Method and device for improving resolution of seismic imaging profile

Abstract

The embodiment of the invention relates to a seismic imaging profile resolution improving method and device, and the method comprises the steps: collecting and analyzing a plurality of seismic imaging profiles, and constructing a plurality of seismic velocity models; obtaining a first target seismic imaging profile based on the seismic convolution model and the plurality of seismic velocity models; by randomly setting different migration imaging parameters, carrying out migration imaging on the plurality of seismic velocity models by using a reverse time migration imaging method to obtain a second target seismic imaging profile; expanding a seismic imaging profile with diversity characteristics by using adaptive synthesis sampling, and adding seismic noise to the seismic imaging profile; designing an initial generative adversarial network in combination with the generative adversarial network, and training to obtain an initial resolution improvement model; and carrying out transfer learning on the initial resolution improvement model to obtain a target resolution improvement model. Therefore, the signal-to-noise ratio and the resolution ratio of the offset profile can be improved, the calculation efficiency is high, and no parameter dependence exists.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of seismic exploration, and particularly to a method and apparatus for improving the resolution of seismic imaging profiles. Background Technology

[0002] Migration imaging often faces challenges such as complex subsurface geological conditions, inaccurate velocity models, and noise, which limit the resolution of the migrated imaging profiles. To address this, current mainstream methods involve processing the migration profiles using techniques such as deconvolution, time-varying spectral extension, and spectral whitening to improve resolution. Deconvolution improves temporal resolution by compressing seismic wavelets, but this can lead to undesirable effects such as noise amplification. Time-varying spectral extension divides seismic data into different time windows, performs continuous wavelet transforms on the data in each window, and then performs spectral extension on each window, reconstructing the seismic data using inverse wavelet transforms. However, this method can lead to accuracy loss with multiple transformations. Spectral whitening smooths the seismic data spectrum, especially in the high-frequency range, to compensate for frequency distortion caused by subsurface structures. This technique helps to balance the energy distribution of different frequency components, improving seismic data resolution and making subsurface structure details more clearly visible. However, this method typically focuses on the amplitude information of the seismic data, and may not be sensitive enough to some geological structures, especially those subtle features derived from phase information. Furthermore, while smoothing the spectrum improves high-frequency resolution, it may lose some low-frequency information.

[0003] In recent years, with the advancement of artificial intelligence technology, resolution improvement methods based on deep learning have also been developed. These methods improve the resolution of seismic migration imaging by introducing deep learning networks such as convolutional neural networks (CNNs), generative adversarial networks (GANs), and residual networks (ResNets). Convolutional neural networks (CNNs) learn image features and introduce upsampling operations to achieve effective image reconstruction from low to high resolution. However, convolutional layers are insufficient in capturing global information of certain migration profiles, and deep CNNs have a large number of parameters, which can easily lead to overfitting. Residual networks (ResNets) solve the gradient vanishing problem by introducing residual connections, enabling deeper networks to effectively learn and transmit high-level features in the image, thereby improving resolution. However, residual structures focus more on larger-scale features when transmitting information and are not sensitive enough to small-scale details.

[0004] Patent application CN2020103655636 discloses a multi-dictionary image super-resolution method based on a Gaussian mixture model. This method first extracts features from low-resolution images using stationary wavelet transform, extracts residual features from high-resolution images, and obtains training sample pairs through overlapping sampling of corresponding regions. It then classifies the training sample pairs using a Gaussian mixture model and learns a corresponding dictionary pair for each class. In the reconstruction stage, multiple dictionaries are used simultaneously for super-resolution reconstruction of the image, and an improved global optimization method is used to further enhance the reconstruction quality. This method not only trains better and more generalizable dictionaries but also avoids the problem of a single global dictionary pair failing to reconstruct image patches with diverse structures, thus enabling better super-resolution reconstruction of low-resolution images. However, using a Gaussian mixture model to classify training samples and learn corresponding dictionaries may face the risk of overfitting, especially with a limited number of training samples. Overfitting may cause the model to perform poorly on unseen data.

[0005] Patent application CN202110213988X discloses a method, apparatus, medium, and electronic device for seismic bandwidth widening. The method includes: acquiring seismic data; determining seismic wavelet data based on well logging data; retrieving the odd and even components of the reflection coefficient based on the seismic wavelet data and the seismic data; determining a target reflection coefficient based on the odd and even components; and determining the seismic bandwidth widening result based on the target reflection coefficient and the broadband wavelet. This method can recover low-frequency and high-frequency information from the seismic data and compensate for this information in the seismic data, thereby widening the seismic data bandwidth. However, the method's determination of the target reflection coefficient based on the odd and even components may be subject to certain errors due to limitations in the inversion algorithm and the complexity of underground structures.

[0006] Patent application CN2017102274480 discloses an image super-resolution reconstruction method based on wavelet transform and convolutional neural networks. In the training phase, a high-resolution image Ih in the training dataset is Gaussian filtered and downsampled to generate a low-resolution image II. A single-scale two-dimensional discrete wavelet transform is performed on Ih to extract four frequency components: low-frequency component FLL, horizontal low-frequency and vertical high-frequency component FLH, horizontal high-frequency and vertical low-frequency component FHL, and diagonal high-frequency component FHH. II is used as input data, and the four frequency components of Ih are used as labels to train four convolutional neural network models. In the super-resolution reconstruction phase, the low-resolution image II is input into the four trained convolutional neural network models to generate the four frequency components of the high-resolution image, and a single-scale two-dimensional discrete wavelet inverse transform is performed on these components to generate the high-resolution image Ih. This method performs super-resolution reconstruction of images at different frequencies, fully utilizing the learning ability of convolutional neural networks and significantly enhancing the super-resolution reconstruction effect. However, during the training phase, this method may cause irreversible loss of information due to Gaussian filtering and downsampling of high-resolution images to generate low-resolution images, which in turn affects the network's performance in the super-resolution stage. Summary of the Invention

[0007] In view of this, in order to solve the above-mentioned technical problems or some of the technical problems, the present invention provides a method and apparatus for improving the resolution of seismic imaging profiles.

