Processing method, device and equipment for seismic data

Abstract

The application discloses a seismic data processing method, device and equipment, wherein the method comprises the following steps: acquiring the seismic data of the common shot gather of a target area, wherein the seismic data comprises source-end ghost wave data and receiver-end ghost wave data; performing correction processing on the seismic data to obtain corrected seismic data; inputting the normalized corrected seismic data into a pre-trained ghost wave suppression model to output first seismic data after receiver-end ghost wave suppression and second seismic data after source-end ghost wave suppression; and performing inverse normalization processing and inverse correction processing on the second seismic data to obtain target seismic data after ghost wave suppression. The application can distinguish the source-end and receiver-end ghost wave data in the seismic data, can simultaneously realize effective suppression of the receiver-end ghost wave noise and the source-end ghost wave noise, and can quickly, efficiently and accurately obtain the target seismic data after ghost wave suppression.

Description

Technical Field

[0001] The present application relates to the field of oil exploration technology, and in particular to a method, device and equipment for processing seismic data. Background Art

[0002] Ghost wave fields include source ghost wave fields and detector-side ghost wave fields. Source ghost wave fields typically occur when the earthquake source is located below the sea surface during marine seismic acquisition. Due to the strong wave impedance created by the seawater-air interface, the sea surface reflection coefficient is very close to -1. Therefore, at the sea surface, the upgoing portion of the source wave field becomes a downgoing wave field, known as the source ghost wave field. Detector-side ghost wave fields are the sea surface reflection of the entire upgoing wave field. The presence of ghost waves can severely limit the bandwidth of seismic exploration and reduce the signal-to-noise ratio of seismic data.

[0003] Currently, research on ghost suppression relies on traditional algorithms, such as those based on convolution, ray tracing, and wave equations, as well as neural network-based approaches, such as deep convolutional neural networks and residual neural networks. However, none of these approaches distinguish between source-side ghosts and receiver-side ghosts, and thus fail to effectively suppress source-side ghosts.

[0004] Currently, no effective solution has been proposed to suppress ghost noise in seismic data. Summary of the Invention

[0005] The purpose of the embodiments of the present application is to provide a method, apparatus and device for processing seismic data to solve the problem of being unable to effectively suppress ghost noise in seismic data.

[0006] To solve the above technical problems, the first aspect of this specification provides a method for processing seismic data, comprising:

[0007] Acquire seismic data of a common shot gather in a target area, wherein the seismic data includes ghost wave data at the source end and ghost wave data at the receiver end;

[0008] performing correction processing on the seismic data to obtain corrected seismic data;

[0009] After normalizing the corrected seismic data, the data are input into a pre-trained ghost wave suppression model to output the first seismic data after ghost wave suppression at the detector end and the second seismic data after ghost wave suppression at the source end;

[0010] The second seismic data is subjected to inverse normalization processing and inverse correction processing to obtain target seismic data of the target area after ghost wave suppression.

[0011] In some embodiments, after normalizing the corrected seismic data, the data is input into a pre-trained ghost suppression model, and the output of the first seismic data after the ghost waves are suppressed at the detector end and the second seismic data after the ghost waves are suppressed at the source end include:

[0012] After the corrected seismic data is normalized, it is input into a pre-trained ghost wave suppression model, and the first seismic data after the ghost wave is suppressed at the detector end is output;

[0013] The first seismic data is input into a pre-trained ghost wave suppression model, and the second seismic data after the ghost waves at the source end are output.

[0014] In some embodiments, training the ghost suppression model includes:

[0015] Acquiring historical seismic data of a common shot gather, wherein the historical seismic data includes historical ghost wave data at the source end and historical ghost wave data at the geophone end;

[0016] Correcting the historical seismic data to obtain corrected seismic data;

[0017] A ghost wave suppression method based on sparse inversion is used to suppress the historical ghost wave data at the geophone end in the corrected seismic data, and the corrected seismic data after ghost wave suppression is obtained.

[0018] The calibrated seismic data after ghost wave suppression is sparsely sampled as label data to simulate the common detection point gather data after ghost wave suppression at the source end.

[0019] Normalizing the corrected seismic data and the labeled data, and using the normalized corrected seismic data and the labeled data as training data;

[0020] The training data is input into the initial neural network model for training to obtain the ghost wave suppression model.

[0021] In some embodiments, correcting the historical seismic data to obtain corrected seismic data includes:

[0022] Depth difference correction is performed on the historical seismic data, and single-shot seismic data that meets preset conditions is selected from the historical seismic data after depth difference correction as corrected seismic data according to a preset ratio.

[0023] In some embodiments, inputting training data into an initial neural network model for training to obtain the ghost wave suppression model includes:

[0024] Inputting the normalized corrected seismic data into the initial neural network model to obtain the actual output of the initial neural network model, and calculating the distance between the actual output and the label data;

[0025] The parameters of the initial neural network model are adjusted based on the distance and the actual output until the distance meets a preset distance threshold, and the training is stopped to obtain the ghost wave suppression model.

[0026] In some embodiments, the initial neural network model includes a generator and a discriminator;

[0027] Accordingly, the normalized corrected seismic data is input into the initial neural network model to obtain the actual output of the initial neural network model, and the distance between the actual output and the label data is calculated, including:

[0028] Inputting the normalized corrected seismic data into the generator, performing feature extraction on the normalized corrected seismic data using the generator, and generating noise-free seismic data based on the feature extraction result;

[0029] Inputting the noise-free seismic data and the label data into the discriminator, determining whether the noise-free seismic data matches the label data, and feeding back the determination result to the generator;

[0030] Calculating the distance between the noise-free seismic data and the label data, and feeding the distance back to the generator;

[0031] Accordingly, adjusting the parameters of the initial neural network model based on the distance and the actual output includes:

[0032] Based on the judgment result and the distance, parameters of the generator and the discriminator are adjusted.

[0033] In some embodiments, the generator includes N encoding modules, N decoding modules, N skip connection layers and residual modules, and the encoding modules correspond one to one to the decoding modules;

[0034] The encoding module is used to encode the input data, and the decoding module is used to

[0035] The skip connection layer is used to connect the output of the encoding module and the input of the decoding module symmetrically distributed with the encoding module;

[0036] The residual module is used to connect the input of the generator and the output of the Nth decoding module.

