A seismic data denoising model establishing method, a denoising method and related devices

Abstract

The application discloses a seismic data denoising model establishing method, a seismic data denoising method and related devices. The seismic data denoising model establishing method comprises the following steps: based on multiple sets of original seismic data, denoised seismic data corresponding to the original seismic data is obtained by combining a denoising method and a synthetic seismic method according to a set method ratio, as label data, and a sample set is obtained; the original seismic data of each sample in the sample set is subjected to wavelet frequency division processing to obtain multi-band data, and a converted sample set is obtained; a selected deep neural network model is trained by using the converted sample set, and a seismic data denoising model is obtained, which is used for denoising processing of input seismic data. Through the established seismic data denoising model, multiple seismic noises can be suppressed, and on the basis of obviously improving the noise suppression efficiency, the noise recognition precision is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of seismic data processing technology in oil and gas geophysical exploration, and particularly to a method for establishing a seismic data denoising model, a denoising method, and related devices. Background Technology

[0002] Seismic data is inevitably subject to various noise interferences during the acquisition process. In recent years, the exploration environment has become increasingly complex, the exploration difficulty has continued to increase, and the demand for detailed exploration has continued to rise. Therefore, the demand for efficient and high-fidelity noise reduction of seismic data has become increasingly strong.

[0003] Earthquake noise is complex and diverse. Based on its morphology, it can be broadly classified into random noise and coherent noise. For random noise, conventional noise suppression algorithms can be used, including fx-domain predictive filtering, wavelet transform, and curvelet transform. For coherent noise such as surface waves and linear interference, conventional algorithms can be used, including FK filtering, FKK domain filtering, anomalous amplitude attenuation, KL transform, and tilt filtering. The general idea behind these conventional algorithms is to utilize the differences between signal and noise in a specific data domain (time-space domain, frequency-space domain, FK domain, etc.) to achieve signal-to-noise separation.

[0004] Conventional noise suppression algorithms abstract information about noise propagation patterns, physical characteristics, and morphological features into mathematical language, separating signals from noise within a specific data space (such as the frequency domain or curve domain). This mathematical abstraction process typically involves theoretical assumptions. While these assumptions can characterize certain features of seismic data to some extent, they cannot be precisely satisfied, thus limiting the effectiveness of denoising algorithms. Furthermore, conventional denoising algorithms are usually designed for specific types of noise, requiring multiple algorithms to work together in practice. Each algorithm requires parameter tuning, making the noise suppression process highly subjective and inefficient when processing massive datasets characterized by "two-wide-one-high" (wide azimuth, wide bandwidth, high density) data. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide a method for establishing a seismic data denoising model, a denoising method and related apparatus to overcome or at least partially solve the above problems, thereby significantly improving the noise suppression efficiency and effectively enhancing the noise identification accuracy.

[0006] In a first aspect, embodiments of the present invention provide a method for establishing a seismic data denoising model, comprising:

[0007] Based on multiple sets of original seismic data, denoised seismic data corresponding to the original seismic data are obtained by combining denoising methods and synthesizing seismic signals according to a set method ratio. A sample is composed of a set of original seismic data and the corresponding denoised seismic data as label data, resulting in a sample set containing multiple samples.

[0008] The original seismic data of each sample in the sample set is processed by wavelet frequency division to obtain multi-band data, resulting in the transformed sample set.

[0009] The selected deep neural network model is trained using the transformed sample set to obtain a seismic data denoising model, which is used to denoise the input seismic data.

[0010] Secondly, embodiments of the present invention provide a method for denoising seismic data, comprising:

[0011] The original seismic data to be denoised is input into the seismic data denoising model established according to the above method, and the denoised seismic data is obtained according to the output of the model.

[0012] Thirdly, embodiments of the present invention provide an apparatus for establishing a seismic data denoising model, comprising:

[0013] The sample set establishment module is used to obtain denoised seismic data corresponding to the original seismic data by combining denoising methods and synthesizing seismic signals according to a set method ratio based on multiple sets of original seismic data. A sample is composed of a set of original seismic data and the corresponding denoised seismic data as label data, and a sample set containing multiple samples is obtained.

