Intelligent denoising method and device for multi-type noise and related equipment

By constructing a denoising convolutional neural network model suitable for multiple types of seismic noise, it is expanded into a three-dimensional architecture, combined with multiple data preprocessing methods and acceleration technologies, the denoising efficiency and fidelity of seismic data in the existing technology is solved. The problem is to achieve efficient intelligent noise denoising and signal-to-noise ratio improvement for multiple types of noise.

CN118915157APending Publication Date: 2024-11-08CHINA NAT PETROLEUM CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310506426.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-06
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing seismic data denoising technology has limitations in terms of efficiency and fidelity, especially in the application of three-dimensional actual data denoising, poor generalization ability, high parameter adjustment requirements, low calculation efficiency, and cannot be effectively applied to the overall multi-type noise. Denoising.

Method used

Based on the denoising convolutional neural network, the basic denoising network model suitable for multiple types of seismic noise is constructed. By training the noise-containing seismic data and noise data, it is expanded into a three-dimensional network architecture, combining constraint tomography static correction, geometric diffusion compensation, and frequency division Partitioned surface wave suppression and other methods use BN and RL technologies to speed up the training process and achieve intelligent noise denoising.

Benefits of technology

It realizes intelligent suppression of multiple types of seismic noise, enhances the continuity of the reflection axis of the target layer and signal energy, avoids distortion of the denoising result, greatly improves the denoising efficiency, and effectively suppresses surface waves, linear coherent noise and random interference. Noise broadens the effective frequency band and improves the signal-to-noise ratio.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118915157A_ABST
    Figure CN118915157A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of seismic data processing, and discloses an intelligent denoising method and device for multi-type noise and related equipment, and the method comprises the steps: constructing a basic denoising network model for multi-type seismic noise based on a denoising convolutional neural network; training the basic denoising network model by using noisy seismic data and seismic noise data to obtain an intelligent denoising network model; and de-noising actual noisy seismic data by using the intelligent de-noising network model to obtain an intelligent noise suppression result. The method is suitable for intelligent suppression of various earthquake noises, the continuity and signal energy of a reflection axis of a target layer are enhanced, distortion of a denoising result is effectively avoided, and meanwhile, the denoising efficiency is greatly improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of seismic data processing, and in particular relates to an intelligent denoising method, device and related equipment for multiple types of noise. Background Art

[0002] Existing seismic data often have the characteristics of "high signal-to-noise ratio, high resolution, and high fidelity". High signal-to-noise ratio is the basis of high resolution and high fidelity. Without a high signal-to-noise ratio, high resolution is out of the question and high fidelity loses its meaning. De-noising technology is an important way to improve the signal-to-noise ratio of seismic data and plays an important role in seismic data processing. As the problems to be solved in seismic exploration become more and more complex and the amount of collected data increases, the processing accuracy and efficiency requirements of denoising technology are also getting higher and higher.