[0008] In a first aspect, embodiments of the present invention provide a method for improving the resolution of seismic imaging profiles, comprising:

[0009] Collect and analyze multiple seismic imaging profiles to construct multiple seismic velocity models;

[0010] Based on the seismic convolution model and the multiple seismic velocity models, the seismic imaging profile of the first target is obtained;

[0011] By randomly setting different migration imaging parameters, the reverse time migration imaging method is used to perform migration imaging on the multiple seismic velocity models to obtain the second target seismic imaging profile.

[0012] For the second target seismic imaging profile, adaptive synthetic sampling is used to expand the seismic imaging profile with diverse characteristics, and seismic noise is added to each of the resulting seismic imaging profiles.

[0013] Design an initial generative adversarial network that combines a generator and a discriminator to achieve denoising and resolution enhancement;

[0014] The seismic imaging profile is used as input for network training to train the initial generative adversarial network, resulting in an initial resolution improvement model.

[0015] The initial resolution improvement model is transferred and learned by using actual seismic imaging profile data of the target work area to obtain a target resolution improvement model applicable to the target work area.

[0016] The target resolution enhancement model is used to process the actual seismic imaging profile of the target work area to obtain the target seismic imaging profile.

[0017] In one possible implementation, the method further includes:

[0018] Theoretical seismic profiles are obtained from existing seismic migration imaging data, and different actual seismic profiles are obtained from existing seismic imaging profiles of various types of oilfields.

[0019] Based on the theoretical and actual seismic profiles, a seismic texture structure model corresponding to each actual seismic profile is constructed.

[0020] Multiple seismic velocity models are constructed based on the seismic texture structure model and the rock physical parameter characteristics of the work area corresponding to the seismic texture structure model.

[0021] In one possible implementation, the method further includes:

[0022] Based on the aforementioned multiple seismic velocity models, multiple reflection coefficient models are obtained through a first formula, wherein the first formula is:

[0023] Where R(x) is the reflection coefficient model and v(x) is the seismic velocity model;

[0024] The high-frequency wavelet is convolved with the multiple reflection coefficient models using the seismic convolution model theory to obtain multiple seismic imaging profiles of the first target.

[0025] In one possible implementation, the method further includes:

[0026] Different migration imaging parameters are randomly set for each of the seismic velocity models to obtain a set of imaging parameter combinations corresponding to each of the seismic velocity models;

[0027] Based on each set of migration imaging parameters, the finite difference numerical simulation method is used to obtain the simulated seismic record of each seismic velocity model under the corresponding set of migration imaging parameters;

[0028] The simulated seismic record is migrated using the reverse time migration imaging method to obtain the corresponding migration imaging profile, wherein the migration imaging profile is the second target seismic imaging profile of the seismic velocity model under the corresponding set of migration imaging parameters.

[0029] In one possible implementation, the method further includes:

[0030] An adaptive synthetic sampling method is used to adaptively sample each second target seismic imaging profile, expanding each second target seismic imaging profile into multiple seismic imaging profiles with diverse characteristics.

[0031] Seismic noise is added to each seismic imaging profile with diverse characteristics to obtain noisy seismic imaging profiles containing Gaussian noise and simulated seismic noise.

[0032] In one possible implementation, the method further includes:

[0033] Based on the noisy seismic imaging profile and the first target seismic imaging profile, a neural network training set required for network training based on semi-supervised learning is constructed.

[0034] The first target seismic imaging profile is input as labeled data into the initial generative adversarial network to obtain the initial prediction model;

[0035] The noisy seismic imaging profile is predicted using the initial prediction model, and the prediction results are analyzed for confidence. Prediction results with confidence scores higher than a preset confidence threshold are added to the label set to obtain the initial resolution improvement model.

[0036] In one possible implementation, the method further includes:

[0037] The initial generative adversarial network is trained using the actual seismic profile and the neural network training set. Transfer learning is carried out starting with the initial resolution improvement model, and the parameters of the initial generative adversarial network are optimized through rounds of training.

[0038] An adaptive learning rate is used to dynamically adjust the learning rate in each round of training to obtain a target resolution improvement model suitable for the target work area.

[0039] Secondly, embodiments of the present invention provide a resolution-enhancing device for seismic imaging profiles, comprising:

[0040] The module is used to collect and analyze various seismic imaging profiles and build multiple seismic velocity models;

[0041] The acquisition module is used to obtain the first target seismic imaging profile based on the seismic convolution model and the multiple seismic velocity models;

[0042] The acquisition module is used to perform migration imaging on the multiple seismic velocity models by randomly setting different migration imaging parameters and using the reverse time migration imaging method to obtain a second target seismic imaging profile.

[0043] The expansion module is used to expand the seismic imaging profile with diverse features for the second target seismic imaging profile using adaptive synthetic sampling, and to add seismic noise to each of the resulting seismic imaging profiles.