[0014] The sample set conversion and establishment module is used to process the original seismic data of each sample in the sample set through wavelet frequency division to obtain multi-frequency band data and thus obtain the converted sample set.

[0015] The model training module is used to train a selected deep neural network model using the transformed sample set to obtain a seismic data denoising model, which is used to denoise the input seismic data.

[0016] Fourthly, embodiments of the present invention provide a computer program product, including a computer program / instruction, wherein the computer program / instruction, when executed by a processor, implements the above-mentioned method for establishing a seismic data denoising model, or implements the above-mentioned method for denoising seismic data.

[0017] Fifthly, embodiments of this disclosure provide a server, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the above-described seismic data denoising model establishment method, or implements the above-described seismic data denoising method.

[0018] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following:

[0019] (1) The seismic data denoising model establishment method provided in this embodiment of the invention constructs a sample set by combining denoising methods and synthesizing seismic signals; the original seismic data of each sample in the sample set is processed by wavelet frequency division to obtain multi-frequency band data, resulting in a transformed sample set; a selected deep neural network model is trained using the transformed sample set to obtain a seismic data denoising model, which is used to denoise the input seismic data. The established seismic data denoising model can suppress various types of seismic noise, significantly improving noise suppression efficiency and effectively enhancing noise identification accuracy.

[0020] (2) Conventional combined denoising methods have relatively poor denoising effects, but good amplitude preservation; synthetic seismic signal methods have relatively good denoising effects, but they suppress high-frequency information and have poor amplitude preservation. The seismic data denoising model establishment method provided in this embodiment of the invention constructs a sample set by combining denoising methods and synthetic seismic signal methods, which ensures the denoising fidelity and denoising effect of the final seismic data denoising model.

[0021] (3) The earthquake data denoising model establishment method provided in this embodiment of the invention does not directly train the selected deep neural network model using the sample set. Instead, it first performs wavelet frequency division processing on the original earthquake data of each sample in the sample set to achieve multi-band learning of data features, which makes the model learning effect better and can better denoise earthquake data.

[0022] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0023] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0024] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0025] Figure 1 This is a flowchart of the method for establishing a seismic data denoising model in Embodiment 1 of the present invention;

[0026] Figure 2 This is a schematic diagram of the seismic phase axis picked up manually in Embodiment 1 of the present invention;

[0027] Figure 3 This is a schematic diagram illustrating the dominant frequency and amplitude of local seismic data in Embodiment 1 of the present invention;

[0028] Figure 4 This is a schematic diagram of the process for creating a sample set in Embodiment 1 of the present invention;

[0029] Figure 5 This is a schematic diagram of the multi-band semi-supervised seismic denoising network in Embodiment 1 of the present invention;

[0030] Figure 6 This is a schematic diagram of the multi-band semi-supervised seismic denoising network in Embodiment 1 of the present invention;

[0031] Figure 7 The simulated earthquake data is from Embodiment 2 of the present invention;

[0032] Figure 8 This is the effective signal of the simulated earthquake data in Embodiment 2 of the present invention;

[0033] Figure 9 This is the denoising result in Embodiment 2 of the present invention;

[0034] Figure 10 This refers to the actual noise in the simulated earthquake data in Embodiment 2 of the present invention;

[0035] Figure 11 The noise identified in Embodiment 2 of the present invention;

[0036] Figure 12 This refers to actual single-shot data from a work area in Embodiment 3 of the present invention;

[0037] Figure 13 for Figure 12 The denoising results of the conventional composite process for actual single-shot data are shown below.

[0038] Figure 14 for Figure 12 The actual single-shot data shown is the denoising result of Embodiment 3 of the present invention;

[0039] Figure 15 for Figure 12 The noise identified by the conventional composite process for denoising actual single-shot data is shown below.

[0040] Figure 16 for Figure 12 The noise identified in the denoising of the actual single-shot data shown in Embodiment 3 of this invention;

[0041] Figure 17 This is a seismic overlay profile of a certain block before denoising in Embodiment 4 of the present invention;

[0042] Figure 18 for Figure 17The denoising results of the conventional composite process for the earthquake stacked profile are shown.

