Imaging domain ghost wave suppression method based on deep learning

Abstract

The invention relates to the technical field of marine seismic data ghost wave suppression processing, in particular to an imaging domain ghost wave suppression method based on deep learning. Comprising the following steps: S1, constructing an earthquake training sample set, artificially synthesizing a fault velocity model, and synthesizing a ghost wave training sample set by using a reverse time migration RTM algorithm based on the fault velocity model; s2, constructing a U-net neural network; s3, training a U-net neural network by using the synthesized ghost wave training sample set; and S4, inputting the seismic data of the target work area into the U-Net neural network for ghost wave intelligent suppression processing to obtain seismic data after ghost wave suppression. According to the invention, ghost waves can be more effectively and stably removed or suppressed, and interference of the ghost waves on effective reflection signals is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of ghost wave suppression processing of offshore seismic data, and more specifically, to an imaging domain ghost wave suppression method based on deep learning. Background Art

[0002] In the process of broadband marine streamer seismic data acquisition, in order to avoid the influence of clutter noise caused by complex sea surface, the source and the detector are usually placed at a certain depth below the sea surface. Due to the huge difference in wave impedance between seawater and air, a strong reflection interface (reflection coefficient is approximately equal to -1) is formed between the two, forming a free surface. This makes the detector located below the sea surface receive free surface reflection waves from the source and the detector in addition to the reflection waves from the underground interface. Since the sinking depth of the source and the detector in the same work area is fixed, this free surface reflection wave usually follows the underground interface reflection wave on the seismic shot collection record or processed seismic image, so it is vividly called "ghost wave". Ghost waves and effective reflection waves will interfere with each other, resulting in a longer duration of effective reflection waves, greatly reducing the vertical resolution of seismic images. In addition, ghost waves have a notch effect, which will cause zero value points to appear periodically in the spectrum of the seismic signal, resulting in the loss of certain specific frequency components (especially high-frequency and low-frequency information), thereby narrowing the effective frequency band of the seismic, which brings great difficulties to subsequent seismic inversion and data interpretation. Therefore, removing or suppressing ghost waves is an important step in the processing of marine seismic data.

[0003] The cost of seismic data acquisition is relatively high. In the actual exploration process, the cost needs to be controlled. It is not possible to repeat trial and error or repeat acquisition. It is necessary to develop corresponding processing technology to maximize the accuracy of existing seismic data. Conventional ghost wave suppression technology has problems of numerical instability and low computational efficiency during the calculation process due to the trap effect of ghost waves and complex theoretical derivation. In addition, the actual exploration environment is complex and diverse, and traditional methods have certain limitations. In order to overcome these problems, it is necessary to continuously research and develop new methods and technologies to better meet the challenges in marine exploration. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art in effectively and stably suppressing ghost waves, and to provide an imaging domain ghost wave suppression method based on deep learning, which can more effectively and stably remove or suppress ghost waves and reduce their interference with effective reflection signals.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A method for suppressing ghost waves in an imaging domain based on deep learning is provided, comprising the following steps:

[0007] S1. Construct an earthquake training sample set, artificially synthesize a fault velocity model, and synthesize a ghost wave training sample set based on the fault velocity model using the reverse time migration (RTM) algorithm;

[0008] S2. Build a U-net neural network;

[0009] S3. Use the above synthesized ghost wave training sample set to train the U-net neural network;

[0010] S4. Input the seismic data of the target work area into the U-Net neural network for intelligent ghost wave suppression processing to obtain the seismic data after ghost wave suppression.

[0011] According to the above technical means, the present invention provides a method for suppressing ghost waves in the imaging domain based on deep learning. The method uses a deep neural network to find and extract the nonlinear mapping relationship between seismic images containing ghost waves and those without ghost waves, thereby removing or suppressing ghost waves in the imaging domain. This method can more effectively and stably remove or suppress ghost waves and reduce their interference with effective reflection signals.

[0012] Preferably, the step S1 comprises:

[0013] S11. Generate horizontal layered velocity model V 1 (X, Z): Randomly generate a one-dimensional velocity sequence V(Z), where the number, thickness and velocity of the strata are randomly selected from a reasonable range determined by the actual geological conditions of the work area, and expand the one-dimensional velocity sequence V(Z) horizontally into a two-dimensional horizontal layered velocity model V 1 (X,Z);

[0014] S12. Generate inclined fold model In the horizontal layered velocity model V 1 (X, Z), the vertical displacement field M(X, Z) is used to add the formation dip and fold structure to obtain the inclined fold model

[0015]

[0016] S13. Generate a true fault velocity model Using fault displacement fields in the inclined fold model Fault structures are added to obtain a true fault velocity model that includes formation dip, folds, and fault structures.

