Marine seismic multiple suppression method and device, electronic equipment and storage medium

Abstract

The invention discloses a marine seismic multiple suppression method and device, electronic equipment and a storage medium, and relates to the technical field of data processing, and the method comprises the steps: constructing a marine stratum model; performing forward modeling on the ocean stratum model to generate initial seismic shot gather data with and without multiple waves; expanding the initial seismic shot gather data to obtain target seismic shot gather data; constructing a deep learning model; training a deep learning model according to the target seismic shot gather data to obtain a trained deep learning model; and suppressing multiple waves in the seismic data by using the trained deep learning model. According to the method, the initial seismic shot gather data is generated through forward modeling and is expanded into the target seismic shot gather data, so that rich and diversified sample data is obtained; according to the method, the deep learning model including the depth separable convolution structure is constructed, the parameter quantity of the deep learning model is reduced, a lightweight model is realized, and multiple waves in seismic data can be accurately suppressed according to the trained deep learning model.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, electronic device and storage medium for suppressing multiple waves of marine earthquakes. Background Art

[0002] In marine seismic exploration, due to the existence of strong impedance surfaces such as the seafloor interface and sea level, the collected seismic data often inevitably contain a large number of different types of multiple wave interference. As a kind of noise with strong energy and wide distribution, multiple waves will hinder the acquisition of effective wave information, affect the resolution and signal-to-noise ratio of seismic profiles, and cause great difficulties in seismic data velocity analysis, pre-stack and post-stack migration, thus forming false and unreliable seismic imaging results, which in turn affects the accuracy of seismic data interpretation. Summary of the invention

[0003] The main purpose of the embodiments of the present application is to propose a method, device, electronic device and storage medium for suppressing multiple waves of marine earthquakes, so as to suppress multiple waves in seismic data.

[0004] To achieve the above purpose, an embodiment of the present application provides a method for suppressing marine seismic multiple waves, the method comprising the following steps: Constructing ocean stratigraphic models; Performing forward simulation on the marine stratum model to generate initial seismic shot gather data; wherein the initial seismic shot gather data includes first seismic data with multiple waves and second seismic data without multiple waves; Expanding the initial seismic shot gather data to obtain target seismic shot gather data; Constructing a deep learning model; wherein the deep learning model includes a feature extraction network and a feature fusion network, and the feature extraction network includes a deep separable convolution structure; Training the deep learning model according to the target seismic shot gather data to obtain the trained deep learning model; The trained deep learning model is used to suppress multiple waves in seismic data.

[0005] In some embodiments, constructing the ocean stratum model comprises the following steps: Construct the marine stratum model of seawater layers at different depths; wherein the marine stratum model includes a horizontal layer model, an inclined stratum model, a fault model and a salt dome model.

[0006] In some embodiments, the forward modeling of the marine stratum model to generate initial seismic shot gather data comprises the following steps: The marine stratum models are divided into regional type, boundary type and coupled type respectively; Setting boundary conditions of the free surface of seawater for the regional type, boundary type and coupling type in each of the marine stratum models respectively; wherein the boundary conditions include reflection boundaries and absorption boundaries; Finite difference method forward simulation is performed on each of the marine stratum models using different observation models to generate the first seismic data with multiple waves based on the reflection boundary, and the second seismic data without multiple waves based on the absorption boundary.

[0007] In some embodiments, the step of expanding the initial seismic shot gather data to obtain target seismic shot gather data comprises at least one of the following steps: adding Gaussian white noise of different intensities to the initial seismic shot gather data; Alternatively, performing random trace deletion operations of different proportions on the initial seismic shot gather data; Alternatively, the initial seismic shot gather data are rotated at different angles.

[0008] In some embodiments, constructing a deep learning model comprises the following steps: Construct the deep learning model according to the Mobilenet-Unet network structure; The depth-wise separable convolution structure is added to the deep learning model; wherein the depth-wise separable convolution structure includes a depth-wise convolution structure and a point-wise convolution structure, the depth-wise convolution structure is used to extract spatial features, and the point-by-point convolution structure is used to extract channel features.

[0009] In some embodiments, training the deep learning model according to the target seismic shot gather data to obtain the trained deep learning model comprises the following steps: Initializing the input data size, output data size, and parameter weights of the deep learning model; The target seismic shot gather data are labeled and standardized to unify the data size to the input data size to obtain training data; wherein the first seismic data are labeled as training samples, and the second seismic data are labeled as training labels; Determining a loss function and an optimization function for training the deep learning model; The initialized deep learning model is trained using the training data and according to the loss function and the optimization function until the trained deep learning model is obtained after a preset condition is met.

[0010] In some embodiments, the method of suppressing multiple waves in seismic data using the trained deep learning model comprises the following steps: The seismic data with multiple waves are input into the trained deep learning model to suppress the multiple waves in the input data and obtain the seismic data after the multiple waves are suppressed.