[0043] Figure 19 for Figure 17 The denoising result of the earthquake superimposed profile shown in Embodiment 4 of the present invention;

[0044] Figure 20 This is a schematic diagram of the structure of the seismic data denoising model establishment device in an embodiment of the present invention. Detailed Implementation

[0045] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0046] It should be understood that the terminology used in this invention is merely for describing particular embodiments and is not intended to limit the invention. Furthermore, with respect to numerical ranges in this invention, it should be understood that each intermediate value between the upper and lower limits of the range is also specifically disclosed. Every smaller range between any stated value or intermediate value within a stated range, and any other stated value or intermediate value within said range, is also included in this invention. The upper and lower limits of these smaller ranges may be independently included or excluded from the range.

[0047] Unless otherwise stated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. While only preferred methods and materials have been described herein, any methods and materials similar or equivalent to those described herein may be used in the implementation or testing of this invention. All references to this specification are incorporated by way of citation to disclose and describe methods and / or materials associated with those references. In the event of any conflict with any incorporated reference, the content of this specification shall prevail.

[0048] Deep learning, a rapidly developing machine learning algorithm in recent years, has demonstrated its powerful potential for nonlinear representation and deep data mining through revolutionary successes in fields such as image classification and speech recognition. In the field of seismic exploration, deep learning technology has been rapidly applied in areas such as seismic fault identification, first arrival picking, and attribute recognition. This invention's embodiments utilize deep learning for seismic denoising.

[0049] To address the problems of low noise identification accuracy, poor adaptability, and low efficiency in existing seismic data noise suppression techniques, this invention provides a method for establishing a seismic data denoising model, a denoising method, and related devices. This method can suppress various types of seismic noise, significantly improving noise suppression efficiency while effectively enhancing noise identification accuracy.

[0050] Example 1

[0051] Embodiment 1 of the present invention provides a method for establishing a seismic data denoising model, the process of which is as follows: Figure 1 As shown, it includes the following steps:

[0052] Step S11: Based on multiple sets of original seismic data, denoised seismic data corresponding to the original seismic data are obtained by combining denoising methods and synthesizing seismic signals according to a set method ratio. A sample is formed by using a set of original seismic data and the corresponding denoised seismic data as label data, resulting in a sample set containing multiple samples.

[0053] Raw seismic data refers to data that has been denoised, as opposed to denoised seismic data. Raw seismic data is not limited to the most original acquired seismic data; any seismic data containing noise that needs to be suppressed can be considered raw seismic data. A set of raw seismic data can be considered raw single-shot seismic data.

[0054] The method ratio can be set to 1:1, meaning that 50% of the original seismic data from multiple sets are obtained by combining denoising methods, and 50% are obtained by synthesizing seismic signals. Optionally, the method ratio can also be other ratios, which are not limited in this embodiment.

[0055] I. Denoising seismic data corresponding to the original seismic data is obtained by combining denoising methods.

[0056] The actual seismic data is denoised using conventional seismic noise suppression techniques (a combination of techniques such as anomalous amplitude suppression, FKK domain filtering, and tilt filtering). The denoised seismic data is then used as label data, forming a sample dataset together with the original seismic data before denoising.

[0057] 2. Denoising seismic data corresponding to the original seismic data is obtained by synthesizing seismic signals.

[0058] 1. Calibrate the phase axis of the original seismic data and pick up the dominant frequency data and amplitude data of the original seismic data.

[0059] Visible seismic phase axes (e.g., in-shot) in the original seismic data can be calibrated through manual interaction. Figure 2 ), while picking up the dominant frequency and amplitude of the original seismic data (e.g. Figure 3 ).

[0060] 2. Based on the relevant information of the phase axis, as well as the main frequency data and amplitude data, the seismic signal is synthesized through numerical simulation and used as denoised seismic data.

[0061] (1) Using the equivalent velocity and equivalent time of the phase axis and the shot-receiver distance of the original seismic data, the arrival time data of the seismic signal is obtained according to the propagation law of the reflected signal.