[0044] The design module is used to combine generative adversarial networks to design an initial generative adversarial network for a generator and a discriminator with denoising and resolution enhancement functions;

[0045] The training module is used to train the initial generative adversarial network by taking the seismic imaging profile as input for network training, thereby obtaining an initial resolution-enhanced model.

[0046] The training module is used to perform transfer learning on the initial resolution improvement model using actual seismic imaging profile data of the target work area, so as to obtain a target resolution improvement model suitable for the target work area.

[0047] The processing module is used to process the actual seismic imaging profile of the target work area based on the target resolution improvement model to obtain the target seismic imaging profile.

[0048] Thirdly, embodiments of the present invention provide an electronic device, including: a processor and a memory, wherein the processor is configured to execute a resolution improvement program for seismic imaging profiles stored in the memory, so as to implement the resolution improvement method for seismic imaging profiles described in the first aspect above.

[0049] Fourthly, embodiments of the present invention provide a storage medium, comprising: the storage medium storing one or more programs, the one or more programs being executable by one or more processors to implement the method for improving the resolution of seismic imaging profiles described in the first aspect above.

[0050] The seismic imaging profile resolution enhancement scheme provided in this embodiment of the invention involves collecting and analyzing various seismic imaging profiles to construct multiple seismic velocity models; obtaining a first target seismic imaging profile based on the seismic convolution model and the multiple seismic velocity models; obtaining a second target seismic imaging profile by randomly setting different migration imaging parameters and using reverse time migration imaging to perform migration imaging on the multiple seismic velocity models; expanding the seismic imaging profiles with diverse characteristics using adaptive synthetic sampling for the second target seismic imaging profiles, and adding seismic noise to each group of seismic imaging profiles; designing an initial generative adversarial network (GAN) with a generator and discriminator that has denoising and resolution enhancement functions by combining a generative adversarial network (GAN); training the initial GAN ​​using the seismic imaging profiles as input to obtain an initial resolution enhancement model; performing transfer learning on the initial resolution enhancement model using actual seismic imaging profile data of the target work area to obtain a target resolution enhancement model suitable for the target work area; and processing the actual seismic imaging profiles of the target work area based on the target resolution enhancement model to obtain the target seismic imaging profile. Compared to existing methods for improving the resolution of migration imaging profiles, which suffer from problems such as model overfitting, large errors, and low network performance in the super-resolution stage, this proposed method addresses these issues. By applying sub-networks for denoising and resolution enhancement to improve the quality of seismic imaging profiles, and establishing a mapping relationship between low-quality and high-quality seismic imaging profiles, high-quality post-stack seismic profiles are obtained. This approach simultaneously improves the signal-to-noise ratio and resolution of the migration profiles, offering high computational efficiency, no parameter dependence, and applicability to any complex geological and geophysical environment, demonstrating significant potential. Attached Figure Description

[0051] Figure 1 A flowchart illustrating a method for improving the resolution of seismic imaging profiles provided in an embodiment of the present invention;

[0052] Figure 2 A schematic diagram of a salt dome model provided in an embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of a low-quality seismic imaging profile obtained from a salt dome model using the reverse time migration imaging method.

[0054] Figure 4 This is a schematic diagram of a high-quality seismic imaging profile obtained from a salt dome model using a method to improve the resolution of the local seismic imaging profile.

[0055] Figure 5 This is a schematic diagram of low-quality seismic data obtained by adding simulated seismic noise using Gaussian distribution and convolution of Ricker wavelet with Gaussian noise matrix.

[0056] Figure 6This is a schematic diagram of high-quality seismic data obtained by adding simulated seismic noise using a method to improve the resolution of the local seismic imaging profile.

[0057] Figure 7 (a) is a schematic diagram of an actual seismic imaging profile extracted from actual data from Western Data.

[0058] Figure 7 (b) is from Figure 7 (a) Schematic diagram of the seismic imaging profile obtained by improving resolution;

[0059] Figure 8 (a) is Figure 7 (a) is a partial view of the seismic imaging profile extracted from the actual seismic imaging profile of Western Data.

[0060] Figure 8 (b) is from Figure 7 (b) shows a partial view of the seismic imaging profile extracted from the improved resolution seismic imaging profile.

[0061] Figure 9 (a) is Figure 7 (a) shows the spectrum of the actual seismic imaging profile of the western data obtained by spectrum analysis.

[0062] Figure 9 (b) is from Figure 7 (b) shows the spectrum of the improved seismic imaging profile obtained using spectral analysis;

[0063] Figure 10 A schematic diagram of a device for improving the resolution of seismic imaging profiles provided in an embodiment of the present invention;

[0064] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0067] Figure 1 This is a flowchart illustrating a method for improving the resolution of seismic imaging profiles according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method specifically includes:

[0068] S11. Collect and analyze various seismic imaging profiles to construct multiple seismic velocity models.

[0069] In this embodiment of the invention, different typical theoretical seismic profiles are collected by extensively reading relevant literature on seismic migration imaging; different typical actual seismic profiles are extracted by extensively reviewing typical seismic imaging profiles of various types of oilfields at home and abroad; seismic geological structure models corresponding to the typical theoretical seismic profiles and typical actual seismic profiles are constructed; based on the seismic geological models, the rock physical parameter characteristics of the work area corresponding to the model are obtained, and finally a large number of seismic velocity models with universal characteristics are established.

[0070] S12. Based on the seismic convolution model and the multiple seismic velocity models, obtain the first target seismic imaging profile.

