Ocean vibroseis synchronous correction and signal separation method and device
By using a priori constraint inversion framework based on convolutional neural networks (CNN), Doppler correction and aliasing signal separation of marine controllable seismic sources were achieved, solving the problem of simultaneous processing of Doppler effect and aliasing signals in marine seismic data acquisition, and improving the reliability and generalization ability of data processing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JILIN UNIVERSITY
- Filing Date
- 2026-04-24
- Publication Date
- 2026-05-26
AI Technical Summary
Existing technologies struggle to simultaneously correct for the Doppler effect and separate aliased signals during marine seismic data acquisition, leading to a decline in data processing quality and reliability.
An inversion framework based on convolutional neural network (CNN) prior constraints is adopted, and the inversion iterative process is embedded in a plug-and-play manner. The objective function is minimized to demix and correct the measured aliased co-detector point gather. A pre-trained denoiser is used for Doppler correction and aliased signal separation.
It effectively eliminates the Doppler effect caused by ship motion, solves the problems of aliasing and phase distortion in seismic data acquired from multiple sources, and improves the reliability and generalization ability of marine seismic data processing.
Smart Images
Figure CN122085359A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of marine controlled-source seismic signal processing technology, and more specifically, to a method and apparatus for synchronous correction and signal separation of marine controlled-source seismic signals. Background Technology
[0002] In marine seismic data acquisition, traditional airgun seismic sources can significantly impact the marine environment and organisms due to the instantaneous high-pressure pulses they generate. In particular, such pulsed signals can damage the hearing of marine mammals, thereby altering their behavioral patterns and even affecting population numbers in the long term. In contrast, controlled marine seismic sources significantly reduce interference with the marine ecosystem by emitting non-pulsed seismic signals with comparable energy but longer duration and gradually varying frequency. In actual maritime operations, the controlled seismic source moves continuously with the source vessel and achieves continuous reception and recording through simultaneous or delayed excitation, thereby effectively improving acquisition efficiency and reducing operating costs.
[0003] However, the aforementioned operational methods also introduce new problems. On the one hand, the motion of the seismic source vessel introduces Doppler frequency shift into the received signal, causing phase distortion; on the other hand, the random delay excitation method of multiple sources causes the signals from each source to overlap in the recording, generating aliasing noise. These two factors together affect the quality and reliability of subsequent data processing. Therefore, it is urgent to synchronously correct for the Doppler effect from the mixed seismic records and achieve effective separation of multi-source signals.
[0004] For Doppler frequency shift correction, existing technologies mainly employ two approaches: one is to perform phase correction using fk-domain filtering after correlation, and the other is to perform compensation through two-dimensional deconvolution before correlation. While the post-correlation filtering method can effectively handle linear frequency sweeps and suppress aliasing artifacts, it is difficult to directly apply to nonlinear frequency sweep signals and is susceptible to endpoint effects, thus limiting correction accuracy. Furthermore, this method may introduce spurious frequency interference in the frequency domain processing, affecting the subsequent interpretation of seismic data. Although the pre-correlation deconvolution method performs well in terms of computational efficiency and correction effect, its separation from aliasing noise is usually a separate process, making it difficult to achieve simultaneous optimization and efficient coordination of Doppler frequency shift correction and aliasing noise separation under complex and variable actual acquisition conditions. In addition, while existing inversion-based separation techniques have some applicability in processing complex signals, they are not deeply integrated with Doppler correction, limiting the overall recovery performance and reliability in shipborne mobile source aliasing data scenarios.
[0005] Therefore, existing technologies struggle to simultaneously process and optimize Doppler effect correction and aliased signal separation when dealing with signal aliasing and phase distortion caused by continuous excitation from multiple sources and the continuous movement of the source vessel in marine seismic acquisition. Thus, a new technical solution is urgently needed to simultaneously correct the Doppler effect and effectively separate aliased signals. Summary of the Invention
[0006] To address the aforementioned technical problems, this disclosure provides a method and apparatus for synchronous correction and signal separation of a marine controllable seismic source.
[0007] According to one aspect of this disclosure, a method for synchronous correction and signal separation of a marine controllable seismic source is provided, the method comprising: Acquire raw seismic records from a marine data acquisition terminal, the raw seismic records including ocean-controlled source superimposed data; The measured aliased co-detector gathers were obtained using the original seismic records. By embedding a pre-trained prior model into the inversion iterative process in a plug-and-play manner to minimize the predetermined objective function, the measured aliased co-detector point gathers are unmixed and corrected to obtain usable single-source seismic data. The prior model includes multiple denoisers, each of which employs a convolutional neural network (CNN) structure. The prior model is trained in the following manner: Forward modeling yields simulated aliased co-detector point gathers and their corresponding simulated single-source co-detector point gathers, constructing aliased single-source training pairs; bandpass filtering is used to convert a predetermined natural image dataset into pseudo-seismic data, and Gaussian white noise of different noise levels is added to the pseudo-seismic data to construct pseudo-data training pairs; and the parameters of the prior model are determined using the aforementioned aliased single-source training pairs and pseudo-data training pairs.
[0008] In some implementations, the pseudo-seismic data is obtained in the following manner: Identify publicly available natural image datasets; The image data in the publicly available natural image dataset is transformed into the frequency-wavenumber fk domain using a two-dimensional Fourier transform to obtain a two-dimensional complex spectrum; After removing large-dipping events from the two-dimensional complex spectrum, a filter is applied to obtain a filtered two-dimensional complex spectrum, which retains only the components related to the seismic data. The filtered two-dimensional complex spectrum is subjected to a two-dimensional inverse Fourier transform to obtain the pseudo-seismic data.