[0003] Conventional denoising methods have certain limitations in denoising efficiency and fidelity. To address this problem, Gao Han (2018) compared the application effects of deep learning denoising methods based on convolutional denoising neural network (DnCNN), dilated convolutional neural network (Dilate-DnCNN), residual neural network (Res-Net), and encoder-decoder network (Encoder-Decoder Net). Mai Hao (2019) proposed the Seis-Res Net seismic data noise suppression neural network model. The algorithm uses a 41-layer deep residual neural network structure, breaking through the limitation of the number of layers of the deep learning algorithm network, making the network learn the seismic signal characteristics more fully, and the waveform information of the seismic effective signal is restored. Gu Hang (2019) found that residual learning based on the DnCNN network can perform adaptive denoising on data with different signal-to-noise ratios, with the characteristics of automatic feature extraction and blind denoising; and based on DnCNN, a symmetric dilated convolutional neural network (SDCNN) was designed, which can make full use of the field information of seismic signals to distinguish signals from noise. Yu Kun (2020) used hyperparameters to train the BP neural network and verified that the denoising fidelity of this method was good, but the setting of hyperparameters and the number of training samples would affect the accuracy of the method to a certain extent. In response to the above defects, Wang Peng (2020) developed a seismic data denoising method based on geological structure guidance and convolutional neural network, which made up for the problem of insufficient actual data training samples of convolutional neural network and avoided the process of manual optimization of parameters in traditional denoising. Cui Shaohua et al. (2020) used a convolutional neural network improved based on LeNet-5 for intelligent denoising. This method can be applied to the horizontal and inclined phase axis denoising of two-dimensional pre-stack and post-stack seismic data, and the effect is better than that of the BP algorithm. Zhang Chaoming et al. (2022) proposed a Constrained-DnCNN deep denoising model based on DnCNN and constrained convolution, which uses constrained convolution to provide low-dimensional features and pass them to the deep network to suppress the interference of original seismic data on noise residuals and improve the network training effect through skip connections. Song Hui et al. (2020) proposed an unsupervised seismic data denoising algorithm based on convolutional denoising autoencoder to denoise and reconstruct seismic data. However, this method can only be applied to the random noise suppression of two-dimensional post-stack seismic data. Wang Feng (2020) proposed a method for simultaneous denoising and reconstruction of three-dimensional seismic data, which first divides the three-dimensional data into two-dimensional seismic profiles, performs simultaneous denoising and reconstruction on them, and finally combines them into three-dimensional seismic data. Chen Jie et al. (2020) studied the denoising and reconstruction problems of three-dimensional seismic data based on a data-driven tight framework, and it can be applied to the post-stack random noise suppression of three-dimensional synthetic data and three-dimensional actual data, but the computational efficiency is low.Zhao Binghui et al. (2021) applied convolutional neural networks to suppress aliasing noise in the process of source separation in multi-source seismic mixed acquisition. Zhang Yan et al. (2022) proposed a deep learning denoising algorithm that includes a noise distribution estimation subnet and a denoising subnet. This method has certain robustness in 2D pre-stack wave suppression and 2D post-stack random noise suppression, and can also improve the signal-to-noise ratio of seismic data.

[0004] Although the method proposed by Gu Hang (2019) also improved the DnCNN network and can make full use of the information in the field of seismic signals to distinguish signals from noise, the generalization ability of this method in the application of three-dimensional actual data denoising is currently very poor; although the method proposed by Mai Hao (2019) broke through the limitation of the number of network layers of the deep learning algorithm, the algorithm can only be applied to the suppression of two-dimensional post-stack random noise and coherent noise, and has high requirements for parameter adjustment; although the methods proposed by Wang Feng (2020) and Chen Jie et al. (2020) can simultaneously denoise and reconstruct three-dimensional data, the algorithm is not mature, or can only be applied to synthetic data or has low computational efficiency; although the method proposed by Zhang Yan et al. (2022) has certain robustness in two-dimensional pre-stack wave suppression and two-dimensional post-stack random noise suppression, and can improve the signal-to-noise ratio of seismic data, it cannot be applied to the overall denoising of three-dimensional data blocks.

[0005] In view of the above problems, it is necessary to provide an intelligent denoising method, device and related equipment suitable for intelligent suppression of various types of seismic noise, which can enhance the continuity and signal energy of the reflection axis of the target layer, effectively avoid distortion of the denoising results, and greatly improve the denoising efficiency. Summary of the invention

[0006] In order to overcome the defects of the above-mentioned prior art, the purpose of the present invention is to provide an intelligent denoising method, device and related equipment suitable for intelligent suppression of various types of seismic noise, which can enhance the continuity and signal energy of the reflection axis of the target layer, effectively avoid distortion of the denoising results, and greatly improve the denoising efficiency.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] An intelligent denoising method for multiple types of noise includes the following steps:

[0009] Based on the denoising convolutional neural network, a basic denoising network model for multiple types of seismic noise is constructed;

[0010] The basic denoising network model is trained using noisy seismic data and seismic noise data to obtain an intelligent denoising network model;

[0011] The intelligent denoising network model is used to denoise the actual noisy seismic data and obtain the intelligent noise suppression result.

