A method and device for jointly suppressing free surface multiple waves and ghost waves

Abstract

The present invention provides a method and device for jointly suppressing free surface multiple waves and ghost waves, belonging to the field of exploration seismic data processing, including: adding a self-attention mechanism in the upsampling path and downsampling path of the original UNet++ network, adding a residual convolution block between the convolution layers on the jump path between the encoder and the decoder to construct an improved UNet++ network; obtaining a first seismic data set containing free surface multiple waves and ghost waves and a second seismic data set not containing free surface multiple waves and ghost waves; using the first seismic data set as a training sample and the second seismic data set as a label to train the improved UNet++ network to obtain an interference wave suppression model; obtaining seismic data to be processed, and inputting the seismic data to be processed into the interference wave suppression model for suppression processing. The method can realize the joint suppression of free surface multiple waves and ghost waves and improve the interference wave suppression efficiency.

Description

Technical Field

[0001] The invention belongs to the field of exploration seismic data processing, and in particular relates to a method and a device for jointly suppressing free surface multiple waves and ghost waves. Background Art

[0002] In deepwater seismic exploration, free surface multiple waves and ghost waves are two common and difficult-to-suppress interference signals, which reduce the bandwidth and resolution of the data and affect the quality of seismic imaging and inversion. Among them, free surface multiple waves refer to waves formed by multiple reflections of seismic waves between the free surface (such as the sea surface or lake surface) and the underground medium, and ghost waves are false waves caused by multiple reflections of waves between the seawater and the seabed interface. Suppressing these interference waves can improve the quality of seismic data, making the data clearer and easier to interpret.

[0003] At present, ghost waves and free surface multiple waves in seismic data are usually suppressed separately. Separate suppression requires the use of different algorithms and processing steps, which often requires a lot of computing resources and time, resulting in low interference wave suppression efficiency. Therefore, how to achieve the joint suppression of ghost waves and free surface multiple waves and improve the suppression efficiency of interference waves has become a problem to be solved. Summary of the invention

[0004] In order to solve the problem of low suppression efficiency of interference waves in seismic data, the present invention provides a method and device for jointly suppressing free surface multiple waves and ghost waves.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A free surface multiple wave and ghost wave combined suppression method, comprising:

[0007] Add self-attention mechanisms to the upsampling and downsampling paths of the original UNet++ network, and add residual convolution blocks between the convolutional layers on the skip path between the encoder and decoder to build an improved UNet++ network;

[0008] A first seismic data set containing free surface multiple waves and ghost waves and a second seismic data set without free surface multiple waves and ghost waves are obtained; the first seismic data set is used as a training sample and the second seismic data set is used as a label to train an improved UNet++ network to obtain an interference wave suppression model;

[0009] The seismic data to be processed are obtained and input into the interference wave suppression model. In the feature extraction stage, a convolution operation is performed on the input seismic data through a convolution layer to obtain a feature map, the size of the feature map is reduced through a downsampling operation, and the feature map is enhanced through a self-attention mechanism. In the feature fusion stage, the size of the feature map is increased through upsampling, and the feature map is enhanced through a self-attention mechanism. The enhanced feature map is spliced ​​with the feature map from the corresponding encoder path, and the spliced ​​feature map is added to the feature map processed by multiple convolution layers by a residual convolution block. After continuous upsampling and splicing, the output layer outputs seismic data without free surface multiple waves and ghost waves.

[0010] Optionally, the improved UNet++ network is trained based on an optimization objective function, and the optimization objective function is:

[0011] ;

[0012] in, represents label data without free surface multiples and ghost waves, represents the output of the improved UNet++ network, Indicates the number of time sampling points, Indicates the number of channels, Indicates that the sampling point is The number of channels is does not include label data for free surface multiples and ghost waves. Indicates that the sampling point is The number of channels is is the output of the improved UNet++ network, N represents the total amount of seismic data, N= .