[0011] To achieve the above-mentioned purpose, another aspect of the embodiment of the present application provides a marine seismic multiple wave suppression device, the device comprising: A first model building unit, used for building an ocean stratum model; A data simulation unit, used for forward modeling the marine stratum model to generate initial seismic shot gather data; wherein the initial seismic shot gather data includes first seismic data with multiple waves and second seismic data without multiple waves; A data expansion unit, used for expanding the initial seismic shot gather data to obtain target seismic shot gather data; A second model construction unit is used to construct a deep learning model; wherein the deep learning model includes a feature extraction network and a feature fusion network, and the feature extraction network includes a deep separable convolution structure; A model training unit, used for training the deep learning model according to the target seismic shot gather data to obtain the trained deep learning model; A multiple wave suppression unit is used to suppress multiple waves in seismic data using the trained deep learning model.

[0012] To achieve the above-mentioned purpose, another aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned method for suppressing multiple waves of marine earthquakes when executing the computer program.

[0013] To achieve the above objectives, another aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for suppressing multiple waves of marine earthquakes is implemented.

[0014] The embodiments of the present application include at least the following beneficial effects: The present application can construct an ocean stratigraphic model; forward model the ocean stratigraphic model to generate initial seismic shot data; wherein the initial seismic shot data includes a first seismic data with multiple waves and a second seismic data without multiple waves; expand the initial seismic shot data to obtain target seismic shot data; construct a deep learning model; wherein the deep learning model includes a feature extraction network and a feature fusion network, and the feature extraction network includes a deep separable convolution structure; train the deep learning model according to the target seismic shot data to obtain a trained deep learning model; use the trained deep learning model to suppress multiple waves in the seismic data. The present application simulates an ocean stratigraphic model and then forward model the initial seismic shot data, and expands it into target seismic shot data, thereby obtaining rich and diverse sample data; constructs a deep learning model including a deep separable convolution structure, which can reduce the number of parameters of the deep learning model and realize a lightweight model, and uses a variety of target seismic shot data to train a lightweight deep learning model, and then according to the trained deep learning model, multiple waves in the seismic data can be accurately and efficiently suppressed. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1 A schematic diagram of a process for suppressing multiple waves of an ocean earthquake provided in an embodiment of the present application; Figure 2 An example diagram of an ocean stratum model provided in an embodiment of the present application; Figure 3 An example diagram of seismic data generated by different boundary conditions in the forward modeling process provided in the embodiment of the present application; Figure 4 An example diagram of the expanded seismic data provided in the embodiment of the present application; Figure 5 The Mobilenet-Unet network structure diagram provided for the embodiment of the present application; Figure 6 The standard convolution structure and depth-separable convolution structure diagram provided in the embodiments of the present application; Figure 7 An example flow chart of a method for suppressing multiple waves of marine earthquakes provided in an embodiment of the present application; Figure 8 A graph of loss values ​​corresponding to the training set and the validation set during the deep learning model training provided in the embodiment of the present application; Fig. 9 This is a diagram showing the multiple wave suppression effect of the deep learning model provided in the embodiment of the present application in the test set; Fig.10 A schematic diagram of the structure of a marine seismic multiple wave suppression device provided in an embodiment of the present application; Fig.11 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. They are only examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the attached claims.

[0018] It is understood that the terms "first", "second", etc. used in this application can be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another concept. For example, without departing from the scope of the embodiment of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein can be interpreted as "at the time of" or "when" or "in response to determination".

[0019] The terms "at least one", "multiple", "each", "any", etc. used in this application, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0021] Before describing the embodiments of the present application in detail, some related technologies involved in the embodiments of the present application are first described as follows: In marine seismic exploration, due to the existence of strong impedance surfaces such as the seafloor interface and sea level, the collected seismic data often inevitably contain a large number of different types of multiple wave interference. As a kind of noise with strong energy and wide distribution, multiple waves will hinder the acquisition of effective wave information, affect the resolution and signal-to-noise ratio of seismic profiles, and lead to great difficulties in seismic data velocity analysis, pre-stack and post-stack migration, thus forming false and unreliable seismic imaging results, and further affecting the accuracy of seismic data interpretation. Therefore, multiple wave identification and removal technology has always been a key factor restricting the accuracy of marine seismic data processing. How to effectively identify multiple types of multiple waves and adopt appropriate suppression methods in the process of marine seismic data processing so that it can retain effective seismic wave information to the maximum extent is of great significance to improving the accuracy of subsequent marine seismic data interpretation. In particular, with the emergence of wide-azimuth and high-density marine seismic data acquisition methods, it is often necessary to suppress multiple waves for massive seismic exploration data. It is particularly important to study new fast and intelligent multiple wave suppression methods.

[0022] At present, traditional multiple wave suppression methods are divided into two categories: filtering methods based on the difference in seismic attributes between effective waves and multiple waves, and predictive subtraction methods based on wave equation theory. The filtering method mainly uses the differences in apparent velocity between primary waves and multiple waves, the differences in dip angles on stacked sections, the differences in frequency domain, and the differences in periodicity, and suppresses multiple waves through different transformation methods. Such as the commonly used predictive deconvolution, beam filtering method, and Radon domain filtering method. The method based on wave theory uses data-driven or model-driven methods to construct multiple wave models in the form of original data, and then uses adaptive subtraction algorithms to achieve multiple wave suppression of seismic data. For example, the model-driven wave field extrapolation method and the data-driven inverse scattering series method, feedback iteration method, and virtual phase axis method. And the widely used surface-related multiple removal (SRME) method and the emerging sparse inversion primary wave estimation method (EPSI). In addition, deep learning methods have also made certain progress in the application of multiple wave removal. Through deep convolutional neural networks and network model training, a mapping between seismic data with multiple waves and seismic data with multiple waves removed is formed, thereby achieving direct removal of multiple waves from seismic data. For example, generative adversarial networks, improved U-net networks, and neural network models with self-attention mechanisms are used to construct deep learning network models for multiple wave removal.