[0017] S14. Generate RTM images: Use the RTM algorithm to transform the real fault velocity model Convert to seismic images, and obtain seismic images with and without ghost waves;

[0018] S15. Generate an earthquake training sample set.

[0019] Preferably, the step S12 comprises: the vertical displacement field M(X,Z) is obtained by M 1 (X,Z) and M 2 (X,Z) consists of two parts:

[0020] M(X,Z)=M 1 (X,Z)+M 2 (X,Z)(1)

[0021] In formula (1), M 1 (X,Z) is a straight line equation used to simulate the formation dip, and its expression is:

[0022] M 1 (X,Z)=tanα·(XX c )(2)

[0023] Where α represents the formation inclination; V 1 (X c ,Z) represents the middle track of the velocity model;

[0024] M 2 (X,Z) is composed of N one-dimensional Gaussian functions, which are used to simulate the underground fold structure. Its expression is as follows:

[0025]

[0026] Among them, Z max Indicates the maximum depth of the velocity model; b k 、c k and σ k They represent the amplitude, center position and half-length of the kth Gaussian function, and their sizes are all randomly generated. By superimposing and summing N different Gaussian functions, multiple different fold structures are added to the velocity model.

[0027] Using the vertical displacement field M(X,Z) above, we can generate a horizontal layered velocity model V 1 Add the formation dip angle and fold structure to (X,Z) to generate the inclined fold model V 2 (X,Z):

[0028]

[0029] Preferably, the step S13 comprises:

[0030] S131. Determine the fault reference line: Use the center point position of the reference line and fault dip angle θ define a fault reference line; wherein the center point position P and the fault dip angle θ are randomly generated within the set range;

[0031] S132. Coordinate system transformation: using the randomly determined reference line center point position and the fault dip angle θ will transform the global coordinate system Convert to the local coordinate system (x, z):

[0032]

[0033] Where R is the rotation matrix:

[0034]

[0035] After the coordinate system conversion, point P is the origin of the local coordinate system; the positive direction of the x-axis points to the dip; the z-axis is the normal of the reference line, and its positive direction points to the hanging wall of the fault; the fault reference line is the z-axis of the local coordinate system;

[0036] S133. Generate fault curve: randomly generate a series of disturbance points near the fault reference line z=0, and then use these disturbance points to interpolate and construct the fault curve z f =f(x);

[0037] S134. Calculate the fault displacement field D(x,z): First, calculate the dip displacement field Dx(x,z) in two-dimensional space, and then calculate the normal displacement field Dz(x,z); the fault displacement field D(x,z) is a two-dimensional vector field defined in the local coordinate system (x,z), which consists of two parts: the dip displacement field Dx(x,z) and the normal displacement field Dz(x,z):

[0038] D(x,z)=[D x (x,z)u x ,D z (x,z)u z ](7)

[0039] Among them, u x and u z Represent the unit vectors in the x and z directions respectively;

[0040] S135. Use fault displacement field to simulate fault structure: According to the calculated fault displacement field D(x,z), the inclined fold model Add fault structure to get fault model

[0041]

[0042] Where T represents the transposition symbol.

[0043] Preferably, the step S134 includes:

[0044] First, calculate the projection displacement field d(x; z = 0) of the inclination displacement field Dx(x, z) on the reference line:

[0045]

[0046] Among them, d max represents the maximum displacement of the center point, which is the origin of the local coordinate system; r(x) is the normalized radius:

[0047]

[0048] Among them, l x represents the half length of the fault line;

[0049] Then, the fault displacement of each point in the two-dimensional space is calculated using the projected displacement field d(x; z=0) on the fault reference line; a boundary γ is defined so that Dx(x,z) is 0 when it reaches or exceeds the boundary γ;

[0050] For the upper plate f(x)≤z≤f(x)+γ, the calculation expression of Dx(x,z) is:

[0051] D x (x,z)=λ·d(x;z=0)·ζ(x,z)(11)

