Method, device, equipment, medium and program for establishing depth domain velocity model
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA PETROLEUM & CHEMICAL CORP
- Filing Date
- 2024-12-26
- Publication Date
- 2026-06-26
AI Technical Summary
Existing methods for establishing depth domain velocity models have low reliability, making it difficult to guarantee the accuracy and efficiency of the models.
By acquiring pre-stack seismic shot gather data, updating the velocity using a pre-trained depth-domain initial velocity model, and combining a self-attention mechanism and forward modeling of the acoustic equation, a L2-norm target loss function is established for iterative inversion to obtain the depth-domain velocity model.
It improves the efficiency and accuracy of establishing depth domain velocity models and enhances the depth domain velocity analysis capabilities of seismic data.
Smart Images
Figure CN122283879A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to geophysical exploration technology and neural network technology, and in particular to a method, apparatus, device, medium and program for establishing a depth domain velocity model. Background Technology
[0002] Velocity modeling is an important part of seismic data processing, and its goal is to establish a model that can accurately describe the velocity distribution of the subsurface medium. This model is crucial for subsequent depth migration imaging, seismic inversion, and seismic interpretation. Currently, traditional velocity modeling algorithms mainly include (1) ray tomography-based modeling algorithms, which are a method of reconstructing the subsurface velocity structure by inverting the travel time of seismic waves. They are usually based on ray theory and gradually improve the velocity model through an iterative optimization process. (2) Waveform inversion modeling algorithms, which are highly nonlinear inversion methods that obtain the optimal velocity model by minimizing the difference between the simulated wavefield and the actual observed data. However, in actual processing, they are limited by data quality and computational load, and their large-scale practical application still has certain limitations.
[0003] In recent years, with the continuous development of artificial intelligence, algorithms for velocity modeling using technologies such as deep learning have seen some progress. The advantage of this method is that it can learn patterns from large amounts of data, thereby quickly predicting velocity models. In general, the existing intelligent velocity modeling techniques mainly focus on directly establishing the relationship between input seismic waveform data or an initial velocity model and a high-precision velocity model. Based on deep neural networks, they use a data-driven approach to mine the relationship between the high-precision velocity model and data such as seismic waveforms. While these methods have some modeling effectiveness, their reliability is difficult to guarantee to a large extent due to the lack of constraints from physical information. Summary of the Invention
[0004] This disclosure provides a method, apparatus, device, medium, and program for establishing a depth domain velocity model, in order to solve the problem of low reliability in existing methods for establishing depth domain velocity models.
[0005] Firstly, this disclosure provides a method for establishing a depth-domain velocity model, including:
[0006] Acquire pre-stack seismic shot gather data for the current region, and output velocity update values based on the pre-stack seismic shot gather data using a pre-trained depth domain initial velocity model.
[0007] The updated speed is obtained based on the speed update amount;
[0008] The updated velocity is subjected to forward modeling of the acoustic wave equation based on preset forward modeling parameters to obtain a forward modeling record;
[0009] Based on the forward modeling records and the pre-acquired observation records, a L2 target loss function is established, and the L2 target loss function is used to iteratively invert the initial velocity model in the depth domain.
[0010] After reaching the preset number of iterations, the depth domain velocity model is confirmed. The depth domain velocity model is then used to perform depth domain velocity analysis based on the seismic data of the target area to obtain velocity data.
[0011] In some embodiments, the step of using a pre-trained depth-domain initial velocity model to output velocity updates based on the pre-stack seismic shot gather data includes:
[0012] The seismic pre-stack shot gather data is continuously convolved using the first and second convolution units in the depth domain initial velocity model to obtain convolution features;
[0013] The seismic pre-stack shot gather data is deactivated using the first and second deactivation units in the depth domain initial velocity model to obtain deactivation characteristics.
[0014] The inactive features are weighted and mapped using the self-attention mechanism layer in the initial velocity model of the depth domain to obtain the mapped features.
[0015] The mapping features are subjected to continuous deconvolution processing to obtain the speed update amount.
[0016] In some embodiments, the step of using the self-attention mechanism layer in the initial velocity model of the depth domain to perform weighted mapping processing on the inactive features to obtain mapped features includes:
[0017] The inactive features are mapped into a query matrix, a key matrix, and a value matrix using the self-attention mechanism layer in the deep domain initial velocity model.
[0018] The attention score is calculated using the query matrix and the key matrix, and the attention score is normalized to obtain the attention weight.
[0019] The value matrix is weighted and summed using the attention weights to obtain a weighted value matrix.
[0020] The weighted matrix is linearly transformed and mapped back to the original space to obtain the mapped features.
[0021] In some embodiments, performing forward modeling of the acoustic wave equation on the updated velocity according to preset forward modeling parameters to obtain a forward modeling record includes:
[0022] A grid model is generated based on the size and resolution of the current region;
[0023] The finite difference method is used to transform the pre-defined acoustic wave equation into a discrete difference equation.