[0037] In some embodiments, the discriminator includes multiple convolution modules and discrimination modules, which are used to perform convolution processing on the data input into the convolution module. The convolution module is used to extract features from the input data. The discrimination module is used to judge whether the noise-free seismic data matches the label data based on the feature extraction results of multiple convolution modules, and use the judgment result as the output of the discriminator.

[0038] A second aspect of this specification provides a seismic data processing device, comprising:

[0039] A data acquisition module is used to acquire seismic data of a common shot gather in a target area, wherein the seismic data includes ghost wave data at the source end and ghost wave data at the detector end;

[0040] A data preprocessing module, configured to perform correction processing on the seismic data to obtain corrected seismic data;

[0041] A ghost wave suppression module is used to normalize the corrected seismic data, input the data into a pre-trained ghost wave suppression model, and output the first seismic data after ghost wave suppression at the detector end and the second seismic data after ghost wave suppression at the source end;

[0042] The target data determination module is used to perform inverse normalization processing and inverse correction processing on the second seismic data to obtain target seismic data after ghost waves are suppressed in the target area.

[0043] The third aspect of this specification provides an electronic device, comprising: a memory and a processor, wherein the processor and the memory are communicatively connected to each other, the memory stores computer instructions, and the processor implements the steps of any one of the methods described in the first aspect by executing the computer instructions.

[0044] A fourth aspect of this specification provides a computer storage medium, wherein the computer storage medium stores computer program instructions, and when the computer program instructions are executed, the steps of any one of the methods described in the first aspect are implemented.

[0045] The seismic data processing method provided in the embodiments of this specification obtains seismic data of a common shot point gather in a target area, wherein the seismic data includes ghost wave data at the source end and ghost wave data at the detector end; corrects the seismic data to obtain corrected seismic data; normalizes the corrected seismic data and inputs them into a pre-trained ghost wave suppression model to output first seismic data after ghost wave suppression at the detector end and second seismic data after ghost wave suppression at the source end; denormalizes and decorrects the second seismic data to obtain target seismic data after ghost wave suppression in the target area. In this application, by correcting the seismic data, the seismic data corresponding to the detector end in the seismic data is corrected to the source end, and the processed seismic data is input into a pre-trained ghost wave suppression model, which can simultaneously achieve effective suppression of the ghost wave noise at the detector end and the ghost wave noise at the source end in the seismic data. In addition, this application can distinguish the ghost wave data at the source end and the detector end in the seismic data, and suppress ghost waves through a pre-trained ghost wave suppression model. It is not limited by the seismic data observation system, reduces manual intervention in the ghost wave suppression process, and can quickly, efficiently and accurately achieve the suppression of ghost wave noise at the source end and the detector end. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the implementation methods of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the implementation methods or the description of the prior art. Obviously, the drawings described below are only some implementation methods recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0047] Figure 1 FIG2 is a schematic diagram of a training method for a ghost wave suppression model provided in an embodiment of the present application;

[0048] Figure 2 FIG2 is a schematic diagram of the initial neural network model data processing flow provided by an embodiment of the present application;

[0049] Figure 3 Shown is a schematic diagram of a generator provided by an embodiment of the present application;

[0050] Figure 4 Shown is a schematic diagram of an identifier provided by an embodiment of the present application;

[0051] Figure 5 FIG2 is a schematic diagram of a method for processing seismic data provided by an embodiment of the present application;

[0052] Figure 6 Schematic diagram of a forward layered model provided in an embodiment of the present application;

[0053] Figure 7 Shown are single shot seismic data before and after ghost wave suppression of the forward layered model provided in an embodiment of the present application;

[0054] Figure 8 Shown is a schematic diagram of single shot seismic data before and after processing provided by an embodiment of the present application;

[0055] Figure 9 Shown is a schematic diagram of the corresponding frequency spectra before and after processing of single-shot seismic data provided by an embodiment of the present application;

[0056] Figure 10 FIG2 is a schematic diagram of a seismic data processing device provided in an embodiment of the present application;

[0057] Figure 11 Shown is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0059] As mentioned above, existing common-receiver gathers at the source end are sparsely sampled. Directly suppressing source-end ghost data based on wavefield propagation on the seismic data corresponding to these sparsely sampled common-receiver gathers introduces aliasing during the suppression process, making it impossible to effectively suppress source-end ghost data. Common-shot gathers, on the other hand, are densely sampled. The source-end and ghost data in these common-shot gathers carry underground information, making the source-end ghost data in these common-shot gathers more complex and ineffective in suppressing source-end ghost data. Furthermore, existing technologies fail to distinguish between source-end ghost data and receiver-end ghost data during ghost suppression.

[0060] In order to solve the above problems, the applicant found that there is a height difference in the position setting of the source and the detector in actual marine seismic acquisition. The ghost wave data at the source end of the measured common detection point data set and the ghost wave data at the detector end of the common shot point data set were analyzed. It was found that the height difference between the source and the detector in space is reflected in the time of the ghost wave data of the two. Therefore, this application utilizes the relationship between the source-end ghost wave data of the common-detection point gather and the detector-end ghost wave data of the common-shot point gather. First, the seismic data of the common-shot point gather is corrected, the seismic data is corrected to the depth of the source end, and the ghost wave of the detector end after correction is attenuated. Then, the attenuated seismic data of the detector end is sparsely sampled so that the spatial sampling of the seismic data of the detector end is the same as that of the seismic data at the source end. The seismic data of the common-shot point gather that are sparsely sampled before and after correction and ghost wave suppression are used to simulate the seismic data of the sparsely sampled common-detector gather before and after source-end ghost wave attenuation to obtain a suitable training pair, and finally a ghost wave suppression model is established through deep learning methods.

[0061] Furthermore, in actual application, the present application corrects the seismic data, corrects the seismic data corresponding to the detector end in the seismic data to the source end, and inputs the processed seismic data into a pre-trained ghost wave suppression model, thereby achieving effective suppression of the ghost wave noise at the detector end and the ghost wave noise at the source end in the seismic data. In addition, the present application can distinguish the ghost wave data at the source end and the detector end in the seismic data, and suppresses ghost waves through a pre-trained ghost wave suppression model, which is not restricted by the seismic data observation system, reduces manual intervention in the ghost wave suppression process, and can quickly, efficiently and accurately suppress the ghost wave noise at the source end and the detector end.