[0071] Using a large number of universally applicable seismic velocity models v(x), a large number of corresponding reflection coefficient models are obtained through formula (1), where formula (1) is:

[0072]

[0073] By combining the theory of seismic convolution model, the high-frequency wavelet w(t) is convolved with the obtained reflection coefficient model R(x) to obtain a large number of first-target seismic imaging profiles (high-quality seismic imaging profiles). This profile will be used as label data for subsequent network training.

[0074] The specific formula for the earthquake convolution model theory is shown in formula (2):

[0075]

[0076] S13. By randomly setting different migration imaging parameters, the multiple seismic velocity models are migrated using the reverse time migration imaging method to obtain the second target seismic imaging profile.

[0077] For each seismic velocity model v(x), different migration imaging parameters are randomly set to form a set of imaging parameter combinations for subsequent migration imaging of the model; here, in order to make the method universal, 20 sets of random parameters are selected for each velocity model.

[0078] The offset parameters specifically include: the dominant frequency f of the source wavelet. Rand spatial location of the epicenter x s =(xs ,y s ,z s )′; where different offset imaging parameters are randomly set, the specific features are as follows:

[0079] The dominant frequency f of the random source wavelet Rand Specifically, it can be expressed as formula (3):

[0080] f Rand =Rand(·)×(f max -f min )+f min (3)

[0081] Random spatial location of the seismic source x s,Rand Represented as formula (4):

[0082]

[0083] The subscript "Rand" represents random sampling, the subscript "max" represents the maximum value, and the subscript "min" represents the minimum value. Rand(·) represents the random sampling function, and its value range is 0 to 1.

[0084] Using each set of migration imaging parameters for each model, the simulated seismic record of the model under that set of parameters is obtained using the finite difference numerical simulation method; by applying the reverse time migration imaging method to migrate the simulated seismic record, the resulting migration imaging profile is the second target seismic imaging profile (low-quality seismic imaging profile) of the model under that set of parameters.

[0085] S14. For the second target seismic imaging profile, adaptive synthetic sampling is used to expand the seismic imaging profile with diverse characteristics, and seismic noise is added to each of the resulting seismic imaging profiles.

[0086] For each set of migration imaging parameters corresponding to each seismic velocity model, an adaptive synthetic sampling method is used to adaptively sample the seismic migration imaging profile, expanding each original migration imaging profile into N new low-quality seismic migration imaging profiles with different characteristics. The superscript "i" can be any N integers from 1 to N, representing the index number of the newly expanded low-quality seismic migration imaging profile.

[0087] The adaptive synthetic sampling method refers to extracting each offset imaging profile. The amplitude, phase, and spectrum characteristic data are obtained, and these characteristic data are preprocessed. Adaptive synthetic sampling is then performed on the preprocessed characteristic data, the extracted features are classified, and then randomly combined to expand into N new low-quality seismic migration imaging profiles with different characteristics. The adaptive synthetic sampling method specifically includes the following five steps:

[0088] Step (1): Calculate the number of samples available for synthesis, which can be specifically expressed as:

[0089] G=(m l -m s )β (5)

[0090] Each offset imaging profile The number of majority class samples for amplitude, phase, and spectral characteristics is m. l The number of minority class samples is m s β∈[0,1] random numbers, if β equals 1, the ratio of positive to negative after sampling is 1:1;

[0091] Step (2): Calculate the majority class percentage among K-nearest neighbors, expressed mathematically as:

[0092] r m =Δ m / K (6)

[0093] Where, Δ i Let m be the number of majority class samples in the K-nearest neighbors, where m = 1, 2, 3, ..., m s ;

[0094] Step (3): For r m Standardized, the mathematical expression is as follows:

[0095]

[0096] Step (4): Based on the majority class proportion in the K-nearest neighbors, calculate the number of types of samples from each minority class that can be used to generate new low-quality seismic migration imaging profiles. The mathematical expression is as follows:

[0097]

[0098] Step (5): Generate new low-quality seismic migration imaging profiles with different characteristics based on the SMOTE algorithm. The mathematical expression is as follows:

[0099]

[0100] Where, x i It is the i-th sample in the minority class, xzi It is x i A minority class sample is randomly selected from the K nearest neighbors, where zi∈[1,g] and λ∈[0,1].

[0101] The characteristic data such as amplitude, phase and spectrum are preprocessed, specifically including zero-mean normalization and data normalization of the characteristic data such as amplitude, phase and spectrum;

[0102] Specifically, zero-mean normalization can be expressed as:

[0103] y new =yy a (10)

[0104] Among them, y new This is the centralized data, y is the original data, y a It is the mean of the data;

[0105] Specifically, data normalization can be expressed as:

[0106]

[0107] Among them, y nor These are the normalized data, where min(y) is the minimum value of the data and max(y) is the maximum value of the data.

[0108] New low-quality seismic migration imaging profiles with different characteristics were obtained. By adding appropriate seismic noise, we obtain noisy, low-quality seismic migration imaging profiles containing both Gaussian noise and simulated seismic noise. Specifically: based on new low-quality seismic migration imaging profiles with different characteristics. Gaussian noise was added using a Gaussian distribution, and simulated earthquake noise was added by convolving a Ricker wavelet with a Gaussian noise matrix.