[0013] Optionally, the improved UNet++ network is trained to obtain an interference wave suppression model, including:

[0014] The first seismic data set is input into the improved UNet++ network, the prediction results are output through forward propagation, the prediction results are compared with the second seismic data set, and the loss value and gradient are calculated according to the optimization objective function;

[0015] Back propagation is performed based on the loss value and gradient, and the network parameters of the improved UNet++ network are updated through the Adam optimizer to reduce the loss value;

[0016] The normal propagation and reverse propagation processes are iterated until the number of iterations reaches a preset number, and then the interference wave suppression model is obtained.

[0017] Optionally, acquiring a first seismic data set containing free surface multiple waves and ghost waves and a second seismic data set not containing free surface multiple waves and ghost waves comprises:

[0018] Establish the wave equation as a mathematical model for seismic wave propagation and discretize the space and time domains;

[0019] The spatial and temporal partial derivatives in the wave equation are discretized by the finite difference method to generate a discretized wave equation;

[0020] Absorbing boundary conditions are introduced at the boundaries of the model, and the source function is set to simulate the excitation process of the source. During the simulation, the second seismic data set without free surface multiple waves and ghost waves is obtained by maintaining the uniformity and stability of the medium parameters. By changing the distribution of the medium parameters and adjusting the source function and initial conditions, factors that produce ghost waves are introduced to obtain the first seismic data set containing free surface multiple waves and ghost waves.

[0021] The present invention also provides a free surface multiple wave and ghost wave combined suppression device, comprising:

[0022] A building block for adding self-attention mechanisms in the upsampling and downsampling paths of the original UNet++ network, and adding residual convolution blocks between convolutional layers on the skip path between the encoder and decoder to build an improved UNet++ network;

[0023] A training module is used to obtain a first seismic data set containing free surface multiple waves and ghost waves and a second seismic data set without free surface multiple waves and ghost waves; the first seismic data set is used as a training sample and the second seismic data set is used as a label to train an improved UNet++ network to obtain an interference wave suppression model;

[0024] The processing module is used to obtain the seismic data to be processed, input the seismic data to be processed into the interference wave suppression model, and in the feature extraction stage, convolve the input seismic data through the convolution layer to obtain a feature map, reduce the size of the feature map through the downsampling operation, and enhance the feature map through the self-attention mechanism; in the feature fusion stage, increase the size of the feature map through upsampling, and enhance the feature map through the self-attention mechanism; splice the enhanced feature map with the feature map from the corresponding encoder path, and add the spliced ​​feature map with the feature map processed by multiple convolution layers by the residual convolution block; after continuous upsampling and splicing, the output layer outputs seismic data without free surface multiple waves and ghost waves.

[0025] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned free surface multiple wave and ghost wave joint suppression method when executing the program.

[0026] The present invention also provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned free surface multiple wave and ghost wave joint suppression method is implemented.

[0027] The free surface multiple wave and ghost wave combined suppression method provided by the present invention has the following beneficial effects:

[0028] The present invention trains an improved UNet++ network through a first seismic data set containing free surface multiple waves and ghost waves and a second seismic data set not containing free surface multiple waves and ghost waves, and jointly suppresses free surface multiple waves and ghost waves in seismic data through the trained model; the improved UNet++ network adds a self-attention mechanism in the upsampling path and the downsampling path of the original UNet++ network, and the self-attention mechanism can calculate the correlation between various parts of the input seismic data, so as to better understand the relationship between long-distance features in the seismic signal. For seismic data, especially when suppressing free surface multiple waves and ghost waves, the self-attention mechanism can effectively highlight the effective signal and reduce the interference of free surface multiple waves and ghost waves on the overall data; adding residual convolution blocks between convolution layers on the jump path between the encoder and the decoder can keep the main features of the seismic signal from being weakened in the deep network, and suppress free surface multiple waves and ghost waves at the same time, so as to improve the recognition and extraction capabilities of the effective signal; therefore, the present invention can more effectively jointly suppress free surface multiple waves and ghost waves in seismic data through the improved UNet++ network, and improve the suppression efficiency of interference waves. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiment of the present invention and its design scheme, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0030] Figure 1 A schematic flow chart of a method for jointly suppressing free surface multiple waves and ghost waves provided by an embodiment of the present invention;

[0031] Figure 2 A schematic diagram of the structure of an original UNet++ provided in an embodiment of the present invention;

