Seismic data denoising method, control device and storage medium

Abstract

The embodiment of the invention provides a seismic data denoising method, a control device and a storage medium, and belongs to the technical field of oil-gas exploration. The method comprises the following steps: acquiring actual seismic data of a preset area; and performing noise suppression processing on the actual seismic data through a two-channel five-dimensional denoising model to obtain a denoised seismic data volume, the two-channel five-dimensional denoising model being obtained by training a two-channel five-dimensional denoising neural network, the two-channel five-dimensional denoising neural network comprises a Fourier transform layer, a five-dimensional denoising neural network layer and an inverse Fourier transform layer, the output channel of the Fourier transform layer is two channels, and the input channel of the inverse Fourier transform layer is two channels. High-dimensional information of'height-width-one-height 'data is fully utilized, and a two-channel five-dimensional denoising neural network is trained based on a five-dimensional space (spatial three-dimensional, offset and azimuth) and frequency real and imaginary part two channels, so that the characteristics of noise samples are learned, and the noise recognition precision of a two-channel five-dimensional denoising model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas exploration, and in particular to a method, a control device and a storage medium for seismic data denoising. Background Art

[0002] In recent years, the target of oil and gas exploration has been continuously shifting to complex structures, stratigraphic lithology, carbonate rocks and other fields. The application scale of "two wides and one high" (i.e., "wide azimuth, wide bandwidth, high density") exploration technology has grown rapidly. The problem that follows is that the volume of seismic data has grown exponentially, while the data quality (for example, signal-to-noise ratio) has stagnated. Although massive seismic data contains richer underground information, it has brought unprecedented challenges to processing technology. Affected by the acquisition method, complex underground structures, etc., seismic noise presents complex and diverse characteristics. These noises seriously affect the imaging quality of seismic data, and thus mislead reservoir prediction and well deployment.

[0003] According to the noise form, seismic noise can be roughly divided into random noise and coherent noise. For random sound, the existing noise suppression algorithms include fx domain prediction filtering, wavelet transform, curvelet transform, etc.; for coherent noise such as surface waves and linear interference, the existing algorithms include FK filtering, FKK domain filtering, abnormal amplitude attenuation, KL transform, dip filtering, etc. The existing noise suppression algorithms abstract the propagation law, physical characteristics, morphological characteristics and other information of noise into mathematical language, and complete the separation of signal and noise in the data space (for example, frequency domain, curvelet domain, etc.). Although this type of noise suppression method based on mathematical models has a certain noise suppression effect, the disadvantages of the widely used "two widths and one height" data include: 1. In the process of mathematical abstraction, theoretical assumptions are usually required; although the conditions of the theoretical assumptions can characterize certain characteristics of seismic data to a certain extent, they cannot be accurately met in actual data processing, especially for low signal-to-noise ratio data, which leads to the limitations of the denoising algorithm. 2. Conventional denoising algorithms are usually designed for a specific type of noise. In actual production, multiple denoising algorithms are required to complete the task. Each denoising algorithm requires parameter debugging. The noise suppression process is highly subjective. When processing massive data represented by "two wide and one high" seismic data, the efficiency is relatively low. Summary of the invention

[0004] The purpose of an embodiment of the present invention is to provide a method for seismic data denoising, which addresses the problems of low noise recognition accuracy, poor adaptability, and inability to fully utilize "two wide and one high" seismic data information in existing noise suppression technologies.

[0005] In order to achieve the above-mentioned purpose, an embodiment of the present invention provides a method for seismic data denoising, and the method for seismic data denoising comprises: acquiring actual seismic data of a preset area; and performing noise suppression processing on the actual seismic data through a dual-channel five-dimensional denoising model to obtain a denoised seismic data body, the dual-channel five-dimensional denoising model is obtained by training a dual-channel five-dimensional denoising neural network, the dual-channel five-dimensional denoising neural network comprises a Fourier transform layer, a five-dimensional denoising neural network layer and an inverse Fourier transform layer, the output channel of the Fourier transform layer is dual-channel, and the input channel of the inverse Fourier transform layer is dual-channel.

