Method for suppressing random noise of seismic data based on stratigraphic dip constraint

Abstract

The invention discloses a method for suppressing random noise of seismic data based on stratigraphic dip constraint, and belongs to the technical field of signal processing. According to the method, the structure of the denoising convolutional neural network DnCNN is optimized, dip angle information related to stratum features is introduced to serve as training network constraints of the DnCNN, damage of a conventional DnCNN noise suppression method to effective signals in seismic data is eliminated, and effective protection of geological features such as fractures and fractured-vuggy bodies is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, and particularly relates to a method for suppressing random noise in seismic data based on formation dip angle constraint. Background Technique

[0002] Earthquake is the ground motion caused by seismic waves released from the seismic source, which is a collection of different frequencies and different amplitudes (or intensities) within a limited time range. Studying the propagation law of seismic waves in the formation can identify the underground geological structure. However, under the interference background of different topographies and geological conditions, seismic signal data is distorted by some external factors, and the seismic signal data is severely intertwined with each other. Random noise often seriously affects the signal-to-noise ratio of complex seismic signals. At the same time, weak seismic signals are always masked by random noise in the exploration of deep geophysical targets.

[0003] The existence of random noise will affect subsequent processing and interpretation work. For example, in high-resolution processing, while enhancing high-frequency effective signals, the noise mixed in the high-frequency band will also be enhanced, resulting in a low signal-to-noise ratio of the seismic data after high-resolution processing, and even some false reflection information will appear. Therefore, the seismic signal information cannot be directly used. Removing random noise from seismic signal data and enhancing the reliability of seismic data play an important role in effectively analyzing geological structures and reasonably interpreting earthquakes.

[0004] An important link in seismic data processing is to improve the signal-to-noise ratio. To improve the signal-to-noise ratio of seismic signals, noise needs to be eliminated. The noise in seismic data mainly includes two parts: random noise and coherent noise. Generally speaking, the appearance of coherent noise has certain regularity and can be removed by targeted methods. Random noise has no regularity and is relatively difficult to remove. In seismic data processing, effective suppression of random noise can significantly improve the signal-to-noise ratio of seismic data. How to effectively suppress random noise interference and restore effective seismic signals remains a key issue in high-precision seismic exploration.

[0005] In the prior art, the characteristics of seismic data are usually combined to use wavelet transform method to eliminate random noise therein, so as to achieve the purpose of improving the signal-to-noise ratio. For example, the patent (CN114325821A) proposes a method and system for suppressing strong scattering noise in pre-stack seismic data based on a 3D-SNACNN network, constructs a 3D-SNACNN network, selects three-dimensional seismic data and rearranges them into OVT domain data as a data set for network training; uses a three-dimensional continuous wavelet fast algorithm to denoise the selected OVT domain seismic data to obtain corresponding clean data, divides the clean data and the network training data set into a number of three-dimensional data that meet the input requirements of the 3D-SNACNN network in the same way, and then selects some data from them to form training sample pairs, and sends the training sample pairs to the 3D-SNACNN network for training. After the training is completed, the 3D-SNACNN network is used to process the seismic data in the test set to suppress various random noises in the three-dimensional seismic data; the patent (CN116127285A) proposes an improved wavelet threshold denoising method, which decomposes the noisy signal through several layers of wavelets to obtain the corresponding coefficients of the high-frequency wavelets of each layer. The amplitude of the corresponding analysis wavelet coefficient is lower for useful signals and higher for noise. The improved threshold function is applied to process the wavelet coefficients of each layer, and the coefficients smaller than the threshold are retained and those higher than the threshold are removed. Then these wavelet coefficients are reconstructed using inverse wavelet transform and output as the processed denoised signal. The present invention constructs a new threshold function to solve the defects of the traditional wavelet denoising algorithm in denoising.

[0006] However, there are two main threshold functions in wavelet threshold denoising: hard threshold function and soft threshold function. The denoised signal obtained by the hard threshold function has better approximation, but it will produce additional oscillations; the soft threshold function changes the wavelet coefficients. Although the obtained denoised signal has better smoothness, there is a large error with the original signal and the original signal characteristics cannot be completely maintained. Therefore, the above patent uses wavelet threshold denoising, which still has insurmountable defects.