[0023] Disadvantages of existing technology: Traditional multiple wave suppression methods have certain shortcomings. Predictive deconvolution uses the regularity of the time difference period of multiple waves in seismic data, and has the advantages of simple prediction compensation settings and fast running speed. For multiple waves generated by rugged seabed, because they do not fully conform to the periodicity, the multiple wave suppression effect is not good. The τ-p domain predictive deconvolution developed later has improved it, but in actual shallow water data applications, there is a risk of damaging effective signals; the high-precision Radon transform method is based on the difference between the effective wave velocity and the multiple wave velocity to suppress in the Radon domain. It is suitable for attenuating multiple waves in medium and long offsets, but has a poor suppression effect on near-channel multiple waves. The SRME (Surface-Related Multiple Elimination) method is a multiple wave suppression technology widely used in marine seismic data processing in recent years. This technical method does not require prior information such as velocity and layer position. It predicts multiple waves through data-driven prediction and performs adaptive subtraction on the gather to achieve the effect of removing multiple waves. It has a good attenuation effect on long-period multiple waves related to the free surface. However, the SRME method has a strong dependence on data integrity and the prediction effect of near-channel multiple waves is not ideal. The sparse inversion primary wave estimation method EPSI (Estimation of Primaries by Sparse Inversion) uses the information of the entire wave field for solution. Since there is no process of adaptively subtracting multiple waves, it can reduce the damage to the valid signal overlapping with the multiple waves. At present, the algorithm requires huge memory consumption and computing time, so it cannot be applied to actual data in industry. In general, the multiple wave model subtraction method of wave theory has great difficulty in predicting multiple waves in complex media, and the attenuation effect of multiple waves between layers is poor. Therefore, in actual processing, it is not suitable for all types of data, and the processing effect is limited. Although the multiple wave removal method based on deep learning methods has been developed to a certain extent, and has been improved in intelligence and multiple wave processing efficiency, these deep learning models often have a large number of parameters and a large amount of model calculation. In addition, each time a new type of multiple wave is encountered, the deep learning network is retrained with the help of transfer learning, and the generalization performance is insufficient.

[0024] Through the investigation of the background knowledge of marine seismic multiple wave suppression methods, it can be found that although there are many studies on the suppression of marine seismic exploration multiple wave methods, most of the methods based on traditional filtering or wave equation theory have certain limitations. Although the method of suppressing multiple waves based on deep learning has also made some progress, it has the disadvantages of high complexity of deep convolutional neural network models and poor generalization application. Overall, there is a lack of a convenient, low-computation, general, intelligent and efficient multiple wave suppression method in marine seismic multiple wave processing methods.

[0025] The embodiment of the present application provides a method, device, electronic device and storage medium for suppressing multiple waves of marine seismic. The technical solution of the present application includes: constructing a marine stratum model; forward modeling the marine stratum model to generate initial seismic shot data; wherein the initial seismic shot data includes first seismic data with multiple waves and second seismic data without multiple waves; expanding the initial seismic shot data to obtain target seismic shot data; constructing a deep learning model; wherein the deep learning model includes a feature extraction network and a feature fusion network, and the feature extraction network includes a deep separable convolution structure; training the deep learning model according to the target seismic shot data to obtain a trained deep learning model; using the trained deep learning model to suppress multiple waves in the seismic data. The present application simulates the marine stratum model to generate initial seismic shot data forward, and expands it into target seismic shot data, thereby obtaining rich and diverse sample data; constructing a deep learning model including a deep separable convolution structure, which can reduce the number of parameters of the deep learning model, realize a lightweight model, and use a variety of target seismic shot data to train a lightweight deep learning model, and then according to the trained deep learning model, multiple waves in the seismic data can be accurately and efficiently suppressed.

[0026] The embodiment of the present application provides a method for suppressing multiple waves of marine earthquakes, and relates to the field of data processing technology. The method for suppressing multiple waves of marine earthquakes provided in the embodiment of the present application can be applied to a terminal, can also be applied to a server, and can also be software running in a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, and a car terminal, etc., but is not limited to this; the server side can be configured as an independent physical server, or it can be configured as a server cluster or a distributed system composed of multiple physical servers, and can also be configured to provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms and other basic cloud computing services. The cloud server, the server can also be a node server in a blockchain network; the software can be an application that implements the method for suppressing multiple waves of marine earthquakes, etc., but is not limited to the above forms.

[0027] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0028] Reference Figure 1 The present application embodiment provides a method for suppressing multiple waves of an ocean earthquake. The method may include but is not limited to steps S100 to S150, which are as follows: S100: Construct ocean stratigraphic model.

