A method and system for intelligently suppressing noise in seismic data

Abstract

The present invention relates to the technical field of seismic data processing for geological exploration, and in particular to a method and system for intelligently suppressing noise in seismic data. The method comprises the following steps: constructing a first seismic data noise suppression dataset based on effective seismic signals and random noise; constructing a first-level noise suppression model based on the first seismic data noise suppression dataset, and then constructing a second seismic data noise suppression dataset; constructing a second-level noise suppression model based on the second seismic data noise suppression dataset using a complementary set empirical mode decomposition method and a noise reduction autoencoder; using the first-level noise suppression model and the second-level noise suppression model to reduce noise in seismic data, and obtaining noise-suppressed seismic data through data reconstruction. The present invention performs multi-level noise reduction on seismic data by constructing the first-level noise suppression model and the second-level noise suppression model, thereby maintaining the continuity and high-frequency characteristics of the event axis of the seismic data and improving the noise suppression effect on the seismic data.

Description

Technical Field

[0001] The present invention relates to the technical field of geological exploration seismic data processing, and in particular to a method and system for intelligently suppressing seismic data noise. Background Art

[0002] Seismic exploration, a key tool for oil and gas exploration, relies on accurately extracting valuable geological information from seismic data. However, in practice, seismic data is inevitably subject to interference from various noise sources, including the surface environment, instrument errors, and subsurface heterogeneity. This noise not only severely reduces the signal-to-noise ratio of seismic data but can also compromise the continuity of events, posing significant challenges to subsequent geological interpretation and oil and gas resource assessment.

[0003] While traditional methods for suppressing noise in seismic data, such as filtering and denoising algorithms, have achieved some success in reducing noise, their effectiveness is often limited in complex and variable noise environments. In particular, when the signal and noise spectra overlap, traditional methods struggle to effectively remove the noise while maintaining signal integrity. This often results in the loss of high-frequency signal features, while low-frequency noise may persist, further complicating the analysis and interpretation of seismic data.

[0004] In recent years, rapid advances in machine learning have ushered in new hope for seismic data noise suppression. The application of advanced technologies such as generative adversarial networks (GANs) and deep convolutional neural networks (DnCNNs) has been particularly noteworthy. However, while pursuing higher noise suppression effectiveness, ensuring the continuity of seismic data events and preventing the loss of high-frequency features remain pressing challenges in this field. Summary of the Invention

[0005] In view of the defects in the prior art, the present invention provides a method and system for intelligently suppressing noise in seismic data.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a method for intelligently suppressing noise in seismic data, the method comprising the following steps: constructing a first seismic data noise suppression dataset for model training based on effective seismic signals and random noise; constructing a first-level noise suppression model based on the first seismic data noise suppression dataset, and then constructing a second seismic data noise suppression dataset for model training; constructing a second-level noise suppression model based on the second seismic data noise suppression dataset using a complementary set empirical mode decomposition method and a noise reduction autoencoder; using the first-level noise suppression model and the second-level noise suppression model to reduce noise in the seismic data, and obtaining the noise-suppressed seismic data through data reconstruction. The present invention performs multi-level noise reduction on the seismic data by constructing the first-level noise suppression model and the second-level noise suppression model, thereby maintaining the continuity and high-frequency characteristics of the event axis of the seismic data and improving the noise suppression effect on the seismic data.

[0007] Optionally, constructing a first seismic data noise suppression dataset for model training based on effective seismic signals and random noise includes the following steps:

[0008] Determine the effective seismic signal, and superimpose different random noises on the effective seismic signal to obtain multiple different channels of earthquake measured simulation signals;

[0009] The measured earthquake simulation signal is cut into a plurality of segment signals, and the first seismic data noise suppression data set is constructed using the effective earthquake signal and the segment signals.

[0010] Optionally, constructing a first-level noise suppression model based on the first seismic data noise suppression dataset, and then constructing a second seismic data noise suppression dataset for model training comprises the following steps:

[0011] Constructing a primary noise suppression model, and using the first seismic data noise suppression dataset to complete training and verification of the primary noise suppression model, thereby obtaining the first-level noise suppression model;

[0012] Performing first-level noise reduction on each of the segment signals in the first seismic data noise suppression dataset using the first-level noise suppression model to obtain corresponding first-level noise reduction data;

[0013] The second seismic data noise suppression data set for model training is constructed using the seismic effective signal and the first-level noise reduction data.