[0062] The arrival time data of the seismic signal is obtained using the following formula (1):

[0063]

[0064] In formula (1), t is the arrival time of the seismic signal, v is the equivalent velocity, t0 is the equivalent time, and x is the shot-receiver distance, which can be calculated based on the coordinates of the shot point and the receiver point in the seismic trace.

[0065] (2) Based on the arrival time of each signal in the arrival time data and the amplitude in the amplitude data corresponding to that time, determine the reflection coefficient and obtain the reflection coefficient data.

[0066] Based on the arrival times of the aforementioned seismic signals, the amplitude of the seismic signals is placed at the corresponding time points (this value is determined by manually picking 'a' and the random dynamic range 'ε'). a (Determined jointly), the reflection coefficient data is obtained:

[0067] r(t)=a(t)+ε a (2)

[0068] In formula (2), r(t) is the reflection coefficient at time t, a(t) is the amplitude at time t in the amplitude data, and ε a For the random dynamic coefficient of amplitude.

[0069] (3) Construct sub-wavelengths based on the main frequency data.

[0070] Construct a Rayleigh wavelet, a zero-phase wavelet, or a mixed-phase wavelet based on the master frequency data.

[0071] This invention uses three wavelets for numerical simulation:

[0072] i. Lake wavelet:

[0073] ii. Zero-phase wavelet:

[0074] iii. Mixed phase wavelet:

[0075] In formulas (3)-(5), f0 is the wavelet dominant frequency, which is the dominant frequency f in the picked dominant frequency data. p and random dynamic range ε fJoint decision:

[0076] f0 = f p +ε f (6)

[0077] When calculating synthetic seismic signals from different raw seismic data, the Rick wavelet, zero-phase wavelet, and mixed-phase wavelet can be used respectively to ensure that the resulting sample set contains labeled data obtained from various wavelets, thus eliminating the influence of wavelet type.

[0078] (4) Perform convolution operation on the wavelet and reflection coefficient data to synthesize the seismic signal.

[0079] See Figure 4 The diagram shown illustrates the overall process for creating a simulated data sample set.

[0080] In some embodiments, the test set and training set can be divided into a certain proportion, for example, the test set accounts for 10% of the total number of samples.

[0081] Step S12: The original seismic data of each sample in the sample set is processed by wavelet frequency division to obtain multi-band data, and the transformed sample set is obtained.

[0082] Step S13: Train the selected deep neural network model using the transformed sample set to obtain the seismic data denoising model, which is used to denoise the input seismic data.

[0083] In some embodiments, a wavelet frequency division processing module may be embedded in a selected deep neural network model. The deep neural network may be a DNN network, and the deep neural network model with the embedded wavelet frequency division processing module is, in terms of characteristics and function, a multi-band semi-supervised seismic denoising network model. Its network is as follows: Figure 5 As shown.

[0084] The network structure of the multi-band semi-supervised seismic denoising network model is as follows: Figure 6 As shown, this is a U-shaped network structure, which includes a feature extraction layer, a downsampling layer, an upsampling layer, and an output layer.

[0085] The downsampling layer is a Haar wavelet decomposition layer, and the upsampling layer is a Haar wavelet reconstruction layer;

[0086] The feature extraction layer contains three sets of convolutional layers with 64 channels and a filter size of 3. Each set of convolutional layers contains the ReLU activation function.

[0087] The output layer contains a set of convolutional layers with 1 channel and a filter size of 3.

[0088] In addition, skip connections are added between feature maps of the same scale to avoid feature loss.

[0089] After inputting the sample set into the multi-band semi-supervised seismic denoising network model, the original seismic data in the sample set is used as the network input data. After wavelet frequency division, multi-band data is obtained, which is then used as the multi-channel input of the deep neural network. After processing by the deep neural network, the denoised seismic data is obtained.

[0090] The training process of a deep neural network model is an iterative optimization process. The network parameters of the selected deep neural network model are iteratively optimized using the transformed sample set until the number of iterations reaches the set maximum number of iterations. At this point, the pre-constructed loss function also reaches its minimum value.

