocessing steps. Furthermore, the processing method provided in this embodiment of the invention can effectively reduce noise interference: the introduction of noise can be effectively controlled during the forward modeling process, ensuring that the signal in the training data retains its original characteristics as much as possible, thereby reducing the impact of noise interference during the training process.
[0150] In the above embodiments, a method for processing seismic data is provided. Correspondingly, the present invention also provides a seismic data processing apparatus. The seismic data processing apparatus provided in the embodiments of the present invention can implement the above-described seismic data processing method. The seismic data processing apparatus can be implemented by software, hardware, or a combination of both. For example, the seismic data processing apparatus may include integrated or separate functional modules or units to perform the corresponding steps in the above methods.
[0151] Please refer to Figure 7 This diagram illustrates a seismic data processing apparatus provided by some embodiments of the present invention. Since the apparatus embodiments are substantially similar to the method embodiments, the description is relatively simple; relevant details can be found in the description of the method embodiments. The apparatus embodiments described below are merely illustrative.
[0152] like Figure 7 As shown, the seismic data processing device 700 may include:
[0153] The acquisition module 701 is used to acquire seismic profile data of the target area and preprocessed actual seismic data of the target area. The seismic profile data includes high-frequency seismic profile data and low-frequency seismic profile data.
[0154] The generation module 702 is used to generate a training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area.
[0155] The training and optimization module 703 is used to input the training data in the training set into the multi-scale augmented cascaded residual network for training and model optimization until the current conditions meet the stopping iteration conditions, then stop the iteration and optimization, and obtain the optimized training model.
[0156] The processing module 704 is used to process the seismic data to be processed in the target area based on the optimized training model to obtain the processed seismic data.
[0157] In some embodiments of the present invention, the stopping iteration conditions include: the model has converged, or the current iteration number has reached a preset iteration number.
[0158] In some embodiments of the present invention, the acquisition module 701 is further configured to: acquire a first model before generating a training set based on seismic profile data of the target area and preprocessed actual seismic data of the target area; and acquire the reflection coefficient sequence, high-frequency seismic waveform and first noise term at high frequency obtained from the interpretation of well logging data.
[0159] Processing module 704 is also used for:
[0160] The reflection coefficient sequence, high-frequency seismic waveform, and the first noise term at high frequency are input into the first model for processing, and high-frequency seismic profile data are output; and
[0161] The acquisition module 701 is also used to: acquire a second model before generating a training set based on seismic profile data of the target area and preprocessed actual seismic data of the target area; and acquire a reflection coefficient sequence, low-frequency seismic waveform and a second noise term at low frequency.
[0162] Processing module 704 is also used for:
[0163] The reflection coefficient sequence, low-frequency seismic waveform, and second noise term are input into the second model for processing, and low-frequency seismic profile data are output.
[0164] In some embodiments of the present invention, the acquisition module 701 is further configured to: acquire the multi-scale augmented cascaded residual network and acquire the cascaded residual structure of the multi-scale augmented cascaded residual network before inputting the training data in the training set into the multi-scale augmented cascaded residual network for training and model optimization, wherein the multi-scale augmented cascaded residual network includes scale convolution kernels.
[0165] The seismic data processing device 700 may further include:
[0166] Capture module (in) Figure 7 (Not shown in the image) is used to capture detailed information from seismic profile data using multi-scale convolution kernels;
[0167] Enhancement module (in) Figure 7 (Not shown in the image), used to enhance the learning performance of the model to be optimized through a cascaded residual structure.
[0168] In some embodiments of the present invention, the processing module 704 is specifically used for:
[0169] Acquire seismic data to be processed for the target area;
[0170] The seismic data to be processed in the target area is denoised to obtain denoised seismic data.
[0171] The denoised seismic data is input into the optimized training model for further processing to obtain the processed seismic data.
[0172] In some embodiments of the present invention, the acquisition module 701 is further configured to: acquire actual seismic data of the target area before generating a training set based on seismic profile data of the target area and preprocessed actual seismic data of the target area;
[0173] The seismic data processing device 700 may further include:
[0174] Preprocessing module (in) Figure 7 (Not shown in the image) is used to preprocess actual seismic data according to a preset method to obtain preprocessed actual seismic data.
