A device and method for processing three-dimensional seismic data

Abstract

The present invention provides a device and method for processing three-dimensional seismic data. The method part includes the following steps: sensor detectors arranged on the surface or in a well record wave field signals; a source vehicle moves and excites according to a preset grid, and a recording system synchronously stores seismic trace data; original three-dimensional seismic data blocks are input into a generator, and processed three-dimensional seismic data blocks are output; data blocks output by the generator or real high-quality data blocks are input through a discriminator to distinguish local statistical characteristics of generated data from real data; data is preprocessed; and processing results are evaluated using quantitative indicators and qualitative comparisons. The present invention can effectively reduce noise in seismic data, repair data missing areas, and enhance detailed features of the data to improve the reliability of subsequent interpretation and imaging, avoid the problem of excessive deviation of original data caused by errors in the detector itself, and improve the protection effect of the detector.

Description

Technical Field

[0001] The present invention relates to the field of geophysical exploration and computer artificial intelligence technology, and in particular to a device and method for processing three-dimensional seismic data. Background Art

[0002] In geophysical exploration, seismic data acquisition involves placing a seismic source on the surface or underwater to generate artificial seismic waves and recording the signals reflected by these seismic waves at the interfaces of different underground strata. In this way, important data about stratum interfaces, rock and soil properties, and geological structures can be obtained.

[0003] In geophysical exploration, the acquisition and processing of seismic data is an important means of obtaining information about underground geological structures. In recent years, generative adversarial networks (GANs) have achieved remarkable results in fields such as image processing and speech signal enhancement. Through adversarial training of generators and discriminators, GANs can effectively learn the distribution characteristics of data. However, existing three-dimensional seismic data processing solutions face many challenges, such as noise interference, data loss, and resolution degradation during data acquisition. Traditional processing methods have certain limitations, making it difficult for processing results to achieve the expected accuracy. Existing processing methods (such as traditional filtering and interpolation algorithms) show limitations when processing complex underground structures, making it difficult to strike a balance between data quality and computational efficiency. In addition, the acquisition process of three-dimensional seismic data also faces problems such as insufficient accuracy, difficulty in acquisition, and easy damage. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a three-dimensional seismic data processing device and method to solve the problems raised in the above-mentioned background technology. The present invention can effectively reduce the noise in the seismic data, repair the data missing areas, and enhance the detailed features of the data to improve the reliability of subsequent interpretation and imaging, avoid the problem of excessive deviation of the original data due to the error of the detector itself, and improve the protection effect of the detector.

[0005] In order to achieve the above-mentioned purpose, the present invention is implemented through the following technical solutions: a three-dimensional seismic data processing device, comprising a processing device body, the processing device body comprising a seismic data recorder, a verification module, a signal line and a detector, the verification module is installed on the side of the seismic data recorder, and the seismic data recorder is connected to a signal line, the signal line passes through the interior of the verification module, the signal line is connected to multiple detectors, a shell is installed on the top of the detector, and a plug-in rod is inserted at the bottom of the detector, a reeling column is provided on the top of the shell, the bottom end of the shell is used to press on the ground, the plug-in rod is used to be embedded under the ground, a verification compartment is opened inside the verification module, and each of the detectors is pushed vertically into the interior from the open end of the verification compartment.

[0006] Furthermore, the detector includes an outer shell, a receiving sleeve, a plug-in rod and a fixing rod. The bottom of the outer shell is integrally formed with a receiving sleeve, lifting grooves are provided on both sides of the outer shell, the top side of the outer shell is integrally formed with a fixing plate, and the side of the receiving sleeve is mounted with an impact plate.

[0007] Furthermore, a transmission sleeve is integrally formed on the top of the plug-in rod, a movable plate is installed on the side of the transmission sleeve, a fixed rod is inserted at the end of the movable plate, a spring is sleeved on the top of the fixed rod, and the movable plate extends outward from the inside of the lifting slot.

[0008] Furthermore, the bottom of the receiving sleeve is in an open state, the bottom end of the plug rod passes downward from the bottom of the receiving sleeve, the two ends of the spring are respectively connected to the surfaces of the movable plate and the fixed plate, and the transmission sleeve is embedded in the interior of the shell.

[0009] Furthermore, the verification module includes a motor and a vibration simulation chamber, a drive shaft is inserted into the interior of the vibration simulation chamber, a motor is installed at one end of the drive shaft, a cam is integrally formed on the surface of the drive shaft, and a pressure rod is also inserted into the interior of the vibration simulation chamber, and end plates are integrally formed at both ends of the pressure rod.