[0009] In some implementations, the filtering is achieved using a filter as shown in the following formula: , in, Indicates frequency, Indicates wave number, This indicates the predetermined apparent speed threshold. This represents the filter response.
[0010] In some implementations, the prior model is trained using the loss function shown below: , in, Represents the loss function. Represents the parameters of the prior model. This represents the total number of training samples. For training sample index, This indicates the first common receiver gather or the first pseudo-seismic data with added noise in the mixed seismic data. Noisy input data This indicates the first point in the corresponding simulated single-source common-receiver gather or clean pseudo-seismic data. A clean target data, Indicates the prior model to The predicted output.
[0011] In some implementations, the forward modeling yields simulated aliased common detector gathers and their corresponding simulated single-source common detector gathers, including: The wave equation is used to independently simulate the seismic records containing Doppler phase shifts generated by each of a pre-selected multiple seismic sources at a fixed receiver point, in order to obtain the simulated single-source common receiver point gather; Within the common receiver point domain of the multiple seismic sources, the simulated single-source common receiver point gathers are superimposed with random time delays within a predetermined range to generate the simulated aliased common receiver point gathers.
[0012] In some implementations, the objective function is expressed as follows: , in, This indicates the available single-source seismic data. This represents the measured aliasing common detector gather. Represents the aliasing operator. This represents the Doppler distortion operator. Represents the regularization parameter. Represents the regularization term, This represents the inverse Fourier transform. This represents the solution that minimizes the objective function.
[0013] In some implementations, minimizing a predetermined objective function by embedding a pre-trained prior model in a plug-and-play manner into the inversion iterative process to unmix and correct the measured aliased co-detector point gather includes: Iterative updates are performed using the following formula: , , in, Indicating the corresponding prior model A noise denoiser, Indicates the first Noise level of the next iteration This represents the Doppler distortion operator. Describes the inverse of the Doppler distortion operator. Represents the aliasing operator. This represents the transpose of the aliasing operator. Represents the identity matrix. Indicates after the first The current solution obtained after the next iteration update; Indicates the first The current solution obtained after the next iteration update. This indicates the measured aliasing common detector gather. This represents the intermediate result obtained after denoising in the k-th iteration. Indicates the iteration number. N represents the total number of iterations, when hour, , This indicates the available single-source seismic data.
[0014] According to one aspect of this disclosure, a marine controllable seismic source synchronous correction and signal separation device is provided, comprising: The data acquisition unit is used to acquire raw seismic records from the marine data acquisition terminal, the raw seismic records including marine controlled source superimposed data; The data processing unit is used to obtain the measured aliased co-detector point gathers using the original seismic records; The unmixing and correction unit is used to minimize a predetermined objective function by embedding a pre-trained prior model into the inversion iterative process in a plug-and-play manner to unmix and correct the measured overlapping co-detector point gathers to obtain usable single-source seismic data. The prior model includes multiple denoisers, each of which employs a convolutional neural network (CNN) structure. The prior model is trained in the following manner: Forward modeling yields simulated aliased co-detector point gathers and their corresponding simulated single-source co-detector point gathers, constructing aliased single-source training pairs; filtering techniques are used to convert a predetermined natural image dataset into pseudo-seismic data, and Gaussian white noise of different noise levels is added to the pseudo-seismic data to construct pseudo-data training pairs; and the parameters of the prior model are determined using the aforementioned aliased single-source training pairs and pseudo-data training pairs.
[0015] According to one aspect of this disclosure, an electronic device is provided, the electronic device including a processor and a memory, the memory storing a computer program that, when executed by the processor, causes the processor to perform the methods described above.
[0016] According to one aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when run by a processor, causes the processor to perform the methods described above.
[0017] The embodiments disclosed herein employ an inversion framework based on prior constraints of a convolutional neural network (CNN) to achieve synchronous correction and separation, which can effectively eliminate the Doppler effect caused by ship motion, solve the problem of aliasing in multi-source seismic data acquisition, and solve the problem of signal aliasing and phase distortion caused by continuous excitation of multiple seismic sources and continuous movement of the source ship in marine seismic acquisition, while improving the reliability and generalization ability of marine seismic data processing. Attached Figure Description
[0018] Figure 1 This is a structural example diagram of a system applicable to embodiments of this disclosure; Figure 2 This is a schematic flowchart of the marine controllable seismic source synchronous correction and signal separation method provided in the embodiments of this disclosure; Figure 3 An example diagram of ideal seismic data (i.e., simulated single-source common-detector point gather) acquired by a single source when the ship is stationary, obtained through forward modeling of embodiments of this disclosure; Figure 4 This is an example diagram of a simulated aliased common detector point gather obtained through forward modeling of an embodiment of this disclosure at a ship speed of 2.5 m / s; Figure 5 for Figure 4 Simulated aliasing common receiver gather at a ship speed of 2.5 m / s Figure 3 Residual plot of ideal seismic data (i.e., simulated single-source common-detector point gather) acquired by China Shipbuilding when stationary; Figure 6 This is a flowchart illustrating the process of converting a natural image into pseudo-seismic data according to an embodiment of this disclosure. Figure 6 'a' is a natural image example. Figure 6 b is Figure 6 Figure a shows a temporal representation of a natural image after suppressing vertical features. Figure 6 c is Figure 6 The pseudo-seismic data view of the natural image shown in figure a Figure 6 d is Figure 6 A schematic diagram of the fk spectrum of the natural image shown in figure a. Figure 6 e is Figure 6 Schematic diagram of the fk spectrum of b. Figure 6 f is Figure 6 A schematic diagram of the fk spectrum of c; Figure 7 This is an example diagram of usable single-source seismic data obtained through embodiments of this disclosure; Figure 8 This is a residual diagram of usable single-source seismic data and ideal data obtained through embodiments of this disclosure; Figure 9 This is a schematic diagram of the structure of the marine controllable seismic source synchronous correction and signal separation device provided in the embodiments of this disclosure; Figure 10 This is a schematic structural block diagram of an electronic device provided in an embodiment of this disclosure. Detailed Implementation
[0019] Hereinafter, exemplary embodiments according to the present disclosure will be described in detail with reference to the accompanying drawings. It is obvious that the described embodiments are merely some embodiments of the present disclosure, and not all embodiments of the present disclosure, and it should be understood that the present disclosure is not limited to the exemplary embodiments described herein.