[0175] In some embodiments of the present invention, the preset methods include: noise reduction method and data standardization processing.
[0176] In some embodiments of the present invention, the generation module 702 is further configured to: after obtaining the optimized training model, generate a test set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area;
[0177] The seismic data processing device 700 may further include:
[0178] Evaluation module (in) Figure 7 (Not shown in the image) is used to evaluate the performance of multi-scale enhanced cascaded residual networks based on test data in the test set, and to obtain the evaluation results.
[0179] In some embodiments of the present invention, the evaluation results include at least: evaluation results related to resolution improvement effect and evaluation results related to noise reduction performance.
[0180] In some embodiments of the present invention, the evaluation metrics used to evaluate the performance of multi-scale enhanced cascaded residual networks based on test data in the test set include at least: signal-to-noise ratio, structural similarity, and visual effect.
[0181] In some embodiments of the present invention, the seismic data processing apparatus 700 provided in the present invention is based on the same inventive concept and has the same beneficial effects as the seismic data processing method provided in the foregoing embodiments of the present invention.
[0182] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform a combination Figure 1 The method described.
[0183] According to another embodiment, an electronic device is also provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements a combination... Figure 1 The method described.
[0184] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium.
[0185] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for processing seismic data, characterized in that, The method includes: Acquire seismic profile data of the target area and preprocessed actual seismic data of the target area, wherein the seismic profile data includes high-frequency seismic profile data and low-frequency seismic profile data; A training set is generated based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area. The training data in the training set is input into a multi-scale augmented cascaded residual network for training and model optimization until the current conditions meet the stopping iteration conditions, at which point the iteration and optimization are stopped, and the optimized training model is obtained. The optimized training model is used to process the seismic data to be processed in the target area to obtain processed seismic data.
2. The processing method according to claim 1, characterized in that, The stopping iteration conditions include: the model has converged, or... The current iteration count has reached the preset iteration count.
3. The processing method according to claim 1, characterized in that, Before generating the training set based on the seismic profile data of the target region and the preprocessed actual seismic data of the target region, the method further includes: Obtain the first model; Obtain the reflection coefficient sequence, high-frequency seismic waveform, and first noise term at high frequency obtained from the interpretation of well logging data; The reflection coefficient sequence, the high-frequency seismic waveform, and the first noise term at high frequency are input into the first model for processing, and the high-frequency seismic profile data are output; and Obtain the second model; Obtain the reflection coefficient sequence, low-frequency seismic waveform, and the second noise term at low frequency; The reflection coefficient sequence, the low-frequency seismic waveform, and the second noise term are input into the second model for processing, and the low-frequency seismic profile data are output.
4. The processing method according to claim 1, characterized in that, Before inputting the training data from the training set into the multi-scale augmented cascaded residual network for training and model optimization, the method further includes: Obtain the multi-scale enhanced cascaded residual network and obtain the cascaded residual structure of the multi-scale enhanced cascaded residual network, wherein the multi-scale enhanced cascaded residual network includes scale convolution kernels; The multi-scale convolution kernel captures detailed information from seismic profile data. The cascaded residual structure enhances the learning performance of the model to be optimized.
5. The processing method according to claim 1, characterized in that, The process of processing the seismic data to be processed in the target area based on the optimized training model yields processed seismic data, including: Acquire the seismic data to be processed in the target area; The seismic data to be processed in the target area is denoised to obtain denoised seismic data. The noise-reduced seismic data is input into the optimized training model for further processing to obtain the processed seismic data.
6. The processing method according to claim 1, characterized in that, Before generating the training set based on the seismic profile data of the target region and the preprocessed actual seismic data of the target region, the method further includes: Obtain actual seismic data for the target area; The actual earthquake data is preprocessed according to a preset method to obtain the preprocessed actual earthquake data.
7. The processing method according to claim 6, characterized in that, The preset methods include: noise reduction and data standardization.
8. The processing method according to claim 1, characterized in that, After obtaining the optimized trained model, the method further includes: A test set is generated based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area. The performance of the multi-scale enhanced cascaded residual network is evaluated based on the test data in the test set, and the evaluation results are obtained.