[0006] Optionally, training the dual-channel five-dimensional denoising neural network includes: based on the three-dimensional spatial, offset and azimuth five-dimensional attributes of the seismic data, constructing a five-dimensional data set through sample seismic data and / or simulated seismic data, and dividing the five-dimensional data set into a five-dimensional training set and a five-dimensional verification set; and constructing the dual-channel five-dimensional denoising neural network, training the dual-channel five-dimensional denoising neural network through the five-dimensional training set, and verifying the dual-channel five-dimensional denoising neural network through the five-dimensional verification set, to obtain a trained dual-channel five-dimensional denoising model.

[0007] Optionally, the data for constructing the five-dimensional data set through the sample seismic data composition includes: denoising the sample seismic data through a preset seismic noise suppression strategy, using the denoised sample seismic data as a label signal, using the sample seismic data as input data, and constructing the data for the five-dimensional data set through the input data and the label data composition.

[0008] Optionally, the data for constructing the five-dimensional data set through the simulated seismic data include: calculating the arrival time and longitudinal wave reflection coefficient of the seismic signal from the shot point to the receiving point based on the simulated seismic data to generate a simulated seismic signal; generating simulated noise data by simulating random noise, using the simulated seismic signal as a label signal, using the simulated seismic signal and the simulated random noise as input data, and constructing the data for the five-dimensional data set through the input data and the label data.

[0009] Optionally, the preset seismic noise suppression strategy includes one or a combination of abnormal amplitude suppression, FKK domain filtering, and dip filtering.

[0010] Optionally, the calculation of the arrival time of the seismic signal from the shot point to the receiving point includes: acquiring the coordinates of the shot point, the coordinates of the receiving point, the longitudinal wave velocity and the azimuth in the simulated seismic data; and calculating the arrival time of the seismic signal based on a given time by using the acquired coordinates of the shot point, the coordinates of the receiving point, the longitudinal wave velocity and the azimuth.

[0011] Optionally, generating a simulated seismic signal according to the seismic signal arrival time and the longitudinal wave reflection coefficient includes: calculating the reflection coefficient r(t) of the formation placed at the corresponding time point when the seismic signal arrives at time t by the following formula:

[0012]

[0013] Among them, a(t) is the amplitude fitting vector; e(t) is the dynamic adjustment parameter, R PP is the longitudinal wave reflection coefficient; θ is the incident angle of seismic longitudinal wave; is the earthquake P-wave azimuth;

[0014] The simulated seismic signal is calculated by convolution of the reflection coefficient r(t) and the seismic wavelet.

[0015] Optionally, the generating of simulated noise data based on simulated random noise includes: performing low-pass filtering on Gaussian noise data generated by simulating random noise by using a Ricker wavelet having a main frequency of a preset frequency to generate the simulated noise data.

[0016] Optionally, the five-dimensional denoising neural network layer includes multiple layers of five-dimensional convolutional layers, and the construction of each five-dimensional convolutional layer includes: based on given five-dimensional arrays f and g, defining the five-dimensional convolutional layer:

[0017]

[0018] Among them, j1, j2, j3, j4, j5 represent the elements of array f, i1-j1, i2-j2, i3-j3, i4-j4, i5-j5 represent the elements of array g;

[0019] Transform the defined five-dimensional convolutional layer:

[0020]

[0021] Optionally, the five-dimensional denoising neural network layer includes 7 five-dimensional convolutional layers, the first layer of the five-dimensional convolutional layer includes a set of five-dimensional convolutions with an input channel number of 2, an output channel number of 64, and a filter size of 3, the second to sixth layers respectively include a set of five-dimensional convolutions with an input channel number of 64, an output channel number of 64, and a filter size of 3 and a ReLU activation function, and the seventh layer includes a set of five-dimensional convolutional layers with an input channel number of 64, an output channel number of 2, and a filter size of 3.

[0022] Optionally, the loss function of the dual-channel five-dimensional denoising neural network is expressed by the following formula:

[0023]

[0024] Where K is the number of samples in the five-dimensional training set, {x k ,y k} is a set of training sample data, x k To simulate the noise data, y k is the label data, Θ represents the network parameters; ||·|| F is the Frobenius norm.

[0025] An embodiment of the present invention also provides a control device for seismic data denoising, the control device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the computer program to implement the above-mentioned method for seismic data denoising.

[0026] An embodiment of the present invention further provides a machine-readable storage medium, on which instructions are stored, and the instructions enable a machine to execute the above-mentioned method for seismic data denoising.