[0020] To address the signal aliasing and phase distortion problems caused by continuous excitation from multiple sources and the continuous movement of the source vessel in marine seismic acquisition, this disclosure provides the following method and apparatus for synchronous correction and signal separation of marine controllable seismic sources, which adopts an inversion framework based on prior constraints of convolutional neural networks (CNN) to achieve synchronous correction and separation.
[0021] Figure 1 An example architecture diagram of a system to which embodiments of this disclosure apply is shown. See also... Figure 1 The applicable system of this disclosure embodiment may include: a marine data acquisition terminal and a data processing terminal. The marine data acquisition terminal is used to generate aliased seismic waves, record aliased wave fields reflected from the subsurface, and form raw seismic data. The data processing terminal is used to perform marine controllable source synchronous correction and signal separation on the raw seismic data from the marine data acquisition terminal using the method provided in this disclosure embodiment.
[0022] The data acquisition and processing terminals at sea can communicate with each other via satellite, shipborne network or other means, and data exchange between them can also be achieved through mobile hard drives or other means.
[0023] The marine data acquisition terminal can include a source vessel and a receiving platform. The source vessel carries marine controllable source equipment, which may include, but is not limited to, exciters, hydraulic / electric units, and control systems. The source vessel is responsible for generating multiple sets of continuous scanning signals according to a preset time sequence while moving, thus producing a mixed wave field. The receiving platform can be, but is not limited to, a towed vessel carrying, towing, or deploying hydrophone arrays (i.e., receivers), or a seabed node deployment vessel, used to record the mixed wave field reflected from the subsurface, forming raw seismic data. The raw seismic data is usually recorded by shot gather or continuous time. After cross-correlation processing and gather rearrangement, the measured mixed common receiver gather can be obtained from the raw seismic data.
[0024] The data processing end can be implemented as, but is not limited to, a server cluster consisting of a large number of CPU nodes and GPU nodes.
[0025] In practical applications, the source vessel moves and operates at sea according to a pre-set aliasing acquisition sequence (e.g., simultaneous excitation by multiple vessels), generating aliased seismic waves. The receiving platform records these wave fields, forming raw seismic data (i.e., ocean-controlled source aliasing data before correlation, hereinafter referred to as raw seismic records). The raw seismic data is transmitted to the data processing terminal via satellite, mobile hard drive, or shipborne network. The data processing terminal uses the method of this embodiment to perform ocean-controlled source synchronization correction and signal separation on these raw seismic data.
[0026] The specific implementation methods of the embodiments disclosed herein will be described in detail below.
[0027] Figure 2 This diagram illustrates a flowchart of a method for synchronous correction and signal separation of a marine controllable seismic source provided in an embodiment of this disclosure. This method can be achieved through the aforementioned... Figure 1 This is executed by the data processing unit in the system shown. See also Figure 2 The marine controllable seismic source synchronous correction and signal separation method of this disclosure includes the following steps: Step 201: Obtain the raw seismic records from the marine data acquisition terminal. The raw seismic records include ocean-controlled source superimposed data. For example, it is possible to obtain oceanic controlled-source seismic records collected from the seabed node (OBN) field experiment conducted in the Beibu Gulf.
[0028] Step 202: Obtain the measured mixed-detector point gather using the original seismic records; For example, measured aliased common detector gathers can be generated through aliasing processing.
[0029] Step 203 involves minimizing a predetermined objective function by embedding a pre-trained prior model into the inversion iterative process in a plug-and-play manner to unmix and correct the measured overlapping co-receiver point gathers, thereby obtaining usable single-source seismic data. The prior model includes multiple denoisers, each employing a convolutional neural network (CNN) structure. The prior model is trained as follows: Forward modeling yields simulated aliased co-detector point gathers and their corresponding simulated single-source co-detector point gathers, constructing aliased single-source training pairs; filtering techniques are used to convert a predetermined natural image dataset into pseudo-seismic data, and Gaussian white noise of different noise levels is added to the pseudo-seismic data to construct pseudo-data training pairs; and the parameters of the prior model are determined using the aforementioned aliased single-source training pairs and pseudo-data training pairs.
[0030] In specific applications, the prior model can employ any neural network architecture applicable to the embodiments of this disclosure. For example, the prior model can employ, but is not limited to, an Image Restoration Convolutional Neural Network (IRCNN), which includes 25 denoisers for handling different noise levels.
[0031] In some examples, each denoiser in the prior model can adopt a symmetric encoder-decoder structure and be trained using mean squared error (MSE) as the loss function. Each denoiser is trained at different noise levels, thus allowing it to be embedded as a denoiser in the iterative optimization framework of step 203.