[0014] Optionally, the constructing of a primary noise suppression model and using the first seismic data noise suppression dataset to complete training and verification of the primary noise suppression model, thereby obtaining the first-level noise suppression model comprises the following steps:

[0015] Construct the feature extraction and enhancement module, the feature multi-scale integration module and the feature fusion output module in sequence;

[0016] Downsampling the output of the feature extraction and enhancement module as the input of the feature multi-scale integration module, using the output of the feature extraction and enhancement module as the first input of the feature fusion output module, and upsampling the output of the feature multi-scale integration module as the second input of the feature fusion output module to obtain the primary noise suppression model;

[0017] The first seismic data noise suppression data set is divided into a training set and a validation set, thereby completing the training and validation of the primary noise suppression model, and finally obtaining the first-level noise suppression model.

[0018] Optionally, the feature extraction and enhancement module satisfies the following relationship:

[0019] ,

[0020] Wherein, F is the input of the feature extraction and enhancement module, is the output of the feature extraction and enhancement module, Indicates that the convolution operation is performed using the first type of convolution layer. 、 、 and These are all reference symbols set for the convenience of writing. Indicates the use of ReLU activation function, Indicates the use of Sigmoid activation function, Indicates maximum pooling. Indicates average pooling.

[0021] Optionally, the feature multi-scale integration module satisfies the following relationship:

[0022] ,

[0023] in, is the output of the feature multi-scale integration module, The i-th reference symbol is set for the convenience of writing. , Indicates that the convolution operation is performed using the second type of convolution layer. Indicates that the third type of convolution layer is used for convolution operation. Indicates the use of ReLU activation function, Indicates the use of Sigmoid activation function, Indicates maximum pooling. Indicates average pooling. Indicates downsampling. Indicates upsampling. is the output of the feature extraction and enhancement module, Indicates that and Perform channel splicing.

[0024] Optionally, the feature fusion output module satisfies the following relationship:

[0025] ,

[0026] in, is the output of the feature fusion output module, Indicates the use of hyperbolic tangent activation function, Indicates that the convolution operation is performed using the first type of convolution layer. The jth reference symbol is set for the convenience of writing. , Indicates the use of ReLU activation function, Indicates the use of Sigmoid activation function, Indicates maximum pooling. Indicates average pooling. Indicates upsampling. is the output of the feature extraction and enhancement module, is the output of the feature multi-scale integration module, Indicates that and Perform channel splicing.

[0027] Optionally, constructing a secondary noise suppression model based on the second seismic data noise suppression dataset using a complementary set empirical mode decomposition method and a noise reduction autoencoder comprises the following steps:

[0028] Using the multiple IMF components of the first-level denoised data obtained by the complementary set empirical mode decomposition method as input to the denoising autoencoder to obtain the original second-level noise suppression model;

[0029] The second seismic data noise suppression data set is divided into a training set and a validation set, thereby completing the training and validation of the original secondary noise suppression model to obtain the secondary noise suppression model.

[0030] Optionally, the denoising of seismic data using the primary noise suppression model and the secondary noise suppression model, and obtaining the noise-suppressed seismic data through data reconstruction comprises the following steps:

[0031] Performing first-level noise reduction on seismic data using the first-level noise suppression model to obtain first-level noise reduction data;

[0032] Performing secondary noise reduction on seismic data using the secondary noise suppression model to obtain a set of noise-reduced IMF components;

[0033] The obtained noise-reduced IMF components are used to perform data reconstruction to obtain noise-suppressed seismic data.

[0034] In a second aspect, the present invention provides a system for intelligently suppressing noise from seismic data, the system comprising: a data acquisition device, a data output device, a processor and a storage device, the storage comprising a computer-readable storage medium, the computer-readable storage medium storing a computer program, the computer program comprising program instructions, which, when executed by the processor, enable the processor to implement a method for intelligently suppressing noise from seismic data according to the present invention.