[0012] Furthermore, a basic denoising network model for multiple types of seismic noise is constructed based on the denoising convolutional neural network, including:

[0013] Modify the number of convolutional layers and activation functions of the denoising convolutional neural network so that it can be applied to the field of earthquake denoising;

[0014] The modified denoising convolutional neural network is expanded into a three-dimensional network architecture to obtain the basic denoising network model.

[0015] Furthermore, the modified denoising convolutional neural network is expanded into a three-dimensional network architecture to obtain a basic denoising network model, including:

[0016] The input and output sample dimensions, convolution kernel dimensions, and step size dimensions of the modified denoising convolutional neural network are expanded from two dimensions to three dimensions.

[0017] Furthermore, the basic denoising network model is trained using the noisy seismic data and the seismic noise data to obtain an intelligent denoising network model, including:

[0018] De-noising the original noisy seismic data before and after stacking to obtain seismic noise data;

[0019] Normalizing the original noisy seismic data and the seismic noise data according to the minimum amplitude value and the maximum amplitude value of the original noisy seismic data and the seismic noise data;

[0020] The normalized original noisy seismic data and noise data are used to create a training sample data set;

[0021] The basic denoising network model is trained using the training sample data set, the model training parameters in the basic denoising network model are determined, and the intelligent denoising network model is obtained.

[0022] Furthermore, the pre-stack and post-stack original noisy seismic data are denoised to obtain seismic noise data, including:

[0023] Constrained tomographic static correction and geometric diffusion compensation are used to preprocess the pre-stack and post-stack raw noisy seismic data.

[0024] The noise suppression method combination of frequency-division and partition-division surface wave suppression, frequency-division abnormal amplitude suppression, irregular coherent noise suppression, ringing suppression, and random noise suppression is used to suppress the noise of the pre-processed original noisy seismic data to obtain seismic noise data.

[0025] The original noisy seismic data and seismic noise data are segmented according to the training sample size to form training and test sample data sets.

[0026] Furthermore, the basic denoising network model is trained using noisy seismic data and seismic noise data to obtain an intelligent denoising network model, further comprising:

[0027] Use BN and RL techniques to speed up the training process and improve the denoising performance of the intelligent denoising network model.

[0028] Furthermore, the intelligent denoising network model is used to denoise the actual noisy seismic data to obtain intelligent noise suppression results, including:

[0029] The normalized actual noisy seismic data is input into the intelligent denoising network model, and the predicted noise data containing various seismic noises is output;

[0030] The residual learning strategy is adopted to perform subtraction on the actual noisy seismic data and the predicted noisy data containing various seismic noises to obtain the intelligent noise suppression result and / or output the predicted residual image.

[0031] Furthermore, the seismic noise includes one or more of surface wave noise, linear coherent noise, random noise and abnormal interference noise.

[0032] In a second aspect, the present invention discloses an intelligent denoising device for multiple types of seismic noise, comprising:

[0033] A basic denoising network model building module is used to build a basic denoising network model for multiple types of seismic noise based on a denoising convolutional neural network;

[0034] An intelligent denoising network model training module is used to train a basic denoising network model according to noisy seismic data and seismic noise data to obtain an intelligent denoising network model;

[0035] The intelligent denoising module is used to denoise the normalized actual noisy seismic data through an intelligent denoising network model to obtain the intelligent noise suppression result of the actual noisy seismic data.

[0036] In a third aspect, the present invention discloses a computer-readable storage medium having computer-readable instructions stored thereon. When the instructions are executed by a processor, the processor executes the above method.

[0037] In a fourth aspect, the present invention discloses a computer device, comprising:

[0038] processor;

[0039] a memory storing a computer program,

[0040] When the computer program is executed by the processor, the processor executes the above method.