[0032] Each training sample in the aliased single-source training pair may include an aliased co-detector point gather and its corresponding simulated single-source co-detector point gather. The aliased co-detector point gather is the input, and its corresponding simulated single-source co-detector point gather is the label. Each training sample in the pseudo-data training pair may include noisy pseudo-data and its corresponding clean pseudo-data. The clean pseudo-data is the pseudo-seismic data obtained by transforming a predetermined natural image dataset using filtering techniques. The noisy pseudo-data is the data obtained by adding Gaussian white noise of different noise levels to the pseudo-seismic data. The noisy pseudo-data is the input, and the clean pseudo-data is the label.
[0033] In practical applications, each denoiser in the prior model is trained independently using the aforementioned method. For example, aliased single-source training pairs and pseudo-data training pairs can be mixed in a pre-defined ratio to form a unified sample set. This sample set is then used to train, validate, and test each denoiser in the prior model to determine its parameters. Alternatively, pseudo-data training pairs can be used to form a training set. This training set is then used to train each denoiser in the prior model to initially determine its parameters. Finally, the dataset formed by aliased single-source training pairs is used to fine-tune the parameters of each denoiser in the prior model to meet validation and testing requirements. This disclosure does not limit the specific training method for the prior model.
[0034] The prior model can be trained using the loss function shown in the following equation: , in, Represents the loss function. Represents the parameters of the prior model. This represents the total number of training samples. For training sample index, This indicates the first common receiver gather or the first pseudo-seismic data with added noise in the mixed seismic data. Noisy input data This indicates the first point in the corresponding simulated single-source common-receiver gather or clean pseudo-seismic data. A clean target data, Indicates the prior model to The predicted output.
[0035] Those skilled in the art should understand that the aforementioned loss function is merely an example. In specific applications, any other applicable loss function may be used as needed, and this disclosure does not impose any limitations on such applications.
[0036] In some implementations, forward modeling to obtain simulated aliased co-detector gathers and their corresponding simulated single-source co-detector gathers may include: using wave equation finite difference to independently simulate seismic records containing Doppler phase shifts generated by each of a pre-selected plurality of sources at a fixed detector point to obtain simulated single-source co-detector gathers; and superimposing each simulated single-source co-detector gather within the co-detector domain of the plurality of sources by a random time delay within a predetermined range to generate simulated aliased co-detector gathers.
[0037] The acquisition process can be combined with the Doppler effect for forward modeling. The wave equation finite difference technique can be used to forward model the seismic record as the source moves with the ship. This process is simulated in the computer using discrete grid points with a spacing of 5m between two adjacent grid points. When the ship moves, a certain phase distortion will occur, resulting in the common-detector point domain seismic record after Doppler frequency shift (i.e., the simulated single-source common-detector point gather mentioned above). In the common-detector point domain, seismic data containing the Doppler effect are artificially mixed to simulate the aliased and phase-distorted seismic data emitted by the two sources with random time delays within 0 to 1s, thereby obtaining the simulated aliased common-detector point gather.
[0038] When modeling the marine acquisition process, the finite difference technique of the wave equation was used to simulate the seismic wave field of the source ship under continuous movement. This method uses a spatial discrete grid for wave field calculation, setting a fixed grid spacing to accurately characterize the phase distortion caused by the ship's motion. During this process, changes in ship speed cause relative motion between the source and the detector, thus introducing the Doppler effect into the synthesized record. Finally, common-detector point domain seismic data containing Doppler phase shift is generated (i.e., the simulated single-source common-detector point gather mentioned above). To further simulate the actual excitation situation of multiple sources, the seismic records of two sources are mixed in the common-detector point domain, where the excitation time delays of the two sources are randomly distributed within the range of 0 to 1 second, thereby synthesizing seismic data with both aliasing characteristics and phase distortion (i.e., the simulated aliased common-detector point gather mentioned above).
[0039] Figure 3 This is an example diagram of ideal seismic data (i.e., simulated single-source common-detector point gathers) acquired by a single source when the ship is stationary, obtained through forward modeling according to embodiments of this disclosure. Figure 4 This is an example diagram of a simulated aliasing common receiver point gather obtained through forward modeling of an embodiment of this disclosure at a ship speed of 2.5 m / s. Figure 5 for Figure 4 Simulated aliasing common receiver gather at a ship speed of 2.5 m / s Figure 3 Residual plot of ideal seismic data (i.e., simulated single-source common-detector point gather) acquired by China Shipbuilding when stationary with a single source.
[0040] Combination Figure 3 and Figure 4 The ideal data acquired by a single source when the ship is stationary and the data acquired by a dual source are respectively simulated using the simulated single-source common detector gather when the ship is moving at a speed of 2.5 m / s.
[0041] Given the known structure and parameters of the subsurface medium, theoretical calculations were used to study the propagation patterns of seismic waves in order to synthesize seismic records. Based on a tilt velocity model, the velocity of the first medium layer was 1500 m / s (simulating the speed of sound in seawater), the velocity of the second medium layer was 1550 m / s, and the velocity of the third medium layer was 1600 m / s. The seismic source was located at the center of the sea surface, and the shot spacing was 5 m. Geophones were arranged on the same plane as the seismic source, and the source vessel moved at a speed of 2.5 m / s.
[0042] In the simulation, the wave equation finite difference technique is used for numerical forward modeling, and the seismic wave field during the ship's movement is approximated by a discrete grid. Since the derivative is replaced by the difference, numerical discretization may introduce dispersion error, affecting the resolution. Therefore, it is necessary to reasonably adjust the parameters to suppress the dispersion phenomenon, so as to obtain an accurate simulated common-source common-source point domain seismic record (i.e., simulated single-source common-source point gather).