[0029] It is understandable that the marine strata include different landforms and geological structures. This embodiment can construct a corresponding marine strata model for a specific marine strata to simulate the marine strata in the study area.

[0030] Further, step S100 may include step S101: S101: constructing the marine stratum model of seawater layers at different depths; wherein the marine stratum model includes a horizontal layer model, an inclined stratum model, a fault model and a salt dome model.

[0031] Exemplarily, in order to obtain seismic data containing multiple waves under different marine strata conditions and acquire a multi-category training data set, this embodiment can construct marine strata models of seawater layers at different depths based on actual marine strata models, such as horizontal layered models, inclined strata models, fault models, salt dome models and other marine strata models. Figure 2 The optional marine stratum model example diagrams provided for this embodiment are, from left to right and from top to bottom, example diagrams of a horizontal layered model, an inclined stratum model, a fault model, and a salt dome model.

[0032] S110: forward modeling the marine stratum model to generate initial seismic shot gather data; wherein the initial seismic shot gather data includes first seismic data with multiple waves and second seismic data without multiple waves.

[0033] Specifically, the forward modeling of this embodiment may be to apply seismic waves to the simulated ocean stratum model, and then collect seismic echoes as initial seismic shot gather data.

[0034] The simulated seismic echoes may include multiple waves or not include multiple waves. In this embodiment, the seismic echoes including multiple waves may be used as the first seismic data, and the seismic data without multiple waves may be used as the second seismic data.

[0035] Further, step S110 may include steps S111 to S113: S111: dividing each of the marine stratum models into a regional type, a boundary type and a coupled type; S112: setting boundary conditions of the free surface of seawater for the regional type, boundary type and coupling type in each of the marine stratum models respectively; wherein the boundary conditions include reflection boundaries and absorption boundaries; S113: Perform finite difference forward simulation on each of the marine stratum models using different observation models, thereby generating the first seismic data with multiple waves based on the reflection boundary, and generating the second seismic data without multiple waves based on the absorption boundary.

[0036] Specifically, when the forward modeling of this embodiment generates seismic data, the region of the marine stratigraphic model can be discretized according to the calculation needs, for example, divided into regional type, boundary type and coupled type. The regional type numerical solution is mainly the finite difference method and the finite element method; the boundary type numerical solution is mainly the boundary element method; the coupled type numerical solution is mainly the finite element-boundary element coupling method. In the numerical solution of partial differential equations, the finite difference method is one of the most commonly used methods. Due to its easy understanding and simple use, it has developed rapidly and has been widely used in numerical simulation of earthquakes. The finite difference method usually discretizes the wave equation and approximates the wave equation through the discretized finite difference equation. In essence, it is an approximate method that replaces differentials with differences.

[0037] This embodiment uses the finite difference method forward modeling for the numerical simulation of seismic shot records in the process of marine seismic exploration. In the finite difference forward modeling of marine seismic exploration, the setting of the boundary conditions of the seawater free surface will determine whether there are multiple wave reflections. Therefore, based on the above settings, this fourteenth example can generate seismic data containing multiple waves by setting the boundary conditions of the seawater free surface as a reflection boundary in the forward modeling process, and generate seismic data without multiple waves by setting the boundary conditions of the seawater free surface as an absorption boundary. Exemplarily, in this embodiment, examples of seismic data generated by different boundary conditions in the forward modeling process are as follows: Figure 3 shown.

[0038] In some embodiments, the observation system of the forward simulation is designed with reference to the method of towline acquisition in actual marine seismic exploration. For each marine geological model, different observation systems are set for forward simulation. For a simple horizontal layered model, considering that the type of multiple waves is also simple, this embodiment can be set to obtain seismic data by using a fixed detector and a mobile source. For a complex marine geological model (such as an inclined stratum model, a fault model, and a salt dome model), forward simulation is performed by using a fixed detector and a mobile source and a detector moving with the source, respectively, to obtain multiple wave seismic data under different observation systems.

[0039] S120: Expanding the initial seismic shot gather data to obtain target seismic shot gather data.

[0040] It can be understood that in order to expand the data volume and enrich the data types, the present embodiment can expand the initial seismic shot gather data by means of data augmentation, and the expanded data is used as the target seismic shot gather data.

[0041] Further, step S120 may include at least one of steps S121 to S123: S121: adding Gaussian white noise of different intensities to the initial seismic shot gather data; S122: performing random trace missing operations of different proportions on the initial seismic shot gather data; S123: Rotating the initial seismic shot gather data at different angles.

[0042] Specifically, the initial seismic shot data obtained by forward modeling is generally of high quality, while the situation in actual marine seismic exploration is often more complicated. When collecting marine seismic data, there will be noise interference or data missing in the seismic records due to marine environmental factors or damage to the detector. There are certain differences between the seismic records simulated by forward modeling and the seismic records actually collected. In order to reduce the gap between the two and increase the number of training samples, this embodiment performs sample augmentation operations on the initial seismic shot data generated by forward modeling. According to the characteristics of actual marine seismic exploration data acquisition, Gaussian white noise of different intensities is added to the initial seismic shot data obtained by forward modeling, and random trace missing operations of different proportions are performed on the initial seismic shot data to increase the complexity of the training data; at the same time, the initial seismic shot data is rotated 90° and 180° to further expand the number of samples.