[0041] Technical effects and advantages of the present invention:

[0042] 1. The method of the present invention is suitable for intelligent suppression of various types of seismic noise, enhancing the continuity and signal energy of the reflection axis of the target layer, and effectively avoiding distortion of denoising results, while greatly improving denoising efficiency.

[0043] 2. The method of the present invention can effectively suppress surface waves, linear coherent noise and random interference noise of various types of seismic noise, broaden the effective frequency band and greatly improve the signal-to-noise ratio of seismic data.

[0044] 3. The method of the present invention can ensure that there is no cross-layer phenomenon before and after denoising of two-dimensional seismic data, with good fidelity, and provide support for the detailed interpretation work of subsequent interpreters.

[0045] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flow chart of the intelligent denoising method for multiple types of seismic noise of the present invention;

[0047] Figure 2 It is a schematic diagram of the extended three-dimensional intelligent denoising network architecture of the present invention;

[0048] Figure 3 A schematic diagram of pre-stack noisy seismic data A according to an embodiment of the present invention;

[0049] Figure 4 for Figure 3 Schematic diagram of the result after the mid-surface wave and linear coherent noise are suppressed by intelligent noise;

[0050] Figure 5 for Figure 3 and Figure 4 Schematic diagram of spectrum analysis;

[0051] Figure 6 Schematic diagram of pre-stack noisy seismic data B according to an embodiment of the present invention;

[0052] Figure 7 for Figure 6 Schematic diagram of the result after random noise is suppressed by intelligent noise;

[0053] Figure 8 for Figure 6 and Figure 7 Schematic diagram of spectrum analysis;

[0054] Fig. 9 A schematic diagram of post-stack noisy seismic data A according to an embodiment of the present invention;

[0055] Fig.10 for Fig. 9 Schematic diagram of the result after the random noise and the noise that was not removed before stacking are suppressed by intelligent noise;

[0056] Fig.11 Schematic diagram of post-stack noisy seismic data B according to an embodiment of the present invention;

[0057] Fig.12 for Fig.11 Schematic diagram of the result after the random noise and the noise that was not removed before stacking are suppressed by intelligent noise;

[0058] Fig.13 A schematic diagram of three-dimensional post-stack noisy seismic data A according to an embodiment of the present invention;

[0059] Fig.14 for Fig.13 Schematic diagram of the frequency-wavenumber spectrum of the results after intelligent noise suppression of 3D post-stack noisy seismic data;

[0060] Fig.15 for Fig.13 Schematic diagram of the result after being suppressed by intelligent noise;

[0061] Fig.16 for Fig.15 Schematic diagram of the frequency-wavenumber spectrum of the results after intelligent noise suppression of 3D post-stack noisy seismic data;

[0062] Fig.17 Schematic diagram of three-dimensional post-stack noisy seismic data B according to an embodiment of the present invention;

[0063] Fig.18 for Fig.17 Schematic diagram of the results of intelligent noise suppression of 3D post-stack noisy seismic data;

[0064] Fig.19 It is a structural schematic diagram of the intelligent denoising device of the present invention;

[0065] Fig. 20 It is a schematic diagram of the structure of the computer device of the present invention. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] The present invention provides an intelligent denoising method for multiple types of noise, such as Figure 1 As shown, the method comprises the following steps:

[0068] Based on the denoising convolutional neural network, a basic denoising network model for multiple types of seismic noise is constructed;

[0069] The basic denoising network model is trained using noisy seismic data and seismic noise data to obtain an intelligent denoising network model;

[0070] The intelligent denoising network model is used to denoise actual noisy seismic data to obtain intelligent noise suppression results. 。

[0071] in , Seismic noise includes one or more of surface wave noise, linear coherent noise, random noise and abnormal interference noise.