[0052] Among them, λ represents the percentage of the upper plate displacement in the projected displacement field d(x; z=0), which is randomly selected between [0,1]; ζ(x,z) is a nonlinear attenuation function, which attenuates the dip displacement field Dx(x,z) away from the fault curve along the normal direction, and decays to 0 when it reaches the traction radius γ; the nonlinear attenuation function ζ(x,z) is expressed as:

[0053]

[0054] For the lower plate f(x)-γ≤z≤f(x), the calculation expression of Dx(x,z) is:

[0055] D x (x,z)=(λ-1)·d(x;z=0)·ζ(x,z)(13)

[0056] After calculating the inclination displacement field Dx(x,z) in two-dimensional space, the normal displacement field Dz(x,z) can be directly obtained using the following formula:

[0057] D z (x,z)=f(x+D x (x,z))-f(x)(14).

[0058] Preferably, steps S131 to S135 can be used to obtain the oblique fold model V 2(X, Z) to simulate a fault; repeat steps S131 to S135 in the same velocity model to generate a fault velocity model containing multiple faults

[0059]

[0060] Preferably, the step S14 comprises: first, defining an offshore seismic data observation system; randomly selecting one of the conventional horizontal towed cable type, up-and-down cable type, variable depth cable type and deep towed type as a seismic data acquisition method, wherein the sinking depths of the seismic source and the geophone are randomly selected within a set range; then, using boundary conditions with and without free surfaces, respectively, RTM imaging is performed to obtain seismic images with and without ghost waves.

[0061] Preferably, step S15 comprises: randomly selecting images with a size of 128×128 pixels at the same position from the calculated RTM images with and without ghost waves, and using the images containing ghost waves as earthquake training samples, and the images without ghost waves as label images; repeatedly randomly selecting multiple 128×128 training samples and labels from the two RTM images to obtain a small earthquake training sample set.

[0062] Preferably, in step S2, the process of processing data by the U-net neural network includes: after the seismic sample with ghost waves is input into the U-net, it first undergoes a series of downsampling steps in the encoder path to help the U-net capture hierarchical features; wherein the seismic image with an input size of 128×128 is gradually reduced to 64×64, 32×32, 16×16, 8×8 and 4×4, and the number of channels is gradually increased to 128, 256, 512, 1024, 2048 and 4096; then, a series of upsampling processes are performed in the decoder path to restore the spatial size; at the same time, the jump connection combines the high-level semantic features in the decoder with the low-level details in the encoder, so that the U-net neural network can effectively aggregate rough global context information and detailed local features; finally, a 1×1 convolutional layer outputs a suppressed ghost wave image of the same size as the input;

[0063] Then use the mean square error, i.e. L 2 Norm, as a loss function:

[0064]

[0065] Where n represents the total number of pixels in the input seismic image; p i represents the suppressed ghost wave image output by the U-net neural network; l i Represents a ghost-free labeled image.

[0066] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-described method when executing the computer program.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] The present invention provides an imaging domain ghost wave suppression method based on deep learning, which uses a deep learning model to learn complex features and patterns from a large amount of training data, and therefore may have more advantages in dealing with complex situations. In addition, the present invention is highly flexible and can establish training sample sets containing different ghost wave features for different work areas and exploration environments to adapt to the actual complex and changeable geological conditions. In addition, the present invention can automatically suppress ghost waves without the need for ghost wave delay time estimation, reducing the need for manual intervention and improving efficiency. In particular, the present invention has the potential to be directly applied to three-dimensional situations, while the prior art is limited by the sparse Crossline acquisition system. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a flow chart of the imaging domain ghost wave suppression method based on deep learning of the present invention;

[0070] Figure 2 It is a schematic diagram of the process of constructing a seismic training sample set in the imaging domain ghost wave suppression method based on deep learning of the present invention. DETAILED DESCRIPTION

[0071] The present invention will be further described below in conjunction with specific implementation modes.

[0072] Embodiment 1

[0073] This embodiment is the first embodiment of the imaging domain ghost suppression method based on deep learning, and includes the following steps:

[0074] S1. Construct an earthquake training sample set, artificially synthesize a fault velocity model, and synthesize a ghost wave training sample set based on the fault velocity model using the reverse time migration (RTM) algorithm;

[0075] S2. Build a U-net neural network;

[0076] S3. Use the above synthesized ghost wave training sample set to train the U-net neural network;

[0077] S4. Input the seismic data of the target work area into the U-Net neural network for intelligent ghost wave suppression processing to obtain the seismic data after ghost wave suppression.