[0010] Furthermore, the end plate is embedded in the inner wall of the vibration simulation chamber, an oscillation plate is installed on the side of the pressure-bearing rod, and an elastic sheet is attached to the side of the end plate. The elastic sheet has an overall arc-shaped structure, and the other end of the elastic sheet rests on the inner wall of the vibration simulation chamber, and the end of the oscillation plate is used to rest on the impact plate of each detector.

[0011] A method for processing three-dimensional seismic data, comprising the following steps:

[0012] Step 1: Sensor detectors arranged on the surface or in the well record wave field signals, and multiple detectors are combined according to rules to suppress ground interference;

[0013] Step 2: The source vehicle moves and excites according to the preset grid, and the recording system synchronously stores the seismic trace data;

[0014] Step 3: Input the original 3D seismic data block into the generator, and output the processed 3D seismic data block;

[0015] Step 4: The discriminator inputs the data block output by the generator or the real high-quality data block to distinguish the local statistical characteristics of the generated data from the real data;

[0016] Step 5: Preprocess the data and use the training set to perform adversarial training on the GAN, optimizing the generator and discriminator alternately.

[0017] Step 6: Use quantitative indicators and qualitative comparison to evaluate the processing results and perform visualization.

[0018] Furthermore, the core architecture of the discriminator includes an input layer, a dimensional convolution layer, a batch normalization layer and a fully connected layer, where the number of convolution kernels is set to 16 or 32 for extracting different types of features; the step size is set to (1,1,1) or (2,2,2) for downsampling the data while extracting features to reduce the spatial dimension of the data.

[0019] Furthermore, the training process of the adversarial network is also included, which combines the adversarial loss with the seismic data characteristic constraints, including the adversarial loss, the reconstruction loss and the total loss of the generator. The gradient penalty adopts Wasserstein GAN, and the gradient penalty term needs to be added to stabilize the training. The total loss of the generator L G It is the weighted sum of adversarial loss and reconstruction loss, and the formula is:

[0020] LG=L Gadv +γL Grec

[0021] Among them, γ is a hyperparameter used to control the weight of reconstruction loss in the total loss. By adjusting the value of γ, it is used to balance the emphasis of the generator between deceiving the discriminator and generating content similar to real data.

[0022] Furthermore, it also includes the data training process, including: pre-training the generator using only content loss: β = 1, α = 0, quickly converging to a rough solution; jointly training the generator and discriminator, gradually increasing the adversarial loss weight: α = 0.1, β = 0.9; introducing a multi-scale discriminator to improve the ability to discriminate details; using the Adam optimizer: β1 = 0.5, β2 = 0.999, and dynamically adjusting the initial learning rate.

[0023] Beneficial effects of the present invention:

[0024] 1. This 3D seismic data processing device can provide protection for each detector during the acquisition process, reducing the difficulty of 3D seismic data acquisition operations. It can also quickly verify each detector through a verification module to ensure the processing validity of the final 3D seismic data and avoid the problem of excessive deviation of the original data caused by the detector's own errors.

[0025] 2. This method for 3D seismic data processing has enhanced noise reduction capabilities, effectively suppressing both random and systematic noise in seismic data. GAN can accurately complete missing data caused by equipment or environmental issues during the acquisition process.

[0026] 3. This method for processing 3D seismic data can effectively enhance data features and strengthen the detailed structure of seismic data, facilitating subsequent geological interpretation and modeling. Furthermore, compared to traditional methods, GAN demonstrates higher computational efficiency when processing complex 3D data. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of a method for processing three-dimensional seismic data according to the present invention;

[0028] Figure 2 This is a schematic structural diagram of a three-dimensional seismic data processing device according to the present invention;

[0029] Figure 3 Schematic diagram of the structure of the detector part of the present invention;

[0030] Figure 4 This is a disassembled diagram of the detector of the present invention;

[0031] Figure 5 It is a structural diagram of the verification module part of the present invention;

[0032] Figure 6 This is an internal breakdown diagram of the verification module of the present invention;

[0033] In the figure: 1. Seismic data recorder; 2. Verification module; 3. Signal line; 4. Detector; 5. Housing; 6. Retracting column; 7. Receiving sleeve; 8. Impact plate; 9. Connecting rod; 10. Transmission sleeve; 11. Lifting slot; 12. Fixed plate; 13. Threaded sleeve; 14. Threaded rod; 15. Movable plate; 16. Fixed rod; 17. Spring; 18. Storage compartment; 19. Verification compartment; 20. Motor; 21. Vibration simulation chamber; 22. Drive shaft; 23. Cam; 24. Pressure rod; 25. End plate; 26. Elastic sheet; 27. Oscillation plate. DETAILED DESCRIPTION