[0027] Through the above technical solution, the embodiment of the present invention uses a dual-channel five-dimensional denoising model to suppress noise in actual seismic data, fully utilize the high-dimensional information of seismic data, and effectively improve the recognition accuracy of noise. By constructing a dual-channel five-dimensional denoising neural network, the high-dimensional information of "high-width-high" data is fully utilized, based on five-dimensional space (three-dimensional space, offset, azimuth) and dual channels of real and imaginary frequency parts, the characteristics of noise samples in the five-dimensional training set are trained and learned to improve the noise recognition accuracy of the dual-channel five-dimensional denoising model, provide high-quality basic data for subsequent seismic imaging and reservoir prediction, and help improve the quality and efficiency of oil and gas exploration and processing.

[0028] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0030] Figure 1 is a flow chart of a method for seismic data denoising provided by an embodiment of the present invention;

[0031] Figure 2 This is a schematic diagram of an example dual-channel five-dimensional denoising neural network architecture;

[0032] Figure 3 is a schematic diagram of the composition of an example five-dimensional data set;

[0033] Figure 4This is a schematic diagram of the noisy data of a single shot in Example 1;

[0034] Figure 5 is a schematic diagram of an effective seismic signal of Example 1;

[0035] Figure 6 is a schematic diagram of the denoising result based on a two-dimensional convolutional neural network in Example 1;

[0036] Figure 7 is a schematic diagram of a denoising result based on an embodiment of the present invention in Example 1;

[0037] Figure 8 This is a schematic diagram of Example 1 for noise recognition based on a two-dimensional convolutional neural network;

[0038] Fig. 9 is a schematic diagram of Example 1 for identifying noise based on an embodiment of the present invention;

[0039] Fig.10 Example 2: Schematic diagram of the noisy part of the data of a single shot;

[0040] Fig.11 Example 2: Schematic diagram of some valid seismic signals;

[0041] Fig.12 This is a partial schematic diagram of the denoising result of Example 2 based on a two-dimensional convolutional neural network;

[0042] Fig.13 is a partial schematic diagram of the denoising result of Example 2 based on an embodiment of the present invention;

[0043] Fig.14 is a partial schematic diagram of Example 2 for noise recognition based on a two-dimensional convolutional neural network; and

[0044] Fig.15 It is a partial schematic diagram of Example 2 for identifying noise based on an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0046] Figure 1 This is a flow chart of a method for seismic data denoising provided by an embodiment of the present invention. Please refer to Figure 1 , the method for seismic data denoising may include the following steps:

[0047] Step S110: Acquire actual earthquake data of a preset area.

[0048] Step S120: performing noise suppression processing on the actual seismic data through a dual-channel five-dimensional denoising model to obtain a denoised seismic data volume.

[0049] Among them, the dual-channel five-dimensional denoising model is obtained by training a dual-channel five-dimensional denoising neural network, and the dual-channel five-dimensional denoising neural network includes a Fourier transform layer, a five-dimensional denoising neural network layer and an inverse Fourier transform layer. The output channel of the Fourier transform layer is dual-channel, and the input channel of the inverse Fourier transform layer is dual-channel.

[0050] For seismic data with the characteristics of "two widths and one height" (i.e., "wide azimuth, wide bandwidth, high density"), a five-dimensional data volume can be formed according to the three-dimensional space, offset distance, and azimuth to extract rich underground information. However, conventional seismic data denoising algorithms are usually developed from the traditional two-dimensional or three-dimensional data denoising processing, which makes it difficult to fully utilize the information of the data in five dimensions, resulting in waste of seismic data information and reducing the accuracy of seismic denoising.

[0051] Therefore, the embodiment of the present invention constructs a five-dimensional data set and trains a dual-channel five-dimensional denoising neural network to obtain a dual-channel five-dimensional denoising model to perform noise suppression processing on actual seismic data, fully utilize the high-dimensional information of the seismic data, and effectively improve the noise recognition accuracy.

[0052] Preferably, training the dual-channel five-dimensional denoising neural network may include steps S121-S122.

[0053] Step S121: Based on the three-dimensional spatial, offset and azimuth five-dimensional attributes of the seismic data, a five-dimensional data set is constructed through sample seismic data and / or simulated seismic data, and the five-dimensional data set is divided into a five-dimensional training set and a five-dimensional verification set.