[0078] Wherein, the step S1 comprises:

[0079] S11. Generate horizontal layered velocity model V 1 (X, Z): Randomly generate a one-dimensional velocity sequence V(Z), where the number, thickness and velocity of the strata are randomly selected from a reasonable range determined by the actual geological conditions of the work area, and expand the one-dimensional velocity sequence V(Z) horizontally into a two-dimensional horizontal layered velocity model V 1 (X,Z);

[0080] S12. Generate inclined fold model V 2 (X,Z): In the horizontal layered velocity model V 1 (X, Z), the vertical displacement field M(X, Z) is used to add the formation dip and fold structure to obtain the inclined fold model

[0081]

[0082] In step S12, the vertical displacement field M(X,Z) is obtained by 1 (X,Z) and M 2 (X,Z) consists of two parts:

[0083] M(X,Z)=M 1 (X,Z)+M 2 (X,Z)(1)

[0084] In formula (1), M 1 (X,Z) is a straight line equation used to simulate the formation dip, and its expression is:

[0085] M 1 (X,Z)=tanα·(XX c )(2)

[0086] Where α represents the formation inclination; V 1 (X c ,Z) represents the middle track of the velocity model;

[0087] M 2 (X,Z) is composed of N one-dimensional Gaussian functions, which are used to simulate the underground fold structure. Its expression is as follows:

[0088]

[0089] Among them, Z max Indicates the maximum depth of the velocity model; b k 、c k and σ k They represent the amplitude, center position and half-length of the kth Gaussian function, and their sizes are all randomly generated. By superimposing and summing N different Gaussian functions, multiple different fold structures are added to the velocity model.

[0090] Using the vertical displacement field M(X,Z) above, we can generate a horizontal layered velocity model V 1 Add formation dip angle and fold structure to (X,Z) to generate inclined fold model

[0091]

[0092] S13. Generate a true fault velocity model Using fault displacement fields in the inclined fold model Fault structures are added to obtain a true fault velocity model that includes formation dip, folds, and fault structures. The step S13 comprises:

[0093] S131. Determine the fault reference line: Use the center point position of the reference line and fault dip angle θ define a fault reference line; wherein the center point position P and the fault dip angle θ are randomly generated within the set range;

[0094] S132. Coordinate system transformation: using the randomly determined reference line center point position and the fault dip angle θ will transform the global coordinate system Convert to the local coordinate system (x, z):

[0095]

[0096] Where R is the rotation matrix:

[0097]

[0098] After the coordinate system conversion, point P is the origin of the local coordinate system; the positive direction of the x-axis points to the dip; the z-axis is the normal of the reference line, and its positive direction points to the hanging wall of the fault; the fault reference line is the z-axis of the local coordinate system;

[0099] S133. Generate fault curve: randomly generate a series of disturbance points near the fault reference line z=0, and then use these disturbance points to interpolate and construct the fault curve z f =f(x);

[0100] S134. Calculate the fault displacement field D(x,z): The fault displacement field D(x,z) is a two-dimensional vector field defined in the local coordinate system (x,z). It consists of two parts: the dip displacement field Dx(x,z) and the normal displacement field Dz(x,z):

[0101] D(x,z)=[D x (x,z)u x ,D z (x,z)u z ](7)

[0102] Among them, u x and u z Represent the unit vectors in the x and z directions respectively;

[0103] Since this embodiment only considers the seamless sliding of the two fault blocks of the upper and lower plates along the fault curve, the normal displacement field Dz(x,z) can be calculated based on the dip displacement field Dx(x,z) and the fault curve z f =f(x) is calculated. Therefore, in order to calculate the two-dimensional fault displacement field D(x,z), the projection displacement field d(x;z=0) of the dip displacement field Dx(x,z) on the reference line is first calculated:

[0104]

[0105] Among them, d max represents the maximum displacement of the center point, which is the origin of the local coordinate system; r(x) is the normalized radius:

[0106]

[0107] Among them, l x represents the half length of the fault line;

[0108] Then, the fault displacement at each point in the two-dimensional space is calculated using the projected displacement field d(x; z=0) on the fault reference line; since the actual fault displacement is the largest at the fault core (fault curve) and the displacement at a certain distance away from the core (traction radius) is 0, a boundary γ is defined so that Dx(x,z) is 0 when it reaches or exceeds the boundary γ;