[0032] Figure 3A schematic diagram of the structure of an improved UNet++ provided in an embodiment of the present invention;

[0033] Figure 4 A schematic diagram of a velocity model provided by an embodiment of the present invention;

[0034] Figure 5 A comparison diagram of the suppression effects of an improved UNet++ network and an SRME model provided in an embodiment of the present invention;

[0035] Figure 6 A single-channel amplitude diagram corresponding to seismic data provided by an embodiment of the present invention;

[0036] Figure 7 A schematic diagram of a salt velocity model provided by an embodiment of the present invention;

[0037] Figure 8 A comparison diagram of the suppression effects of an improved UNet++ network and an original UNet++ network provided by an embodiment of the present invention;

[0038] Fig. 9 Another single-channel amplitude diagram corresponding to seismic data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to enable those skilled in the art to better understand the technical solution of the present invention and implement it, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the scope of protection of the present invention.

[0040] In the description of the present invention, it is to be understood that the terms “center”, “longitudinal”, “lateral”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “top”, “bottom”, “inside”, “outside”, “axial”, “radial”, “circumferential”, etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the technical solutions of the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0041] In addition, the terms "first", "second", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance. In the description of the present invention, it should be noted that, unless otherwise clearly specified or limited, the terms "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. In the description of the present invention, unless otherwise specified, "plurality" means two or more, which will not be described in detail here.

[0042] Based on the technical problem of low suppression efficiency of interference waves (free surface multiple waves and ghost waves) in the prior art, the present invention provides a method for jointly suppressing surface multiple waves and ghost waves, such as Figure 1 As shown, the method comprises the following steps:

[0043] S1, adds self-attention mechanism in the upsampling path and downsampling path of the original UNet++ network, and adds residual convolution blocks between the convolution layers on the skip path between the encoder and decoder to build an improved UNet++ network.

[0044] The UNet++ network is a neural network structure that is widely used in the field of computer vision, especially in semantic segmentation and instance segmentation tasks. It is improved and optimized based on the UNet network. By introducing multi-level feature fusion and jump connection, the accuracy of the segmentation results and the ability to retain details are significantly improved. The UNet++ network can be used to suppress multiple waves in seismic data. The UNet++ network learns the features of the primary wave (effective wave) in the seismic data and uses the primary wave features to reconstruct the primary wave and suppress multiple waves.

[0045] If the joint suppression of multiple waves and ghost waves is to be achieved, the network needs to have more powerful feature extraction and reconstruction capabilities. Based on this, the present invention improves the original UNet++ network.

[0046] Figure 2 A schematic diagram of the structure of an original UNet++ network provided in an embodiment of the present invention, Figure 3 A schematic diagram of the structure of an improved UNet++ network provided in an embodiment of the present invention.

[0047] For example, Figure 2 and Figure 3As shown, the present invention adds a self-attention mechanism (represented by RA in the figure) to the downsampling path and upsampling path of the original UNet++ network. By placing the self-attention mechanism at these key positions, the network can more intelligently focus on the most relevant features in the input data, enhance the network's ability to capture complex seismic reflection information, especially when processing free surface multiple waves and ghost waves, thereby improving the ability to extract important information from seismic data.

[0048] Secondly, the present invention adds a 1×1 residual convolution block between the convolution layers on the jump path of the encoder and the decoder, and the residual convolution block is embedded in the middle layer of the network to construct a deep residual network structure. The residual convolution block works through a "jump connection", that is, the input is directly added to the output after being processed by multiple convolution layers. By adding residual convolution blocks, the network can avoid the problem of gradient vanishing or explosion while increasing the depth, keep the main features of the seismic signal in the deep network from being weakened, and suppress free surface multiple waves and ghost waves, thereby improving the ability to recognize and extract effective signals.

[0049] It should be understood that the self-attention mechanism can improve the network's ability to capture the characteristics of the input data, and the residual convolution block allows the network to learn the residual between the input and output, rather than directly learning the expected complete output, thereby facilitating information transmission in the network, so that even very deep networks can be effectively trained. This method improves the training efficiency and accuracy of the network.