[0054] Please refer to Figure 3 , a five-dimensional data set (for example, represented as N1×N2×N3×N4×N5) can be constructed by sample seismic data and / or simulated seismic data. The data (or elements) of the five-dimensional data set may include label data and input data. The embodiment of the present invention provides three ways to obtain data of the five-dimensional data set.

[0055] 1. Constructing the data of the five-dimensional data set through the sample seismic data composition.

[0056] Preferably, constructing the data of the five-dimensional data set through the sample seismic data composition may include: denoising the sample seismic data through a preset seismic noise suppression strategy, using the denoised sample seismic data as a label signal, using the sample seismic data as input data, and constructing the data of the five-dimensional data set through the input data and the label data composition.

[0057] The preset seismic noise suppression strategy preferably in the embodiment of the present invention includes one or a combination of abnormal amplitude suppression, FKK domain filtering, and dip filtering.

[0058] To illustrate by way of example, the spatial three-dimensional, offset and azimuth information in the sample seismic data is obtained, and after the sample seismic data is denoised by the above-mentioned noise suppression technology, the sample seismic data is arranged in the order of "shot line-shot point-receiving line-receiving point" to form the data of the five-dimensional data set. The denoised data of the five-dimensional data set can be used as (pseudo) label data, and the original sample seismic data is used as input data, and the data of the five-dimensional data set is constructed by the input data and the label data.

[0059] 2. Construct the data of the five-dimensional data set through the simulated seismic data composition.

[0060] Preferably, the data for constructing the five-dimensional data set by forming the simulated seismic data includes: calculating the arrival time and longitudinal wave reflection coefficient of the seismic signal from the shot point to the receiving point based on the simulated seismic data to generate a simulated seismic signal; generating simulated noise data by simulating random noise. The simulated seismic signal is used as a label signal, the simulated seismic signal and the simulated random noise are used as input data, and the data for constructing the five-dimensional data set is formed by forming the input data and the label data.

[0061] Preferably, the calculation of the arrival time of the seismic signal from the shot point to the receiving point includes: obtaining the coordinates of the shot point, the coordinates of the receiving point, the longitudinal wave velocity and the azimuth in the simulated seismic data; and calculating the arrival time of the seismic signal based on a given time by using the acquired coordinates of the shot point, the coordinates of the receiving point, the longitudinal wave velocity and the azimuth.

[0062] Taking an example to illustrate, an orthogonal observation system is designed in the manner of "shot line-shot point-receiving line-receiving point" to obtain the coordinates of all shot points and receiving points in the simulated seismic data; by giving a time t0, for example, t0 = 0, 1, 2, ... (ms), according to the propagation law of seismic signals in anisotropic media, the arrival time of the simulated seismic signal (i.e., the arrival time of the seismic signal) is calculated, as shown in the following formula:

[0063]

[0064] Where α is the longitudinal wave velocity, the default value is 2300, for example, and x is the offset (or shot distance). is the azimuth,

[0065] The offset x can be calculated based on the shot point coordinates and receiving point coordinates in the seismic trace header, as shown in the following formula:

[0066]

[0067] Among them, (x src ,y src ) and (x rec ,y rec ) are the coordinates of the shot point and the receiving point, which can be obtained through the orthogonal observation system.

[0068] Continuing with the above example, for example, the (longitudinal wave) reflection coefficient R is calculated using the formula for the longitudinal wave reflection coefficient in anisotropic media derived by Ruger (1998) PP , the formula is as follows:

[0069]

[0070] Among them, R PP is the longitudinal wave reflection coefficient, θ is the incident angle, is the azimuth, α is the longitudinal wave velocity, β is the shear wave velocity, Z (Z = ρα) is the wave impedance, ρ is the formation density, G (G = ρβ 2 ) is the shear wave tangential modulus, γ, δ and ε are anisotropic parameters, Δ[·] represents the difference of a certain parameter between the upper and lower interfaces, Represents the mean value of a parameter between the upper and lower interfaces.

[0071] The incident angle θ can be calculated by the following formula:

[0072]

[0073] Among them, t0 can be regarded as the arrival time of the seismic signal at offset x=0.