[0043] The method of random time delay excitation from two sources is adopted. The time delay between the two sources is between 0 and 1 second. The phase-distorted seismic data (i.e., simulated aliased co-detector point gather) acquired by the two sources is obtained by using the aliasing operator.
[0044] Combination Figure 3 Ideal data for forward simulation and Figure 4 The pseudo-separation data can be observed with Doppler phase distortion and aliasing on the detector offset recording track. Figure 5 for Figure 4 and Figure 3 The subtraction results clearly show the Doppler frequency shift and aliasing caused by the movement of the seismic source.
[0045] In some implementations, pseudo-earthquake data can be obtained through the following steps: determining a publicly available natural image dataset; converting the image data in the publicly available natural image dataset to the frequency-wavenumber fk domain using a two-dimensional Fourier transform to obtain a two-dimensional complex spectrum; applying a filter to the two-dimensional complex spectrum after removing large-dipping events to obtain a filtered two-dimensional complex spectrum, which retains only the components related to the earthquake data; and performing a two-dimensional inverse Fourier transform on the filtered two-dimensional complex spectrum to obtain the pseudo-earthquake data.
[0046] Publicly available natural image datasets can include, but are not limited to, the ImageNet image dataset, the Waterloo exploration image dataset, and self-constructed geological image datasets. The selection of publicly available natural image datasets can be flexible.
[0047] In the fk domain, the low-spectral density of large-dipping events is typically concentrated near the wavenumber axis. Using a filter can remove the vertical distribution features from natural images, effectively preserving spectral components more similar to seismic data. After removing large-dipping events, a filter inverse transform back to the time domain yields pseudo-seismic data. The filter is implemented using the following equation: , in, Indicates frequency; Indicates wave number; This represents the predetermined apparent velocity threshold, which depends on the minimum apparent velocity of the effective wave in the target seismic data and is used to control the spectral range of the fk filter. This represents the filter response.
[0048] Figure 6 An example diagram illustrating the conversion of a public natural image into pseudo seismic data according to an embodiment of this disclosure is shown; Figure 6 a shows an example image from a publicly available natural image dataset. Figure 6 b is Figure 6 c shows Figure 6 The example image shown in figure is a visualization of the pseudo-seismic data obtained through image conversion. Figure 6 d is Figure 6 e is Figure 6 The diagram shows the corresponding fk spectrum of the pseudo-seismic data obtained by converting the example image shown in Figure a.
[0049] Public natural image datasets offer advantages such as large scale and diverse types, containing rich texture and structural information, providing ample data resources for denoising device training and thus alleviating the problem of insufficient MVib data. However, natural images and seismic data still differ in structural characteristics, making direct training difficult to effectively characterize the features of MVib data. Therefore, through... Figure 6 As can be seen, the present invention preprocesses natural images into seismic-like data for training the denoiser, which can significantly improve the generalization ability of PNIF. In some embodiments, when adding Gaussian white noise of different noise levels to the pseudo-seismic data, Gaussian white noise with noise levels between 2 and 50 and intervals of 2 can be added to the pseudo-seismic data.
[0050] In step 202, by performing cross-correlation processing on the original measured ocean-controlled source aliasing data contained in the original seismic record, the correlated aliasing co-detector point gather can be obtained. This correlated aliasing co-detector point gather is the measured aliasing co-detector point gather mentioned above. This cross-correlation processing method can be implemented using existing technologies, and this disclosure does not impose specific limitations on it.
[0051] In step 203, the Doppler effect correction and hybrid source separation are modeled as an inversion optimization problem. Minimizing the aforementioned objective function enables the dealiasing and correction of marine controllable source data.
[0052] The objective function in step 203 can be expressed as follows: , in, This indicates the available single-source seismic data (i.e., the correlated pre-seismic data after decorrection and unmixing). This indicates the measured aliased common receiver gather (i.e., the aliased phase-distorted seismic data acquired by the movement of the mixed source). Represents the aliasing operator. This represents the Doppler distortion operator. Represents the regularization parameter. Represents the regularization term, This represents the inverse Fourier transform. This represents the solution that minimizes the objective function.
[0053] The aliasing operator can be used to describe the aliasing pattern of multiple seismic sources. In specific applications, the aliasing operator can be pre-defined according to the application scenario. If the excitation time delay of each seismic source is known, the corresponding aliasing operator can be constructed. The Doppler distortion operator is used to compensate for phase distortion. Under ideal acquisition conditions, the Doppler effect caused by the motion of the acquisition system can be described by an extended convolution model. A regularization term is introduced into the model. To impose prior constraints, while regularization parameters Used to balance the weights between data fidelity terms and regularization terms. In specific applications, the regularization term... It can be pre-designed or implicitly defined by deep learning from a prior model. When using a plug-and-play approach, the regularization term... It cannot be explicitly written out; it can be implicitly implemented through the output of the denoiser. The inverse Fourier transform is used to convert fk-domain data back to the time domain, representing the sparse inverse transform.
[0054] In order to recover the relevant foreshock data after uncorrection and unmixing The aforementioned objective function uses the L1 norm to constrain the data fidelity term and introduces a regularization term. Constraints are imposed on the solution through regularization parameters. Balancing data fidelity and regularization, inverse Fourier transform Promotes sparsity of solutions.