[0062] The following first introduces the training process of the ghost wave suppression model in the embodiment of the present application with reference to the accompanying drawings.

[0063] Figure 1 FIG. 1 is a schematic diagram of a training method for a ghost wave suppression model provided in an embodiment of the present application. Figure 1 As shown, the training method includes:

[0064] S101: Acquire historical seismic data of a common shot gather, wherein the historical seismic data includes historical ghost wave data at the source end and historical ghost wave data at the detector end.

[0065] It can be understood that the seismic data of the common shot point gather in any work area may include ghost wave data. Based on the different formation positions of the ghost wave field of the seismic data, the ghost wave data can be divided into source end ghost wave data and detector end ghost wave data. Among them, the common shot point gather can be understood as a collection of seismic records of each channel at the same shot point (source), and the seismic data of the common receiver point gather can be simulated based on the historical seismic data of the common shot point gather. Among them, the common receiver point gather can be understood as a gather composed of all channels of the same receiving point (i.e., receiver point) but different shot points arranged in the order of the common depth point number. The historical seismic data of the common shot point gather can be the seismic data actually collected at the detector end.

[0066] In some embodiments, the historical seismic data obtained in step S101 may be historical seismic data in the target area, that is, a specific work area where ghost wave suppression is required, or may be historical seismic data records in other work areas, or may be historical seismic data in multiple work areas. That is, in some embodiments, the ghost wave suppression model may be based on the historical seismic data in the work area where ghost wave suppression is required, and the training data is generated to train the initial neural network model. In other embodiments, the ghost wave suppression model may also be based on the historical seismic data in any other work area, and the training data is generated to train the initial neural network model. In other embodiments, the ghost wave suppression model may also be based on the historical seismic data in multiple work areas, and the training data is generated to train the initial neural network model. The multiple work areas may include the work area corresponding to the target area, or may not include the work area corresponding to the target area. This application does not impose any restrictions on this.

[0067] In order to achieve a better ghost wave suppression effect, historical seismic data in a specific work area that requires ghost wave suppression can be used to generate training data, and the initial neural network model can be trained to obtain a ghost wave suppression model for the work area. The historical seismic data can be the seismic data that requires ghost wave suppression. That is, for each seismic data that requires ghost wave suppression, the model training method in the embodiment of the present application can be used to train a ghost wave suppression model that is more suitable for this ghost wave suppression, thereby suppressing the ghost wave data at the detector end and the ghost wave data at the source end in the seismic data, and obtaining the target seismic data after ghost wave suppression processing.

[0068] S102: Correcting the historical seismic data to obtain corrected seismic data.

[0069] It can be understood that the correction processing of historical seismic data is to correct the historical seismic data based on the relationship between the seismic data of the common detection point gather and the seismic data of the common shot point gather, correct the historical seismic data to the source depth, and obtain corrected seismic data, so as to use the historical seismic data to simulate the seismic data of the common detection point gather, and realize the ghost wave suppression model that can effectively suppress the ghost wave data at the source end and the ghost wave data at the detector end through training with historical seismic data.

[0070] It can be understood that using historical seismic data from common shot point gathers to simulate historical seismic data at the source end can avoid the introduction of false frequencies by sparsely sampled seismic data at the source end when performing ghost wave suppression based on wave field propagation. At the same time, there will be no underground information carried by the ghost wave data at the source end in the common shot point gathers, which can more effectively suppress the ghost wave data at the source.

[0071] In some embodiments, correcting the historical seismic data to obtain corrected seismic data includes:

[0072] Depth difference correction is performed on the historical seismic data, and single-shot seismic data that meets preset conditions is selected from the historical seismic data after depth difference correction as corrected seismic data according to a preset ratio.

[0073] In some embodiments, depth difference correction of historical seismic data may include time correction of the historical seismic data based on a time difference between the historical seismic data and the historical seismic data from a source-end common-receiver gather. Furthermore, single-shot seismic data that meets preset conditions may be selected from the time-corrected historical seismic data according to a preset ratio as the correction seismic data.

[0074] In some embodiments, representative single-shot seismic records from the time-corrected historical seismic data can be selected according to a preset ratio and combined as the corrected seismic data. Representative single-shot seismic records can include, for example, seismic records that are significantly affected by ghost waves within the work area, and seismic records that are less affected by ghost waves within the work area.

[0075] S103: Using a ghost wave suppression method based on sparse inversion, ghost wave suppression is performed on historical ghost wave data at the geophone end in the corrected seismic data to obtain ghost wave suppressed corrected seismic data.

[0076] It can be understood that the corrected seismic data after ghost wave suppression can be used to simulate the seismic data after the ghost wave data at the source end is suppressed, and the corrected seismic data can be used to simulate the seismic data before the ghost wave data at the source end is suppressed. It can also be understood as the seismic data after the ghost wave data at the detector end is suppressed and before the ghost wave data at the source end is suppressed.

[0077] In some embodiments, performing ghost suppression on the correction data using a sparse inversion-based ghost suppression method may include:

[0078] The ghost waves in the corrected seismic data are regarded as the downgoing wave field formed by the primary wave field reflected downward by the sea surface. The primary wave field in the corrected seismic data is obtained, and the ghost wave operator is used to describe the wave field extension process of the ghost wave. The ghost wave suppression calculation formula is obtained. Then, the algorithm is used to perform plane wave decomposition on the corrected seismic data to obtain the ghost waves in the corrected seismic data. The ghost waves in the corrected seismic data are then removed to obtain the corrected seismic data after ghost wave suppression.

[0079] For example, in some embodiments, historical earthquake data may be represented as S Max (P+GS+GD), correct the historical seismic data, and select N shot seismic data that meet the preset conditions as the corrected seismic data according to the preset ratio, which can be expressed as S N-Train (P+GS+GD). The ghost wave suppression method based on sparse inversion is used to suppress the ghost wave at the detector end of the corrected seismic data, and the ghost wave S N-Train (GD), the corrected seismic data after ghost wave suppression can be expressed as S N-Tag (P+GS).

