Method and device for reconstructing high-resolution seismic imaging profile and storage medium

Abstract

The embodiment of the invention provides a method and device for reconstructing a high-resolution seismic imaging profile and a storage medium, and belongs to the field of exploration geophysics. According to the method, a deep learning technology and seismic imaging method cross fusion method is utilized, a seismic imaging profile training set is generated firstly, neural network training is performed according to the obtained seismic imaging profile training set, and a reconstructed high-resolution seismic imaging profile is obtained. According to the embodiment of the invention, when the high-resolution seismic imaging profile is reconstructed, the seismic data acquisition cost and the seismic data processing calculation cost are effectively reduced without prior knowledge, and the method has the advantage of high automation degree.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of exploration geophysics, and in particular to a method, device and storage medium for reconstructing a high-resolution seismic imaging profile. Background Art

[0002] As the objectives of seismic exploration become more and more complex, the requirements for exploration accuracy and spatial sampling rate are also getting higher and higher. However, due to the constraints of seismic data acquisition costs, geological structure modeling accuracy, and imaging algorithm efficiency, it is difficult to obtain high-resolution underground images.

[0003] Super-resolution image reconstruction is a way to map low-resolution seismic imaging profiles to high-resolution seismic imaging profiles. Among the existing technologies, common image reconstruction methods include interpolation-based image reconstruction methods, which achieve resolution improvement by estimating the pixel values ​​at specific positions between image pixels, but are prone to image detail degradation. Another image reconstruction method uses the sparsity of the image in the frequency domain to achieve high-resolution image reconstruction. At the same time, in order to adapt to image reconstruction in different situations, the image can be sparsely represented by Radon transform, Curvelet transform, and Wavelet transform. There is also a convex set projection method that solves the image super-resolution problem through iteration. The advantage of this method is that the form is flexible, but the algorithm has a high computational cost and has multiple solutions. Model-driven reconstruction methods all have assumptions, and only with rich prior knowledge can a satisfactory image be reconstructed. Summary of the invention

[0004] In view of this, one of the technical problems solved by the embodiments of the present application is to provide a method, device and storage medium for reconstructing high-resolution seismic imaging profiles, cross-integrating deep learning technology with seismic imaging methods, fully mining data information, and improving the quality of seismic imaging profiles while avoiding degradation of image details.

[0005] In a first aspect, an embodiment of the present application provides a method for reconstructing a high-resolution seismic imaging profile, comprising:

[0006] Generate a training set of seismic imaging profiles;

[0007] A neural network training is performed based on the obtained seismic imaging section training set to obtain a reconstructed high-resolution seismic imaging section.

[0008] Preferably, the step of generating a seismic imaging profile training set is:

[0009] Generate velocity model of velocity interface;

[0010] Thinning the obtained velocity model;

[0011] Reverse time migration imaging is performed on the unthinned velocity model and the thinned velocity model respectively to obtain the seismic imaging profile training set.

[0012] Preferably, the formula of the velocity interface is:

[0013]

[0014] Where z represents the depth of the velocity interface and x represents the horizontal coordinate. T is the amplitude and phase, k and b are the slope and intercept of the velocity interface.

[0015] Preferably, the formula for reconstructing a high-resolution seismic imaging section based on the input seismic imaging section is specifically:

[0016] HR=N θ [LR]

[0017] Among them, HR is the reconstructed seismic imaging section, LR is the input seismic imaging section, N θ is a convolutional neural network, and θ represents the network learning parameters.

[0018] Preferably, the formula for reconstructing a high-resolution seismic imaging section based on the input seismic imaging section is specifically:

[0019] HR=N θ [LR]

[0020] Among them, HR is the reconstructed seismic imaging section, LR is the input seismic imaging section, N θ is a convolutional neural network, and θ represents the network learning parameters.