[0034] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0035] See also Figures 1 to 6 The present invention provides the following technical solution: a three-dimensional seismic data processing device, comprising a processing device body, comprising a seismic data recorder 1, a calibration module 2, a signal line 3, and geophones 4. The calibration module 2 is mounted on the side of the seismic data recorder 1, and the seismic data recorder 1 is connected to the signal line 3, which passes through the interior of the calibration module 2 and connects to multiple geophones 4. A housing 5 is mounted on top of each geophone 4, and a plug-in rod 9 is inserted at the bottom of each geophone 4. A retracting column 6 is provided on the top of the housing 5. The bottom end of the housing 5 is configured to press against the ground, and the plug-in rod 9 is configured to be embedded in the ground. A calibration chamber 19 is defined within the calibration module 2, and each geophone 4 is vertically pushed into the chamber from the open end. This three-dimensional seismic data processing device can be used to calibrate modules acquired during the initial seismic data acquisition process, thereby improving the accuracy of the data processing process.

[0036] When the invention is used, multiple detectors 4 are directly embedded in the ground, and the buffer structure on the outer shell 5 is used to provide protection for each detector 4 during the acquisition process, thereby reducing the difficulty of the three-dimensional seismic data acquisition operation, and being able to quickly verify each detector 4 through the verification module 2 to ensure the effectiveness of the final three-dimensional seismic data processing and avoid the problem of excessive deviation of the original data caused by the error of the detector 4 itself.

[0037] In this embodiment, the detector 4 includes a housing 5, a receiving sleeve 7, a plug rod 9, and a fixed rod 16. The bottom of the housing 5 is integrally formed with the receiving sleeve 7. Lifting slots 11 are defined on both sides of the housing 5. A fixing plate 12 is integrally formed on the top side of the housing 5, and an impact plate 8 is attached to the side of the receiving sleeve 7. A transmission sleeve 10 is integrally formed on the top of the plug rod 9. A movable plate 15 is mounted on the side of the transmission sleeve 10. The end of the movable plate 15 is inserted through the fixed rod 16. A spring 17 is sleeved on the top of the fixed rod 16. The movable plate 15 extends outward from the interior of the lifting slot 11. The bottom of the receiving sleeve 7 is open, and the bottom end of the plug rod 9 extends downward from the bottom of the receiving sleeve 7. The ends of the spring 17 are connected to the surfaces of the movable plate 15 and the fixed plate 12, respectively. The transmission sleeve 10 is embedded in the interior of the housing 5.

[0038] Specifically, the detector 4 itself has a double-layer structure, and the bottom plug-in rod 9 is partially in a separated state through the top movable plate 15 and the fixed plate 12. The spring 17 applies downward pressure to push the movable plate 15 downward. This pushing effect causes the tip part of the bottom of the plug-in rod 9 to pass downward from the bottom of the receiving sleeve 7, and maintain this state to install the entire detector 4. During installation, the bottom plug-in rod 9 will be embedded in the ground. When there are hard obstacles such as stones under the ground, the plug-in rod 9 is blocked by the bottom and cannot continue to move downward. At this time, the movable plate 15 will move upward relative to the fixed plate 12, and the spring 17 will be compressed, thereby avoiding excessive impact force from the top during installation, which will cause damage to the tip part of the plug-in rod 9, thereby protecting the end of the plug-in rod 9.

[0039] In this embodiment, the verification module 2 includes a motor 20 and a vibration simulation chamber 21. A drive shaft 22 is inserted into the interior of the vibration simulation chamber 21. The motor 20 is installed at one end of the drive shaft 22. A cam 23 is integrally formed on the surface of the drive shaft 22. A pressure rod 24 is also inserted into the interior of the vibration simulation chamber 21. The two ends of the pressure rod 24 are integrally formed with end plates 25. The end plate 25 is embedded in the inner wall of the vibration simulation chamber 21. An oscillation plate 27 is installed on the side of the pressure rod 24. An elastic sheet 26 is attached to the side of the end plate 25. The elastic sheet 26 has an overall arc-shaped structure, and the other end of the elastic sheet 26 rests against the inner wall of the vibration simulation chamber 21. The end of the oscillation plate 27 is used to rest against the impact plate 8 of each detector 4.