[0007] In addition, in the field of artificial intelligence, the denoising convolutional neural network DnCNN is often used in the suppression of seismic random noise. However, in areas with complex geological structures, the DnCNN denoising method cannot effectively protect geological features such as overlaps and faults related to structures while suppressing noise, which affects the amplitude preservation and fidelity of seismic data.

[0008] Therefore, it is necessary to provide a method that can effectively suppress the random noise of seismic data. Summary of the invention

[0009] In view of the problems existing in the prior art, the present invention provides a method for suppressing random noise in seismic data based on formation dip angle constraint. By optimizing the structure of the denoising convolutional neural network DnCNN and introducing dip angle information related to formation characteristics as the training network constraint of DnCNN, the destruction of effective signals in seismic data by the conventional DnCNN noise suppression method is eliminated, and the effective protection of geological characteristics such as faults and fracture-vug bodies is realized.

[0010] To achieve the above object, in a first aspect, the present invention provides a method for suppressing random noise based on formation dip angle constraint, including the following steps:

[0011] Step S1: Collect seismic data samples containing random noise;

[0012] Step S2: Estimate the noise level of the seismic data containing random noise;

[0013] Step S3: Calculate the formation dip angle;

[0014] Step S4: Train the denoising convolutional neural network DnCNN to obtain a DnCNN network constrained by the formation dip angle;

[0015] Step S5: Denoise the seismic data containing random noise.

[0016] In a preferred embodiment, in step S1, the seismic data containing random noise is a three-dimensional seismic data volume after seismic data processing and migration imaging.

[0017] In a preferred embodiment, the seismic data containing random noise is used to train the denoising convolutional neural network DnCNN and suppress noise.

[0018] In a preferred embodiment, in step S2, the noise level of the seismic data containing random noise in step 1 is estimated, and the noisy seismic data is characterized as:

[0019] y i =s i +n i

[0020] where s i is the effective signal and n i is the noise;

[0021] Calculate the minimum eigenvalue of the covariance matrix of the noisy seismic data:

[0022]

[0023] where λ min (∑y) is the minimum eigenvalue of the covariance matrix of the noisy seismic data, and λ min(∑x) is the minimum eigenvalue of the effective signal block variance matrix;

[0024] Calculate the texture intensity of seismic data:

[0025]

[0026] where, is the covariance matrix; tr() is the trace of the matrix;

[0027] Set a threshold r, define the seismic data with texture intensity less than r as weak texture blocks, and obtain the noise level estimate M using the covariance matrix of the input seismic data.

[0028] In a preferred embodiment, in step S3, calculate the formation dip angle of the seismic data with random noise in step S1 using plane wave decomposition;

[0029]

[0030] where, U is the seismic data, σ x is the formation dip angle, C(σ x ) is the formation dip angle operator, and e is the identity matrix.

[0031] In a preferred embodiment, in step S4, input the seismic data with random noise in step S1, the noise level evaluated in step S2, and the formation dip angle calculated in step S3 into the denoising convolutional neural network DnCNN for training:

[0032] s = F(y, M, Θ, θ)

[0033] where, s is the effective seismic information, M is the estimated noise level estimate; Θ is the value of the trainable model parameter, and θ is the formation dip angle.

[0034] In a preferred embodiment, in step S5, use the DnCNN network constrained by the formation dip angle in step S4 to denoise the seismic data with random noise; more preferably, by using the noise level evaluated in step S2 and the formation dip angle calculated in step S3 as the input values of the DnCNN network constrained by the formation dip angle in step S4 at the same time, train a denoising model to perform seismic random noise suppression processing.