[0021] In the second aspect, based on the same inventive concept, the present application also proposes a device for reconstructing a high-resolution seismic imaging profile, comprising:

[0022] The training set module is connected to the seismic imaging section module and is used to generate a seismic imaging section training set and send the generated training set to the seismic imaging section module.

[0023] The seismic imaging profile module is connected to the training set module and is used to perform neural network training according to the received training set data to obtain a reconstructed high-resolution seismic imaging profile.

[0024] One of the technical problems solved by the embodiments of the present application is to provide a method for reconstructing high-resolution seismic imaging profiles, cross-integrating deep learning technology with seismic imaging methods, fully mining data information, and improving the quality of seismic imaging profiles while avoiding degradation of image details. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0026] Figure 1 A schematic flow chart of a method for reconstructing a high-resolution seismic imaging profile is shown as an exemplary embodiment;

[0027] Figure 2 It is a partial formation velocity model generated when generating the seismic imaging section training set;

[0028] Figure 3 A high-resolution seismic imaging profile obtained by dividing the same formation velocity model;

[0029] Figure 4 The low-resolution seismic imaging profile obtained by dividing the same formation velocity model;

[0030] Figure 5 A schematic diagram of the convolutional neural network structure used in the embodiment of the present application;

[0031] Figure 6 The loss function curve used in the embodiment of the present application;

[0032] Figure 7 A complex model used in the embodiments of the present application;

[0033] Figure 8(a) is a low-resolution profile obtained by seismic imaging processing;

[0034] Figure 8(b) is a high-resolution profile obtained by seismic imaging processing;

[0035] FIG8( c ) is an interpolation profile obtained after encrypting pixels using a linear interpolation method in an embodiment of the present application;

[0036] Figure 8(d) is the predicted profile obtained after high-resolution reconstruction of Figure 8(a) using a neural network;

[0037] Figure 9(a) is a low-resolution cross-section after local magnification;

[0038] Figure 9(b) is a high-resolution cross section after local magnification;

[0039] Figure 9(c) is the interpolation profile after local magnification;

[0040] Figure 9(d) is the predicted section after partial enlargement;

[0041] Fig.10 It is a curve representation of single-channel data extracted at a horizontal distance of 1 km in Figure (bd);

[0042] Fig.11 For Fig.10 The wave number spectrum obtained by Fourier transform of the data in;

[0043] Fig.12 A device for reconstructing high-resolution seismic imaging profiles;

[0044] Fig.13 It is an electronic device for reconstructing high-resolution seismic imaging profiles. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the embodiments of the present application should fall within the scope of protection of the embodiments of the present application.

[0046] At present, the rapid update and iteration of high-performance computing hardware and the explosive growth of data have promoted the application of artificial intelligence in image reconstruction. Dong et al. pioneered a super-resolution convolutional neural network model to improve image quality. Subsequently, a deconvolution layer was added at the end of the network to sample the image, avoiding bicubic interpolation before the image is input into the network to accelerate network convergence. In order to solve the gradient explosion problem, Kim et al. proposed an ultra-deep network model based on a residual learning strategy to strengthen the learning of high-frequency information in the image. Ledig et al. introduced a generative adversarial network into image super-resolution reconstruction to improve the perception effect. The network contains a generator and a discriminator, and is trained in a game-theory manner.

[0047] Based on the above image processing method, the embodiment of the present application proposes a method of cross-integrating deep learning technology and seismic imaging method to fully mine data information and improve the quality of seismic imaging profiles.

[0048] The present application embodiment provides a method for reconstructing a high-resolution seismic imaging profile, comprising the following steps:

[0049] Generate a training set of seismic imaging profiles;

[0050] A neural network training is performed based on the obtained seismic imaging section training set to obtain a reconstructed high-resolution seismic imaging section.

[0051] This application uses a supervised learning strategy to train a neural network, and the strength of the network's generalization ability depends largely on the data set. The premise for generating a seismic imaging profile training set is to design a rich velocity model to obtain a rich data set. It should be noted that the velocity model can be a two-dimensional model.