[0035] The present invention has at least the following beneficial effects:

[0036] 1. In the feature extraction and enhancement module, feature multi-scale integration module, and feature fusion output module constructed by this method, the integrity of the input information is maintained by directly transferring shallow feature information to a deeper level through identity mapping. The residual error between input and output is effectively minimized, ensuring the stability and coherence of information during deep transmission. This helps to accurately extract the multi-scale features of seismic data, thereby improving the noise suppression effect of seismic data.

[0037] 2. This method first performs preliminary feature extraction and feature enhancement on seismic data through the feature extraction and enhancement module, so that more abundant feature information can be extracted subsequently; secondly, this method uses the feature multi-scale integration module to further extract the multi-scale features of the seismic data, realize the deep extraction of feature information, enrich the feature level, and preliminarily integrate the extracted feature information to obtain feature information of a level lower than the feature map extracted by the feature extraction and enhancement module; finally, this method uses the feature fusion output module to finally fuse the feature information extracted by the first two modules to obtain the first-level noise reduction data, thereby retaining the effective signal in the earthquake to the greatest extent, thereby improving the signal-to-noise ratio and event continuity of the seismic data.

[0038] 3. After obtaining the first-level denoised data, this method further constructs a second-level noise suppression model to perform second-level denoising on the first-level denoised data, so as to solve the problems of stubborn residual low-frequency noise and loss of high-frequency features that may exist in the first-level denoised data due to the overlap of seismic signals and noise spectra, thereby further improving the noise suppression effect of seismic data.

[0039] 4. The present invention provides a system compatible with the present method, which can stably operate the present method, improve the efficiency of noise suppression on seismic data, and improve the practicality of the present method. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 A schematic flow chart of a method for intelligently suppressing noise in seismic data according to an embodiment of the present invention;

[0042] Figure 2 A simplified structural diagram of a primary noise suppression model according to an embodiment of the present invention;

[0043] Figure 3 A simplified structural diagram of a feature extraction and enhancement module according to an embodiment of the present invention;

[0044] Figure 4 This is a simplified structural diagram of a feature multi-scale integration module according to an embodiment of the present invention;

[0045] Figure 5 This is a simplified structural diagram of a feature fusion output module according to an embodiment of the present invention;

[0046] Figure 6 Schematic diagram of the framework of a system for intelligently suppressing noise in seismic data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] Specific embodiments of the present invention will be described in detail below. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the present invention. In the following description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, it will be apparent to one of ordinary skill in the art that these specific details are not necessarily required to practice the present invention. In other instances, well-known circuits, software, or methods are not specifically described to avoid obscuring the present invention.

[0048] Throughout this specification, references to "one embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with the embodiment or example is included in at least one embodiment of the present invention. Therefore, appearances of the phrases "in one embodiment," "in an embodiment," "an example," or "an example" in various places throughout this specification are not necessarily all referring to the same embodiment or example. Furthermore, the particular features, structures, or characteristics may be combined in any suitable combinations and / or subcombinations in one or more embodiments or examples. Furthermore, those of ordinary skill in the art will appreciate that the figures provided herein are for illustrative purposes only and are not necessarily drawn to scale.

[0049] It should be noted in advance that, in an optional embodiment, except for independent explanations, the same symbols or letters appearing in all formulas have the same meanings and values.

[0050] In an alternative embodiment, see Figure 1 The present invention provides a method for intelligently suppressing noise in seismic data, the method comprising the following steps:

[0051] S1. Constructing a first seismic data noise suppression dataset for model training based on effective seismic signals and random noise.

[0052] The seismic data referred to in this embodiment refers to the measured seismic signals collected by sensors. Step S1 specifically includes the following steps:

[0053] S11. Determine the effective seismic signal, and superimpose different random noises on the effective seismic signal to obtain multiple different channels of earthquake measured simulation signals.

[0054] Specifically, in this embodiment, the Ricker wavelet is used as a clean signal without noise, i.e., an effective seismic signal. Then, different random noises are superimposed on the effective seismic signal to obtain multiple different channels of measured seismic simulation signals, i.e., measured seismic simulation signals collected by sensors at different locations.

[0055] S12. Cut the measured earthquake simulation signal into a plurality of segment signals, and use the effective earthquake signal and the segment signals to construct the first seismic data noise suppression dataset.