[0050] It should be understood that when building an improved UNet++ network, in addition to determining the network architecture, it is also necessary to select appropriate activation functions, target optimization functions and optimizers, as well as set parameters such as learning rate, momentum, number of iterations, and network input and output sizes.

[0051] In the present invention, supervised learning is adopted to optimize the objective function as follows:

[0052] ;

[0053] in, represents label data without free surface multiples and ghost waves, represents the output of the improved UNet++ network, Indicates the number of time sampling points, Indicates the number of channels, Indicates that the sampling point is The number of channels is does not include label data for free surface multiples and ghost waves. Indicates that the sampling point is The number of channels is is the output of the improved UNet++ network, N represents the total amount of seismic data, N= .

[0054] Optionally, in this embodiment of the present invention, the learning rate is 0.0001, the number of iterations is 20000, and the input size and output size of the network are both 512 512.

[0055] The back propagation process is set to minimize the optimization objective function through the gradient descent algorithm, and the Adam optimizer of the momentum method is used to update the network parameters. Optionally, the LeakyReLU activation function is used in the improved UNet++ network, which retains some negative axis values ​​and alleviates the problem that the activation value is too small and the neuron parameters cannot be updated.

[0056] S2, obtaining a first seismic data set containing free surface multiple waves and ghost waves and a second seismic data set without free surface multiple waves and ghost waves; using the first seismic data set as a training sample and the second seismic data set as a label, training the improved UNet++ network to obtain an interference wave suppression model.

[0057] After determining the network structure and setting the network parameters through the above S2 process, start training the improved UNet++ network.

[0058] First, obtain the seismic data set used to train the network. The seismic data in the seismic data set is two-dimensional seismic data. The two-dimensional seismic data usually exists in the form of a matrix, which is similar to the structure of image data. Each data point represents the amplitude or other attributes of the seismic wave at a specific location and depth.

[0059] Optionally, a first seismic data set containing free surface multiple waves and ghost waves can be obtained by acquiring historical actual seismic data, and then the free surface multiple waves and ghost waves are suppressed by existing suppression methods to obtain a second seismic data set that does not contain free surface multiple waves and ghost waves.

[0060] It is also possible to simulate a first seismic data set containing free surface multiples and ghost waves and a second seismic data set without free surface multiples and ghost waves by the finite difference method, as follows:

[0061] Step 1: Establish a mathematical model of seismic wave propagation. First, based on the physical characteristics of seismic wave propagation, establish a wave equation as the mathematical model of seismic wave propagation. This equation describes the dynamic process of wave propagation in the medium; and discretize the space and time domains to define the grid unit size (space step) and time step. The selection of these parameters needs to meet certain stability conditions to ensure the accuracy of the simulation results.

[0062] Step 2: Discretize the partial derivatives of the wave equation using the finite difference method. This usually involves replacing the spatial and temporal derivatives in the wave equation with the ratio of the difference between the function values ​​at adjacent grid points and the grid step size. By discretizing the partial derivatives, a discretized wave equation is obtained. This equation describes the propagation of the wave on the discrete grid.

[0063] Step 3: Set initial conditions, including the initial state of the medium (such as velocity, density, etc.) and the initial distribution of the wave field (usually zero); introduce absorbing boundary conditions at the model boundary to reduce the boundary reflection of the wave; apply the seismic source function at the specified position in the model to simulate the excitation process of the source. The source function can be a pulse function, a sine wave or other forms of waveform.

[0064] To obtain seismic data without multiple waves and ghost waves, the following conditions must be met during the simulation:

[0065] 1. Maintain the uniformity and stability of medium parameters to avoid multiple waves and ghost waves.

[0066] 2. Ensure that the boundary conditions can effectively absorb waves and prevent waves from reflecting at the boundaries to form multiple waves.

[0067] 3. Avoid introducing factors that produce ghost waves in the source function and initial conditions.

[0068] To obtain seismic data containing multiple waves and ghost waves, the following conditions must be ensured during the simulation:

[0069] 1. By changing the distribution of medium parameters (such as velocity, density, etc.), inhomogeneities and interfaces are introduced to generate multiple waves.