[0035] In a second aspect, the present invention provides a storage medium, in which a computer-executable program is stored, and the program is suitable for implementing a method for suppressing random noise based on formation dip angle constraint when executed, and the method includes the following steps:

[0036] Step S1: Collect seismic data samples with random noise;

[0037] Step S2: Noise level estimation of seismic data with random noise;

[0038] Step S3: Formation dip calculation;

[0039] Step S4: Train the denoising convolutional neural network DnCNN to obtain the DnCNN network constrained by formation dip;

[0040] Step S5: Denoising processing of seismic data with random noise.

[0041] The beneficial effects of the present invention are as follows:

[0042] 1. By optimizing the structure of the denoising convolutional neural network DnCNN, the present invention introduces the dip information related to formation characteristics as the training network constraint of DnCNN to eliminate the damage to the effective signals in seismic data by the conventional DnCNN noise suppression method, and realizes the effective protection of geological characteristics such as faults and fracture-vug bodies.

[0043] 2. The present invention realizes the suppression of seismic random noise based on artificial intelligence, and effectively protects the high-dip formation structure while improving the signal-to-noise ratio of seismic data. Brief Description of the Drawings

[0044] Figure 1 is a schematic flow chart of the method for suppressing random noise of seismic data based on formation dip constraint of the present invention;

[0045] Figure 2 is seismic data with random noise;

[0046] Figure 3 is the formation dip calculated by plane wave decomposition;

[0047] Figure 4 is a schematic structural diagram of the DnCNN network constrained by formation dip;

[0048] Figure 5 is the random noise suppression result of the DnCNN network constrained by formation dip;

[0049] Figure 6 is the residual after random noise suppression of the DnCNN network constrained by formation dip. Detailed Embodiments

[0050] Embodiment

[0051] The schematic flow chart of the method for suppressing random noise of seismic data based on formation dip constraint of the present invention is as shown in the appendix Figure 1 as follows, and the specific process is as follows:

[0052] Step S1: Collect seismic data samples with random noise. The seismic data with random noise is a three-dimensional seismic data volume after seismic data processing and migration imaging, as shown in the appendix Figure 2 as follows;

[0053] Step S2: Estimation of the noise level of the seismic data with noise:

[0054] Estimate the noise level of the seismic data with random noise in Step 1, and characterize the seismic data with noise as:

[0055] y i = s i + n i

[0056] where s i is the effective signal and n i is the noise;

[0057] Calculate the minimum eigenvalue of the covariance matrix of the seismic data with noise:

[0058]

[0059] where λ min (∑y) is the minimum eigenvalue of the covariance matrix of the seismic data with noise, and λ min (∑x) is the minimum eigenvalue of the variance matrix of the effective signal block;

[0060] Calculate the texture intensity of the seismic data:

[0061]

[0062] where is the covariance matrix; tr() is the trace of the matrix;

[0063] Set a threshold r, define the seismic data with texture intensity less than r as weak texture blocks, and obtain the noise level estimate M using the covariance matrix of the input seismic data;

[0064] Step S3: Calculate the formation dip angle:

[0065] Use plane wave decomposition to calculate the formation dip angle of the seismic data with random noise in Step S1, and the result is as shown in the appendix Figure 3 as follows;

[0066]

[0067] where U is the seismic data, σ x is the formation dip angle, C(σ x ) is the formation dip angle operator, and e is the identity matrix;

[0068] Step S4: Train the denoising convolutional neural network DnCNN to obtain the DnCNN network constrained by formation dip angle;

[0069] Input the seismic data with random noise in Step S1, the noise level estimated in Step S2, and the formation dip angle calculated in Step S3 into the denoising convolutional neural network DnCNN for training. The DnCNN network constrained by formation dip angle is as shown in the appendix Figure 4 as follows:

[0070] s = F(y, M, Θ, θ)

[0071] where s is the seismic effective information, M is the estimated noise level; Θ is the value of the trainable model parameter, and θ is the formation dip angle;

[0072] Step S5: Denoising processing of the seismic data with random noise;