[0074] The azimuth angle φ can be calculated by the following formula:

[0075]

[0076] Parameters such as α, β, ρ, γ, δ, and ε can be set according to geological information. The default values ​​of the parameters at the interface are as follows: α (km / s) is 2.300 on the interface and 2.580 below the interface; β (km / s) is 1.060 on the interface and 1.256 below the interface; ρ (g / cm 3 ) is, for example, 2.20 on the interface, and 2.30 below the interface; γ is, for example, 0 on the interface, and 0 below the interface; δ is, for example, 0 on the interface, and 0 below the interface; ε is, for example, 0.05 on the interface, and 0.15 below the interface.

[0077] Preferably, generating a simulated seismic signal according to the arrival time of the seismic signal and the longitudinal wave reflection coefficient comprises: calculating the arrival time t of the seismic signal by the following formula, and placing the reflection coefficient r(t) of the formation at the corresponding time point:

[0078]

[0079] Among them, a(t) is the amplitude fitting vector; e(t) is the dynamic adjustment parameter, RPP is the P-wave reflection coefficient; θ is the seismic P-wave incident angle; is the earthquake P-wave azimuth;

[0080] The simulated seismic signal is calculated by convolution of the reflection coefficient r(t) and the seismic wavelet.

[0081] Continuing with the above example, based on formula (6), the reflection coefficient of the formation is placed at the corresponding time point according to the arrival time of the seismic signal. Among them, a(t) decreases with time t, and the default value is, for example, 1; the default value of e(t) is, for example, a random number between (-0.5, 0.5). The seismic signal s is generated by the following formula:

[0082] s=w(t)*r(t) (7)

[0083] Wherein, w(t) is the seismic wavelet, and the default value is preferably the Ricker wavelet (as shown in the following formula); * is the convolution operator.

[0084]

[0085] Among them, f0 is the main frequency of the wavelet.

[0086] Preferably, step S1214 may include: performing low-pass filtering on Gaussian noise data generated by simulating random noise by using a Ricker wavelet having a main frequency of a preset frequency to generate the simulated noise data.

[0087] As an example, a low-frequency random noise n is simulated and simulated noise data d is generated. In order to simulate the real seismic random noise, it is preferred to use a Ricker wavelet w with a main frequency of 60 Hz. f0 =60 Low-pass filter the Gaussian noise to generate low-frequency random noise n, as shown in the following formula:

[0088] n=randn()*w f0＝60 (9)

[0089] Among them, randn() is the Gaussian noise generating function.

[0090] The data constituting the five-dimensional data set are obtained through numerical simulation, that is, the simulated seismic signal is used as the label signal, the simulated seismic signal and the simulated random noise are used as input data, and the input data and the label data are used to construct the data of the five-dimensional data set, which can enrich the characteristics of the training data to improve the denoising ability of the network training.

[0091] 3. Construct the data of the five-dimensional data set by combining the sample seismic data and the simulated seismic data.

[0092] Through the above one and two, we can obtain the sample seismic data, simulated seismic signals and simulated random noise after denoising. The sample seismic data and simulated seismic signals after denoising are used as label signals, and the sample seismic data, simulated seismic signals and simulated random noise before denoising are used as input data. The data of the five-dimensional data set is constructed by the input data and the label data, which can further enrich the characteristics of the training data.

[0093] Step S122: construct the dual-channel five-dimensional denoising neural network, train the dual-channel five-dimensional denoising neural network through the five-dimensional training set, and verify the dual-channel five-dimensional denoising neural network through the five-dimensional verification set to obtain a trained dual-channel five-dimensional denoising model.

[0094] In the field of seismic exploration, deep learning technology has been rapidly applied in the fields of seismic fault identification, first arrival picking, attribute recognition, etc. In addition, the application of deep learning in seismic denoising has also been gradually expanded. However, the current denoising technology based on deep learning is usually based on two-dimensional or three-dimensional convolution. The mainstream deep learning framework does not support high-dimensional (four-dimensional or higher) convolution. Therefore, the embodiment of the present invention provides a dual-channel five-dimensional denoising neural network to suppress noise in seismic data. Please refer to Figure 2 The preferred dual-channel five-dimensional denoising neural network in the embodiment of the present invention includes a Fourier transform layer, a five-dimensional denoising neural network layer and an inverse Fourier transform layer. The output channel of the Fourier transform layer is dual-channel, and the input channel of the inverse Fourier transform layer is dual-channel.