[0080] S104: Sparsely sample the corrected seismic data after ghost wave suppression and use them as label data to simulate the common detection point gather data after ghost wave suppression at the source end.

[0081] In some embodiments, considering that the seismic data of the common detection point gather at the source end are sparsely sampled, sparse sampling of the corrected seismic data after ghost wave suppression can better simulate the historical seismic data of the common detection point gather at the source end.

[0082] In some embodiments, the corrected seismic data may be subjected to the same sparse sampling as the corrected seismic data after ghost wave suppression, and the sparsely sampled corrected seismic data may be subjected to subsequent normalization processing, and then used as training data for model training.

[0083] S105: normalizing the corrected seismic data and the label data, and using the normalized corrected seismic data and the label data as training data.

[0084] It can be understood that in order to prevent subsequent training data from exceeding the data processing range of the initial neural network model, the corrected seismic data and the label data can be normalized to reduce the order of magnitude of the corrected seismic data and the label data.

[0085] In some embodiments, normalizing the corrected seismic data and the label data may include:

[0086] Determine the maximum and minimum values ​​in the corrected seismic data and the label data, and then based on the determined maximum and minimum values ​​in the corrected seismic data and the label data, use a preset normalization formula to normalize the corrected seismic data and the label data.

[0087] In some embodiments, the maximum and minimum values ​​in the corrected seismic data and the label data can be determined by the following formula:

[0088] Max S=Max[S N-Train P+GS+GD,S N-Tag P+GS] Formula (1)

[0089] Min S=Min S N-Train P+GS+GD,S N-Tag P+GS] Formula (2)

[0090] Among them, Max S can represent the maximum value of the corrected seismic data and the label data, Min S can represent the minimum value of the corrected seismic data and the label data, and S N-Train P+GS+GD can represent the corrected seismic data, S N-Tag P+GS can represent label data.

[0091] In some embodiments, the normalized corrected seismic data and label data can be calculated using the following formula:

[0092]

[0093]

[0094] Among them, S N-Train-Norm (PGSGD) can represent the normalized corrected seismic record, S N-Tag-Norm (P+GS) can represent the normalized label data.

[0095] S106: Inputting the training data into the initial neural network model for training to obtain the ghost wave suppression model.

[0096] It can be understood that the initial neural network model is a deep learning model, such as a convolutional neural network model, a recurrent neural network model, a long short-term memory network model, etc., and this application does not impose any restrictions on this.

[0097] In some embodiments, inputting training data into an initial neural network model for training to obtain the ghost wave suppression model includes:

[0098] input the normalized corrected seismic data into the initial neural network model to obtain an actual output of the initial neural network model, and calculate a distance between the actual output and the label data;

[0099] adjust parameters of the initial neural network model based on the distance and the actual output until the distance meets a preset distance threshold, stop training, and obtain the ghost wave suppression model.

[0100] In some embodiments, the initial neural network model can be a convolutional neural network model, for example, a Wasserstein Generative Adversarial Networks (WGAN) model. The model can include a generator and a discriminator, wherein the generator can be used to capture the distribution characteristics of the input training data, and generate a new data distribution based on the captured distribution characteristics, and output the generated data distribution to the discriminator; the discriminator can be used to distinguish whether the input data is real data or generated data, that is, to judge whether the input data approximates the real training data.

[0101] In some embodiments, the initial neural network model includes a generator and a discriminator;

[0102] Correspondingly, input the normalized corrected seismic data into the initial neural network model to obtain an actual output of the initial neural network model, and calculate a distance between the actual output and the label data, including:

[0103] input the normalized corrected seismic data into the generator, perform feature extraction on the normalized corrected seismic data using the generator, and generate noise-free seismic data based on the feature extraction result; the feature extraction includes extracting the distribution characteristics of the normalized corrected seismic data and denoising, and the noise-free seismic data can be understood as the data generated by the generator;

[0104] input the noise-free seismic data and the label data into the discriminator, judge whether the noise-free seismic data matches the label data, and feed back the judgment result to the generator;

[0105] calculate a distance between the noise-free seismic data and the label data, and feed back the distance to the generator;

[0106] Correspondingly, adjust parameters of the initial neural network model based on the distance and the actual output, including:

[0107] adjust parameters of the generator and the discriminator based on the judgment result and the distance.

[0108] In some embodiments, a WGAN model can be constructed based on the principle of ghost wave suppression and the difficulties in suppressing ghost wave data at the source end. As an initial neural network model, the initial neural network model can be composed of a generator and a discriminator. To prevent network degradation, the generator can include a residual network. When constructing the loss function of the initial neural network model, the Wasserstein distance can be introduced into the loss function to ensure the stability of the training process.

[0109] In some embodiments, the training data can be input into the constructed initial neural network model to learn and train the model and continuously optimize the parameters of the initial neural network model. When the distance (i.e., error) between the actual output of the initial neural network model and the label data is less than a preset distance threshold, it indicates that the ghost wave suppression model training is completed, that is: S N-Tag-Norm =Net(S N-Train-Norm ; w) where w can represent the parameters of the initial neural network model, which can include the weights and biases of the generator and discriminator.

[0110] The initial neural network model constructed in the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0111] Figure 2 The figure shows a schematic diagram of the initial neural network model data processing flow provided by the embodiment of the present application. Figure 2 As shown, the initial neural network model may include a generator 201 and a discriminator 202.

[0112] Generator 201 can be used to capture the distribution characteristics of the normalized corrected seismic data z in the input training data, generate a new data distribution based on the captured distribution characteristics, and output the generated data distribution to discriminator 202; discriminator 202 can be used to determine whether the input data is real data or generated data G(z), that is, to determine whether the input data is close to the real training data. Discriminator 202 can be a binary divider to determine whether the input data is real data or generated data. For example, discriminator 202 can output true when it determines that the input data is real data, which can be represented by "1", and output false when it determines that the input data is generated data, which can be represented by "0". The initial neural network model can iteratively optimize the parameters of generator 201 and discriminator 202 based on the output of the discriminator and the distance between the data G(z) output by generator 201 and the label data x. The initial neural network model can continuously adjust the parameters to make the data distribution generated by generator 201 as close as possible to the real data distribution. The training is terminated when it is determined that the distance between the data output by the generator 201 and the label data meets the preset distance threshold, and a ghost wave suppression model is obtained.