[0109] Adding noise can be specifically represented as:

[0110] y ng =y l +y s ×R g +y m (12)

[0111] Among them, y n This is a low-resolution seismic imaging profile with Gaussian noise added. l This is the low-resolution seismic imaging profile obtained by the above-mentioned extended method, y s R represents the standard deviation of the seismic imaging profile. gIt is a random number matrix of the same size as the seismic imaging profile, which follows a standard Gaussian distribution (i.e., mean 0, variance 1); y m The Gaussian distribution for the simulated seismic noise is as follows:

[0112]

[0113] Where R is a random variable, μ is the mean, and σ is the mean. 2 For variance, its probability density function can be specifically expressed as:

[0114]

[0115] Where d is the vector dimension and δ is the mean covariance;

[0116] Simulated earthquake noise y m Specifically, it can be expressed as:

[0117] y m ={[1-2(πf0t)} 2 ]exp(-(πf0t) 2 )}*R g (15)

[0118] Where t is time (s) and f0 is the main frequency (Hz).

[0119] For each set of migration imaging parameters of each seismic velocity model, the corresponding low-quality seismic imaging profile Adaptive synthetic sampling is used to adaptively sample the seismic migration imaging profile to obtain each migration imaging profile. The corresponding N new low-quality seismic migration imaging profiles with different characteristics Using the above noise-adding method, a noisy, low-quality seismic migration imaging profile is obtained.

[0120] S15. Design an initial generative adversarial network that combines a generator and a discriminator with denoising and resolution enhancement functions.

[0121] Generative Adversarial Networks (GANs) were used as the backbone network architecture for training to improve seismic resolution. Generator and discriminator sub-networks were designed to perform both denoising and resolution improvement. The GAN network architecture is based on the traditional GAN ​​architecture, with improvements to the generator G and the inclusion of a discriminator D. The generator G consists of two networks, designed for denoising and resolution improvement respectively, incorporating a gradient adjustment mechanism to form an artificial intelligence network architecture. The discriminator D consists of multiple convolutional blocks. Specifically, the generator G network architecture is as follows: the denoising module incorporates a ResNet residual network, consisting of 34 convolutional layers connected in a direct-connection manner and skipping every three layers. Except for the input layer, which is a 7×7 convolutional layer, all other convolutional layers are 3×3. After the input layer, the data is connected to the other convolutional layers via pooling layers, and after the network output layer, the data is output via an average pooling layer.

[0122] The resolution enhancement module introduces a fully convolutional neural network (FCN), which consists of 24 convolutional layers interspersed with 8 max pooling layers. In the FCN network, except for the 2×2 convolutional layers, the kernel size of the 6th, 7th, and 8th convolutional layers is all 1*1.

[0123] Specifically, the adversarial loss L of the generator G network can be expressed as:

[0124] L = L G +L R +L F (16)

[0125] Where L G For counter-loss, L R L is the training loss for the ResNet denoising module. F Improve the resolution module loss for FCN.

[0126] Among them, the adversarial loss L G Specifically, it can be expressed as:

[0127]

[0128] Where x i For real earthquake model data, z i This is low signal-to-noise ratio, low-resolution seismic imaging profile data.

[0129] The training loss of the ResNet denoising module is L. R Specifically, it can be expressed as:

[0130]

[0131] Where y r It is a high signal-to-noise ratio seismic imaging profile. It is the probability predicted by the model.

[0132] Among them, the FCN resolution enhancement module suffers a loss of L. F Specifically, it can be expressed as:

[0133]

[0134] Where y j This is low signal-to-noise ratio seismic data. Predict earthquake data for the network.

[0135] The gradient adjustment mechanism specifically adds a gradient penalty term to the loss function in the artificial intelligence network. The gradient penalty term is specifically the Lipschitz condition. Using this condition, when the gradient of the loss function is greater than 1, the artificial intelligence network is constrained to make the gradient of the loss function less than or equal to 1.

[0136] S16. Using the seismic imaging profile as input for network training, the initial generative adversarial network is trained to obtain an initial resolution improvement model.

[0137] The resulting noisy, low-quality seismic migration imaging profile As input for network training, the constructed GAN network is trained to form an initial intelligent network model for improving the quality of seismic imaging profiles. Specifically, based on noisy, low-quality seismic migration imaging profiles... and high-quality seismic imaging profiles Constructing a large-scale training set required for training an intelligent resolution-enhancing network based on semi-supervised learning; and using high-quality seismic imaging profiles. Labeled data is fed into a GAN network to train an initial prediction model. This initial prediction model is then used to analyze noisy, low-quality seismic migration imaging profiles. Prediction, and the prediction results Confidence analysis was performed, and high-confidence prediction results were added to the label set to form an initial intelligent network model for improving the quality of seismic imaging profiles.

[0138] The training process is as follows: During training, the training set is fed into the ResNet and FCN networks in the generator G for training. The predicted high-resolution, high-signal-noise ratio seismic imaging profile data output by G is then fed into the discriminator D. In the discriminator D, a gradient adjustment mechanism is applied to calculate the adversarial loss L, where L is expressed as:

[0139] L = L G +L R +L F (20)

[0140] L G For counter-attack losses:

[0141]

[0142] Where, x i For real earthquake model data, z i This is low signal-to-noise ratio, low-resolution seismic imaging profile data;

[0143] The training loss of the ResNet denoising module is L. R for:

[0144]

[0145] Among them, y r It is a high signal-to-noise ratio seismic imaging profile. It is the model's predicted probability;

[0146] FCN improves resolution module loss L F for:

[0147]

[0148] Among them, y j This is low signal-to-noise ratio seismic data. Predict earthquake data for the network.

[0149] S17. The initial resolution improvement model is transferred to the actual seismic imaging profile data of the target work area to obtain a target resolution improvement model applicable to the target work area.