[0055] Inspired by the Iterative Shrinkage-Threshold Algorithm (ISTA), the aforementioned objective function can be effectively solved as follows: , in, Indicates after the first The current solution obtained after the nth iteration update (that is, the currently recovered single-source seismic data, i.e., the solution obtained at the nth iteration update) The unmixed and corrected data obtained after the next iteration N represents the iteration number (i.e., the current iteration round), and N represents the total number of iterations. This represents the inverse Fourier transform. Indicates Fourier transform, This represents the transpose of the aliasing operator. Describes the inverse of the Doppler distortion operator. After the first The current solution obtained after the nth iteration update (i.e., the solution at the nth iteration update) The unmixed and corrected data obtained after the next iteration.
[0056] When the sparsity constraint imposed on the separation process of the previous expression is implemented using a soft threshold function, The soft threshold function is expressed as follows: , in, For symbolic functions, Indicates the first The threshold parameter of the soft thresholding function during the next iteration is usually... , It is a fixed constant used to control the strength of the sparse constraint. The larger the size, the stronger the contraction, and the sparser the result.
[0057] Therefore, in the process of minimizing the aforementioned objective function, aliasing and phase distortion in the objective function can be resolved through the iterative update steps shown in the following equation: , , Within the inversion framework, the iteration threshold gradually decays to a small coefficient associated with the noise component during each iteration. Based on this principle, the minimization of the aforementioned objective function can be addressed using the iterative update steps shown in the following equation to resolve aliasing and phase distortion within the objective function: , , in, Indicating the corresponding prior model A noise denoiser, Indicates the first Noise level of the next iteration This represents the Doppler distortion operator. Describes the inverse of the Doppler distortion operator. Represents the aliasing operator. This represents the transpose of the aliasing operator. Represents the identity matrix. Indicates after the first The current solution obtained after the next iteration update; Indicates the first The current solution obtained after the next iteration update. This indicates the measured aliasing common detector gather. This represents the intermediate result obtained after denoising in the k-th iteration. Indicates the iteration number. N represents the total number of iterations, when hour, , This indicates the available single-source seismic data.
[0058] in, The threshold parameter for its relationship with the sparse representation coefficients is omitted in the text. The difference. Noise level An adaptive scheduling approach is used to determine that, as iterations proceed, the data recovered through the aforementioned objective function will gradually converge towards the target. In some examples, an exponential decay strategy can be used to determine the noise level in real time. This reduces the effective noise level in each iteration. For example, the noise level... It can be calculated using the following formula: , in, This represents the initial noise level, which increases with the number of iterations. It gradually approaches a very small initial noise lower limit. Initial noise level Initial noise lower limit All settings can be flexibly configured as needed, and can also be modified, reconfigured, or adaptively adjusted based on actual scenarios. Regarding noise levels... The specific method for determining the method is not limited in the embodiments disclosed herein.
[0059] During the inversion iteration process, as the recovered data gradually converges towards the target value, the noise variance... By gradually reducing the noise level through an exponential decay strategy, the noise level is gradually reduced in each iteration.
[0060] The specific iterative process of step 203 is illustrated below.
[0061] The iterative process of step 203 may include the following steps a1 to a5: Step a1, Initialization; Set initial solution , usually take At the same time, a noise level sequence is set. Given the maximum number of iterations N, initialize the iteration number k to 0.
[0062] Step a2: Reset k = k + 1, and determine whether the reset k is less than N; Step a3: If k is less than N, start processing for the current round and return to step a2 after processing for the current round is completed; According to the aforementioned formula, the solution from the previous round... Input to noise level is noise level The intermediate result obtained after denoising in the first iteration is obtained in the denoiser. The intermediate results obtained after denoising in the first iteration are used through the aforementioned formula. Alternation operators Doppler distortion operator get .
[0063] In this iterative process, the noise level The value can be gradually reduced according to a pre-designed strategy (e.g., an exponential decay strategy). This "coarse-to-fine" scheduling allows the aforementioned iterative process to converge stably to the minimum value of the aforementioned objective function.
[0064] Step a4: If k is not less than N, i.e., k = N, then output... And end the current process.
[0065] Controllable seismic sources move continuously with the ship, and continuous recording is achieved through simultaneous or delayed excitation, effectively improving acquisition efficiency and reducing costs. However, the motion of the seismic source ship introduces the Doppler effect into the recording, causing signal phase distortion; at the same time, the random delay excitation method of multiple sources introduces aliasing noise into the data, both of which affect the reliability of subsequent data processing. Previous correction methods and the separation of mixed seismic sources are usually independent processes processed in steps, making it difficult to achieve synchronous optimization and efficient coordination under complex and variable actual acquisition conditions. To address this technical problem, this disclosure provides the aforementioned scheme for simultaneous correction and signal separation of marine controllable seismic sources. It combines deep learning with an inversion framework for simultaneous correction and hybrid source separation. The aliasing and Doppler phase distortion processes are described as an inverse problem. A prior model replaces the traditional sparse domain thresholding step. By combining data-driven learning with physical modeling, it enhances flexibility while maintaining the theoretical rigor of the inversion. This improves the overall recovery performance and reliability in scenarios involving aliased data from shipborne mobile seismic sources, addressing the signal aliasing and phase distortion caused by continuous excitation from multiple sources and the continuous movement of the source vessel in marine seismic acquisition. Furthermore, to alleviate the scarcity of training data for marine controllable seismic sources, this disclosure uses filtered public natural images to generate pseudo-seismic data for training the prior model. This pseudo-seismic data possesses earthquake-like characteristics, effectively improving the generalization ability and training accuracy of the prior model.
[0066] Figure 7 An example diagram of usable single-source seismic data obtained through embodiments of this disclosure is shown. Figure 8 A residual plot of usable single-source seismic data and ideal data obtained through embodiments of this disclosure is shown. Figure 7 and Figure 8 As can be seen, the embodiments of this disclosure can effectively correct the Doppler effect and separate aliased signals simultaneously.