[0052] Furthermore, the step of generating a seismic imaging profile training set may be:

[0053] Generate a velocity model for the velocity interface.

[0054] The obtained velocity model is thinned.

[0055] Reverse time migration imaging is performed on the unthinned velocity model and the thinned velocity model respectively to obtain the seismic imaging profile training set.

[0056] After generating the velocity model of the velocity interface, the original model generated has the problem of over-sampling, and the model becomes sparse after sparse sampling. In this application, the generated velocity model is thinned out to remove redundant data, making the model smaller and reducing the amount of calculation. At the same time, the un-thinned velocity model and the thinned velocity model are respectively subjected to reverse time migration imaging, thereby forming the input and output in the data set, and obtaining the seismic imaging profile training set.

[0057] Furthermore, the formula of the velocity interface using the two-dimensional model can be:

[0058]

[0059] Where z represents the depth of the velocity interface and x represents the horizontal coordinate. T is the amplitude and phase, k and b are the slope and intercept of the velocity interface.

[0060] The parameters in formula 1 are generated randomly in accordance with geological laws, jointly determining information such as the undulation, inclination and depth of the interface, and then filling in the sound wave velocity between two adjacent interfaces.

[0061] Furthermore, the formula for reconstructing a high-resolution seismic imaging section based on the input seismic imaging section is specifically:

[0062] HR=N θ [LR]

[0063] Among them, HR is the reconstructed seismic imaging section, LR is the input seismic imaging section, N θ is a convolutional neural network, and θ represents the network learning parameters.

[0064] The input seismic imaging profile LR is represented by a loss function, which is defined in the form of mean square error: HR true is the label. The goal of network training is to find a set of parameters, denoted as θ * ,make Minimum, that is

[0065]

[0066] The mathematical form of the parameter update iteration is

[0067]

[0068] Where η is the learning rate, is the number of batch samples, and j is the number of iterations.

[0069] The present invention reconstructs high-resolution reverse time migration imaging sections from large-surface reverse time migration imaging sections based on deep learning technology, which can effectively reduce the cost of seismic data acquisition and the computational cost of seismic data processing. No prior knowledge is required in the reconstruction process, and it has the advantages of a high degree of automation.

[0070] As mentioned above, when generating a seismic imaging profile training set, we must first design a rich velocity model, which can be According to the public information, some of the formation velocities are as follows Figure 2 shown.

[0071] When generating the training set, different grids are used for the same model to obtain high-resolution and low-resolution sections. For example, all formation velocity models are divided into 320×200 grid points with a grid spacing of 5m, and high-resolution sections are generated using seismic imaging methods. At the same time, the model is divided into 160×100 grid points with a grid spacing of 10m to generate low-resolution seismic imaging sections. Figure 3 and 4 They are seismic imaging sections with different resolutions obtained by using different grid divisions for the same model. Both can reflect the spatial position of the interface, but the pixels of the former are 4 times that of the latter, and the image quality is significantly better than the latter. The two sections with different resolutions are used as training pairs and together constitute the training data.

[0072] Table 1. Test parameters of seismic imaging algorithm

[0073]

[0074]

[0075] Table 1 shows the parameters of the seismic imaging algorithm. The algorithms all adopt the GPU / CPU collaborative parallel strategy. The "high grid density" in Table 1 refers to the traditional method, and the "low grid density" refers to the method of the present invention. It can be seen that under the same test equipment and consistent model parameters, the traditional method takes 11.5 seconds to perform seismic imaging. The traditional method obtains high-quality seismic imaging profiles at the cost of increased computational costs. However, the method of the present invention takes 3.5 seconds to perform seismic imaging, which is less than one-third of the traditional method, greatly reducing the computational cost.