[0113] Specifically, in some embodiments, x represents the label data, p r (x) represents the probability that the note data does not contain ghost waves. For the input normalized corrected seismic data z, the normalized corrected seismic data z satisfies the probability p z (z). The goal of the initial neural network model is to make p z (z) as close to p as possible r (x). For generator 201, its input is to satisfy the probability p z (z) is the normalized corrected seismic data z, and the output is data p g (z). The ultimate goal of the generator 201 is to make the generated data as close to p as possible. r (x). Where data p g (z) is the probability p of the generator 201 satisfying the input z The normalized corrected seismic data z of (z) is denoised while capturing its data features, and then the data generated based on the obtained data features, that is, the generated data p g (z) does not contain ghost waves.

[0114] In some embodiments, the distance between the data output by the generator 201 and the label data can be calculated using the following formula:

[0115]

[0116] Among them, W(p r ,p g ) represents the distance between the distribution of the label data and the distribution of the data output by the generator 201, sup represents the lower bound, E (x,y)～γ [||xy||] represents the expected operation on the real sample data (i.e., label data) and the generated data (i.e., the data output by the generator), Ω(p r ,p g ) indicates that the distribution is p r and p g For each possible joint distribution γ, a real sample data x and a generated data g can be obtained from the combined joint distribution set, and then the expected value between the real sample data x and the generated data g can be used as the distance between the data output by the generator 201 and the label data.

[0117] In some embodiments, the generator includes N encoding modules, N decoding modules, N skip connection layers and residual modules, and the encoding modules correspond one to one to the decoding modules;

[0118] The skip connection layer is used to connect the output of the encoding module and the input of the decoding module symmetrically distributed with the encoding module;

[0119] The residual module is used to connect the input of the generator and the output of the Nth decoding module.

[0120] The following will introduce the structures of the generator and discriminator with reference to the accompanying figures.

[0121] Figure 3 Shown is a schematic diagram of a generator provided in an embodiment of the present application.

[0122] like Figure 3 As shown in Figure 2, the generator adopts a U-shaped network structure including a residual module.

[0123] The encoding module can effectively extract representative features from the data input to the module. In the encoder module, the normalized corrected seismic data can be used as the input of the generator. The input data is processed using the mirror filling method to obtain better convolution results and suppress ghost noise. Each encoding module can include a Conv layer, a BN layer, a ReLU layer, and a downsampling layer. Among them, the Conv layer, the BN layer, and the ReLU layer can form a submodule, that is, Figure 3 The Conv+BN+ReLU submodule in each encoding module can include two Conv+BN+ReLU submodules and a Downsample layer. The downsampling layer is Figure 3 Downsample layer in .

[0124] The decoding module can include a convolution layer, a batch normalization layer, and an activation function layer. The decoding module and the encoding module are mirror-symmetrical to each other. The convolution layer of the decoding module is used to restore signal details and performs the inverse operation of convolution. Each decoding module can include a Conv layer, a BN layer, a ReLU layer, and an upsampling layer. Among them, the Conv layer, the BN layer, and the ReLU layer can form a submodule, that is, Figure 3 The Conv+BN+ReLU submodule in each decoding module can include a Conv+BN+ReLU submodule and an Upsample layer. The downsampling layer is Figure 3 Upsample layer in .

[0125] The skip connection layer can be used to connect the output of the encoding module with the input of the decoding module symmetrically distributed with the encoding module, which is used to reduce the spatial information loss caused by the downsampling process, and can directly transfer the low-level features from the convolution layer (i.e., the downsampling layer) to the deconvolution layer (i.e., the upsampling layer), which is conducive to recovering signal details. Figure 3 Skipconnection in.

[0126] The residual module can be used to connect the input of the generator and the output of the Nth decoding module. The residual module can be used to alleviate the degradation phenomenon of deep convolutional networks and improve the generalization ability of the network.

[0127] The generator can also include an activation function layer for activating the output of the residual module after residual connection using an activation function as the output of the generator. The generator can use a leaky rectified linear unit (Leaky ReLU) activation function, for example Figure 3 Leaky ReLU layer in .

[0128] In some embodiments, the discriminator includes multiple convolution modules and discrimination modules, which are used to perform convolution processing on the data input into the convolution module. The convolution module is used to extract features from the input data. The discrimination module is used to judge whether the noise-free seismic data matches the label data based on the feature extraction results of multiple convolution modules, and use the judgment result as the output of the discriminator.

[0129] Figure 4 Shown is a schematic diagram of an identifier provided in an embodiment of the present application.

[0130] like Figure 4 As shown, the discriminator includes multiple sequentially connected convolution modules, convolution layers and discrimination modules. Each convolution module may include Figure 4 A Conv layer and a Leaky ReLU layer in . The convolution layer can be Figure 4 During the training of the initial neural network model, the data generated by the generator is fed into the discriminator, which calculates the Wasserstein distance between the generator-generated data and the labeled data. Using this distance as a reference, the model parameters are continuously optimized to reduce the Wasserstein distance. A smaller Wasserstein distance indicates a better training of the initial neural network model, and thus better ghost noise suppression. If the input of the discriminator approximates real data, the correct output is 1. If the input differs significantly from the real data and is judged to be data generated by the generator, the correct output of the discriminator is 0. All convolutional modules in the discriminator use the Leaky ReLU activation function and batch normalization (BN).

[0131] The following will further introduce the seismic data processing method provided in the embodiment of the present application with reference to the accompanying drawings.

[0132] Figure 5 FIG. 1 is a schematic diagram of a method for processing seismic data provided by an embodiment of the present application. Figure 5 As shown, the method includes:

[0133] S501: Acquire seismic data of a common shot gather in a target area, wherein the seismic data includes ghost wave data at the source end and ghost wave data at the detector end.

[0134] S502: Correcting the seismic data to obtain corrected seismic data.

[0135] It can be understood that the correction processing performed on the seismic data at the detector end may be depth difference correction processing.