[0091] During the iterative optimization of network parameters, the current values ​​of the network parameters can be adjusted using the backpropagation algorithm based on the loss function, which is:

[0092]

[0093] In formula (7), l(Θ) is the loss function, Θ is the current value of the network parameters; K is the number of samples in the transformed sample set; {x k ,y k} represents a set of training samples, x k The input data is the multi-band data of the k-th sample. y k These represent the current network output and label data of the k-th sample in the transformed sample set, respectively; ||·|| F The F-norm, which is the square root of the sum of the squares of all terms in the matrix, can be represented as: μ is the regularization parameter, with a default value of 0.01; LSSIM() is the similarity function, which minimizes the network's denoising results. With noise removal Local similarity, without the need for label data supervision, can help improve denoising accuracy and reduce damage to the effective signal.

[0094] Furthermore, the aforementioned similarity function is a custom similarity function defined in this embodiment. The definition of the similarity function is as follows:

[0095]

[0096]

[0097]

[0098]

[0099] In formulas (8)-(11), LSSIM(x,y) represents the similarity between the two sets of data x and y, M is the number of local windows, and the default size of the local window can be 10*10; σ xy，i Let σ be the covariance of the i-th local window of x and y. x，i and σ y，i x represents the i-th local window i y i The standard deviation of the data, N is the standard deviation of the i-th local window x. i or y i The number of data in the data, and x represents the i-th local window i y i The mean of the data, x i，j and y i，j x represents the i-th local window i y i The j-th data in the dataset.

[0100] The parameters of the multi-band semi-supervised seismic denoising network are randomly initialized to obtain the initial model of the multi-band semi-supervised seismic denoising network. The training set is input into the initial model of the multi-band semi-supervised seismic denoising network to obtain the output layer data. The loss function shown in formula (7) is calculated, and it is determined whether the current iteration number meets the maximum training number (the default value can be 50). If it does not meet the maximum training number, the network parameters of the current network are adjusted by the backpropagation algorithm until the maximum training number is reached. The network model that has reached the maximum training number is determined as the multi-band semi-supervised seismic denoising network model. The backpropagation algorithm of the multi-band semi-supervised seismic denoising network is the process of obtaining the optimal network parameters by minimizing the loss function shown in formula (7). The minimization process of the loss function can be implemented by the Adam optimization algorithm.

[0101] The seismic data denoising model establishment method provided in Embodiment 1 of this invention constructs a sample set by combining denoising methods and synthetic seismic signal methods. The original seismic data of each sample in the sample set is processed by wavelet frequency division to obtain multi-frequency band data, resulting in a transformed sample set. A selected deep neural network model is trained using the transformed sample set to obtain the seismic data denoising model, which is used to denoise the input seismic data. The established seismic data denoising model can suppress various types of seismic noise, significantly improving noise suppression efficiency and effectively enhancing noise identification accuracy.

[0102] Conventional combined denoising methods have relatively poor denoising effects but good amplitude preservation; synthetic seismic signal methods have relatively good denoising effects but high suppression of high-frequency information and poor amplitude preservation. The seismic data denoising model establishment method provided in this embodiment of the invention constructs a sample set by combining denoising methods and synthetic seismic signal methods, thus ensuring the denoising fidelity and denoising effect of the final seismic data denoising model.

[0103] The earthquake data denoising model establishment method provided in Embodiment 1 of the present invention does not directly train a selected deep neural network model using a sample set. Instead, it first performs wavelet frequency division processing on the original earthquake data of each sample in the sample set to achieve multi-frequency band learning of data features, which makes the model learning effect better and can better denoise earthquake data.

[0104] By redefining the structural similarity parameter and introducing a loss function to achieve semi-supervised learning of seismic data features, this embodiment of the invention achieves the suppression of various types of seismic noise compared to conventional noise suppression techniques.

[0105] Example 2

[0106] Embodiment 2 of the present invention provides a method for denoising seismic data, comprising:

[0107] The original seismic data to be denoised is input into the seismic data denoising model established according to the above method, and the denoised seismic data is obtained according to the output of the model.