[0070] 2. In the boundary conditions, an imperfect absorption layer can be deliberately set to allow part of the waves to be reflected at the boundary to form multiple waves.

[0071] 3. By adjusting the source function and initial conditions, factors that produce ghost waves are introduced, such as the source location, waveform, etc.

[0072] It should be understood that actual offshore seismic data acquisition is costly and resource-intensive, and cannot be easily trial-and-error or repeated acquisition. Therefore, simulating a first seismic data set containing free surface multiple waves and ghost waves and a second seismic data set not containing free surface multiple waves and ghost waves using the finite difference method can reduce cost and resource consumption and improve efficiency.

[0073] Furthermore, the improved UNet++ network is trained based on the first seismic data set and the second data set, and the training process is as follows:

[0074] The first seismic data set is input into the improved UNet++ network, the prediction results are output through forward propagation, the prediction results are compared with the second seismic data set, and the loss value and gradient are calculated according to the optimization objective function; back propagation is performed based on the loss value and gradient, and the network parameters of the improved UNet++ network are updated through the Adam optimizer to reduce the loss value; the normal propagation and back propagation processes are iterated until the number of iterations reaches the preset number, and the interference wave suppression model is obtained.

[0075] S3, obtaining the seismic data to be processed, inputting the seismic data to be processed into the interference wave suppression model, and obtaining seismic data without free surface multiple waves and ghost waves.

[0076] Specifically, through the construction and training of the above-mentioned improved UNet++ network, an interference wave suppression model that can process free surface multiple waves and ghost waves in seismic data is obtained. In actual application, seismic data containing free surface multiple waves and ghost waves are input into the interference wave suppression model, and the interference wave suppression model processes the seismic data. The processing process is as follows: in the feature extraction stage, a convolution operation is performed on the input seismic data through a convolution layer to obtain a feature map, the size of the feature map is reduced through a downsampling operation, and the feature map is enhanced through a self-attention mechanism; in the feature fusion stage, the size of the feature map is increased through upsampling, and the feature map is enhanced through a self-attention mechanism; the enhanced feature map is spliced ​​with the feature map from the corresponding encoder path, and the spliced ​​feature map is added to the feature map processed by multiple convolution layers by a residual convolution block; after continuous upsampling and splicing, the output layer outputs seismic data that does not contain free surface multiple waves and ghost waves.

[0077] It should be understood that the self-attention mechanism, as a separate module, obtains a weight matrix by calculating the autocorrelation of the input feature map, and then performs weighted summation on the input features to obtain an enhanced feature representation, that is, an enhanced feature map.

[0078] Next, the suppression effect of the improved UNet++ network in the present invention is verified by experiments.

[0079] Design a three-layer horizontal layered velocity model, such as Figure 4 As shown, the source and the geophone are placed on the sea level. There are 101 shots in total, 101 channels in each shot, and the shot spacing and channel spacing are both 5m. The source uses a Ricker wavelet with a main frequency of 35Hz, a sampling interval of 1ms, and a record length of 2s.

[0080] The velocity model is used to obtain single shot records, which are then suppressed using the improved UNet++ network and SRME. The suppression results are shown in the figure below. Figure 5 As shown, Figure 5Figure (a) shows the seismic data without free surface multiple waves; Figure 5 (b) shows seismic data containing free surface multiple waves; Figure 5 Figure (c) shows the seismic data without free surface multiple waves predicted by the improved UNet++; Figure 5 Figure (d) shows the data predicted by SRME without free surface multiple waves.

[0081] Extraction Figure 5 The 50th data of the seismic data in the above example is used to analyze the amplitude diagram. Figure 6 As shown, Figure 6 Figure (a) is a single-channel amplitude diagram comparing the real seismic data without free surface multiple waves obtained by forward simulation with the data without free surface multiple waves predicted by the improved UNet++ network; Figure 6 Figure (b) is a single-channel amplitude diagram comparing the actual data without free surface multiple waves obtained by forward simulation with the data without free surface multiple waves predicted by SRME.

[0082] from Figure 5 and Figure 6 It can be concluded that the improved UNet++ network is significantly more effective in suppressing free surface multiple waves than the traditional SRME method.