[0067] Figure 9 A schematic diagram of the marine controllable seismic source synchronous correction and signal separation device provided in an embodiment of this disclosure is shown. See also Figure 9 The marine controllable seismic source synchronous correction and signal separation device provided in this disclosure embodiment may include: The data acquisition unit 901 is used to acquire raw seismic records from the marine data acquisition terminal, the raw seismic records including marine controlled source superimposed data; Data processing unit 902 is used to obtain measured aliased co-detector point gathers using the original seismic record; The unmixing and correction unit 903 is used to minimize a predetermined objective function by embedding a pre-trained prior model into the inversion iterative process in a plug-and-play manner to unmix and correct the measured overlapping co-detector point gathers to obtain usable single-source seismic data. The prior model includes multiple denoisers, each of which employs a convolutional neural network (CNN) structure. The prior model is trained in the following manner: Forward modeling yields simulated aliased co-detector point gathers and their corresponding simulated single-source co-detector point gathers, constructing aliased single-source training pairs; bandpass filtering is used to convert a predetermined natural image dataset into pseudo-seismic data, and Gaussian white noise of different noise levels is added to the pseudo-seismic data to construct pseudo-data training pairs; and the parameters of the prior model are determined using the aforementioned aliased single-source training pairs and pseudo-data training pairs.
[0068] Furthermore, the marine controllable seismic source synchronous correction and signal separation device may also include: a model training unit 904, used to train the prior model in the following manner: Forward modeling yields simulated aliased co-detector point gathers and their corresponding simulated single-source co-detector point gathers, constructing aliased single-source training pairs; filtering techniques are used to convert a predetermined natural image dataset into pseudo-seismic data, and Gaussian white noise of different noise levels is added to the pseudo-seismic data to construct pseudo-data training pairs; and the parameters of the prior model are determined using the aforementioned aliased single-source training pairs and pseudo-data training pairs.
[0069] It should be noted that the model training unit 904 can also be deployed independently of the marine controllable seismic source synchronous correction and signal separation device; that is, the model training unit can be implemented through a separate device. This disclosure does not limit the specific deployment method of the model training unit.
[0070] Furthermore, the model training unit 904 can be specifically used to obtain the pseudo-earthquake data in the following manner: determining a publicly available natural image dataset; converting the image data in the publicly available natural image dataset to the frequency-wavenumber fk domain through a two-dimensional Fourier transform to obtain a two-dimensional complex spectrum; applying a filter to the two-dimensional complex spectrum after removing large-dipping events to obtain a filtered two-dimensional complex spectrum, wherein the filtered two-dimensional complex spectrum retains only the components related to the earthquake data; and performing a two-dimensional inverse Fourier transform on the filtered two-dimensional complex spectrum to obtain the pseudo-earthquake data.
[0071] Furthermore, the model training unit 904 can be used to implement the forward modeling simulation in the following manner: using wave equation finite difference independent simulation of the seismic records containing Doppler phase shifts generated by each of the pre-selected multiple sources at a fixed receiver point to obtain the simulated single-source common receiver point gather; and superimposing the simulated single-source common receiver point gathers within the common receiver point domain of the multiple sources according to a random time delay within a predetermined range to generate the simulated aliased common receiver point gather.
[0072] Furthermore, the demixing correction unit 903 can be specifically used for iterative updates via the following formula: , , in, Indicating the corresponding prior model A noise denoiser, Indicates the first Noise level of the next iteration This represents the Doppler distortion operator. Describes the inverse of the Doppler distortion operator. Represents the aliasing operator. This represents the transpose of the aliasing operator. Represents the identity matrix. Indicates after the first The current solution obtained after the next iteration update; Indicates the first The current solution obtained after the next iteration update. This indicates the measured aliasing common detector gather. This represents the intermediate result obtained after denoising in the k-th iteration. Indicates the iteration number. N represents the total number of iterations, when hour, , This indicates the available single-source seismic data.
[0073] Other technical details regarding the marine controlled seismic source synchronous correction and signal separation device can be found in the preceding methods section and will not be repeated here. In specific applications, the marine controlled seismic source synchronous correction and signal separation device can be implemented as software, hardware, or a combination of both.
[0074] Figure 10 A schematic structural diagram of an electronic device provided according to an embodiment of this disclosure is shown. See also... Figure 10 The electronic device 100 provided in this embodiment may include a processor 101 and a memory 102. The memory 102 stores a computer program. When the computer program is run by the processor 101, the processor 101 performs the aforementioned method for synchronous correction and signal separation of a marine controllable seismic source.
[0075] The processor 101 may be a central processing unit (CPU), a graphics processing unit (GPU), or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the computing unit to perform desired functions.
[0076] The memory 102 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and a processor may execute the program instructions to implement the methods of the various embodiments of this disclosure described above and / or other desired functions.
[0077] Depending on the specific application, the electronic device may also include any other suitable components. For example, see Figure 10 The electronic device may also include a communication component 103, which can be used to communicate with a marine data acquisition terminal.
[0078] In specific applications, this electronic device can be implemented as Figure 1 The data processing terminal in the system shown.
[0079] In addition to the methods and apparatus described above, embodiments of this disclosure may also be computer program products comprising computer program instructions that, when executed by a processor, cause the processor to perform the steps of the methods according to various embodiments of this disclosure as described in the above-described method section of this specification.
[0080] The computer program product can be written in any combination of one or more programming languages to perform the operations of the embodiments of this disclosure. The programming languages include object-oriented programming languages such as Java and C++, as well as conventional procedural programming languages such as C or similar languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.