[0108] In some embodiments, the seismic data denoising model is embedded with a wavelet frequency division processing module, which is used to perform wavelet frequency division processing on the original seismic data to be denoised to obtain multi-frequency band data.

[0109] The actual earthquake data is input into the earthquake data denoising model mentioned above, namely the multi-frequency band semi-supervised earthquake denoising network model, to obtain the data after noise suppression. Figure 7 To simulate earthquake data, Figure 8 To simulate effective signals from seismic data, Figure 9 The noise reduction results of Embodiment 2 of the present invention are very similar, which proves the effectiveness of noise suppression in Embodiment 2 of the present invention. Figure 10 To simulate the real noise in seismic data, Figure 11 The noise identified in Embodiment 2 of the present invention has a very small difference with the noise, and there is almost no effective signal in the noise, which proves that Embodiment 2 of the present invention has high amplitude preservation while effectively suppressing noise.

[0110] Example 3

[0111] Example 3 is an application case of Block A in an oil field. The noise is mainly surface waves, which affects the further processing and interpretation of seismic data. Figure 12This is single-shot data from a specific exploration block. The surface wave scattering phenomenon is quite severe, making noise reduction difficult. Figure 13 The result is a composite denoising effect using multiple conventional denoising techniques, including FKK filtering and anomalous amplitude suppression. Conventional seismic denoising requires experts to design denoising processes based on experience, and each denoising module needs repeated parameter adjustments, resulting in relatively low efficiency. Furthermore, for surface waves with severe scattering, although conventional methods can remove most of the noise, significant noise residue remains. Figure 14 The denoising results of this embodiment of the invention show that the denoising effect is significantly improved, and no parameter adjustment or process design is required, resulting in higher denoising efficiency. Figure 15 and Figure 16 The noise identified by the conventional composite process and by the method described above both showed no significant signal residue. This indicates that the method of the present invention can identify noise efficiently and accurately, and has high amplitude preservation.

[0112] Example 4

[0113] This fourth example is an application case of Block B in an oilfield. The noise is mainly 50Hz interference, abnormal amplitude and surface wave, and the signal-to-noise ratio of the stacked data is low. Figure 17 This is the superimposed profile of this exploration block before noise reduction. Figure 18 The result is a composite denoising effect using multiple conventional denoising techniques, including FKK filtering, abnormal amplitude suppression, and multi-channel tilt filtering. Figure 19 The denoising results of this invention show a significant improvement in denoising effect, with more obvious seismic phase axes, proving that this invention has high denoising fidelity.

[0114] Based on the inventive concept of this invention, embodiments of this invention also provide a seismic data denoising model establishment device, the structure of which is as follows: Figure 20 As shown, it includes:

[0115] The sample set establishment module 201 is used to obtain denoised seismic data corresponding to the original seismic data by combining denoising methods and synthesizing seismic signals according to a set method ratio based on multiple sets of original seismic data. A sample is composed of a set of original seismic data and the corresponding denoised seismic data as label data, and a sample set containing multiple samples is obtained.

[0116] The sample set conversion and establishment module 202 is used to process the original seismic data of each sample in the sample set through wavelet frequency division to obtain multi-frequency band data and thus obtain the converted sample set.

[0117] The model training module 203 is used to train a selected deep neural network model using the transformed sample set to obtain a seismic data denoising model, which is used to denoise the input seismic data.

[0118] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0119] Based on the inventive concept of the present invention, embodiments of the present invention also provide a computer program product, including a computer program / instruction, wherein the computer program / instruction, when executed by a processor, implements the above-mentioned method for establishing a seismic data denoising model, or implements the above-mentioned method for denoising seismic data.

[0120] Based on the inventive concept of the present invention, embodiments of the present invention also provide a server, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the above-mentioned method for establishing a seismic data denoising model, or implements the above-mentioned method for denoising seismic data.