[0109] For the upper plate f(x)≤z≤f(x)+γ, the calculation expression of Dx(x,z) is:

[0110] D x (x,z)=λ·d(x;z=0)·ζ(x,z)(11)

[0111] Among them, λ represents the percentage of the upper plate displacement in the projected displacement field d(x; z=0), which is randomly selected between [0,1]; ζ(x,z) is a nonlinear attenuation function, which attenuates the dip displacement field Dx(x,z) away from the fault curve along the normal direction, and decays to 0 when it reaches the traction radius γ; the nonlinear attenuation function ζ(x,z) is expressed as:

[0112]

[0113] For the lower plate f(x)-γ≤z≤f(x), the calculation expression of Dx(x,z) is:

[0114] Dx (x,z)=(λ-1)·d(x;z=0)·ζ(x,z)(13)

[0115] After calculating the inclination displacement field Dx(x,z) in two-dimensional space, the normal displacement field Dz(x,z) can be directly obtained using the following formula:

[0116] D z (x,z)=f(x+D x (x,z))-f(x)(14);

[0117] S135. Use fault displacement field to simulate fault structure: According to the calculated fault displacement field D(x,z), the inclined fold model Add fault structure to get fault model

[0118]

[0119] Where T represents the transposition symbol.

[0120] In this embodiment, steps S131 to S135 can be used to generate a fold model. Repeat steps S131 to S135 in the same velocity model to generate a fault velocity model containing multiple faults.

[0121] S14. Generate RTM images: Use the RTM algorithm to transform the real fault velocity model The seismic images with and without ghost waves are converted into seismic images. First, an offshore seismic data observation system is defined. One of the conventional horizontal towed cable, up-and-down cable, variable-depth cable and deep towed cable is randomly selected as the seismic data acquisition method, and the depth of the source and the detector is randomly selected within the set range. Then, the boundary conditions with and without free surface are used to analyze the V 3 (X,Z) RTM imaging is performed to obtain seismic images with and without ghost waves.

[0122] S15. Generate an earthquake training sample set. Randomly select images of 128×128 pixels at the same position in the calculated RTM images with and without ghost waves, and use the images with ghost waves as earthquake training samples, and the images without ghost waves as label images; repeatedly randomly select multiple 128×128 training samples and labels in the two RTM images to obtain a small earthquake training sample set.

[0123] By repeating the above steps S11 to S15, a diverse earthquake training sample set containing rich structural information and different ghost wave characteristics for neural network training can be constructed.

[0124] Compared with the prior art, the ghost wave suppression method based on a deep learning algorithm provided in this embodiment is a data-driven method that uses a deep learning model to learn complex features and patterns from a large amount of training data, and therefore may have more advantages in dealing with complex situations. Moreover, the present invention is highly flexible and can establish training sample sets containing different ghost wave features for different work areas and exploration environments to adapt to the actual complex and changeable geological conditions. In addition, the present invention can automatically suppress ghost waves without the need for ghost wave delay time estimation, reducing the need for manual intervention and improving efficiency. In particular, the present invention has the potential to be directly applied to three-dimensional situations, while the prior art is limited by the sparse Crossline acquisition system.

[0125] Embodiment 2

[0126] This embodiment is the second embodiment of the ghost wave suppression method based on the deep learning algorithm. This embodiment is similar to the first embodiment, except that in this embodiment, a processing process of a U-net neural network is provided; in the neural network training stage, a U-net neural network is first constructed. Compared with the existing U-net, the U-net used in this embodiment contains more convolutional layers, more feature maps output by each layer, and more downsampling and upsampling layers. After the seismic sample with ghost waves is input to the U-net, it first undergoes a series of downsampling steps in the encoder path to help the U-net capture hierarchical features. Among them, the seismic image with an input size of 128×128 is gradually reduced to 64×64, 32×32, 16×16, 8×8 and 4×4, and the number of channels is gradually increased to 128, 256, 512, 1024, 2048 and 4096. Subsequently, a series of upsampling processes are performed in the decoder path to restore the spatial size. At the same time, the skip connection combines the high-level semantic features in the decoder with the low-level details in the encoder, enabling the U-net neural network to effectively aggregate coarse global context information and detailed local features. Finally, a 1×1 convolutional layer outputs a suppressed ghost wave image of the same size as the input.