[0150] As an optimization, based on actual data from the work area, starting with the initial intelligent network model for improving seismic imaging profile quality, transfer learning is conducted to ultimately form an optimized intelligent network model for improving seismic imaging profile quality suitable for a specific work area. A transfer learning mechanism is introduced, utilizing the correlation between different data sets, to apply the network model trained on existing synthetic data to different but correlated actual data, strengthening the network model's ability to learn the features of actual data. Using a small number of new data samples for transfer learning on the original network model can effectively improve the resolution of new data. An adaptive learning rate adjustment method is introduced; the initial learning rate (lr) is set to 0.0001 in the first 20 rounds, and in subsequent rounds, the learning rate is dynamically adjusted to lr based on the network convergence. i =0.1×lr i-1 .

[0151] The parameters of the generator G and discriminator D in the GAN are continuously optimized through numerous training rounds. The convergence of the network model is determined based on the changes in the network loss calculated during training. If convergence is achieved, the network model is output to the next stage; otherwise, the parameters are reselected and the GAN network is trained again.

[0152] The trained model is saved, and finally an optimized intelligent network model for improving the quality of seismic imaging profiles is formed, which is suitable for specific work areas.

[0153] S18. Based on the target resolution improvement model, the actual seismic imaging profile of the target work area is processed to obtain the target seismic imaging profile.

[0154] Based on the obtained target resolution enhancement model, the actual seismic imaging profiles of the target work area are processed to obtain the target seismic imaging profiles. This is achieved by applying denoising and resolution enhancement subnetworks to improve the quality of the seismic imaging profiles; establishing a mapping relationship between low-quality and high-quality seismic imaging profiles to obtain high-quality post-stack seismic profiles; and using adversarial training between the generator and discriminator of a GAN to enable the generator to learn and generate realistic high-resolution images, further improving the quality of the migrated profiles. Furthermore, denoising and resolution enhancement modules are embedded in the generator network, enabling it to learn and generate imaging results with higher signal-to-noise ratio and better resolution. This approach has stronger applicability to accurate velocity models and higher computational efficiency in handling complex geological structures and 3D problems. Research in this field has significant practical implications for applications such as seismic exploration and subsurface structure interpretation.

[0155] The method for improving the resolution of seismic imaging profiles provided in this invention involves collecting and analyzing various seismic imaging profiles to construct multiple seismic velocity models; obtaining a first target seismic imaging profile based on the seismic convolution model and the multiple seismic velocity models; obtaining a second target seismic imaging profile by randomly setting different migration imaging parameters and using reverse time migration imaging to perform migration imaging on the multiple seismic velocity models; expanding the seismic imaging profiles with diverse characteristics using adaptive synthetic sampling for the second target seismic imaging profiles, and adding seismic noise to each group of seismic imaging profiles; designing an initial generative adversarial network (GAN) with a generator and discriminator that has denoising and resolution improvement functions by combining a generative adversarial network (GAN); training the initial GAN ​​using the seismic imaging profiles as input to obtain an initial resolution improvement model; performing transfer learning on the initial resolution improvement model using actual seismic imaging profile data of the target work area to obtain a target resolution improvement model suitable for the target work area; and processing the actual seismic imaging profiles of the target work area based on the target resolution improvement model to obtain the target seismic imaging profile. Compared to existing methods for improving the resolution of migration imaging profiles, which suffer from problems such as model overfitting, large errors, and low network performance in the super-resolution stage, this method addresses these issues. By applying sub-networks for denoising and resolution enhancement to improve the quality of seismic imaging profiles, and establishing a mapping relationship between low-quality and high-quality seismic imaging profiles, it obtains high-quality post-stack seismic profiles. This approach simultaneously improves the signal-to-noise ratio and resolution of the migration profiles, offering high computational efficiency, no parameter dependence, and applicability to any complex geological and geophysical environment, demonstrating significant potential.

[0156] To further illustrate the feasibility and effectiveness of this invention, three examples are provided below:

[0157] Example 1:

[0158] Figure 2 This is a salt dome model. The model contains 2001×501 grids, with a spatial spacing of 10m in each direction. The seismic source is located at (4000m, 100m), the time sampling interval is 1ms, an explosive source is used to generate seismic waves, and the wavelet used is the Ricker wavelet with a dominant frequency of 15Hz. Figure 3 It utilizes the reverse time migration imaging method and Figure 2 The image shows a low-quality seismic imaging profile obtained from the salt dome model. Figure 4 This is a high-quality seismic imaging profile obtained using methods to enhance the resolution of the local seismic imaging profile. From Figure 3 and Figure 4 As can be seen, the resolution enhancement method of this seismic imaging profile produces high-quality seismic imaging profiles with good denoising and resolution enhancement effects, proving the feasibility and effectiveness of the resolution enhancement method of this seismic imaging profile.

[0159] Example 2:

[0160] Figure 5 It is low-quality seismic data obtained by adding simulated seismic noise using Gaussian distribution and convolution of Ricker wavelet with Gaussian noise matrix; Figure 6 The high-quality seismic data obtained using the resolution enhancement method of the local seismic imaging profile demonstrates the feasibility of the resolution enhancement method.

[0161] Example 3:

[0162] Figure 7 These are seismic imaging profiles extracted from actual Western Digital data and obtained from actual seismic imaging profiles using methods to enhance the resolution of local seismic imaging profiles: where, Figure 7 (a) is a seismic imaging profile extracted from actual data from Western Digital. Figure 7 (b) is to utilize the present invention from Figure 7 (a) shows the seismic imaging profile obtained by increasing the resolution.