[0081] Furthermore, embodiments of this disclosure also provide a computer-readable storage medium having a computer program stored thereon, which, when run by a processor, causes the processor to perform the steps in the above-described method for synchronous correction and signal separation of a marine controllable seismic source.
[0082] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.
[0083] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.
Claims
1. A method for synchronous correction and signal separation of a marine controllable seismic source, characterized in that, The method includes: Acquire raw seismic records from a marine data acquisition terminal, the raw seismic records including ocean-controlled source superimposed data; The measured aliased co-detector gathers were obtained using the original seismic records. By embedding a pre-trained prior model into the inversion iterative process in a plug-and-play manner to minimize the predetermined objective function, the measured aliased co-detector point gathers are unmixed and corrected to obtain usable single-source seismic data. The prior model includes multiple denoisers, each of which employs a convolutional neural network (CNN) structure. The prior model is trained in the following manner: Forward modeling yields simulated aliased co-detector point gathers and their corresponding simulated single-source co-detector point gathers, constructing aliased single-source training pairs; bandpass filtering is used to convert a predetermined natural image dataset into pseudo-seismic data, and Gaussian white noise of different noise levels is added to the pseudo-seismic data to construct pseudo-data training pairs; and the parameters of the prior model are determined using the aforementioned aliased single-source training pairs and pseudo-data training pairs.
2. The method according to claim 1, characterized in that, The pseudo-seismic data was obtained in the following way: Identify publicly available natural image datasets; The image data in the publicly available natural image dataset is transformed into the frequency-wavenumber fk domain using a two-dimensional Fourier transform to obtain a two-dimensional complex spectrum; After removing large-dipping events from the two-dimensional complex spectrum, a filter is applied to obtain a filtered two-dimensional complex spectrum, which retains only the components related to the seismic data. The filtered two-dimensional complex spectrum is subjected to a two-dimensional inverse Fourier transform to obtain the pseudo-seismic data.
3. The method according to claim 2, characterized in that, The filtering is implemented using the filter shown in the following formula: ， in, Indicates frequency, Indicates wave number, This indicates the predetermined apparent speed threshold. This represents the filter response.
4. The method according to claim 1, characterized in that, The prior model is trained based on the loss function shown in the following formula: ， in, Represents the loss function. Represents the parameters of the prior model. This represents the total number of training samples. For training sample index, This indicates the first common receiver gather or the first pseudo-seismic data with added noise in the mixed seismic data. Noisy input data This indicates the first point in the corresponding simulated single-source common-receiver gather or clean pseudo-seismic data. A clean target data, Indicates the prior model to The predicted output.
5. The method according to claim 1, characterized in that, The forward modeling yields simulated aliased common-source common-source gathers and their corresponding simulated single-source common-source gathers, including: The wave equation is used to independently simulate the seismic records containing Doppler phase shifts generated by each of a pre-selected multiple seismic sources at a fixed receiver point, in order to obtain the simulated single-source common receiver point gather; Within the common receiver point domain of the multiple seismic sources, the simulated single-source common receiver point gathers are superimposed with random time delays within a predetermined range to generate the simulated aliased common receiver point gathers.
6. The method according to claim 1, characterized in that, The objective function is expressed as follows: ， in, This indicates the available single-source seismic data. This represents the measured aliasing common detector gather. Represents the aliasing operator. This represents the Doppler distortion operator. Represents the regularization parameter. Represents the regularization term, This represents the inverse Fourier transform. This represents the solution that minimizes the objective function.
7. The method according to claim 6, characterized in that, The step of minimizing a predetermined objective function by embedding a pre-trained prior model into the inversion iterative process in a plug-and-play manner to unmix and correct the measured aliased co-detector point gather includes: Iterative updates are performed using the following formula: ， ， in, Indicating the corresponding prior model A noise denoiser, Indicates the first Noise level of the next iteration This represents the Doppler distortion operator. Describes the inverse of the Doppler distortion operator. Represents the aliasing operator. This represents the transpose of the aliasing operator. Represents the identity matrix. Indicates after the first The current solution obtained after the next iteration update; Indicates the first The current solution obtained after the next iteration update. This indicates the measured aliasing common detector gather. This represents the intermediate result obtained after denoising in the k-th iteration. Indicates the iteration number. N represents the total number of iterations, when hour, , This indicates the available single-source seismic data.
8. A marine controllable seismic source synchronous correction and signal separation device, characterized in that, include: The data acquisition unit is used to acquire raw seismic records from the marine data acquisition terminal, the raw seismic records including marine controlled source superimposed data; The data processing unit is used to obtain the measured aliased co-detector point gathers using the original seismic records; The unmixing and correction unit is used to minimize a predetermined objective function by embedding a pre-trained prior model into the inversion iterative process in a plug-and-play manner to unmix and correct the measured overlapping co-detector point gathers to obtain usable single-source seismic data. The prior model includes multiple denoisers, each of which employs a convolutional neural network (CNN) structure. The prior model is trained in the following manner: Forward modeling yields simulated aliased co-detector point gathers and their corresponding simulated single-source co-detector point gathers, constructing aliased single-source training pairs; bandpass filtering is used to convert a predetermined natural image dataset into pseudo-seismic data, and Gaussian white noise of different noise levels is added to the pseudo-seismic data to construct pseudo-data training pairs; and the parameters of the prior model are determined using the aforementioned aliased single-source training pairs and pseudo-data training pairs.
9. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing a computer program that, when executed by the processor, causes the processor to perform the method as described in any one of claims 1-7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to perform the method as described in any one of claims 1-7.