[0040] Specifically, after each detector 4 is placed back into the calibration chamber 19 with the help of the signal line 3, the two adjacent detectors 4 are placed side by side, and the motor 20 is started at this time. The motor 20 drives the drive shaft 22 and the cam 23 on the surface of the drive shaft 22 to rotate. The side of the cam 23 will periodically collide with the side pressure rod 24 during the rotation process. The pressure rod 24 is pushed by the elastic sheet 26 in the initial state, causing the side oscillation plate 27 to partially move to an area away from each detector 4. After being collided with the cam 23, the entire pressure rod 24 is pushed continuously. At this time, the end plate 25 translates toward the position of the detector 4, and the elastic sheet 26 is compressed, and finally the oscillation plate 27 collides with the impact plate 8 on the side of each detector 4. The collision effect acts on the sensor inside the detector 4, and the data collected by each detector 4 is compared to determine whether the collection results of each detector 4 are accurate and consistent, so as to achieve the purpose of synchronous calibration and detection of multiple detectors 4.

[0041] This embodiment also provides a method for processing three-dimensional seismic data, comprising the following steps:

[0042] Step 1: The sensor detectors 4 arranged on the surface or in the well record the wave field signal, and multiple detectors 4 are combined according to a rule to suppress ground interference;

[0043] Step 2: The source vehicle moves and excites according to the preset grid, and the recording system synchronously stores the seismic trace data;

[0044] Step 3: Input the original 3D seismic data block into the generator, and output the processed 3D seismic data block;

[0045] Step 4: The discriminator inputs the data block output by the generator or the real high-quality data block to distinguish the local statistical characteristics of the generated data from the real data;

[0046] Step 5: Preprocess the data and use the training set to perform adversarial training on the GAN, optimizing the generator and discriminator alternately.

[0047] Step 6: Use quantitative indicators and qualitative comparison to evaluate the processing results and perform visualization.

[0048] In this embodiment, the core architecture of the discriminator includes an input layer, a dimensional convolution layer, a batch normalization layer, and a fully connected layer, wherein the number of convolution kernels is set to 16 or 32 for extracting different types of features; the step size is set to (1,1,1) or (2,2,2) for downsampling the data while extracting features to reduce the spatial dimension of the data.

[0049] This embodiment also includes an adversarial network training process, which combines adversarial loss with seismic data characteristic constraints, including adversarial loss, reconstruction loss, and total generator loss. The gradient penalty adopts Wasserstein GAN, and a gradient penalty term needs to be added to stabilize training.

[0050] The reconstruction loss is used to measure the difference between the denoised data generated by the generator and the real clean data, ensuring that the generated data is consistent with the real data in terms of content. The commonly used reconstruction loss is the Mean Squared Error Loss (MSE). Assume that the real clean data is y and the denoised data generated by the generator is The reconstruction loss The formula is:

[0051]

[0052] in:

[0053] n is the total number of elements in the data.

[0054] y i is the i-th element of the real clean data.

[0055] is the i-th element of the denoised data generated by the generator.

[0056] The total loss L of the generator G It is the weighted sum of adversarial loss and reconstruction loss, and the formula is:

[0057] L G =L Gadv +γL Grec

[0058] Among them, γ is a hyperparameter used to control the weight of reconstruction loss in the total loss. By adjusting the value of γ, it is used to balance the emphasis of the generator between deceiving the discriminator and generating content similar to real data.

[0059] This embodiment also includes a data training process, including: pre-training the generator using only content loss: β = 1, α = 0, quickly converging to a rough solution; jointly training the generator and discriminator, gradually increasing the adversarial loss weight: α = 0.1, β = 0.9; introducing a multi-scale discriminator to improve the ability to discriminate details; using the Adam optimizer: β1 = 0.5, β2 = 0.999, and dynamically adjusting the initial learning rate.

[0060] The basic principles, main features and advantages of the present invention are shown and described above. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention.