[0095] The five-dimensional denoising neural network layer preferably in the embodiment of the present invention includes multiple layers of five-dimensional convolutional layers, and the construction of each five-dimensional convolutional layer includes: based on given five-dimensional arrays f and g, defining the five-dimensional convolutional layer:

[0096]

[0097] Among them, j1, j2, j3, j4, j5 represent the elements of array f, i1-j1, i2-j2, i3-j3, i4-j4, i5-j5 represent the elements of array g;

[0098] In order to be compatible with common deep learning frameworks and implement GPU acceleration, the defined five-dimensional convolutional layer is transformed as follows:

[0099]

[0100] The pseudo code for formula (11) is as follows:

[0101] Initialization: Input five-dimensional array f (size is N1×N2×N3×N4×N5) and g, sum=0

[0102] Loop i4 executes from 1 to N4:

[0103] Loop i5 executes from 1 to N5:

[0104] sum←sum+Conv3d(i1,i2,i3)

[0105] Output sum.

[0106] The five-dimensional denoising neural network layer preferably in the embodiment of the present invention includes 7 five-dimensional convolutional layers, the first layer of the five-dimensional convolutional layer includes a set of five-dimensional convolutions with an input channel number of 2, an output channel number of 64, and a filter size of 3, the 2nd to 6th layers respectively include a set of five-dimensional convolutions with an input channel number of 64, an output channel number of 64, and a filter size of 3 and a ReLU activation function, and the 7th layer includes a set of five-dimensional convolutional layers with an input channel number of 64, an output channel number of 2, and a filter size of 3.

[0107] Please refer to Figure 2 ,The dual-channel five-dimensional denoising neural network includes Fourier transform layer FFT, inverse Fourier transform layer iFFT and 7 five-dimensional convolution layers. Among them, the number of input channels of the Fourier transform layer FFT is 1, the number of output channels is 2 (i.e., real part Real and imaginary part Imag dual channels), and there are no training parameters, that is, the input data of the five-dimensional training set is Fourier transformed along the time direction to obtain seismic data in the frequency domain; the first layer of the five-dimensional convolution layer includes a set of five-dimensional convolutions with an input channel number of 2, an output channel number of 64, and a filter size of 3; the 2nd to 6th layers respectively include a set of five-dimensional convolutions with an input channel number of 64, an output channel number of 64, and a filter size of 3 and a ReLU activation function; the 7th layer includes a set of five-dimensional convolutions with an input channel number of 64, an output channel number of 2, and a filter size of 3; the number of input channels of the inverse Fourier transform layer iFFT is 2 (i.e., real part Real and imaginary part Imag dual channels), the number of input channels is 1, and there are no training parameters, that is, the dual-channel data output by the network is inverse Fourier transformed along the time direction to restore the seismic data in the time domain.

[0108] Preferably, the loss function of the dual-channel five-dimensional denoising neural network is expressed by the following formula:

[0109]

[0110] Where K is the number of samples in the five-dimensional training set, {x k ,y k} is a set of training sample data, x k To simulate the noise data, y k is the label data, Θ represents the network parameters; ||·|| F is the Frobenius norm.

[0111] Preferably, step S122 may also include: randomly initializing the parameters of the dual-channel five-dimensional denoising neural network to obtain an initial model of the dual-channel five-dimensional denoising neural network; training the model parameters of the dual-channel five-dimensional denoising neural network through the training set, and verifying the dual-channel five-dimensional denoising neural network through the five-dimensional verification set.

[0112] Input the seismic data of the five-dimensional training set constructed by the simulated noise data and the label data into the initial model of the dual-channel five-dimensional denoising neural network, obtain the data of the output layer, calculate the loss function shown in formula (12), and judge whether the current number of iterations meets the maximum number of training times (the default value is, for example, 50); if not, adjust the network parameters of the current network through the back propagation algorithm until the maximum number of training times is reached, and determine the network model that reaches the training times as the dual-channel five-dimensional denoising model. Among them, the back propagation algorithm of the dual-channel five-dimensional denoising neural network is a process of obtaining the optimal network parameters by minimizing the loss function (as shown in formula (12)). The minimization process of the objective function (as shown in formula (12)) can be achieved by the Adam optimization algorithm.