[0136] In some embodiments, the seismic data can be time-corrected, and the detector-end seismic data in the seismic data can be corrected to the source end, and then the detector-end ghost wave data can be suppressed by the ghost wave suppression model trained in the previous text.

[0137] S503: After normalizing the corrected seismic data, the data are input into pre-trained ghost wave suppression models to output the first seismic data after ghost wave suppression at the detector end and the second seismic data after ghost wave suppression at the source end.

[0138] It can be understood that the ghost wave suppression model can be used Figure 1 The model obtained by training the model training method in . The specific training process and related descriptions can be understood by referring to the content of the previous part, which will not be repeated here.

[0139] It can be understood that the normalization processing of the source end seismic data and the corrected detector end seismic data can be performed using the above formulas (1) to (4), which will not be described in detail here.

[0140] In some embodiments, after normalizing the corrected seismic data, the data is input into a pre-trained ghost suppression model, and the output of the first seismic data after the ghost waves are suppressed at the detector end and the second seismic data after the ghost waves are suppressed at the source end include:

[0141] After the corrected seismic data is normalized, it is input into a pre-trained ghost wave suppression model, and the first seismic data after the ghost wave is suppressed at the detector end is output;

[0142] The first seismic data is input into a pre-trained ghost wave suppression model, and the second seismic data after the ghost waves at the source end are output.

[0143] In some embodiments, the corrected geophone end seismic data can be normalized to obtain the normalized ghost wave shot gather data S in the target area. Max1-Norm (P+GS+GD). The normalized data is input into the ghost wave suppression model to perform the detector-side ghost wave suppression operation, and the first seismic data S after the detector-side ghost wave data is suppressed can be obtained. Max-Norm(P+GS). Then the first seismic data S after suppressing the ghost wave data at the detector end is Max-Norm (P+GS) inputs the ghost wave suppression model again, and performs the source end ghost wave suppression operation to obtain the second earthquake data S after the source end ghost wave data is suppressed. Max-Norm (P).

[0144] S504: performing inverse normalization processing and inverse correction processing on the second seismic data to obtain target seismic data after ghost wave suppression in the target area.

[0145] It can be understood that since the seismic data is corrected to the source end in step S502 and the corrected seismic data is normalized in step S503, it is necessary to perform inverse normalization on the second seismic data to return the data to the initial order of magnitude, and to perform inverse correction on the inverse normalized data, that is, to inversely correct the seismic data corrected to the source end back to the original depth of the geophone end, and obtain the target seismic data S after the seismic data of the target area are subjected to ghost wave suppression at the source end and the geophone end. Max (P).

[0146] In some embodiments, the denormalization process of the second seismic data can be expressed by the following formula:

[0147] S′ Max (P)=S Max-Norm (P)*[MaxS(S)-MinS(S)]+MinS(S) Formula (6)

[0148] Among them, S′ Max (P) can represent the second seismic data after denormalization, S Max-Norm (P) may represent the second seismic data, MaxS(S) may represent the maximum value of the seismic data of the common shot point gather of the target area, and MinS(S) may represent the minimum value of the seismic data of the common shot point gather of the target area.

[0149] The beneficial effects of the embodiments of the present application are described below with reference to the accompanying drawings.

[0150] Figure 6 Shown is a schematic diagram of a forward layered model provided in an embodiment of the present application.

[0151] Figure 7 Shown are single-shot seismic data before and after ghost wave suppression of the forward layered model provided in an embodiment of the present application.

[0152] like Figure 6 As shown, the velocity of the forward layered model provided in the embodiment of the present application remains unchanged in the lateral direction, and the velocity increases with the increase of depth. Figure 6 The horizontal axis represents the horizontal distance, and the vertical axis (i.e.Figure 6 Depth(m)) represents the depth, and different colors correspond to different speeds (i.e. Figure 6 Velocity (m / s) in.

[0153] Figure 7 The use of Figure 6 The forward layered model in the paper is generated using a finite difference scheme for two-dimensional single-shot seismic data with ghost waves. The observation system parameters are: 5m interval between geophones on the horizontal line, 3s sampling time, 1000 sampling points, and 15m focal depth and geophone depth. Figure 7 The data shown in a may be the single shot seismic data among the seismic data obtained in step S501 above.

[0154] The frequency-space domain ghost suppression method is used to Figure 7 After ghost wave suppression processing of the earthquake record shown in a, we can get Figure 7 The seismic data after ghost wave suppression shown in b is obtained by using the frequency-wavenumber domain ghost wave suppression method. Figure 7 After ghost wave suppression processing of the earthquake record shown in a, we can get Figure 7 The seismic data after ghost wave suppression shown in c is obtained by using the Ladong domain ghost wave attenuation method. Figure 7 After ghost wave suppression processing of the earthquake record shown in a, we can get Figure 7 The seismic data after ghost wave suppression shown in d; the seismic data processing method provided by the embodiment of the present application (i.e., the above Figure 5 Corresponding method) Figure 7 After ghost wave suppression processing of the earthquake record shown in a, we can get Figure 7 The target seismic data after ghost wave suppression is shown in e. Figure 7 In the schematic diagrams of seismic data shown in ae of FIG, the horizontal axis represents the horizontal position (i.e. Figure 7 Lateral location [m]), the vertical axis represents time (i.e. Figure 7 Time[s] in .

[0155] Depend on Figure 7 It can be seen from the seismic data shown in ae that the above four methods all suppress the false phase axis after the primary wave in the seismic data well, and the effective reflection wave is clearer and more obvious. However, it can also be clearly seen that compared with the existing ghost wave suppression algorithm, the seismic data processing method of the embodiment of the present application can suppress the ghost wave data in the seismic data at a higher level, and the signal-to-noise ratio and resolution of the overall seismic data are greatly improved.

[0156] In order to further illustrate the application effect of the embodiment of the present application, the processing of sparsely sampled seismic data of shot points in a certain area will be taken as an example for explanation.

[0157] Figure 8 Shown is a schematic diagram of single-shot seismic data before and after processing provided by an embodiment of the present application.

[0158] Figure 9 Shown is a schematic diagram of the corresponding frequency spectra before and after processing of single-shot seismic data provided by an embodiment of the present application.