9. The processing method according to claim 8, characterized in that, The evaluation results include at least: evaluation results related to resolution improvement effect and evaluation results related to noise reduction performance.
10. The processing method according to claim 8, characterized in that, The evaluation metrics used to assess the performance of the multi-scale enhanced cascaded residual network based on the test data in the test set include at least: signal-to-noise ratio, structural similarity, and visual effect.
11. A seismic data processing apparatus, characterized in that, The device includes: The acquisition module is used to acquire seismic profile data of the target area and preprocessed actual seismic data of the target area. The seismic profile data includes high-frequency seismic profile data and low-frequency seismic profile data. The generation module is used to generate a training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area; The training and optimization module is used to input the training data in the training set into the multi-scale augmented cascaded residual network for training and model optimization until the current conditions meet the stopping iteration conditions, at which point the iteration and optimization are stopped, and the optimized training model is obtained. The processing module is used to process the seismic data to be processed in the target area based on the optimized training model, so as to obtain the processed seismic data.
12. The processing apparatus according to claim 11, characterized in that, The stopping iteration conditions include: the model has converged, or... The current iteration count has reached the preset iteration count.
13. The processing apparatus according to claim 11, characterized in that, The acquisition module is also used to: acquire a first model before generating a training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area; and acquire the reflection coefficient sequence, high-frequency seismic waveform and first noise term at high frequency obtained from the interpretation of well logging data. The processing module is also used for: The reflection coefficient sequence, the high-frequency seismic waveform, and the first noise term at high frequency are input into the first model for processing, and the high-frequency seismic profile data are output. as well as The acquisition module is also used to: acquire a second model before generating a training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area; and acquire the reflection coefficient sequence, low-frequency seismic waveform and the second noise term at low frequency. The processing module is also used for: The reflection coefficient sequence, the low-frequency seismic waveform, and the second noise term are input into the second model for processing, and the low-frequency seismic profile data are output.
14. The processing apparatus according to claim 11, characterized in that, The acquisition module is further configured to: acquire the multi-scale augmented cascaded residual network and acquire the cascaded residual structure of the multi-scale augmented cascaded residual network before inputting the training data in the training set into the multi-scale augmented cascaded residual network for training and model optimization, wherein the multi-scale augmented cascaded residual network includes scale convolution kernels. The device further includes: The capture module is used to capture detailed information of seismic profile data through the multi-scale convolution kernel; An enhancement module is used to enhance the learning performance of the model to be optimized through the cascaded residual structure.
15. The processing apparatus according to claim 11, characterized in that, The processing module is specifically used for: Acquire the seismic data to be processed in the target area; The seismic data to be processed in the target area is denoised to obtain denoised seismic data. The noise-reduced seismic data is input into the optimized training model for further processing to obtain the processed seismic data.
16. The processing apparatus according to claim 11, characterized in that, The acquisition module is further configured to: acquire actual seismic data of the target area before generating a training set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area; The processing device further includes: The preprocessing module is used to preprocess the actual seismic data according to a preset method to obtain the preprocessed actual seismic data.
17. The processing apparatus according to claim 16, characterized in that, The preset methods include: noise reduction and data standardization.
18. The processing apparatus according to claim 11, characterized in that, The generation module is further configured to: after obtaining the optimized training model, generate a test set based on the seismic profile data of the target area and the preprocessed actual seismic data of the target area; The processing device further includes: The evaluation module is used to evaluate the performance of the multi-scale enhanced cascaded residual network based on the test data in the test set, and obtain the evaluation results.
19. The processing apparatus according to claim 18, characterized in that, The evaluation results include at least: evaluation results related to resolution improvement effect and evaluation results related to noise reduction performance.
20. The processing apparatus according to claim 18, characterized in that, The evaluation metrics used to assess the performance of the multi-scale enhanced cascaded residual network based on the test data in the test set include at least: signal-to-noise ratio, structural similarity, and visual effect.
21. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1 to 10.
22. An electronic device, characterized in that, The method includes a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method according to any one of claims 1 to 10.