[0043] For example, Figure 4 is the seismic data obtained by expanding this embodiment, wherein: Figure 4 (a)-(c) are seismic data containing multiple waves, seismic data with Gaussian white noise added, and seismic data with random trace missing, respectively; Figure 4(d)-(f) in Figure 4 (a)-(c) in the figure are rotated 90° to obtain: Figure 4 (g)-(i) in Figure 4 (a)-(c) in the figure are obtained by rotating 180°.

[0044] This embodiment can generate multiple types of marine seismic multiple wave suppression training samples: the existing technology suppresses the multiple waves of actual marine seismic data through industrial processes to obtain single wave data as learning labels, which cannot completely remove the multiple waves. The subsequent labeling work is also very trivial and time-consuming. It is not realistic to use processed actual data for labeling training due to high cost. This embodiment can generate seismic shot collection records with and without free surface multiple waves respectively based on the finite difference forward simulation method by setting different free surface multiple wave boundary conditions, and add noise, random trace missing, and rotate 90° and 180° in the deep learning training sample processing to perform sample augmentation in combination with the actual situation in the marine seismic exploration process to construct multiple types of marine seismic multiple wave suppression training samples with multiple categories, accuracy and low labeling cost.

[0045] S130: Construct a deep learning model; wherein the deep learning model includes a feature extraction network and a feature fusion network, and the feature extraction network includes a deep separable convolution structure.

[0046] It can be understood that the feature extraction network can be used to extract features from seismic data, and the feature fusion network can be used to fuse the features of seismic data. The feature extraction network uses a deep separable convolution structure to reduce the number of parameters and calculations of the deep learning model while maintaining high-performance multi-suppression feature extraction.

[0047] Further, step S130 may include steps S131-S132: S131: Construct the deep learning model according to the Mobilenet-Unet network structure; S132: Add the depth-separable convolution structure to the deep learning model; wherein the depth-separable convolution structure includes a depth-wise convolution structure and a point-by-point convolution structure, the depth-wise convolution structure is used to extract spatial features, and the point-by-point convolution structure is used to extract channel features.

[0048] Specifically, this embodiment starts from the idea of ​​lightweight network design, combines the specific problem of marine earthquake multiple wave suppression, and uses the Mobilenet-Unet network structure as a deep learning model. The Mobilenet-Unet network structure of this embodiment is as follows: Figure 5 As shown. Figure 5It can be seen that the Mobilenet-Unet network is a classic "U"-shaped structure, with a feature extraction network on the left and a feature fusion network on the right. Using the deep separable convolution in Mobilenet in the feature extraction process can reduce the number of model parameters and calculations, while maintaining high-performance multiple-suppression feature extraction. In the feature fusion part, retaining the feature graph connection structure of different levels of U-net can better map the signal containing low-dimensional features back to the high-dimensional space and reconstruct the seismic data after multiple wave suppression.

[0049] In order to simplify the deep learning model, this embodiment can introduce deep separable convolution into the deep learning model. Deep separable convolution includes deep convolution and point-by-point convolution, where deep convolution is used to extract spatial features and point-by-point convolution is used to extract channel features. The standard convolution structure and the deep separable convolution structure can be referred to Figure 6 The computation of depthwise separable convolution is very simple. It uses a convolution kernel to perform spatial convolution independently on each channel of the input feature map, and convolves the output of the convolution kernel point by point to combine and concatenate to get the final output. For standard convolution, assuming the convolution kernel size is , the input channel is M, the output channel is N, and the output feature map size is , the standard convolution parameter amount and calculation amount are: ; ; For depth-wise separable convolution, during the depth-wise convolution, the convolution kernel size , the number of convolution kernels is M, each of which needs to be done The number of multiplication and addition operations is , the calculation amount is In point-by-point convolution, the convolution kernel size is , the number of convolution kernels is N, each of which must be executed The number of multiplication and addition operations is , the calculation amount is Therefore, the total number of parameters and total amount of computation for depthwise separable convolution are: ; ; The ratio of the amount of calculation and the amount of parameters can be calculated as: ; ; Compared with traditional convolutional neural networks, the depthwise separable convolution of this embodiment can reduce the number of input channels, thereby effectively reducing the parameters required for the convolution layer, and the running speed is faster than traditional convolution.

[0050] This embodiment can construct a lightweight deep learning model: In the current deep learning network models for suppressing multiple waves of marine earthquakes, most of them only consider the suppression effect of the final multiple waves, without considering how to reduce the training cost and the complexity of the model. Since the different architectures of deep learning convolutional neural network architectures will affect the efficiency and effect of training, and the training process time will also be different in different hardware facilities, it is particularly important to maintain good suppression results of multiple waves of marine earthquakes while reducing the training cost. For this reason, this embodiment starts from the idea of ​​lightweight network design, combines the specific problems of suppressing multiple waves of marine earthquakes, introduces deep separable convolution, and constructs the Mobilenet-Unet network structure as a deep learning model.