[0076] The convolutional neural network structure used in the embodiment of the present application is as follows: Figure 5 As shown in the figure, the feature extraction is performed five times, each of which consists of a 3×3 convolution layer and a ReLu activation function. The last 1×1 convolution layer is used to convert the features into high-resolution seismic imaging profiles. In actual implementation, the network model can be built using the PyTorch framework. The two hyperparameters η and are set to 0.001 and 12 respectively. The loss function curve is as follows Figure 6 As shown in Figure 2, as the number of iterations increases, Gradually converge and become stable.

[0077] In a preferred embodiment, a complex model that does not appear in the training data is used ( Figure 7 ) was tested, with a horizontal distance of 2 km and a vertical depth of 1.5 km. Grids of 200×150 and 400×300 were used for division, and the sections obtained after seismic imaging processing are shown in Figures 8(a) and (b), the latter of which is used as a label to verify the reconstruction effect. After the pixels of Figure 8(a) were encrypted using the linear interpolation method, as shown in Figure 8(c), the section quality did not improve significantly. After Figure 8(a) was reconstructed with high resolution by a neural network, as shown in Figure 8(d), the quality of the reconstructed section was comparable to that of the label. In order to compare the details, the sections in the dotted box in Figure 8 are enlarged in turn as shown in Figure 9. Although there are visual differences between Figure 9(d) and (b), it is significantly better than Figure 9(c), and the sawtooth effect cannot be suppressed by linear interpolation.

[0078] In order to quantitatively evaluate the predicted profile, single-channel data is extracted at the horizontal distance of 1 km in Figure 8 (bd). Fig.10 As shown in the middle curve, the prediction result of the neural network has a higher match with the label in terms of amplitude and phase. Fig.10 The wave number spectrum is obtained by Fourier transforming the data in Fig.11 As shown in the figure, the prediction results and labels also have a high degree of consistency. The above numerical experiments show that the trained convolutional neural network has a certain generalization ability.

[0079] High-resolution seismic imaging profiles are conducive to the effective identification of geological structures and can provide support for the efficient development of resources. However, considering factors such as initial modeling accuracy, computational efficiency, and seismic data acquisition costs, the grid density of the imaging target cannot be arbitrarily increased in the seismic imaging algorithm, which limits the pixel density of the imaging profile and reduces the image quality of the seismic profile. The present invention first designs an underground layered model, and uses a seismic imaging method to generate high- and low-resolution seismic imaging profiles as a training data set. Secondly, a convolutional neural network for reconstructing high-resolution seismic imaging profiles is built, and supervised learning is used to complete network training. The method involved in the present invention does not require prior knowledge and can improve the level of automation of high-resolution seismic imaging profile reconstruction.

[0080] Based on the same inventive concept, the present application also provides a device for reconstructing high-resolution seismic imaging profiles, such as Fig.12 The device includes a training set module 1201 and a seismic imaging profile module 1202, wherein:

[0081] The training set module 1201 is connected to the seismic imaging profile module 1202, and is used to generate a seismic imaging profile training set, and send the generated training set to the seismic imaging profile 1202;

[0082] The seismic imaging profile module 1202 is connected to the training set module 1201 and is used to perform neural network training according to the received training set data to obtain a reconstructed high-resolution seismic imaging profile.

[0083] The present application also provides an electronic device, such as Fig.13 As shown, the electronic device can be implemented as part or all of the electronic device through software, hardware or a combination of both, and the electronic device is used to execute the method for reconstructing a high-resolution seismic imaging profile described in the above method embodiment. Fig.13 As shown, the electronic device 130 includes a memory 1302, a processor 1301, and a computer program stored in the memory 1302 and executable on the processor 1301. The processor 1301 executes the method for reconstructing a high-resolution seismic imaging profile described in the above method embodiment through the computer program.

[0084] In one embodiment, when executing the step of generating a seismic imaging profile training set, the processor 1301 may be configured to execute the following specific steps:

[0085] Generate a velocity model for the velocity interface.

[0086] The obtained velocity model is thinned.