[0061] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A 3D seismic data processing device, comprising a processing device body, characterized in that: The processing device body comprises a seismic data recorder (1), a calibration module (2), a signal line (3) and a detector (4); the calibration module (2) is mounted on the side of the seismic data recorder (1); the seismic data recorder (1) is connected to the signal line (3); the signal line (3) passes through the interior of the calibration module (2); the signal line (3) is connected to a plurality of detectors (4); a housing (5) is mounted on the top of the detector (4); a plug-in rod (9) is plugged into the bottom of the detector (4); the housing (5) A retracting column (6) is provided on the top of the housing (5), the bottom end of the housing (5) is used to press on the ground, the plug rod (9) is used to be embedded in the ground, a calibration chamber (19) is provided inside the calibration module (2), each of the detectors (4) is vertically pushed into the interior from the open end of the calibration chamber (19), the detector (4) comprises a housing (5), a receiving housing (7), a plug rod (9) and a fixing rod (16), the bottom of the housing (5) is integrally formed with a receiving housing (7), and both sides of the housing (5) are provided with lifting The lowering groove (11) is formed with a fixing plate (12) on the top side of the housing (5), and an impact plate (8) is mounted on the side of the receiving housing (7). The calibration module (2) includes a motor (20) and a vibration simulation chamber (21). A driving shaft (22) is inserted into the interior of the vibration simulation chamber (21). The motor (20) is installed at one end of the driving shaft (22). A cam (23) is formed on the surface of the driving shaft (22). A pressure rod (22) is also inserted into the interior of the vibration simulation chamber (21). 4), the two ends of the pressure-bearing rod (24) are integrally formed with end plates (25), the end plates (25) are embedded in the inner wall of the vibration simulation chamber (21), the side of the pressure-bearing rod (24) is installed with an oscillation plate (27), the side of the end plate (25) is attached with an elastic sheet (26), the elastic sheet (26) is an overall arc-shaped structure, and the other end of the elastic sheet (26) is against the inner wall of the vibration simulation chamber (21), and the end of the oscillation plate (27) is used to lean against the impact plate (8) of each detector (4).

2. The 3D seismic data processing device according to claim 1, characterized in that: A transmission sleeve (10) is integrally formed on the top of the plug-in rod (9), a movable plate (15) is installed on the side of the transmission sleeve (10), a fixed rod (16) is inserted at the end of the movable plate (15), a spring (17) is sleeved on the top of the fixed rod (16), and the movable plate (15) is extended outward from the inside of the lifting slot (11).

3. The 3D seismic data processing device according to claim 2, characterized in that: The bottom of the receiving sleeve (7) is in an open state, the bottom end of the plug rod (9) extends downward from the bottom of the receiving sleeve (7), the two ends of the spring (17) are respectively connected to the surfaces of the movable plate (15) and the fixed plate (12), and the transmission sleeve (10) is embedded in the interior of the outer shell (5).

4. A method for processing three-dimensional seismic data using the apparatus as claimed in claim 1, characterized in that: The following steps are involved: Step 1: Sensor detectors arranged on the surface or in the well record wave field signals, and multiple detectors are combined according to rules to suppress ground interference; Step 2: The source vehicle moves and excites according to the preset grid, and the recording system synchronously stores the seismic trace data; Step 3: Input the original 3D seismic data block into the generator, and output the processed 3D seismic data block; Step 4: The discriminator inputs the data block output by the generator or the real high-quality data block to distinguish the local statistical characteristics of the generated data from the real data; Step 5: Preprocess the data and use the training set to perform adversarial training on the GAN, optimizing the generator and discriminator alternately. Step 6: Use quantitative indicators and qualitative comparison to evaluate the processing results and perform visualization.

5. The three-dimensional seismic data processing method according to claim 4, characterized in that: The core architecture of the discriminator includes an input layer, a dimensional convolutional layer, a batch normalization layer, and a fully connected layer. The number of convolution kernels is set to 16 or 32 to extract different types of features; the step size is set to (1, 1, 1) or (2, 2, 2) to downsample the data while extracting features and reduce the spatial dimension of the data.

6. The three-dimensional seismic data processing method according to claim 5, characterized in that: It also includes the training process of the adversarial network, combining the adversarial loss with the characteristics of seismic data, including the adversarial loss, reconstruction loss, and the total loss of the generator. The gradient penalty adopts Wasserstein GAN, and the gradient penalty term needs to be added to stabilize the training. The total loss of the generator LG is the weighted sum of the adversarial loss and the reconstruction loss, and the formula is: LG=L Gadv +γL Grec Among them, γ is a hyperparameter used to control the weight of reconstruction loss in the total loss. By adjusting the value of γ, it is used to balance the emphasis of the generator between deceiving the discriminator and generating content similar to real data.

7. The three-dimensional seismic data processing method according to claim 6, characterized in that: It also includes the data training process, including: pre-training the generator using only content loss: β=1, α=0, quickly converging to a rough solution; jointly training the generator and discriminator, gradually increasing the adversarial loss weight: α=0.1, β=0.9; introducing a multi-scale discriminator to improve the ability to discriminate details; using the Adam optimizer: β1=0.5, β2=0.999, and dynamically adjusting the initial learning rate.

Images