[0051] This embodiment can also use other deep learning convolutional neural networks to construct a deep learning model, such as U-net, Aattention-Unet and other neural networks, but the Mobilenet-Unet network structure of this embodiment has fewer parameters, less calculation, and better generalization.

[0052] S140: Training the deep learning model according to the target seismic shot gather data to obtain the trained deep learning model.

[0053] Specifically, this embodiment can use the target seismic shot gather data as training data to train the deep learning model, so that the deep learning model learns how to suppress multiple waves in the seismic data.

[0054] Further, step S140 may include steps S141 to S144: S141: Initialize the input data size, output data size and parameter weight of the deep learning model; S142: labeling and standardizing the target seismic shot gather data to unify the data size to the input data size, thereby obtaining training data; wherein the first seismic data is labeled as a training sample, and the second seismic data is labeled as a training label; S143: Determine a loss function and an optimization function for training the deep learning model; S144: Utilize the training data and train the initialized deep learning model according to the loss function and the optimization function until the trained deep learning model is obtained after a preset condition is met.

[0055] Specifically, the steps of training the deep learning model in this embodiment may include: 1) Preprocess the training data.

[0056] After obtaining the target seismic shot data and deep learning model, the deep learning model is trained through the training set in the target seismic shot data, and the hyperparameters of the deep learning model are adjusted and the optimal model is selected through the validation set in the target seismic shot data. Before training the deep learning model, the labeled target seismic shot data is preprocessed to prevent the deep learning algorithm from overfitting. The preprocessing includes unifying the seismic data size and data standardization. According to the size of the deep learning film and the original seismic data, the input layer input data size and the output layer output data size of the deep learning model are defined as In data set standardization, the Z-score standardization in data standardization can eliminate the dimensionality effect between different data sets and improve the generalization ability of deep learning models. The calculation formula of Z-score standardization is: ; In the formula x and x 0 is the single shot seismic data before and after standardization, μ is the data x The mean of 0, σ For data x The variance of .

[0057] 2) Define the loss function.

[0058] After completing data preparation and preprocessing, the loss function is defined according to the task of suppressing multiple waves of marine earthquakes. The initial Mobilenet-Unet network is a semantic segmentation neural network used for classification problems. In the task of suppressing multiple waves, it is necessary to obtain seismic data after suppressing multiple waves, which belongs to a regression problem. Therefore, in this embodiment, the mean square error can be selected as the loss function for the convergence of the deep learning model. The definition of the loss function is as follows: ; 3) Initialize training and select optimization function.

[0059] In the initialization of the parameter weights of the deep learning model, the reasonable parameter weights in the deep learning model can be initialized so that the gap between the output value of the deep learning model and the target value is not too large before training. In this embodiment, Xavier can be used to initialize the parameter weights. By using Xavier to make the gradient sizes of each layer of the deep learning model almost the same, the gradient can be prevented from exploding or diffusing in the last layer of the network. In the optimization algorithm, the Adam optimization algorithm is selected and an adaptive learning rate is adopted, that is, each parameter can adopt different learning rates according to the change of its own gradient size during the optimization process. This embodiment also sets an early stopping mechanism to monitor the performance of the model on the validation set during the training process to prevent the deep learning model from overfitting the data during the training process, thereby improving the generalization ability of the deep learning model on new data. Through iterative training, the training period with the smallest loss function in the validation set is selected as the optimal marine seismic multiple wave suppression model of the Mobilenet-Unet model.

[0060] This embodiment trains a deep learning model that can suppress multiple waves: after constructing the target earthquake artillery data and the deep learning model, it is necessary to train the deep learning model through supervised learning, and by continuously optimizing the parameters of the deep learning model, the multiple wave suppression results output by the deep learning model are made as close to the real results as possible. In the task of suppressing multiple waves of marine earthquakes, it is necessary to consider the specific data input size, data normalization method, loss function setting, initialization method and optimization algorithm. In order to accurately fit a deep learning model to suppress multiple waves, this embodiment uses data preprocessing to unify the seismic data size and perform data standardization, and selects a suitable loss function, initialization method and model optimization algorithm according to the category attributes of the multiple wave suppression task.

[0061] S150: Suppress multiple waves in seismic data using the trained deep learning model.

[0062] Understandably, the trained deep learning model is able to identify and suppress multiple waves in seismic data.

[0063] Further, step S150 may include S151: S151: Inputting seismic data with multiple waves into the trained deep learning model to suppress the multiple waves in the input data and obtain seismic data after the multiple waves are suppressed.

[0064] Specifically, the trained deep learning model can identify multiple waves in the input seismic data and suppress them, thereby outputting seismic data without multiple waves or with fewer multiple waves.

[0065] Next, the solution of the embodiment of the present application will be introduced and explained in detail with reference to specific application examples.