[0163] Figure 8 From Figure 7 A partial view of the seismic imaging profile shown, extracted from the seismic imaging profile: [Image of the profile would be inserted here] Figure 8 (a) is Figure 7 (a) shows a partial view of the seismic imaging profile extracted from the actual seismic imaging profile of Western Data. Figure 8 (b) is from Figure 7 (b) shows a partial view of the seismic imaging profile extracted from the improved seismic imaging profile.

[0164] Figure 9 From Figure 7 The spectral graph obtained from the seismic imaging profile shown is obtained using spectral analysis: where, Figure 9 (a) is Figure 7 (a) is a spectrum of the actual seismic imaging profile from Western Data. Figure 9 (b) is from Figure 7 (b) is a spectrum of the seismic imaging profile obtained with improved resolution. Figure 7 (a) and (b) Figure 8 (a) and (b) and Figure 9 As can be seen from (a) and (b), the improved resolution seismic imaging profiles of the actual Western data obtained using the resolution enhancement method of this seismic imaging profile are clearer, the phase axes are more focused, and the frequency band is broadened, proving the effectiveness of the resolution enhancement method of this seismic imaging profile. In conclusion, the resolution enhancement method of this seismic imaging profile has good feasibility and practicality in complex geological and geophysical models.

[0165] Figure 10 A schematic diagram of a resolution-enhancing device for seismic imaging profiles according to an embodiment of the present invention is shown. Figure 10 As shown, the device includes:

[0166] Module 1001 is used to collect and analyze various seismic imaging profiles and construct multiple seismic velocity models. For detailed explanations, please refer to the relevant descriptions in the above method embodiments; they will not be repeated here.

[0167] The acquisition module 1002 is used to obtain a first target seismic imaging profile based on the seismic convolution model and the multiple seismic velocity models. For detailed explanations, please refer to the relevant descriptions in the above method embodiments; they will not be repeated here.

[0168] The acquisition module 1002 is used to perform migration imaging on the multiple seismic velocity models by randomly setting different migration imaging parameters and using the reverse time migration imaging method to obtain a second target seismic imaging profile. For detailed explanations, please refer to the relevant descriptions in the above method embodiments, which will not be repeated here.

[0169] The expansion module 1003 is used to expand the seismic imaging profile with diverse characteristics based on the second target seismic imaging profile using adaptive synthetic sampling, and to add seismic noise to each of the resulting seismic imaging profiles. For detailed explanations, please refer to the relevant descriptions in the above method embodiments, which will not be repeated here.

[0170] Design module 1004 is used to design an initial generative adversarial network (GAN) that combines a generator and a discriminator with denoising and resolution enhancement functions. For detailed explanations, please refer to the relevant descriptions in the above method embodiments; they will not be repeated here.

[0171] Training module 1005 is used to train the initial generative adversarial network by using the seismic imaging profile as input to the network training, thereby obtaining an initial resolution-enhanced model. For detailed explanations, please refer to the relevant descriptions in the above method embodiments; they will not be repeated here.

[0172] Training module 1005 is used to perform transfer learning on the initial resolution improvement model using actual seismic imaging profile data of the target work area, to obtain a target resolution improvement model suitable for the target work area. For detailed explanations, please refer to the relevant descriptions in the above method embodiments; they will not be repeated here.

[0173] Processing module 1006 is used to process the actual seismic imaging profile of the target work area based on the target resolution improvement model to obtain the target seismic imaging profile. For detailed explanations, please refer to the relevant descriptions in the above method embodiments; they will not be repeated here.

[0174] The seismic imaging profile resolution improvement device provided in this embodiment of the invention is used to execute the seismic imaging profile resolution improvement method provided in the above embodiment. Its implementation method and principle are the same. For details, please refer to the relevant description of the above method embodiment, which will not be repeated here.

[0175] Figure 11 An electronic device according to an embodiment of the present invention is shown, such as... Figure 11 As shown, the electronic device may include a processor 1101 and a memory 1102, wherein the processor 1101 and the memory 1102 may be connected via a bus or other means. Figure 11 Taking the example of a connection between China and Israel via a bus.

[0176] Processor 1101 may be a central processing unit (CPU). Processor 1101 may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations thereof.

[0177] The memory 1102, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the methods provided in the embodiments of the present invention. The processor 1101 executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory 1102, thereby implementing the methods in the above-described method embodiments.

[0178] The memory 1102 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor 1101, etc. Furthermore, the memory 1102 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 1102 may optionally include memory remotely located relative to the processor 1101, and these remote memories may be connected to the processor 1101 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0179] One or more modules are stored in memory 1102 and, when executed by processor 1101, perform the methods described in the above method embodiments.

[0180] The specific details of the aforementioned electronic device can be understood by referring to the relevant descriptions and effects in the above method embodiments, and will not be repeated here.

[0181] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by hardware related to computer program instructions. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

[0182] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for improving the resolution of seismic imaging profiles, characterized in that, include: Collect and analyze multiple seismic imaging profiles to construct multiple seismic velocity models; Based on the seismic convolution model and the multiple seismic velocity models, the first target seismic imaging profile (high-quality seismic imaging profile) is obtained; By randomly setting different migration imaging parameters, the multiple seismic velocity models are migrated using the reverse time migration imaging method to obtain a second target seismic imaging profile (low-quality seismic imaging profile). For the second target seismic imaging profile, adaptive synthetic sampling is used to expand the seismic imaging profile with diverse characteristics, and seismic noise is added to each of the resulting seismic imaging profiles. Design an initial generative adversarial network that combines a generator and a discriminator to achieve denoising and resolution enhancement; The seismic imaging profile is used as input for network training to train the initial generative adversarial network, resulting in an initial resolution improvement model. The initial resolution improvement model is transferred and learned by using actual seismic imaging profile data of the target work area to obtain a target resolution improvement model applicable to the target work area. The target resolution enhancement model is used to process the actual seismic imaging profile of the target work area to obtain the target seismic imaging profile.