[0056] Specifically, in this embodiment, after obtaining the measured earthquake simulation signal, each measured earthquake simulation signal can be cut into multiple segment signals according to the set window size, and a segment signal and the earthquake effective signal corresponding to the segment signal are used as a set of seismic signal data, and then multiple sets of seismic signal data are used to construct the first seismic data noise suppression data set.

[0057] S2. Construct a first-level noise suppression model based on the first seismic data noise suppression dataset, and then construct a second seismic data noise suppression dataset for model training.

[0058] Wherein, step S2 specifically includes the following steps:

[0059] S21. Construct a primary noise suppression model, and use the first seismic data noise suppression dataset to complete training and verification of the primary noise suppression model, thereby obtaining the first-level noise suppression model.

[0060] Among them, see Figure 2 、 Figure 3 、 Figure 4 and Figure 5 , step S21 specifically includes the following steps:

[0061] S211, constructing the feature extraction and enhancement module, the feature multi-scale integration module and the feature fusion output module in sequence.

[0062] Specifically, in this embodiment, a feature extraction and enhancement module is constructed. In the feature extraction and enhancement module, a first-class convolutional layer and a ReLU activation function are first used to more carefully extract the local features of the input seismic data, and then a first feature enhancement network FES1 similar to the residual structure is used to enhance the extracted local features to obtain enhanced local features, thereby laying a solid foundation for first-level noise suppression. In the first feature enhancement structure FES1, jump connections are used many times, which enables the network to map part of the information identity to a deeper level when performing deep feature learning, reduce information loss, alleviate the problem of gradient disappearance or explosion in the deep network training process, ensure the stability and coherence of information in the deep transmission process, and help improve the noise suppression effect on seismic data. The function of the feature extraction and enhancement module can be expressed using the following relationship:

[0063] ,

[0064] Among them, F is the input of the feature extraction and enhancement module, is the output of the feature extraction and enhancement module, Indicates that the convolution operation is performed using the first type of convolution layer. 、 、 and These are all reference symbols set for the convenience of writing. Indicates the use of ReLU activation function, Indicates the use of Sigmoid activation function, Indicates maximum pooling. Indicates average pooling. The first type of convolution layer has 64 convolution kernels with a kernel size of 2×2 and a stride of 1. At the same time, the padding parameter of the convolution layer is set to 'same' so that the feature map output by the convolution layer has the same size as the input feature map.

[0065] Furthermore, a feature multi-scale integration module is constructed. The output of the feature extraction and enhancement module is downsampled and input into the second feature enhancement structure FES2, and then the output of the second feature enhancement structure FES2 is downsampled and input into the third feature enhancement structure FES3, thereby extracting the multi-scale features of the seismic data, realizing deep extraction of feature information, and enriching the feature hierarchy. Next, the output of the third feature enhancement structure FES3 is upsampled and channel-joined with the output of the second feature enhancement structure FES2, and a second-class convolutional layer and the second feature enhancement structure FES2 are used to preliminarily integrate the extracted feature information to obtain feature information one level lower than the feature map extracted by the feature extraction and enhancement module. Among them, the second feature enhancement structure FES2 uses the second-class convolutional layer, and the third feature enhancement structure FES3 uses the third-class convolutional layer. The second feature enhancement structure FES2 and the third feature enhancement structure FES3 have the same network structure as the first feature enhancement structure FES1. The second type of convolution layer has 128 convolution kernels, the convolution kernel size is 2×2, and the stride is 1; the third type of convolution layer has 256 convolution kernels, the convolution kernel size is 2×2, and the stride is 1. The function of the feature multi-scale integration module can be expressed using the following relationship:

[0066] ,

[0067] in, is the output of the feature multi-scale integration module, The i-th reference symbol is set for the convenience of writing. , Indicates that the convolution operation is performed using the second type of convolution layer. Indicates that the third type of convolution layer is used for convolution operation. Indicates downsampling. Indicates upsampling. Indicates channel splicing.