[0066] In order to improve the efficiency, accuracy and intelligence of marine earthquake multiple wave suppression, and at the same time consider improving the generalization ability of the training model under the condition of reducing the number of parameters and calculations of the deep convolutional neural network of the deep learning method, this embodiment proposes a marine earthquake multiple wave intelligent suppression method based on a lightweight deep neural network. The marine earthquake multiple wave sample library is generated by finite difference forward simulation under different boundary conditions of different marine geophysical models and sample rotation, noise addition and random trace missing sample expansion methods, and then deep separable convolution is introduced to construct Mobilenet-Unet as the model architecture for multiple wave suppression. While greatly reducing the amount of calculation without sacrificing model performance, the optimal marine earthquake multiple wave artificial intelligence suppression model is finally obtained by determining the mean square error loss as the loss function for training, and the intelligent suppression of marine earthquake multiple waves is realized.

[0067] Reference Figure 7 , this embodiment may include the following steps: Establish different marine stratigraphic models based on marine seismic surveys; Design different observation systems (excitation main frequency, shot spacing, channel spacing, etc.) for different ocean formation models; Two finite difference forward simulations were performed under different boundary conditions to obtain seismic data with and without multiple waves; Based on the actual characteristics of marine seismic data, the number and types of training samples are expanded through sample augmentation; Introducing deep separable convolution to build a Mobilenet-Unet model for multiple wave suppression of marine seismic data, i.e. a deep learning model; Select loss function and optimization function to train the Mobilenet-Unet model for multiple wave suppression of marine seismic data; The optimal Mobilenet-Unet marine seismic data multiple wave suppression model is obtained by combining the augmented training set and validation set; Inputting the marine seismic shot data into the trained deep learning model can obtain seismic data after multiple wave suppression.

[0068] In a more specific embodiment, 15 different marine strata models were established, and 7965 sets of seismic data (including those with and without multiple waves) were generated through finite difference forward simulation and sample augmentation, and 5775 sets were selected as training sets, 1590 sets as validation sets, and 600 sets as test sets. In the construction of the deep learning model, the first layer of the encoding layer first performs full zero filling, two-dimensional convolution, batch normalization and linear correction units in sequence, and then uses deep separable convolution modules of different depths to extract features according to the position of the encoding part. Each deep separable convolution module includes deep convolution, batch normalization and linear correction units, point-by-point convolution, batch normalization and linear correction units. The deep separable convolution modules from the second layer to the fifth layer are 1, 2, 2, and 5 respectively. In the decoding part, 4 upsampling modules are included to restore and decode the low-dimensional features of the primary wave extracted by the encoding part to the original image size, and finally obtain the seismic data after multiple wave suppression. At the same time, a jump connection is used between the encoder and the decoder. That is, in the upsampling process at each level, three splicing layers are used to fuse the feature maps at corresponding positions in the encoding and decoding processes by channel, and the low-resolution feature maps are gradually restored to the same resolution as the input image.

[0069] Xavier was used to initialize the parameter weights during training, and the Adam optimization algorithm was selected for optimization. The initial learning rate was set to 0.001, the number of input samples for the training set and the validation set was 5, and the number of training rounds was set to 500. The early stopping mechanism was used to monitor the performance of the deep learning model on the validation set during training to prevent the deep learning model from overfitting the data during training, thereby improving the generalization ability of the deep learning model on new data. Finally, the training period with the smallest loss function in the validation set was selected as the optimal deep learning model, and the deep learning model was used to suppress multiple waves of marine earthquakes. Among them, the loss value curves corresponding to the training set and the validation set during deep learning model training are as follows: Figure 8 As shown in Figure 2, the multiple wave suppression effect of the deep learning model in the test set is as follows: Fig. 9 shown.

[0070] The method for intelligent suppression of multiple waves of marine earthquakes based on lightweight deep neural networks in this embodiment uses deep learning methods to suppress multiple waves of marine earthquakes, including: preparing training sets and training neural networks, requiring multiple wave samples of as many categories and numbers as possible. Once the optimal model parameter training is completed, the deep neural network can intelligently process seismic data in real time and in batches. This embodiment uses a graphics processor for training. For Mobilenet-Unet, it takes about 5 hours to train a training set containing 5775 earthquake records and a validation set of 1590 earthquake records. When processing 600 groups of samples in the test set, all processing is completed in less than a few minutes, and the error with the actual situation is small. Compared with direct use of traditional commercial software for manual processing, the method of suppressing multiple waves using artificial intelligence in this embodiment has been significantly improved in efficiency and maintained a certain degree of accuracy.

[0071] Reference Fig.10 The embodiment of the present application also provides an ocean earthquake multiple wave suppression device, which can implement the above-mentioned ocean earthquake multiple wave suppression method, and the device includes: A first model building unit, used for building an ocean stratum model; A data simulation unit, used for forward modeling the marine stratum model to generate initial seismic shot gather data; wherein the initial seismic shot gather data includes first seismic data with multiple waves and second seismic data without multiple waves; A data expansion unit, used for expanding the initial seismic shot gather data to obtain target seismic shot gather data; A second model construction unit is used to construct a deep learning model; wherein the deep learning model includes a feature extraction network and a feature fusion network, and the feature extraction network includes a deep separable convolution structure; A model training unit, used for training the deep learning model according to the target seismic shot gather data to obtain the trained deep learning model; A multiple wave suppression unit is used to suppress multiple waves in seismic data using the trained deep learning model.