[0121] Unless otherwise specifically stated, terms such as processing, calculation, operation, determination, display, etc., may refer to the actions and / or processes of one or more processing or computing systems or similar devices that represent the manipulation and conversion of data representing physical (e.g., electronic) quantities within the registers or memory of the processing system into other data similarly representing physical quantities within the memory, registers, or other such information storage, transmission, or display devices of the processing system. Information and signals can be represented using any of a variety of different techniques and methods. For example, data, instructions, commands, information, signals, bits, symbols, and chips mentioned throughout the above description can be represented by voltage, current, electromagnetic waves, magnetic fields or particles, light fields or particles, or any combination thereof.

[0122] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process may be rearranged without departing from the scope of this disclosure. The appended method claims provide elements of various steps in an exemplary order and are not intended to limit the scope to the specific order or hierarchy described.

[0123] In the detailed description above, various features are combined together in a single embodiment to simplify this disclosure. This approach to disclosure should not be construed as reflecting an intention that embodiments of the claimed subject matter require more features than are explicitly stated in each claim. Rather, as reflected in the appended claims, the invention is presented with fewer features than all of the features in a single disclosed embodiment. Therefore, the appended claims are hereby explicitly incorporated into the detailed description, with each claim representing a separate preferred embodiment of the invention.

[0124] Those skilled in the art will also understand that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments herein can be implemented as electronic hardware, computer software, or a combination thereof. To clearly illustrate the interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps described above are generally described in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Those skilled in the art can implement the described functionality in alternative ways for each specific application; however, such implementation decisions should not be construed as departing from the scope of this disclosure.

[0125] The steps of the methods or algorithms described in conjunction with the embodiments herein can be directly embodied in hardware, software modules executed by a processor, or a combination thereof. The software modules can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. The ASIC can reside in a user terminal. Alternatively, the processor and storage medium can exist as discrete components in the user terminal.

[0126] For software implementation, the techniques described in this application can be implemented using modules (e.g., procedures, functions, etc.) that perform the functions described in this application. This software code can be stored in memory units and executed by a processor. The memory units can be implemented within the processor or outside the processor; in the latter case, they are communicatively coupled to the processor via various means, as is well known in the art.

[0127] The foregoing description includes examples of one or more embodiments. It is certainly impossible to describe all possible combinations of components or methods in order to describe the above embodiments, but those skilled in the art will recognize that further combinations and arrangements of the various embodiments are possible. Therefore, the embodiments described herein are intended to cover all such changes, modifications, and variations that fall within the scope of the appended claims. Furthermore, the term "comprising" as used in the specification or claims is interpreted in a manner similar to the term "including," as interpreted when used as a conjunction in the claims. Additionally, the use of any term "or" in the specification of the claims is intended to mean "non-exclusive or."

Claims

1. A method for establishing a seismic data denoising model, characterized in that, include: Based on multiple sets of original seismic data, denoised seismic data corresponding to the original seismic data are obtained by combining denoising methods and synthesizing seismic signals according to a set method ratio. A sample is composed of a set of original seismic data and the corresponding denoised seismic data as label data, resulting in a sample set containing multiple samples. The original seismic data of each sample in the sample set is processed by wavelet frequency division to obtain multi-band data, resulting in the transformed sample set. The network parameters of the selected deep neural network model are iteratively optimized using the transformed sample set until the maximum number of iterations is reached, resulting in a seismic data denoising model, which is used to denoise the input seismic data. During the iterative optimization of the network parameters, the current values ​​of the network parameters are adjusted using a backpropagation algorithm based on a loss function, which is: （1） In formula (1), For loss function, Here, K represents the current value of the network parameters, and K represents the number of samples in the transformed sample set. , These are the current network output and label data of the k-th sample in the transformed sample set, respectively. This indicates taking the F-norm. Here, LSSIM() is the regularization parameter, and LSSIM() is the similarity function. This represents the multi-band data of the k-th sample.