[0083] The above is the performance of the improved UNet++ network in suppressing free surface multiple waves. The effect of the improved UNet++ network in jointly suppressing free surface multiple waves and ghost waves is verified below.

[0084] Select a slice from the 3D SEG SALT model to display, such as Figure 7 As shown, Figure 7 A schematic diagram of a salt velocity model provided by an embodiment of the present invention. The source and the detector are sunk 0m-20m below the sea level, and the seismic record is 1201 shots, 1201 channels per shot, and the shot spacing and channel spacing are both 5m. The source uses a Ricker wavelet with a main frequency of 35Hz, a sampling interval of 0.5ms, and a record length of 2s.

[0085] pass Figure 7 The salt velocity model in the forward multi-shot record is used to extract three representative single-shot records, and then they are suppressed by the improved UNet++ network. The suppression effect is shown in Figure 2. Figure 8 As shown, Figure 8 Figure (a) shows that the source is sunk below the sea level and the receiver depth is set at 0m. The forward modeling results do not contain free surface multiple waves and source ghost waves. Figure 8 The middle (d) figure shows the forward modeling result without free surface multiple waves and cable ghost waves when the source depth is set to 0 m and the geophone is placed below the sea level. Figure 8 Figure (g) in the figure sinks both the source and the geophone below the sea level, and the forward modeling results do not contain free surface multiple waves, source ghost waves, cable ghost waves, or a mixture of source ghost waves and cable ghost waves. Figure 8 Figures (b), (e) and (h) are the compressed forms of the UNet++ network. Figure 8 Figures (c), (f) and (i) are the forms after being compressed by the improved UNet++ network. Figure 8 Figure (j) shows the effect of the UNet++ network suppressing free surface multiple waves and ghost waves. Figure 8 From the arrow in the middle and the rectangle circle, it can be clearly seen that the improved UNet++ network can not only suppress free surface multiple waves, but also suppress source ghost waves, cable ghost waves, and the mixed form of source ghost waves and cable ghost waves, and the suppression effect of the improved UNet++ network is better than that of the UNet++ network.

[0086] Extraction Figure 8 The 30th channel of seismic data in the above example is used to analyze the amplitude diagram. Fig. 9 As shown, Fig. 9 Figures (a), (b), and (c) are single-channel amplitude diagrams comparing the real seismic data without free surface multiple waves and ghost waves obtained by forward simulation and the seismic data without free surface multiple waves and ghost waves predicted by the improved UNet++ network; Fig. 9 Figure (d) is a single-channel amplitude diagram comparing the real data without free surface multiple waves and ghost waves obtained by forward simulation with the data without free surface multiple waves and ghost waves predicted by the UNet++ network; Fig. 9 The image (a) in the figure is extracted from Figure 8 Figure (a) and Figure 8 The 30th data of Figure (c) is Fig. 9 The (b) figure in the figure is extracted from Figure 8 Figure (d) and Figure 8 The data of the 30th track in Figure (f) is Fig. 9 The (c) figure in the figure is extracted from Figure 8 Figure (g) and Figure 8 The data of the 30th track in Figure (i) is Fig. 9 The (d) figure in the figure is extracted from Figure 8 Figure (g) and Figure 8 The data of track 30 in Figure (j) are from Fig. 9It can be seen from the single-channel amplitude diagram that after the improved UNet++ network suppresses free surface multiple waves and ghost waves, the free surface multiple waves, source ghost waves, cable ghost waves, and the mixed forms of source ghost waves and cable ghost waves of the seismic data have been greatly attenuated, and the energy of the notch point has been significantly enhanced, achieving the effect of broadening the frequency band range of seismic data and enhancing the energy of low-frequency signals. From the comparison of the real data and the test data results, it can be seen that the results predicted by the improved UNet++ network are basically consistent with the real data, with a small error. Compared with the improved UNet++ network, the suppression effect of the UNet++ network is obviously not as good as that of the improved UNet++ network.