[0068] Furthermore, a feature fusion output module is constructed. First, the output of the feature multi-scale integration module is upsampled and then channel-joined with the output of the feature extraction and enhancement module; then, a first-class convolutional layer and ReLU activation function are used to reduce the number of channels, and the first feature enhancement network FES1 is further used to refine the feature processing to ensure the effective integration of multi-scale feature information; finally, the first-class convolutional layer and ReLU activation function are used to fuse the feature information, and the first-class convolutional layer and hyperbolic tangent activation function are used to convert the finely processed feature map into first-level noise reduction data, thereby retaining the effective signal in the earthquake to the greatest extent, thereby improving the signal-to-noise ratio and event continuity of the seismic data. The function of the feature fusion output module can be expressed using the following relationship:

[0069] ,

[0070] in, is the output of the feature fusion output module, Indicates the use of hyperbolic tangent activation function, Indicates that the convolution operation is performed using the first type of convolution layer. The jth reference symbol is set for the convenience of writing. .

[0071] S212. Downsample the output of the feature extraction and enhancement module and use it as the input of the feature multi-scale integration module; use the output of the feature extraction and enhancement module as the first input of the feature fusion output module; and upsample the output of the feature multi-scale integration module and use it as the second input of the feature fusion output module to obtain the primary noise suppression model.

[0072] S213. Divide the first seismic data noise suppression data set into a training set and a validation set, thereby completing the training and validation of the primary noise suppression model, and finally obtaining the first-level noise suppression model.

[0073] Specifically, in this embodiment, the first seismic data noise suppression data set is divided into a training set and a validation set in a ratio of 7:3, thereby completing the training and validation of the primary noise suppression model and obtaining a first-level noise suppression model.

[0074] S22. Performing first-level noise reduction on each of the segment signals in the first seismic data noise suppression dataset using the first-level noise reduction model to obtain corresponding first-level noise reduction data.

[0075] Specifically, in this embodiment, a first-level noise suppression model is used to perform first-level noise reduction on each segment signal to obtain corresponding first-level noise reduction data.

[0076] S23. Use the effective seismic signal and the first-level noise reduction data to construct the second seismic data noise suppression data set for model training.

[0077] Specifically, in this embodiment, the second noise-suppressed seismic data dataset is constructed in a manner similar to that of the first noise-suppressed seismic data dataset. In fact, the second noise-suppressed seismic data dataset is obtained by replacing the segmented signals in the first noise-suppressed seismic data dataset with the corresponding first-level noise-reduced data.

[0078] S3. Based on the second seismic data noise suppression dataset, a secondary noise suppression model is constructed using a complementary set empirical mode decomposition method and a noise reduction autoencoder.

[0079] Wherein, step S3 specifically includes the following steps:

[0080] S31. Using the multiple IMF components of the primary denoised data obtained by the complementary set empirical mode decomposition method as input to a denoising autoencoder to obtain an original secondary noise suppression model.

[0081] Specifically, in this embodiment, complementary ensemble empirical mode decomposition (CEMD) is a variant of CEMD that has better stability than CEMD and can adaptively process input data. A denoising autoencoder, based on the autoencoder, adds noise to the input data to train the entire network. The ultimate goal is to reconstruct the original data without noise, eliminate the effects of random destruction, and retain the high-frequency valid signal after modal decomposition. Therefore, this embodiment combines CEMD and denoising autoencoder to construct a two-level noise suppression model.

[0082] S32. Divide the second seismic data noise suppression data set into a training set and a validation set, thereby completing the training and validation of the original secondary noise suppression model to obtain the secondary noise suppression model.

[0083] Specifically, in this embodiment, the second seismic data noise suppression dataset is divided into a training set and a validation set in a ratio of 7:3.

[0084] S4. Use the primary noise suppression model and the secondary noise suppression model to reduce noise on the seismic data, and obtain the noise-suppressed seismic data through data reconstruction.

[0085] Wherein, step S4 specifically includes the following steps:

[0086] S41. Performing first-level noise reduction on the seismic data using the first-level noise suppression model to obtain first-level noise reduction data.

[0087] Specifically, in this embodiment, a first-level noise suppression model is used to perform first-level noise reduction on seismic data collected in real time to obtain corresponding first-level noise reduction data, thereby achieving preliminary suppression of noise.

[0088] S42. Perform secondary noise reduction on the seismic data using the secondary noise suppression model to obtain a set of noise-reduced IMF components.

[0089] Specifically, in this embodiment, since the first-level denoised data obtained in step S41 may contain stubborn residual low-frequency noise and loss of high-frequency features due to the overlap of the seismic signal and the noise spectrum, this embodiment uses a second-level noise suppression model to perform second-level denoising on the first-level denoised data obtained in step S41 to obtain a set of denoised IMF components, and then the noise-suppressed seismic data can be obtained through data reconstruction.