[0113] Taking the actual seismic data of a certain area as an example, the actual seismic data is arranged in the order of "shot line-shot point-receiving line-receiving point" to form a five-dimensional data body and then input into the dual-channel five-dimensional denoising model to obtain the data after noise suppression. Figure 4 The noisy data of a single shot in the area (all receiving lines of the shot point are selected) are shown, which contains a lot of random noise and is difficult to denoise; Figure 5 The valid seismic signal without added noise is shown; Figure 6 The denoising results using a two-dimensional convolutional neural network are shown, with a signal-to-noise ratio of 12.4dB; Figure 7 The denoising result of the dual-channel five-dimensional denoising model of the embodiment of the present invention is shown, and its signal-to-noise ratio is 14.1dB; the embodiment of the present invention can effectively improve the denoising signal-to-noise ratio by 12.1%; Figure 8 shows the noise identified using a two-dimensional convolutional neural network; Fig. 9The figure shows the noise identified by the dual-channel five-dimensional denoising model according to the embodiment of the present invention. Fig.10 Part of the noisy data of another single shot is shown (a certain receiving line of the shot point is selected); Fig.11 The corresponding valid seismic signal without added noise is shown; Fig.12 The denoising results using a two-dimensional convolutional neural network are shown, with a signal-to-noise ratio of 12.3dB; Fig.13 The denoising result (selecting a certain arrangement) of the dual-channel five-dimensional denoising model of the embodiment of the present invention is shown, and its signal-to-noise ratio is 15.6dB; the embodiment of the present invention can effectively improve the denoising signal-to-noise ratio by 21.2%; Fig.14 shows the noise identified using a two-dimensional convolutional neural network; Fig.15 The figure shows the noise identified by the dual-channel five-dimensional denoising model of the embodiment of the present invention. The embodiment of the present invention can accurately identify the noise without obvious effective signal residue. The embodiment of the present invention can identify the noise efficiently and accurately, and has high amplitude preservation.

[0114] Accordingly, the embodiment of the present invention uses a dual-channel five-dimensional denoising model to suppress noise in actual seismic data, making full use of the high-dimensional information of seismic data and effectively improving the recognition accuracy of noise. By constructing a dual-channel five-dimensional denoising neural network, the high-dimensional information of the "high-width-high" data is fully utilized, and based on the five-dimensional space (three-dimensional space, offset distance, azimuth) and the dual channels of real and imaginary parts of frequency, the characteristics of noise samples in the five-dimensional training set are trained and learned to improve the noise recognition accuracy of the dual-channel five-dimensional denoising model, providing high-quality basic data for subsequent seismic imaging and reservoir prediction, and helping to improve the quality and efficiency of oil and gas exploration and processing.

[0115] An embodiment of the present invention also provides a control device for seismic data denoising, the control device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the computer program to implement the above-mentioned method for seismic data denoising.

[0116] An embodiment of the present invention further provides a machine-readable storage medium, on which instructions are stored, and the instructions enable a machine to execute the above-mentioned method for seismic data denoising.

[0117] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0118] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0119] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0120] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0121] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0122] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0123] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0124] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0125] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for denoising seismic data, characterized in that: The method for seismic data denoising comprises: Obtaining actual seismic data for a preset area; and The actual seismic data is subjected to noise suppression processing by a dual-channel five-dimensional denoising model to obtain a denoised seismic data volume. The dual-channel five-dimensional denoising model is obtained by training a dual-channel five-dimensional denoising neural network, wherein the dual-channel five-dimensional denoising neural network includes a Fourier transform layer, a five-dimensional denoising neural network layer and an inverse Fourier transform layer, wherein the output channel of the Fourier transform layer is dual-channel, and the input channel of the inverse Fourier transform layer is dual-channel.

2. The method for seismic data denoising according to claim 1, characterized in that: Training the dual-channel five-dimensional denoising neural network includes: Based on the three-dimensional spatial, offset and azimuth five-dimensional attributes of the seismic data, a five-dimensional data set is constructed through sample seismic data and / or simulated seismic data, and the five-dimensional data set is divided into a five-dimensional training set and a five-dimensional validation set; and The dual-channel five-dimensional denoising neural network is constructed, the dual-channel five-dimensional denoising neural network is trained by the five-dimensional training set, and the dual-channel five-dimensional denoising neural network is verified by the five-dimensional verification set to obtain a trained dual-channel five-dimensional denoising model.