2. The method as described in claim 1, characterized in that, The similarity function is defined as follows: （2） （3） （4） （5） In formulas (2)-(5), Let M represent the similarity between two sets of data, x and y, and M be the number of local windows. Let x be the covariance of the i-th local window for x and y. and These are the i-th local windows. , The standard deviation of the data, where N is the standard deviation of the i-th local window. or The number of data in the data, and These are the i-th local windows. , The mean of the data. and These are the i-th local windows. , The j-th data in the dataset.

3. The method as described in claim 1, characterized in that, The deep neural network model adopts a U-shaped network structure, which includes a feature extraction layer, a downsampling layer, an upsampling layer, and an output layer; The downsampling layer is a wavelet decomposition layer, and the upsampling layer is a wavelet reconstruction layer.

4. The method as described in claim 3, characterized in that, The feature extraction layer contains three sets of convolutional layers with 64 channels and a filter size of 3, and each set of convolutional layers contains the ReLU activation function; The output layer contains a set of convolutional layers with 1 channel and a filter size of 3.

5. The method as described in claim 1, characterized in that, Denoising seismic data corresponding to the original seismic data is obtained through synthetic seismic signal methods, specifically including: Calibrate the phase axis of the original seismic data, and pick up the dominant frequency data and amplitude data of the original seismic data; Based on the relevant information of the phase axis, the main frequency data, and the amplitude data, a seismic signal is synthesized through numerical simulation to serve as denoised seismic data.

6. The method as described in claim 5, characterized in that, The process of synthesizing seismic signals through numerical simulation based on the relevant information of the phase axis, the dominant frequency data, and the amplitude data specifically includes: Using the equivalent velocity, equivalent time, and shot-receiver distance of the original seismic data along the same phase axis, the arrival time data of the seismic signal is obtained according to the propagation law of the reflected signal. Based on the arrival time of each signal in the arrival time data and the amplitude in the amplitude data corresponding to that time, the reflection coefficient is determined, and the reflection coefficient data is obtained. Construct sub-wavelengths based on the main frequency data; The wavelet and the reflection coefficient data are convolved to synthesize a seismic signal.

7. The method as described in claim 6, characterized in that, The construction of the sub-wavelength based on the main frequency data specifically includes: Based on the main frequency data, a Lake wavelet, a zero-phase wavelet, or a mixed-phase wavelet can be constructed.

8. The method according to any one of claims 1 to 7, characterized in that, The raw seismic data refers to raw single-shot seismic data.

9. A method for denoising seismic data, characterized in that, include: The original seismic data to be denoised is input into the seismic data denoising model established according to any one of claims 1 to 8, and the denoised seismic data is obtained according to the output of the model.

10. The method as described in claim 9, characterized in that, The seismic data denoising model is embedded with a wavelet frequency division processing module, which is used to perform wavelet frequency division processing on the original seismic data to be denoised, to obtain multi-frequency band data.

11. A device for establishing a seismic data denoising model, characterized in that, include: The sample set establishment module is used to obtain denoised seismic data corresponding to the original seismic data by combining denoising methods and synthesizing seismic signals according to a set method ratio based on multiple sets of original seismic data. A sample is composed of a set of original seismic data and the corresponding denoised seismic data as label data, and a sample set containing multiple samples is obtained. The sample set conversion and establishment module is used to process the original seismic data of each sample in the sample set through wavelet frequency division to obtain multi-frequency band data and thus obtain the converted sample set. The model training module is used to iteratively optimize the network parameters of the selected deep neural network model using the transformed sample set until the number of iterations reaches the set maximum number of iterations, thereby obtaining the seismic data denoising model, which is used to denoise the input seismic data. During the iterative optimization of the network parameters, the current values ​​of the network parameters are adjusted using a backpropagation algorithm based on a loss function, which is: （1） In formula (1), For loss function, Here, K represents the current value of the network parameters, and K represents the number of samples in the transformed sample set. , These are the current network output and label data of the k-th sample in the transformed sample set, respectively. This indicates taking the F-norm. Here, LSSIM() is the regularization parameter, and LSSIM() is the similarity function. This represents the multi-band data of the k-th sample.

12. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the seismic data denoising model establishment method according to any one of claims 1 to 8, or the seismic data denoising method according to claim 9 or 10.

Images