[0159] In this embodiment, the actual marine seismic data is collected by towline, the airgun is set at a depth of 5 meters, the cable is at a depth of 6 meters, the water depth is 500-1000 meters, the track spacing is 12.5 meters, and the shot spacing is 100 meters. Figure 8 The schematic diagram of the earthquake record shown in a. The traditional detector end ghost wave suppression method is used to Figure 8 The ghost wave suppression of the earthquake record shown in a can be obtained Figure 8 Schematic diagram of the earthquake record shown in b; the traditional source end ghost wave suppression method is used to Figure 8 The ghost wave suppression of the earthquake record shown in a can be obtained Figure 8 Schematic diagram of the earthquake record shown in c; the earthquake data processing method provided by the embodiment of the present application is used to Figure 8 The ghost wave suppression of the earthquake record shown in a can be obtained Figure 8 Schematic diagram of the earthquake record shown in d. Figure 8 In the schematic diagrams of seismic data shown in ad, the horizontal axis represents the horizontal distance (i.e. Figure 8 The vertical axis represents the time (i.e. Figure 8 time / s).

[0160] Depend on Figure 8 From b to c in the figure, we can see that the ghost waves at the receiver end and the ghost waves at the source end ( Figure 8 The reflected wave phase is effectively suppressed, and the three phases of "white, black and white" are changed to one "white phase". However, due to the sparse shot sampling, there are still residual ghost waves ( Figure 8 After the ghost wave is suppressed by the seismic data processing method provided in the embodiment of the present application, the residual source end ghost wave is effectively suppressed ( Figure 9 (indicated by the arrow in d).

[0161] Figure 8 The curves of line ①, ②, ③, and ④ are Figure 9The spectrum corresponding to the single shot seismic record shown in ad, where the horizontal axis represents the frequency (i.e. Figure 9 Frequency / Hz), the ordinate represents the amplitude (i.e. Figure 9 Amplitude / dB in ). From Figures 6 to 9 As can be seen in the figure, the frequency band is gradually widened from the previous 10-80 Hz to 5-120 Hz, and the data octave is increased from 3.0 before attenuation to 4.5 after ghost wave suppression. It can be seen that by using the seismic data processing method provided in the embodiment of the present application to suppress ghost waves, the spectrum of the seismic data is significantly expanded in both the low-frequency band and the high-frequency band, and the reflection characteristics are clearer.

[0162] From the above Figure 10 As shown, the data processing method provided in the embodiment of the present application can realize the extraction of ghost wave features in seismic data through the initial neural network model, and the ghost wave suppression model can be trained through training data. The ghost wave suppression model can not only suppress the ghost wave data at the detector end, but also suppress the ghost wave data at the source end in the seismic data, and the denoising effect is better than the traditional ghost wave suppression method, which has a high practical value for promoting the improvement of the quality of marine seismic data processing; in addition, the seismic data processing method provided in the embodiment of the present application can suppress ghost waves without being restricted by the observation system, and can adapt to ghost wave suppression under two-dimensional and three-dimensional observation systems. At the same time, manual intervention is reduced in the attenuation process, and the degree of intelligence is increased. Compared with the currently existing ghost wave attenuation methods, it has higher computational efficiency.

[0163] An embodiment of the present application also provides a device for processing seismic data. Figure 10 FIG. 1 is a schematic diagram of a seismic data processing device provided in an embodiment of the present application. Figure 11 As shown, the seismic data processing device 1000 may include:

[0164] The data acquisition module 1001 is used to acquire seismic data of a common shot gather in a target area, wherein the seismic data includes ghost wave data at the source end and ghost wave data at the detector end.

[0165] The data preprocessing module 1002 is used to perform correction processing on the seismic data to obtain corrected seismic data.

[0166] The ghost wave suppression module 1003 is used to normalize the corrected seismic data, input the pre-trained ghost wave suppression model, and output the first seismic data after the ghost waves are suppressed at the detector end and the second seismic data after the ghost waves are suppressed at the source end.

[0167] The target data determination module 1004 is configured to perform inverse normalization and inverse correction processing on the second seismic data to obtain target seismic data of the target area after ghost waves are suppressed.

[0168] The description and functions of the above modules can be understood by referring to the content of the seismic data processing method section, which will not be repeated here.

[0169] This specification also provides a computer storage medium, wherein the computer storage medium stores computer program instructions, and when the computer program instructions are executed, the steps of the above-mentioned seismic data processing method are implemented.

[0170] This specification also provides a computer program product, comprising a computer program, which implements the steps of the above-mentioned seismic data processing method when executed by a processor.

[0171] The embodiment of the present invention further provides an electronic device, 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 5 The bus connection is taken as an example.

[0172] The processor 1101 may be a central processing unit (CPU). The processor 1101 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0173] 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 seismic data processing method in the embodiments of the present invention. Processor 1101 executes the non-transitory software programs, instructions, and modules stored in memory 1102 to perform various processor functions and data processing, thereby implementing the seismic data processing method in the above-mentioned method embodiment.

[0174] The memory 1102 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created by the processor 1101, etc. In addition, the memory 1102 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 1102 may optionally include a memory remotely located relative to the processor 1101, and these remote memories may be connected to the processor 1101 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0175] The one or more modules are stored in the memory 1102 and when executed by the processor 1101, perform the following steps: ​ The method for processing seismic data in the illustrated embodiment.

[0176] The specific details of the above electronic device can be understood by referring to the corresponding descriptions and effects in the above method embodiments, and will not be repeated here.

[0177] In some embodiments, the electronic device may be a terminal such as a PC (Personal Computer), a tablet computer, a smartphone, a wearable device, or an intelligent robot; or a server. The server may be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services. This application does not impose any restrictions on this.

[0178] This specification also provides a computer storage medium, wherein the computer storage medium stores computer program instructions, and when the computer program instructions are executed, the steps of the above-mentioned seismic data processing method are implemented.

[0179] This specification also provides a computer program product, comprising a computer program, which implements the steps of the above-mentioned seismic data processing method when executed by a processor.

[0180] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium, and when the program is executed, the processes of the above-mentioned embodiment methods can be included. The storage medium can be a magnetic disc, an optical disc, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc. The storage medium can also include a combination of the above-mentioned types of memories.

[0181] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the difference from other embodiments.