[0127] Then use the mean squared error (MSE), that is, L 2 Norm, as a loss function:

[0128]

[0129] Where n represents the total number of pixels in the input seismic image; p i represents the suppressed ghost wave image output by the U-net neural network; l i Represents a ghost-free labeled image.

[0130] The MSE loss function is optimized using the Adam algorithm, where the initial learning rate is 10 -5 , the batch size is set to 10 and the training rounds are set to 100.

[0131] Embodiment 3

[0132] This embodiment provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described in Embodiment 1 or Embodiment 2 when executing the computer program.

[0133] In the specific contents of the above-mentioned specific implementation methods, the various technical features can be combined in any non-contradictory manner. In order to make the description concise, not all possible combinations of the above-mentioned technical features are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0134] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for suppressing ghost waves in imaging domain based on deep learning, characterized in that: The following steps are involved: S1. Construct an earthquake training sample set, artificially synthesize a fault velocity model, and synthesize a ghost wave training sample set based on the fault velocity model using the reverse time migration (RTM) algorithm; S2. Build a U-net neural network; S3. Use the above synthesized ghost wave training sample set to train the U-net neural network; S4. Input the seismic data of the target work area into the U-Net neural network for intelligent ghost wave suppression processing to obtain the seismic data after ghost wave suppression.

2. The imaging domain ghost suppression method based on deep learning according to claim 1, characterized in that: The step S1 comprises: S11. Generate horizontal layered velocity model V1(X,Z): randomly generate a one-dimensional velocity sequence V(Z), in which the number, thickness and velocity of strata are randomly selected from a reasonable range determined by the actual geological conditions of the work area, and expand the one-dimensional velocity sequence V(Z) along the horizontal direction into a two-dimensional horizontal layered velocity model V1(X,Z); S12. Generate inclined fold model Based on the horizontal layered velocity model V1(X,Z), the vertical displacement field M(X,Z) is used to add the formation dip and fold structure to obtain the inclined fold model S13. Generate a true fault velocity model Using fault displacement fields in the inclined fold model Fault structures are added to obtain a true fault velocity model that includes formation dip, folds, and fault structures. S14. Generate RTM images: Use the RTM algorithm to transform the real fault velocity model Convert to seismic images, and obtain seismic images with and without ghost waves; S15. Generate an earthquake training sample set.

3. The imaging domain ghost suppression method based on deep learning according to claim 2, characterized in that: The step S12 includes: the vertical displacement field M(X,Z) is composed of two parts: M1(X,Z) and M2(X,Z): M(X,Z)=M1(X,Z)+M2(X,Z)(1) In formula (1), M1(X, Z) is a straight line equation used to simulate the formation dip, and its expression is: M1(X,Z)=tanα·(X-X c )(2) Where α represents the formation inclination; V1(X c ,Z) represents the middle track of the velocity model; M2(X,Z) is composed of N one-dimensional Gaussian functions and is used to simulate the underground fold structure. Its expression is as follows: Among them, Z max Indicates the maximum depth of the velocity model; b k 、c k and σ k They represent the amplitude, center position and half-length of the kth Gaussian function, and their sizes are all randomly generated. By superimposing and summing N different Gaussian functions, multiple different fold structures are added to the velocity model. Using the vertical displacement field M(X,Z) above, the formation dip and fold structure can be added to the horizontal layered velocity model V1(X,Z) to generate an inclined fold model 4. The imaging domain ghost suppression method based on deep learning according to claim 3 is characterized in that: The step S13 comprises: S131. Determine the fault reference line: Use the center point position of the reference line and the fault dip angle θ define a fault reference line; wherein the center point position P and the fault dip angle θ are randomly generated within the set range; S132. Coordinate system transformation: using the randomly determined reference line center point position and the fault dip angle θ will transform the global coordinate system Convert to the local coordinate system (x, z): Where R is the rotation matrix: After the coordinate system conversion, point P is the origin of the local coordinate system; the positive direction of the x-axis points to the dip; the z-axis is the normal of the reference line, and its positive direction points to the hanging wall of the fault; the fault reference line is the z-axis of the local coordinate system; S133. Generate fault curve: randomly generate a series of disturbance points near the fault reference line z=0, and then use these disturbance points to interpolate and construct the fault curve z f =f(x); S134. Calculate the fault displacement field D(x,z): First, calculate the dip displacement field Dx(x,z) in two-dimensional space, and then calculate the normal displacement field Dz(x,z); the fault displacement field D(x,z) is a two-dimensional vector field defined in the local coordinate system (x,z), which consists of two parts: the dip displacement field Dx(x,z) and the normal displacement field Dz(x,z): D(x,z)=[D x (x,z)u x ,D z (x,z)u z ](7) Among them, u x and u z Represent the unit vectors in the x and z directions respectively; S135. Use fault displacement field to simulate fault structure: According to the calculated fault displacement field D(x,z), the inclined fold model Add fault structure to get fault model Where T represents the transposition symbol.