[0090] S43. Reconstruct data using the obtained noise-reduced IMF components to obtain noise-suppressed seismic data.

[0091] Specifically, in this embodiment, this step is a conventional technical means and will not be described in detail here.

[0092] It should be noted that, in some cases, the actions described in the specification can be performed in a different order and still achieve the desired results. In this embodiment, the order of steps given is only to make the embodiment appear clearer and easier to explain, rather than to limit it.

[0093] In an alternative embodiment, see Figure 6 The present invention provides an intelligent noise suppression system for seismic data, which includes: a data acquisition device A1, a data output device A2, a processor A3 and a storage A4. The storage A4 includes a computer-readable storage medium, and the computer-readable storage medium stores a computer program. The computer program includes program instructions. When the program instructions are executed by the processor A3, the processor A3 executes the contents of steps S1 to S4.

[0094] Specifically, in this embodiment, when the computer program executed by the processor A3 implements the contents described in steps S1-S4, the data acquisition device A1 is used to acquire seismic data, and the data output device A2 is used to output the seismic data after noise suppression.

[0095] In summary, the present invention performs multi-level noise reduction on seismic data by constructing a first-level noise suppression model and a second-level noise suppression model, thereby achieving noise suppression on seismic data. Among them, the method performs preliminary feature extraction and feature enhancement on seismic data through a feature extraction and enhancement module, so that more abundant feature information can be extracted later; the multi-scale features of the seismic data are further extracted through a feature multi-scale integration module, the depth of feature information is extracted, the feature level is enriched, and the extracted feature information is preliminarily integrated to obtain feature information of a level lower than the feature map extracted by the feature extraction and enhancement module; the feature information extracted by the first two modules is finally fused through a feature fusion output module to obtain first-level noise reduction data, thereby retaining the effective signal in the earthquake to the greatest extent, thereby improving the signal-to-noise ratio and event continuity of the seismic data. After obtaining the first-level noise reduction data, the method further constructs a second-level noise suppression model to perform second-level noise reduction on the first-level noise reduction data, so as to solve the problem of stubborn residual low-frequency noise and loss of high-frequency features when the seismic signal overlaps with the noise spectrum in the first-level noise reduction data, thereby further improving the noise suppression effect on the seismic data. In addition, the system can stably run the method, improve the efficiency of noise suppression on seismic data, and improve the practicality of the method.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. A method for intelligently suppressing noise in seismic data, characterized in that: The steps include: Constructing a first seismic data noise suppression dataset for model training based on seismic effective signals and random noise; Construct the feature extraction and enhancement module, the feature multi-scale integration module and the feature fusion output module in sequence; Downsampling the output of the feature extraction and enhancement module as the input of the feature multi-scale integration module, using the output of the feature extraction and enhancement module as the first input of the feature fusion output module, and upsampling the output of the feature multi-scale integration module as the second input of the feature fusion output module to obtain a primary noise suppression model; In the feature extraction and enhancement module, a first-class convolutional layer and ReLU activation function are first used to extract local features of the input seismic data in a relatively detailed manner. Then, a first feature enhancement network FES1 similar to the residual structure is used to enhance the extracted local features. In the first feature enhancement structure FES1, skip connections are used multiple times. The feature extraction and enhancement module satisfies the following relationship: ， Wherein, F is the input of the feature extraction and enhancement module, is the output of the feature extraction and enhancement module, Indicates that the convolution operation is performed using the first type of convolution layer. 、 、 and These are all reference symbols set for the convenience of writing. Indicates the use of ReLU activation function, Indicates the use of Sigmoid activation function, Indicates maximum pooling. Indicates average pooling; The output of the feature extraction and enhancement module is downsampled and input into the second feature enhancement structure FES2, and then the output of the second feature enhancement structure FES2 is downsampled and input into the third feature enhancement structure FES3, the output of the third feature enhancement structure FES3 is upsampled and channel-joined with the output of the second feature enhancement structure FES2, and a second-class convolutional layer and the second feature enhancement structure FES2 are used to preliminarily integrate the extracted feature information to obtain feature information of a level lower than the feature map extracted by the feature extraction and enhancement module; The feature multi-scale integration module satisfies the following relationship: ， in, is the output of the feature multi-scale integration module, The i-th reference symbol is set for the convenience of writing. , Indicates that the convolution operation is performed using the second type of convolution layer. Indicates that the third type of convolution layer is used for convolution operation. Indicates the use of ReLU activation function, Indicates the use of Sigmoid activation function, Indicates maximum pooling. Indicates average pooling. Indicates downsampling. Indicates upsampling. is the output of the feature extraction and enhancement module, Indicates channel splicing; The output of the feature multi-scale integration module is upsampled and then channel-joined with the output of the feature extraction and enhancement module. A first-class convolutional layer and a ReLU activation function are then used to reduce the number of channels. The first feature enhancement network FES1 is further used to refine the feature processing. Finally, the first-class convolutional layer and a ReLU activation function are used to fuse the feature information. The first-class convolutional layer and a hyperbolic tangent activation function are used to convert the finely processed feature map into first-level denoised data. The feature fusion output module satisfies the following relationship: ， in, is the output of the feature fusion output module, Indicates the use of hyperbolic tangent activation function, Indicates that the convolution operation is performed using the first type of convolution layer. The jth reference symbol is set for the convenience of writing. , Indicates the use of ReLU activation function, Indicates the use of Sigmoid activation function, Indicates maximum pooling. Indicates average pooling. Indicates upsampling. is the output of the feature extraction and enhancement module, is the output of the feature multi-scale integration module, Indicates channel splicing; Dividing the first seismic data noise suppression data set into a training set and a validation set, thereby completing the training and validation of the primary noise suppression model, and finally obtaining a first-level noise suppression model; Performing first-level noise reduction on each segment signal in the first seismic data noise suppression dataset using the first-level noise reduction model to obtain corresponding first-level noise reduction data; Constructing a second seismic data noise suppression dataset for model training using the seismic effective signal and the first-level noise reduction data; Based on the second seismic data noise suppression dataset, a secondary noise suppression model is constructed using a complementary set empirical mode decomposition method and a noise reduction autoencoder; The first-level noise suppression model and the second-level noise suppression model are used to reduce noise on seismic data, and the noise-suppressed seismic data are obtained through data reconstruction.