2. The method according to claim 1, characterized in that, The process involves collecting and analyzing multiple seismic imaging profiles to construct several seismic velocity models, including: Theoretical seismic profiles are obtained from existing seismic migration imaging data, and different actual seismic profiles are obtained from existing seismic imaging profiles of various types of oilfields. Based on the theoretical and actual seismic profiles, a seismic texture structure model corresponding to each actual seismic profile is constructed. Multiple seismic velocity models are constructed based on the seismic texture structure model and the rock physical parameter characteristics of the work area corresponding to the seismic texture structure model.

3. The method according to claim 2, characterized in that, The method for obtaining the first target seismic imaging profile based on the seismic convolution model and the multiple seismic velocity models includes: Based on the aforementioned multiple seismic velocity models, multiple reflection coefficient models are obtained through a first formula, wherein the first formula is: in, R ( x () represents the reflection coefficient model. v ( x () represents the earthquake velocity model; The high-frequency wavelet is convolved with the multiple reflection coefficient models using the seismic convolution model theory to obtain multiple seismic imaging profiles of the first target.

4. The method according to claim 3, characterized in that, The process of obtaining a second target seismic imaging profile by randomly setting different migration imaging parameters and using the reverse time migration imaging method to perform migration imaging on the multiple seismic velocity models includes: Different migration imaging parameters are randomly set for each of the seismic velocity models to obtain a set of imaging parameter combinations corresponding to each of the seismic velocity models; Based on each set of migration imaging parameters, the finite difference numerical simulation method is used to obtain the simulated seismic record of each seismic velocity model under the corresponding set of migration imaging parameters; The simulated seismic record is migrated using the reverse time migration imaging method to obtain the corresponding migration imaging profile, wherein the migration imaging profile is the second target seismic imaging profile of the seismic velocity model under the corresponding set of migration imaging parameters.

5. The method according to claim 4, characterized in that, For the second target seismic imaging profile, adaptive synthetic sampling is used to expand seismic imaging profiles with diverse characteristics, and seismic noise is added to each resulting set of seismic imaging profiles, including: An adaptive synthetic sampling method is used to adaptively sample each second target seismic imaging profile, expanding each second target seismic imaging profile into multiple seismic imaging profiles with diverse characteristics. Seismic noise is added to each seismic imaging profile with diverse characteristics to obtain noisy seismic imaging profiles containing Gaussian noise and simulated seismic noise.

6. The method according to claim 5, characterized in that, The step of using the seismic imaging profile as input to train the initial generative adversarial network to obtain an initial resolution-enhancing model includes: Based on the noisy seismic imaging profile and the first target seismic imaging profile, a neural network training set required for network training based on semi-supervised learning is constructed. The first target seismic imaging profile is input as labeled data into the initial generative adversarial network to obtain the initial prediction model; The noisy seismic imaging profile is predicted using the initial prediction model, and the prediction results are analyzed for confidence. Prediction results with confidence scores higher than a preset confidence threshold are added to the label set to obtain the initial resolution improvement model.

7. The method according to claim 6, characterized in that, The process of transferring the initial resolution improvement model to actual seismic imaging profile data of the target work area to obtain a target resolution improvement model suitable for the target work area includes: The initial generative adversarial network is trained using the actual seismic profile and the neural network training set. Transfer learning is carried out starting with the initial resolution improvement model, and the parameters of the initial generative adversarial network are optimized through rounds of training. An adaptive learning rate is used to dynamically adjust the learning rate in each round of training to obtain a target resolution improvement model suitable for the target work area.

8. A device for improving the resolution of seismic imaging profiles, characterized in that, include: The module is used to collect and analyze various seismic imaging profiles and build multiple seismic velocity models; The acquisition module is used to obtain the first target seismic imaging profile based on the seismic convolution model and the multiple seismic velocity models; The acquisition module is used to perform migration imaging on the multiple seismic velocity models by randomly setting different migration imaging parameters and using the reverse time migration imaging method to obtain a second target seismic imaging profile. The expansion module is used to expand the seismic imaging profile with diverse features for the second target seismic imaging profile using adaptive synthetic sampling, and to add seismic noise to each of the resulting seismic imaging profiles. The design module is used to combine generative adversarial networks to design an initial generative adversarial network for a generator and a discriminator with denoising and resolution enhancement functions; The training module is used to train the initial generative adversarial network by taking the seismic imaging profile as input for network training, thereby obtaining an initial resolution-enhanced model. The training module is used to perform transfer learning on the initial resolution improvement model using actual seismic imaging profile data of the target work area, so as to obtain a target resolution improvement model suitable for the target work area. The processing module is used to process the actual seismic imaging profile of the target work area based on the target resolution improvement model to obtain the target seismic imaging profile.

9. An electronic device, characterized in that, include: A processor and a memory, the processor being configured to execute a resolution enhancement program for seismic imaging profiles stored in the memory to implement the resolution enhancement method for seismic imaging profiles according to any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores one or more programs, which can be executed by one or more processors to implement the method for improving the resolution of seismic imaging profiles according to any one of claims 1 to 7.

Images