[0073] Use the DnCNN network constrained by formation dip angle to perform denoising processing on the seismic data with random noise; by taking the noise level evaluated in Step S2 and the formation dip angle calculated in Step S3 as the input values of the DnCNN network constrained by formation dip angle at the same time, a denoising model is trained to perform seismic random noise suppression processing. The random noise suppression result of the DnCNN network constrained by formation dip angle is as shown in the appendix Figure 5 as follows, and the residual after random noise suppression of the DnCNN network constrained by formation dip angle is as shown in the appendix Figure 6 as follows. It can be seen that the signal-to-noise ratio of the seismic data has been significantly improved. Due to the elimination of the influence of random noise, the imaging characteristics of the strike-slip fault zone in the seismic profile and coherent body attributes are clearer, and the identification and description of faults are more reliable. Since formation dip angle constraint is introduced in noise suppression, the characteristics of geological bodies related to formation dip angle such as faults and fracture-vug bodies are effectively protected, and the signal-to-noise ratio is improved on the basis of amplitude preservation.

[0074] The technical solution of the present invention is not limited to the technical means disclosed above, but also includes technical solutions composed of any combination of the above technical features. The above is the specific implementation manner of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A method for suppressing random noise based on formation dip constraint, characterized in that Including the following steps: Step S1: Collect seismic data samples containing random noise; Step S2: Estimate the noise level of the seismic data containing random noise; Step S3: Calculate the formation dip angle; Step S4: Train the denoising convolutional neural network DnCNN to obtain the DnCNN network constrained by the formation dip angle; Step S5: Denoise the seismic data containing random noise.

2. The method according to claim 1, wherein In step S1, the seismic data containing random noise is a three-dimensional seismic data volume after seismic data processing and migration imaging.

3. The method according to claim 1, wherein The seismic data containing random noise is used to train the denoising convolutional neural network DnCNN and suppress noise.

4. The method according to claim 1, wherein In step S2, estimate the noise level of the seismic data containing random noise in step 1, and characterize the seismic data with noise as: y i = s i + n i where s i is the valid signal, and n i is the noise; Calculate the minimum eigenvalue of the covariance matrix of the seismic data with noise; where λ min (∑y) is the minimum eigenvalue of the covariance matrix of noisy seismic data, and λ min (∑x) is the minimum eigenvalue of the variance matrix of the effective signal block; Calculate the texture intensity of the seismic data; Among them, C xi is the covariance matrix; tr() is the trace of the matrix; Set a threshold r, define the seismic data with texture intensity less than r as a weak texture block, and obtain the noise level estimate M using the covariance matrix of the input seismic data.

5. The method according to claim 1, wherein In step S3, calculate the formation dip angle of the seismic data containing random noise in step S1 using plane wave decomposition; Where U is seismic data, σ x is the formation dip angle, C(σ x ) is the formation dip angle operator, and e is the identity matrix.

6. The method according to claim 1, characterized in that, In step S4, input the seismic data containing random noise in step S1, the noise level evaluated in step S2, and the formation dip angle calculated in step S3 into the denoising convolutional neural network DnCNN for training: s = F(y, M, Θ, θ) where s is the effective seismic information, M is the estimated noise level estimate; Θ is the value of the trainable model parameter, and θ is the formation dip angle.

7. The method according to claim 1, characterized in that, In step S5, use the DnCNN network constrained by the formation dip angle in step S4 to denoise the seismic data containing random noise.

8. The method according to claim 1, wherein In step S5, by taking the noise level evaluated in step S2 and the formation dip angle calculated in step S3 as the input values of the DnCNN network constrained by the formation dip angle in step S4 at the same time, a denoising model is trained to perform seismic random noise suppression processing.

9. The method according to claim 8, wherein In step S5, input the seismic data containing random noise into the denoising model, and the denoised seismic data can be output.

10. A storage medium storing a computer-executable program, characterized in that, When the program is executed, it is suitable for implementing a method for suppressing random noise based on formation dip angle constraint. The method includes the following steps: Step S1: Collect seismic data samples containing random noise; Step S2: Estimate the noise level of the seismic data containing random noise; Step S3: Calculate the formation dip angle; Step S4: Train the denoising convolutional neural network DnCNN to obtain the DnCNN network constrained by the formation dip angle; Step S5: Denoise the seismic data containing random noise.

Images