[0072] According to some embodiments of the present invention, when constructing a basic denoising network model, on the basis of the denoising convolutional neural network proposed by (Zhang Kai, 2017), the original network architecture is kept unchanged for parameter adjustment, and the number of convolutional layers, the size of the convolutional kernel, the step size, the activation function and other elements of the denoising convolutional neural network are finely designed for feasibility and necessity according to the requirements of the present invention. For example, the original 17 convolutional layers are reduced to 6 layers, and in addition, the relu activation function of the last convolutional layer is modified to tanh to make it suitable for seismic noise suppression; the input and output sample dimensions, convolutional kernel dimensions, and step size dimensions of the denoising convolutional neural network after parameter adjustment are expanded from two dimensions to three dimensions, so as to obtain a basic denoising network model suitable for noise suppression of three-dimensional seismic data volumes, such as Figure 2 As shown, the three-dimensional basic denoising network model proposed in the embodiment of the present invention has three types of layers:

[0073] (i) Conv+ReLU: For the first layer, there are 64 filters of size 3*3*c, which are used to generate 64 feature maps, and then ReLU is used to process nonlinearity. Where c represents the number of image channels. When c=1, the filter is used for grayscale image denoising; when c=3, the filter is used for color image denoising.

[0074] (ii) Conv+BN+ReLU: For the intermediate layers between the first and last layers, 64 filters of size 3*3*64 are included, and batch normalization is added between convolution and ReLU.

[0075] (iii) Conv+tanh: For the last layer, c filters of size 3*3*64 are used to reconstruct the output, which is used to output the intelligent noise suppression result of 3D post-stack seismic data.

[0076] The difference between the three-dimensional architecture and the two-dimensional network architecture lies in the dimension of the input and output data and the dimension of the convolution kernel. In order to ensure that the size of the input and output feature maps remains unchanged, "same" zero padding is performed before each convolution layer. The basic denoising network model architecture for three-dimensional post-stack data denoising proposed in the embodiment of the present invention has a total of 6 layers, which is shallower than the original author's network depth. The reason is that if the network depth is too deep, the training process will tend to overfit, and if the network depth is too shallow, the extracted features will also become edge features, which is not conducive to denoising.

[0077] According to some embodiments of the present invention, when the basic denoising network model is trained using noisy seismic data and seismic noise data,

[0078] Constrained tomographic static correction and geometric diffusion compensation are used to preprocess the pre-stack and post-stack raw noisy seismic data.

[0079] The noise suppression method combination of frequency-division and partition-division surface wave suppression, frequency-division abnormal amplitude suppression, irregular coherent noise suppression, ringing suppression, and random noise suppression is used to suppress the noise of the pre-processed original noisy seismic data to obtain seismic noise data.

[0080] The original noisy seismic data and seismic noise data are segmented according to the training sample size to form training and test sample data sets.

[0081] In an optional embodiment, the above-mentioned noise suppression technology is used to process the seismic data, and the noisy seismic data and the separated noise data are segmented to generate 3,382,500 pairs of training sample sets of size 32*32*32 and 100 test sets of size 32*32*32. The training set and the test set are strictly non-intersecting.

[0082] In order to ensure the generalization ability of the network model for multi-type seismic noise suppression, the training and test sample data sets need to be normalized first, that is, each sample consists of post-stack seismic data with a value range between 0 and 1. The normalized training and test sample data sets are input into the basic denoising network model in batches, with the batch size set to 128, the number of channels set to 1, and the processed images are grayscale images. The Adaptive Moment Estimation optimization algorithm (Kingma et al., 2014) is used to optimize the network, and the Rectifier Linear Unit (ReLU) BN and RL technology are used to accelerate the training process and improve the denoising performance, improve the stability of the network and the nonlinear approximation ability, the learning rate is set to 0.0001, and the total training cycle is set to 10. One cycle traverses all samples in the training sample set, determines the model training parameters in the basic denoising network model, and obtains the intelligent denoising network model.