[0072] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0073] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned marine earthquake multiple wave suppression method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0074] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0075] See also Fig.11 , Fig.11 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes: The processor 1101 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application; The memory 1102 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1102 can store an operating system and other application programs. When the technical solution provided in the embodiments of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1102, and the processor 1101 calls and executes the marine seismic multiple wave suppression method of the embodiment of the present application; Input / output interface 1103, used to implement information input and output; The communication interface 1104 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.); A bus 1105 that transmits information between various components of the device (e.g., the processor 1101, the memory 1102, the input / output interface 1103, and the communication interface 1104); The processor 1101 , the memory 1102 , the input / output interface 1103 and the communication interface 1104 are connected to each other in communication within the device via the bus 1105 .

[0076] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned marine earthquake multiple wave suppression method is implemented.

[0077] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiments, the functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0078] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0079] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0080] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0081] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0082] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0083] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0084] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0085] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0086] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0087] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0088] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0089] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for suppressing multiple waves of marine earthquakes, characterized in that: The method comprises the following steps: Constructing ocean stratigraphic models; Performing forward simulation on the marine stratum model to generate initial seismic shot gather data; wherein the initial seismic shot gather data includes first seismic data with multiple waves and second seismic data without multiple waves; Expanding the initial seismic shot gather data to obtain target seismic shot gather data; Constructing a deep learning model; wherein the deep learning model includes a feature extraction network and a feature fusion network, and the feature extraction network includes a deep separable convolution structure; Training the deep learning model according to the target seismic shot gather data to obtain the trained deep learning model; Suppressing multiple waves in seismic data using the trained deep learning model; The construction of the ocean stratum model comprises the following steps: Construct the marine stratum model of seawater layers at different depths; wherein the marine stratum model includes a horizontal layer model, an inclined stratum model, a fault model and a salt dome model.

2. The method for suppressing marine seismic multiple waves according to claim 1, characterized in that: The forward modeling of the marine stratum model to generate initial seismic shot gather data comprises the following steps: The marine stratum models are divided into regional type, boundary type and coupled type respectively; Setting boundary conditions of the free surface of seawater for the regional type, boundary type and coupling type in each of the marine stratum models respectively; wherein the boundary conditions include reflection boundaries and absorption boundaries; Finite difference method forward simulation is performed on each of the marine stratum models using different observation models to generate the first seismic data with multiple waves based on the reflection boundary, and the second seismic data without multiple waves based on the absorption boundary.

3. The method for suppressing marine seismic multiple waves according to claim 1, characterized in that: The step of expanding the initial seismic shot gather data to obtain target seismic shot gather data comprises at least one of the following steps: adding Gaussian white noise of different intensities to the initial seismic shot gather data; Alternatively, performing random trace deletion operations of different proportions on the initial seismic shot gather data; Alternatively, the initial seismic shot gather data are rotated at different angles.

4. The method for suppressing marine seismic multiple waves according to claim 1, characterized in that: The construction of the deep learning model includes the following steps: Construct the deep learning model according to the Mobilenet-Unet network structure; The depth-wise separable convolution structure is added to the deep learning model; wherein the depth-wise separable convolution structure includes a depth-wise convolution structure and a point-wise convolution structure, the depth-wise convolution structure is used to extract spatial features, and the point-by-point convolution structure is used to extract channel features.

5. The method for suppressing marine seismic multiple waves according to claim 1, characterized in that: The step of training the deep learning model according to the target seismic shot gather data to obtain the trained deep learning model comprises the following steps: Initializing the input data size, output data size, and parameter weights of the deep learning model; The target seismic shot gather data are labeled and standardized to unify the data size to the input data size to obtain training data; wherein the first seismic data are labeled as training samples, and the second seismic data are labeled as training labels; Determining a loss function and an optimization function for training the deep learning model; The initialized deep learning model is trained using the training data and according to the loss function and the optimization function until the trained deep learning model is obtained after a preset condition is met.

6. The method for suppressing marine seismic multiple waves according to any one of claims 1 to 5, characterized in that: The method of suppressing multiple waves in seismic data by using the trained deep learning model comprises the following steps: The seismic data with multiple waves are input into the trained deep learning model to suppress the multiple waves in the input data and obtain the seismic data after the multiple waves are suppressed.

7. Marine seismic multiple wave suppression device, characterized in that: The device comprises: A first model building unit, used for building an ocean stratum model; A data simulation unit, used for forward modeling the marine stratum model to generate initial seismic shot gather data; wherein the initial seismic shot gather data includes first seismic data with multiple waves and second seismic data without multiple waves; A data expansion unit, used for expanding the initial seismic shot gather data to obtain target seismic shot gather data; A second model construction unit is used to construct a deep learning model; wherein the deep learning model includes a feature extraction network and a feature fusion network, and the feature extraction network includes a deep separable convolution structure; A model training unit, used for training the deep learning model according to the target seismic shot gather data to obtain the trained deep learning model; A multiple wave suppression unit is used to suppress multiple waves in seismic data using the trained deep learning model.

8. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the marine seismic multiple wave suppression method as claimed in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the marine seismic multiple wave suppression method according to any one of claims 1 to 6 is implemented.

Images