[0087] Reverse time migration imaging is performed on the unthinned velocity model and the thinned velocity model respectively to obtain the seismic imaging profile training set.

[0088] In one embodiment, when executing the velocity model for generating the velocity interface, the formula executed by the processor 1301 is:

[0089]

[0090] Where z represents the depth of the velocity interface and x represents the horizontal coordinate. T is the amplitude and phase, k and b are the slope and intercept of the velocity interface.

[0091] In one embodiment, the processor 1301 reconstructs a high-resolution seismic imaging section according to the input seismic imaging section using the formula:

[0092] HR=N θ [LR]

[0093] Among them, HR is the reconstructed seismic imaging section, LR is the input seismic imaging section, N θ is a convolutional neural network, and θ represents the network learning parameters.

[0094] In one embodiment, the formula for the loss function is:

[0095]

[0096] HR true is the label. The goal of network training is to find a set of parameters, denoted as θ * , so that L(θ) is minimized, that is

[0097] The memory 1302 is used to store data generated in the intermediate process and a computer program for executing a velocity model for generating a velocity interface.

[0098] The storage medium and electronic device provided by the embodiments of the present disclosure cross-integrate deep learning technology with seismic imaging methods, fully mine data information, and improve the quality of seismic imaging sections while avoiding degradation of image details.

[0099] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0100] Those of ordinary skill in the art will appreciate that the functional modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0101] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the preferred embodiments of the embodiments of the present application have been described, once the ordinary technicians in this field know the basic creative concepts, they can make additional changes and modifications to these embodiments. Therefore, the attached claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application. Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present application without departing from the spirit and scope of the embodiments of the present application. In this way, if these modifications and variations of the embodiments of the present application fall within the scope of the claims of the embodiments of the present application and their equivalent technologies, the embodiments of the present application are also intended to include these changes and variations.

Claims

1. A method for reconstructing a high-resolution seismic imaging profile, characterized in that: include: Generate a training set of seismic imaging profiles; A neural network training is performed based on the obtained seismic imaging section training set to obtain a reconstructed high-resolution seismic imaging section.

2. The method for reconstructing a high-resolution seismic imaging profile according to claim 1, characterized in that: The steps of generating a seismic imaging profile training set are: Generate velocity model of velocity interface; Thinning the obtained velocity model; Reverse time migration imaging is performed on the unthinned velocity model and the thinned velocity model respectively to obtain the seismic imaging profile training set.

3. The method for reconstructing a high-resolution seismic imaging profile according to claim 2, characterized in that: The formula for the velocity interface is: Where z represents the depth of the velocity interface and x represents the horizontal coordinate. T is the amplitude and phase, k and b are the slope and intercept of the velocity interface.

4. The method for reconstructing a high-resolution seismic imaging profile according to claim 1, characterized in that: The formula for reconstructing a high-resolution seismic imaging section based on the input seismic imaging section is specifically: HR=N θ [LR] Among them, HR is the reconstructed seismic imaging section, LR is the input seismic imaging section, N θ is a convolutional neural network, and θ represents the network learning parameters.

5. The method for reconstructing a high-resolution seismic imaging profile according to claim 4, characterized in that: The formula of the loss function is: HR true is the label. The goal of network training is to find a set of parameters, denoted as θ * , so that L(θ) is minimized, that is 6. A device for reconstructing high-resolution seismic imaging profiles, characterized in that: include: The training set module is connected to the seismic imaging section module and is used to generate a seismic imaging section training set and send the generated training set to the seismic imaging section module. The seismic imaging profile module is connected to the training set module and is used to perform neural network training according to the received training set data to obtain a reconstructed high-resolution seismic imaging profile.

7. A storage medium, characterized in that: The storage medium includes a stored program, wherein the program executes the method for determining a robot safety circuit according to any one of claims 1 to 5 when running.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The processor executes the method for reconstructing high-resolution seismic imaging sections as described in any one of claims 1 to 5 through the computer program.

Images