2. The method for intelligently suppressing noise in seismic data according to claim 1, characterized in that: The method of constructing a first seismic data noise suppression dataset for model training based on seismic effective signals and random noise comprises the following steps: Determine the effective seismic signal, and superimpose different random noises on the effective seismic signal to obtain multiple different channels of earthquake measured simulation signals; The measured earthquake simulation signal is cut into a plurality of segment signals, and the first seismic data noise suppression data set is constructed using the effective earthquake signal and the segment signals.

3. The method for intelligently suppressing seismic data noise according to claim 1, characterized in that: Based on the second seismic data noise suppression dataset, constructing a secondary noise suppression model using a complementary set empirical mode decomposition method and a noise reduction autoencoder includes the following steps: Using the multiple IMF components of the first-level denoised data obtained by the complementary set empirical mode decomposition method as input to the denoising autoencoder to obtain the original second-level noise suppression model; The second seismic data noise suppression data set is divided into a training set and a validation set, thereby completing the training and validation of the original secondary noise suppression model to obtain the secondary noise suppression model.

4. The method for intelligently suppressing noise in seismic data according to claim 1, characterized in that: The method of using the first-level noise suppression model and the second-level noise suppression model to reduce noise on seismic data and obtaining noise-suppressed seismic data through data reconstruction comprises the following steps: Performing first-level noise reduction on seismic data using the first-level noise suppression model to obtain first-level noise reduction data; Performing secondary noise reduction on seismic data using the secondary noise suppression model to obtain a set of noise-reduced IMF components; The obtained noise-reduced IMF components are used to perform data reconstruction to obtain noise-suppressed seismic data.

5. An intelligent noise suppression system for seismic data, characterized in that: The intelligent noise suppression system for seismic data includes: a data acquisition device, a data output device, a processor and a storage device, the storage includes a computer-readable storage medium, the computer-readable storage medium stores a computer program, the computer program includes program instructions, and when the program instructions are executed by the processor, the processor implements the intelligent noise suppression method for seismic data as described in any one of claims 1 to 4.

Images