[0083] According to some embodiments of the present invention, when the intelligent denoising network model is used to denoise actual noisy seismic data, in order to ensure the consistency of the input of the intelligent denoising network model, the actual noisy pre-stack and post-stack seismic data are also normalized based on the minimum amplitude and maximum amplitude of the actual noisy pre-stack and post-stack seismic data, and the normalized actual noisy pre-stack and post-stack seismic data are input into the intelligent denoising network model to obtain predicted noise data, and then the residual learning strategy is adopted to subtract the actual noisy seismic data from the predicted noise data to obtain the intelligent noise suppression result and / or output the predicted residual image.

[0084] According to some embodiments of the present invention, Figure 3 A is a schematic diagram of pre-stack noisy seismic data. Figure 4 for Figure 3 Schematic diagram of the result after the mid-surface wave and linear coherent noise are suppressed by intelligent noise. Figure 3 and Figure 4 The vertical coordinate is time, and the horizontal coordinate is the contact line number, such as Figure 3 and Figure 4 As shown, after the pre-stack noisy seismic data A is denoised by the intelligent denoising method of the present invention, the surface wave and linear coherent noise are effectively suppressed; Figure 5 for Figure 3 and Figure 4 Schematic diagram of spectrum analysis, such as Figure 5 As shown in the figure, the intelligent denoising device effectively suppresses noises such as surface waves and linear interference before stacking, and at the same time plays a role in widening the effective frequency band.

[0085] According to some embodiments of the present invention, Figure 6 This is a schematic diagram of pre-stack noisy seismic data B. Figure 7 for Figure 6 Schematic diagram of the result after random noise is suppressed by intelligent noise. Figure 6 and Figure 7 The vertical coordinate is time, and the horizontal coordinate is the contact survey line number. Figure 6 and Figure 7 As shown, after the pre-stack noisy seismic data B is denoised by the intelligent denoising method of the present invention, the random noise is effectively suppressed; Figure 8 for Figure 6 and Figure 7 Schematic diagram of spectrum analysis, such as Figure 8 As shown, the intelligent denoising method of the present invention effectively suppresses random interference noise before stacking, broadens the effective frequency band, and greatly improves the signal-to-noise ratio of seismic data.

[0086] According to some embodiments of the present invention, Fig. 9 A is a schematic diagram of post-stack noisy seismic data. Fig.10 for Fig. 9Schematic diagram of the result after the random noise and the noise that was not removed before stacking are suppressed by intelligent noise. Fig. 9 and Fig.10 The horizontal coordinate is the contact line number, and the vertical coordinate is the time, such as Fig. 9 As shown in Figure 2, based on the two-dimensional post-stack noisy seismic data, it can be seen that random noise interference is serious and some noise is not suppressed before stacking, making the strong reflection axis blurred and the stratigraphic contact relationship unclear, which will bring difficulties to the accurate seismic geological interpretation, such as Fig.10 As shown, the patented method can effectively suppress random noise and other abnormal interference noise, ensure that there is no cross-layer phenomenon before and after denoising of two-dimensional seismic data, have good fidelity, and effectively improve the signal-to-noise ratio, which can provide support for the detailed interpretation work of subsequent interpreters.

[0087] According to some embodiments of the present invention, Fig.11 This is a schematic diagram of post-stack noisy seismic data B. Fig.12 for Fig.11 Schematic diagram of the result after the random noise and the noise that was not removed before stacking are suppressed by intelligent noise. Fig.11 and Fig.12 The horizontal coordinates are all the connection line numbers, and the vertical coordinates are time. Fig.11 , 12 As shown in the figure, the post-stack noisy seismic data B can improve the signal-to-noise ratio of the seismic profile after intelligent denoising, making the energy of the strong reflection axis more concentrated and not shifted.

[0088] The trained network was migrated and applied to the intelligent denoising of seismic data in other exploration areas, and the post-stack network was applied to the post-stack data, with good denoising effect.