[0087] Based on the same inventive concept, an embodiment of the present invention further provides a free surface multiple wave and ghost wave combined suppression device, exemplarily, the device comprises:

[0088] A building module for adding self-attention mechanisms in the upsampling and downsampling paths of the original UNet++ network, and adding residual convolution blocks between convolutional layers on the skip path between the encoder and decoder to build an improved UNet++ network.

[0089] The training module is used to obtain a first seismic data set containing free surface multiple waves and ghost waves and a second seismic data set without free surface multiple waves and ghost waves; the first seismic data set is used as a training sample and the second seismic data set is used as a label to train the improved UNet++ network to obtain an interference wave suppression model.

[0090] The processing module is used to obtain the seismic data to be processed, input the seismic data to be processed into the interference wave suppression model, and in the feature extraction stage, convolve the input seismic data through the convolution layer to obtain a feature map, reduce the size of the feature map through the downsampling operation, and enhance the feature map through the self-attention mechanism; in the feature fusion stage, increase the size of the feature map through upsampling, and enhance the feature map through the self-attention mechanism; splice the enhanced feature map with the feature map from the corresponding encoder path, and add the spliced ​​feature map with the feature map processed by multiple convolution layers by the residual convolution block; after continuous upsampling and splicing, the output layer outputs seismic data without free surface multiple waves and ghost waves.

[0091] Each module in the above-mentioned free surface multiple wave and ghost wave combined suppression device can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0092] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps in the embodiment of the method for joint suppression of free surface multiple waves and ghost waves. The specific implementation method can be found in the method embodiment, which will not be repeated here.

[0093] Furthermore, the present invention also provides a non-temporary computer-readable storage medium containing instructions, and a computer program is stored on the storage medium. For example, a memory containing instructions, the above instructions can be executed by a processor of a computer device to complete the above method. For example, the non-temporary computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device. When the computer program is executed by the processor, the steps in the embodiment of the free surface multiple wave and ghost wave combined suppression method can be implemented. The specific implementation method can be found in the method embodiment, which will not be repeated here.

[0094] It should be understood by those skilled in the art that embodiments of the present invention may provide methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0095] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0096] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0097] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0098] It should be pointed out that the specific implementation methods described above can enable those skilled in the art to understand the invention more comprehensively, but do not limit the invention in any way. Therefore, although the invention has been described in detail in this specification and embodiments, those skilled in the art should understand that the invention can still be modified or replaced by equivalents; and all technical solutions and improvements that do not deviate from the spirit and scope of the invention are included in the protection scope of the invention. Any figure mark in the claims should not be regarded as limiting the claims involved. Any simple change or equivalent replacement of the technical solution that can be obviously obtained by any person familiar with the art within the technical scope disclosed in the present invention belongs to the protection scope of the present invention.

Claims

1. A method for combined suppression of free surface multiples and ghost waves, characterized in that: include: Add self-attention mechanisms to the upsampling and downsampling paths of the original UNet++ network, and add residual convolution blocks between the convolution layers on the skip path between the encoder and decoder to build an improved UNet++ network; A first seismic dataset containing free-surface multiples and ghost waves and a second seismic dataset without free-surface multiples and ghost waves are obtained; the first seismic dataset is used as training samples and the second seismic dataset is used as labels to train an improved UNet++ network to obtain an interference wave suppression model; Obtain the seismic data to be processed and input it into the interference wave suppression model. In the feature extraction stage, convolution is performed on the input seismic data through a convolutional layer to obtain a feature map. The size of the feature map is reduced through downsampling, and the feature map is enhanced through a self-attention mechanism. In the feature fusion stage, the size of the feature map is increased through upsampling, and the feature map is enhanced through a self-attention mechanism. The enhanced feature map is concatenated with the feature map from the corresponding encoder path, and the residual convolution block adds the concatenated feature map to the feature maps processed by multiple convolutional layers. After continuous upsampling and splicing, the output layer outputs seismic data without free surface multiple waves and ghost waves; The improved UNet++ network is trained based on the optimization objective function, and the optimization objective function is: Where y represents the label data without free surface multiple waves and ghost waves, represents the output of the improved UNet++ network, h represents the number of time sampling points, w represents the number of channels, and y i,j Indicates the label data without free surface multiple waves and ghost waves when the time sampling point is i and the number of channels is j. represents the output of the improved UNet++ network when the time sampling point is i and the number of channels is j, N represents the total amount of seismic data, N = h × w; The improved UNet++ network is trained to obtain the interference wave suppression model, including: The first earthquake dataset is input into the improved UNet++ network, and the prediction results are output through forward propagation. The prediction results are compared with the second earthquake dataset, and the loss value and gradient are calculated according to the optimization objective function. Backpropagation is performed based on the loss value and gradient, and the network parameters of the improved UNet++ network are updated through the Adam optimizer to reduce the loss value; The normal propagation and reverse propagation processes are iterated until the number of iterations reaches a preset number, and then the interference wave suppression model is obtained.