[0182] The system, device, module or unit described in the above embodiments can be specifically implemented by a computer chip or entity, or by a product with certain functions.

[0183] For the convenience of description, the above device is described as various units respectively described by functions. Of course, the functions of each unit can be implemented in the same or more software and / or hardware in the implementation of the present application.

[0184] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software and the necessary general hardware platform. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a ROM / RAM, a magnetic disc, an optical disc, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the method of some parts of the embodiments of the present application.

[0185] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, small computers, large computers, distributed computing environments including any of the above systems or devices, etc.

[0186] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0187] Although the present application has been described through embodiments, those skilled in the art will appreciate that there are many modifications and variations to the present application without departing from the spirit of the present application. It is intended that the appended claims include these modifications and variations without departing from the spirit of the present application.

Claims

1. A method for processing seismic data, characterized in that: include: Acquire seismic data of a common shot gather in a target area, wherein the seismic data includes ghost wave data at the source end and ghost wave data at the receiver end; performing correction processing on the seismic data to obtain corrected seismic data; After the corrected seismic data is normalized, it is input into a pre-trained ghost wave suppression model, and the first seismic data after the ghost wave is suppressed at the detector end is output; the first seismic data is input into the pre-trained ghost wave suppression model, and the second seismic data after the ghost wave is suppressed at the source end is output; Performing an inverse normalization process and an inverse correction process on the second seismic data to obtain target seismic data of the target area after ghost wave suppression; The ghost wave suppression model is trained in the following way: Acquiring historical seismic data of a common shot gather, wherein the historical seismic data includes historical ghost wave data at the source end and historical ghost wave data at the geophone end; Correcting the historical seismic data to obtain corrected seismic data; A ghost wave suppression method based on sparse inversion is used to suppress the historical ghost wave data at the geophone end in the corrected seismic data, and the corrected seismic data after ghost wave suppression is obtained. The calibrated seismic data after ghost wave suppression is sparsely sampled as label data to simulate the common detection point gather data after ghost wave suppression at the source end. Normalizing the corrected seismic data and the labeled data, and using the normalized corrected seismic data and the labeled data as training data; The training data is input into the initial neural network model for training to obtain the ghost wave suppression model.

2. The method according to claim 1, characterized in that Correcting the historical seismic data to obtain corrected seismic data includes: Depth difference correction is performed on the historical seismic data, and single-shot seismic data that meets preset conditions is selected from the historical seismic data after depth difference correction as corrected seismic data according to a preset ratio.

3. The method according to claim 1, characterized in that Inputting the training data into the initial neural network model for training to obtain the ghost wave suppression model, including: Inputting the normalized corrected seismic data into the initial neural network model to obtain the actual output of the initial neural network model, and calculating the distance between the actual output and the label data; The parameters of the initial neural network model are adjusted based on the distance and the actual output until the distance meets a preset distance threshold, and the training is stopped to obtain the ghost wave suppression model.

4. The method according to claim 3, characterized in that The initial neural network model includes a generator and a discriminator; Accordingly, the normalized corrected seismic data is input into the initial neural network model to obtain the actual output of the initial neural network model, and the distance between the actual output and the label data is calculated, including: Inputting the normalized corrected seismic data into the generator, performing feature extraction on the normalized corrected seismic data using the generator, and generating noise-free seismic data based on the feature extraction result; Inputting the noise-free seismic data and the label data into the discriminator, determining whether the noise-free seismic data matches the label data, and feeding back the determination result to the generator; Calculating the distance between the noise-free seismic data and the label data, and feeding the distance back to the generator; Accordingly, adjusting the parameters of the initial neural network model based on the distance and the actual output includes: Based on the judgment result and the distance, parameters of the generator and the discriminator are adjusted.

5. The method according to claim 4, characterized in that The generator includes N encoding modules, N decoding modules, N skip connection layers and residual modules, and the encoding modules correspond one to one to the decoding modules; The skip connection layer is used to connect the output of the encoding module and the input of the decoding module symmetrically distributed with the encoding module; The residual module is used to connect the input of the generator and the output of the Nth decoding module.

6. The method according to claim 4, characterized in that The discriminator includes multiple convolution modules and a discrimination module, which is used to perform convolution processing on the data input to the convolution module. The convolution module is used to extract features from the input data. The discrimination module is used to judge whether the noise-free seismic data matches the label data based on the feature extraction results of multiple convolution modules, and use the judgment result as the output of the discriminator.

7. A seismic data processing device, characterized in that: include: A data acquisition module is used to acquire seismic data of a common shot gather in a target area, wherein the seismic data includes ghost wave data at the source end and ghost wave data at the detector end; A data preprocessing module, configured to perform correction processing on the seismic data to obtain corrected seismic data; The ghost wave suppression module is used to normalize the corrected seismic data, input the pre-trained ghost wave suppression model, and output the first seismic data after the ghost waves are suppressed at the detector end; input the first seismic data into the pre-trained ghost wave suppression model, and output the second seismic data after the ghost waves are suppressed at the source end; a target data determination module, configured to perform inverse normalization processing and inverse correction processing on the second seismic data to obtain target seismic data of the target area after ghost wave suppression; The ghost wave suppression model is trained in the following way: Acquiring historical seismic data of a common shot gather, wherein the historical seismic data includes historical ghost wave data at the source end and historical ghost wave data at the geophone end; Correcting the historical seismic data to obtain corrected seismic data; A ghost wave suppression method based on sparse inversion is used to suppress the historical ghost wave data at the geophone end in the corrected seismic data, and the corrected seismic data after ghost wave suppression is obtained. The calibrated seismic data after ghost wave suppression is sparsely sampled as label data to simulate the common detection point gather data after ghost wave suppression at the source end. Normalizing the corrected seismic data and the labeled data, and using the normalized corrected seismic data and the labeled data as training data; The training data is input into the initial neural network model for training to obtain the ghost wave suppression model.

8. An electronic device, characterized in that: include: A memory and a processor, wherein the processor and the memory are communicatively connected to each other, the memory stores computer instructions, and the processor implements the steps of the method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer storage medium, characterized in that The computer storage medium stores computer program instructions, and when the computer program instructions are executed, the steps of the method according to any one of claims 1 to 6 are implemented.

Images