[0089] According to some embodiments of the present invention, Fig.13 A is a schematic diagram of three-dimensional post-stack noisy seismic data. Fig.15 for Fig.13 Schematic diagram of the result after being suppressed by intelligent noise. Fig.13 and Fig.15 The longitudinal coordinate is the longitudinal survey line number, the transverse coordinate is the contact survey line number, and the vertical coordinate is time. Fig.15 As shown in the figure, the random noise interference of 3D post-stack noisy seismic data is serious, which makes the strong reflection axis blurred and the contact relationship of the strata unclear, which will bring difficulties to the accurate seismic geological interpretation, and some denoising methods will affect the amplitude and phase of the original seismic signal; Fig.17 As shown, the patented method can effectively suppress random noise, ensure that the three-dimensional seismic data remains closed before and after denoising, and the strong reflection axis does not shift, which can provide support for the subsequent interpretation work of interpreters.

[0090] Fig.14 for Fig.13Schematic diagram of the frequency-wavenumber spectrum of the results of intelligent noise suppression of 3D post-stack noisy seismic data. Fig.16 for Fig.15 Schematic diagram of the frequency-wavenumber spectrum of the results of intelligent noise suppression of 3D post-stack noisy seismic data. Fig.14 and Fig.16 The vertical coordinate is the frequency, and the horizontal coordinate is the wave number. Fig.14 and Fig.16 The FK spectrum of the noise predicted by the intelligent denoising network contains almost no effective signal, that is, the random noise and the effective signal are effectively separated, which can ensure the fidelity of the denoising result. The FK spectrum energy of the denoised data is stronger and more concentrated than that of the noisy data.

[0091] According to some embodiments of the present invention, Fig.17 This is a schematic diagram of 3D post-stack noisy seismic data B. Fig.18 for Fig.17 Schematic diagram of the results of intelligent noise suppression of 3D post-stack noisy seismic data. Fig.17 The data size in is 441*901*201, the spatial sampling interval is 10m, the time sampling interval is 4ms, the longitudinal coordinate is the longitudinal survey line number, the transverse coordinate is the contact survey line number, and the vertical coordinate is time. Fig.18 As shown in the figure, the three-dimensional post-stack noisy seismic data B can improve the signal-to-noise ratio of the seismic profile after intelligent denoising, making the energy of the strong reflection axis more concentrated and not shifted, and the stratigraphic contact relationship clearer.

[0092] The intelligent denoising method for multiple types of noise in the embodiment of the present invention can accurately and efficiently predict seismic noise from seismic data containing single or multiple types of noise. For a data volume of 4.54GB, the prediction can be completed in only 3.84 minutes using an NVIDIA V100 GPU. Although manual interpretation can ensure accuracy, it is time-consuming and labor-intensive.

[0093] In a second aspect, the present invention also provides an intelligent denoising device for multiple types of noise, such as Fig.19 As shown, the device comprises:

[0094] A basic denoising network model building module is used to build a basic denoising network model for multiple types of seismic noise based on a denoising convolutional neural network;

[0095] An intelligent denoising network model training module is used to train a basic denoising network model according to noisy seismic data and seismic noise data to obtain an intelligent denoising network model;

[0096] The intelligent denoising module is used to denoise the normalized noisy seismic data through a trained intelligent denoising network model to obtain an intelligent noise suppression result of the noisy seismic data.

[0097] In a third aspect, the present invention further provides a computer-readable storage medium having computer-readable instructions stored thereon, which, when executed by a processor, enables the processor to execute the above method.

[0098] In a fourth aspect, the present invention further provides a computer device, such as Fig. 20 As shown, the computer device includes a processor and a memory, wherein a computer program is stored in the memory, wherein when the computer program is executed by the processor, the processor executes the above method.

[0099] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent denoising method for multiple types of noise, characterized in that: The following steps are involved: Based on the denoising convolutional neural network, a basic denoising network model for multiple types of seismic noise is constructed; The basic denoising network model is trained using noisy seismic data and seismic noise data to obtain an intelligent denoising network model; The intelligent denoising network model is used to denoise actual noisy seismic data to obtain an intelligent noise suppression result.