5. The imaging domain ghost suppression method based on deep learning according to claim 4, characterized in that: The step S134 includes: First, calculate the projection displacement field d(x; z = 0) of the inclination displacement field Dx(x, z) on the reference line: Among them, d max represents the maximum displacement of the center point, which is the origin of the local coordinate system; r(x) is the normalized radius: Among them, l x represents the half length of the fault line; Then, the fault displacement of each point in the two-dimensional space is calculated using the projected displacement field d(x; z=0) on the fault reference line; a boundary γ is defined so that Dx(x,z) is 0 when it reaches or exceeds the boundary γ; For the upper plate f(x)≤z≤f(x)+γ, the calculation expression of Dx(x,z) is: D x (x,z)=λ·d(x;z=0)·ζ(x,z)(11) Among them, λ represents the percentage of the upper plate displacement in the projected displacement field d(x; z=0), which is randomly selected between [0,1]; ζ(x,z) is a nonlinear attenuation function, which attenuates the dip displacement field Dx(x,z) away from the fault curve along the normal direction, and decays to 0 when it reaches the traction radius γ; the nonlinear attenuation function ζ(x,z) is expressed as: For the lower plate f(x)-γ≤z≤f(x), the calculation expression of Dx(x,z) is: D x (x,z)=(λ-1)·d(x;z=0)·ζ(x,z)(13) After calculating the inclination displacement field Dx(x,z) in two-dimensional space, the normal displacement field Dz(x,z) can be directly obtained using the following formula: D z (x,z)=f(x+D x (x,z))-f(x)(14)。 6. The imaging domain ghost suppression method based on deep learning according to claim 5, characterized in that: By using steps S131 to S135, it is possible to Repeat steps S131 to S135 in the same velocity model to generate a fault velocity model V3 (X, Z) containing multiple faults.

7. The imaging domain ghost suppression method based on deep learning according to any one of claims 2 to 6, characterized in that: The step S14 comprises: first, defining an offshore seismic data observation system; randomly selecting one of the conventional horizontal towed cable type, up-and-down cable type, variable depth cable type and deep towed type as a seismic data acquisition method, wherein the depths of the source and the detector are randomly selected within a set range; then, using boundary conditions with and without free surfaces, respectively, RTM imaging is performed to obtain seismic images with and without ghost waves.

8. The imaging domain ghost suppression method based on deep learning according to any one of claims 2 to 6, characterized in that: The step S15 includes: randomly selecting images with a size of 128×128 pixels at the same position from the calculated RTM images with and without ghost waves, and using the images with ghost waves as earthquake training samples, and the images without ghost waves as label images; repeatedly randomly selecting multiple 128×128 training samples and labels from the two RTM images to obtain a small earthquake training sample set.

9. The imaging domain ghost suppression method based on deep learning according to claim 8, characterized in that: In step S2, the process of processing data by the U-net neural network includes: after the seismic sample with ghost waves is input into the U-net, it first undergoes a series of downsampling steps in the encoder path to help the U-net capture hierarchical features; wherein the seismic image with an input size of 128×128 is gradually reduced to 64×64, 32×32, 16×16, 8×8 and 4×4, and the number of channels is gradually increased to 128, 256, 512, 1024, 2048 and 4096; then, a series of upsampling processes are performed in the decoder path to restore the spatial size; at the same time, the skip connection combines the high-level semantic features in the decoder with the low-level details in the encoder, so that the U-net neural network can effectively aggregate the rough global context information and the detailed local features; finally, a 1×1 convolution layer outputs a suppressed ghost wave image of the same size as the input; Then use the mean square error, that is, the L2 norm, as the loss function: Where n represents the total number of pixels in the input seismic image; p i represents the suppressed ghost wave image output by the U-net neural network; l i Represents a ghost-free labeled image.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method described in any one of claims 1 to 9 are implemented.

Images