3. The method for seismic data denoising according to claim 2, characterized in that: The data for constructing the five-dimensional data set by the sample seismic data composition includes: The sample seismic data is denoised using a preset seismic noise suppression strategy. The sample seismic data after denoising is used as a label signal, and the sample seismic data is used as input data. The data of the five-dimensional data set is constructed by combining the input data and the label data.

4. The method for seismic data denoising according to claim 2, characterized in that: The data for constructing the five-dimensional data set by the simulated seismic data composition includes: Based on the simulated seismic data, calculating the arrival time of the seismic signal from the shot point to the receiving point and the longitudinal wave reflection coefficient to generate a simulated seismic signal; By simulating random noise, simulated noise data is generated. The simulated seismic signal is used as a label signal, the simulated seismic signal and the simulated random noise are used as input data, and the data of the five-dimensional data set is constructed by combining the input data and the label data.

5. The method for seismic data denoising according to claim 3, characterized in that: The preset seismic noise suppression strategy includes one or a combination of abnormal amplitude suppression, FKK domain filtering, and dip filtering.

6. The method for seismic data denoising according to claim 4, characterized in that: The calculation of the arrival time of the seismic signal from the shot point to the receiving point includes: Obtaining the coordinates of the shot points, the coordinates of the receiving points, the P-wave velocity and the azimuth in the simulated seismic data; and The arrival time of the seismic signal is calculated based on a given time by obtaining the coordinates of the shot point, the coordinates of the receiving point, the longitudinal wave velocity and the azimuth.

7. The method for seismic data denoising according to claim 6, characterized in that: The generating a simulated seismic signal according to the seismic signal arrival time and the longitudinal wave reflection coefficient comprises: The reflection coefficient r(t) of the formation placed at the corresponding time point when the seismic signal arrives at time t is calculated by the following formula: Among them, a(t) is the amplitude fitting vector; e(t) is the dynamic adjustment parameter, R PP is the longitudinal wave reflection coefficient; θ is the incident angle of seismic longitudinal wave; is the earthquake P-wave azimuth; The simulated seismic signal is calculated by convolution of the reflection coefficient r(t) and the seismic wavelet.

8. The method for seismic data denoising according to claim 4, characterized in that: The step of generating simulated noise data based on simulated random noise includes: The simulated noise data is generated by performing low-pass filtering on Gaussian noise data generated by simulating random noise through a Ricker wavelet whose main frequency is a preset frequency.

9. The method for seismic data denoising according to claim 1, characterized in that: The five-dimensional denoising neural network layer includes multiple layers of five-dimensional convolutional layers, and the construction of each five-dimensional convolutional layer includes: Based on the given five-dimensional arrays f and g, define the five-dimensional convolutional layer: Among them, j1, j2, j3, j4, j5 represent the elements of array f, i1-j1, i2-j2, i3-j3, i4-j4, i5-j5 represent the elements of array g; Transform the defined five-dimensional convolutional layer:

10. The method for seismic data denoising according to claim 1, characterized in that: The five-dimensional denoising neural network layer includes seven five-dimensional convolutional layers. The first layer of the five-dimensional convolution layer includes a set of five-dimensional convolutions with 2 input channels, 64 output channels, and a filter size of 3. The 2nd to 6th layers include a set of five-dimensional convolutions and ReLU activation functions with 64 input channels, 64 output channels, and a filter size of 3. The 7th layer includes a set of five-dimensional convolutional layers with 64 input channels, 2 output channels, and a filter size of 3.

11. The method for seismic data denoising according to claim 1, characterized in that: The loss function of the dual-channel five-dimensional denoising neural network is expressed by the following formula: Where K is the number of samples in the five-dimensional training set, {x k ,y k } is a set of training sample data, x k To simulate noise data, y k is the label data, Θ represents the network parameters; ||·|| F is the Frobenius norm.

12. A control device for seismic data denoising, characterized in that: The control device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the method for seismic data denoising according to any one of claims 1-11.

13. A machine-readable storage medium, characterized in that: The machine-readable storage medium stores instructions, which enable the machine to execute the method for seismic data denoising according to any one of claims 1-11.

Images