2. The intelligent denoising method according to claim 1, characterized in that: The basic denoising network model for multiple types of seismic noise is constructed based on the denoising convolutional neural network, including: Modifying the number of convolutional layers and activation function of the denoising convolutional neural network so that it can be applied to the field of earthquake denoising; The modified denoising convolutional neural network is expanded into a three-dimensional network architecture to obtain a basic denoising network model.

3. The intelligent denoising method according to claim 2, characterized in that: The modified denoising convolutional neural network is expanded into a three-dimensional network architecture to obtain a basic denoising network model, including: The input and output sample dimensions, convolution kernel dimensions, and step size dimensions of the modified denoising convolutional neural network are expanded from two dimensions to three dimensions.

4. The intelligent denoising method according to claim 1, characterized in that: The method of using noisy seismic data and seismic noise data to train the basic denoising network model to obtain an intelligent denoising network model includes: De-noising the original noisy seismic data before and after stacking to obtain seismic noise data; Normalizing the original noisy seismic data and the seismic noise data according to the minimum amplitude value and the maximum amplitude value of the original noisy seismic data and the seismic noise data; The normalized original noisy seismic data and noise data are used to create a training sample data set; The basic denoising network model is trained using the training sample data set, model training parameters in the basic denoising network model are determined, and an intelligent denoising network model is obtained.

5. The intelligent denoising method according to claim 4, characterized in that: The process of denoising the original noisy seismic data before and after stacking to obtain seismic noise data includes: Constrained tomographic static correction and geometric diffusion compensation are used to preprocess the pre-stack and post-stack raw noisy seismic data. The noise suppression method combination of frequency-division and partition-division surface wave suppression, frequency-division abnormal amplitude suppression, irregular coherent noise suppression, ringing suppression, and random noise suppression is used to suppress the noise of the pre-processed original noisy seismic data to obtain seismic noise data. The original noisy seismic data and seismic noise data are segmented according to the training sample size to form training and test sample data sets.

6. The intelligent denoising method according to claim 5, characterized in that: The method of using noisy seismic data and seismic noise data to train the basic denoising network model to obtain an intelligent denoising network model also includes: Use BN and RL techniques to speed up the training process and improve the denoising performance of the intelligent denoising network model.

7. The intelligent denoising method according to claim 1, characterized in that: The method of using the intelligent denoising network model to denoise actual noisy seismic data to obtain an intelligent noise suppression result includes: Inputting the normalized actual noisy seismic data into the intelligent denoising network model, and outputting predicted noise data containing multiple seismic noises; The residual learning strategy is adopted to perform a difference between the actual noisy seismic data and the predicted noisy data containing multiple seismic noises to obtain an intelligent noise suppression result and / or output a predicted residual image.

8. The intelligent denoising method according to any one of claims 1 to 7, characterized in that: The seismic noise includes one or more of surface wave noise, linear coherent noise, random noise and abnormal interference noise.

9. An intelligent denoising device for multiple types of seismic noise, characterized in that: include: A basic denoising network model building module is used to build a basic denoising network model for multiple types of seismic noise based on a denoising convolutional neural network; An intelligent denoising network model training module is used to train the basic denoising network model according to the noisy seismic data and the seismic noise data to obtain an intelligent denoising network model; The intelligent denoising module is used to denoise the normalized actual noisy seismic data through the intelligent denoising network model to obtain the intelligent noise suppression result of the actual noisy seismic data.

10. A computer-readable storage medium, characterized in that: Computer-readable instructions are stored thereon, and when the instructions are executed by a processor, the processor is caused to execute the method according to any one of claims 1 to 8.

11. A computer device, characterized in that: include: processor; a memory storing a computer program, When the computer program is executed by the processor, the processor is caused to execute the method according to any one of claims 1 to 8.