2. The method for combined suppression of free surface multiple waves and ghost waves according to claim 1, characterized in that: Acquiring a first seismic dataset containing free surface multiples and ghost waves and a second seismic dataset excluding free surface multiples and ghost waves, including: Establish the wave equation as the mathematical model of seismic wave propagation and discretize the space and time domains; The spatial and temporal partial derivatives in the wave equation are discretized by the finite difference method to generate a discretized wave equation. Absorbing boundary conditions are introduced at the boundaries of the model, and the source function is set to simulate the excitation process of the source. During the simulation, by maintaining the uniformity and stability of the medium parameters, a second seismic data set without free surface multiple waves and ghost waves is obtained. By changing the distribution of the medium parameters and adjusting the source function and initial conditions, factors that produce ghost waves are introduced, and the first seismic data set containing free surface multiple waves and ghost waves is obtained.

3. A device for combined suppression of free surface multiple waves and ghost waves, characterized in that: include: A building block for adding self-attention mechanisms to the upsampling and downsampling paths of the original UNet++ network, and adding residual convolution blocks between convolutional layers on the skip path between the encoder and decoder to build an improved UNet++ network; A training module is used to obtain a first seismic dataset containing free surface multiples and ghost waves and a second seismic dataset without free surface multiples and ghost waves; the first seismic dataset is used as a training sample and the second seismic dataset is used as a label to train an improved UNet++ network to obtain an interference wave suppression model; The processing module is used to obtain the seismic data to be processed and input the seismic data to be processed into the interference wave suppression model. In the feature extraction stage, the input seismic data is convolved by the convolution layer to obtain a feature map, the size of the feature map is reduced by downsampling, and the feature map is enhanced by the self-attention mechanism. In the feature fusion stage, the size of the feature map is increased by upsampling, and the feature map is enhanced by the self-attention mechanism. The enhanced feature map is spliced ​​with the feature map from the corresponding encoder path, and the residual convolution block adds the spliced ​​feature map to the feature maps processed by multiple convolution layers. After continuous upsampling and splicing, the output layer outputs seismic data without free surface multiple waves and ghost waves; The improved UNet++ network is trained based on the optimization objective function, and the optimization objective function is: Where y represents the label data without free surface multiple waves and ghost waves, represents the output of the improved UNet++ network, h represents the number of time sampling points, w represents the number of channels, and y i,j Indicates the label data without free surface multiple waves and ghost waves when the time sampling point is i and the number of channels is j. represents the output of the improved UNet++ network when the time sampling point is i and the number of channels is j, N represents the total amount of seismic data, N = h × w; The improved UNet++ network is trained to obtain the interference wave suppression model, including: The first earthquake dataset is input into the improved UNet++ network, and the prediction results are output through forward propagation. The prediction results are compared with the second earthquake dataset, and the loss value and gradient are calculated according to the optimization objective function. Backpropagation is performed based on the loss value and gradient, and the network parameters of the improved UNet++ network are updated through the Adam optimizer to reduce the loss value; The normal propagation and reverse propagation processes are iterated until the number of iterations reaches a preset number, and then the interference wave suppression model is obtained.

4. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method for jointly suppressing free surface multiple waves and ghost waves according to any one of claims 1 and 2 is implemented.

5. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for jointly suppressing free surface multiple waves and ghost waves as claimed in any one of claims 1 and 2 is implemented.

Images