[0024] Numerical calculations are performed on each point in the medium based on the difference equation and the updated velocity to obtain the calculation results;
[0025] The wave field value of each grid in the grid model is iteratively updated based on the calculation results and the forward modeling parameters to obtain the forward modeling record.
[0026] In some embodiments, establishing a L2-norm objective loss function based on the forward modeling records and pre-acquired observation records includes:
[0027] The L2-norm objective loss function can be expressed using the following formula:
[0028]
[0029] Where J is the loss value, n represents the number of samples in the forward modeling record, and U k (θ) represents the k-th sample in the forward modeling record. This represents the k-th sample in the observation record.
[0030] In some embodiments, the iterative inversion of the initial velocity model in the depth domain using the L2 norm objective loss function includes:
[0031] The loss value between the forward modeling record and the observation record is calculated using the L2-norm objective loss function;
[0032] After updating the weight parameters of the depth domain initial velocity model based on the loss value, the process returns to the step of outputting the velocity update amount based on the pre-trained depth domain initial velocity model and the seismic pre-stack shot gather data.
[0033] Secondly, this disclosure provides a depth-domain velocity model establishment apparatus, comprising:
[0034] The data acquisition module is used to acquire pre-stack shot gather data for the current region.
[0035] The velocity update module is used to output a velocity update amount based on the pre-stack seismic shot gather data using a pre-trained depth domain initial velocity model, and to obtain the updated velocity based on the velocity update amount.
[0036] The iterative inversion module is used to perform forward modeling of the acoustic wave equation on the updated velocity according to preset forward modeling parameters to obtain forward modeling records. Based on the forward modeling records and pre-acquired observation records, a L2 target loss function is established, and the L2 target loss function is used to perform iterative inversion of the initial velocity model in the depth domain.
[0037] The data analysis module is used to confirm the depth domain velocity model after reaching a preset number of iterations, and then use the depth domain velocity model to perform depth domain velocity analysis based on the seismic data of the target area to obtain velocity data.
[0038] Thirdly, this disclosure 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 of the method described in the foregoing aspects.
[0039] Fourthly, this disclosure provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the methods described in the above aspects.
[0040] Fifthly, this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the methods described in the above aspects.
[0041] This disclosure provides a method, apparatus, device, storage medium, and computer program for establishing a depth domain velocity model. It acquires pre-stack seismic shot gather data for the current region, uses a pre-trained initial depth domain velocity model to output velocity updates based on the pre-stack seismic shot gather data, obtains updated velocities based on the updated velocities, performs forward modeling of the updated velocities using preset forward modeling parameters to obtain forward modeling records, establishes a L2-norm target loss function based on the forward modeling records and pre-acquired observation records, and uses the L2-norm target loss function to iteratively invert the initial depth domain velocity model. After reaching a preset number of iterations, the depth domain velocity model is confirmed. The depth domain velocity model is then used to perform depth domain velocity analysis based on seismic data from the target region to obtain velocity data. This solves the problem of low reliability in existing depth domain velocity model establishment methods and improves the efficiency and accuracy of depth domain velocity model establishment. Attached Figure Description
[0042] The present disclosure will be described in more detail below based on embodiments and with reference to the accompanying drawings:
[0043] Figure 1 A flowchart illustrating a method for establishing a depth domain velocity model according to an embodiment of this disclosure;
[0044] Figure 2 A flowchart illustrating a method for establishing a depth domain velocity model according to an embodiment of this disclosure;
[0045] Figure 3 This is a schematic diagram of the network structure of a neural network model provided in an embodiment of the present disclosure;
[0046] Figure 4 This is a functional block diagram of a depth domain velocity model establishment device provided in an embodiment of the present disclosure.
[0047] In the accompanying drawings, the same parts are referred to by the same reference numerals, and the drawings are not drawn to scale. Detailed Implementation
[0048] To enable those skilled in the art to better understand the technical solutions of this disclosure, and to fully understand and implement the process of how this disclosure applies technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. The embodiments of this disclosure and the various features within them can be combined with each other without conflict, and the resulting technical solutions are all within the protection scope of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort should fall within the protection scope of this disclosure.
[0049] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0050] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0051] Example 1
[0052] Figure 1 This is a flowchart illustrating a method for establishing a depth domain velocity model according to an embodiment of this disclosure. Figure 1 As shown, a method for establishing a depth domain velocity model includes...
[0053] S1. Obtain the pre-stack seismic shot gather data for the current region, and output the velocity update based on the pre-stack seismic shot gather data using a pre-trained depth domain initial velocity model.
[0054] In this embodiment of the invention, the pre-stack seismic shot gather data refers to the raw seismic data collected in the field during seismic exploration. These data record the reflection information of seismic waves generated from the source (shot point) reaching each geophone.
[0055] In this embodiment of the invention, the initial velocity model in the deep domain is a neural network model, and a self-attention mechanism is introduced.
[0056] In this embodiment of the invention, the step of using a pre-trained depth-domain initial velocity model to output velocity updates based on the pre-stack seismic shot gather data includes:
[0057] The seismic pre-stack shot gather data is continuously convolved using the first and second convolution units in the depth domain initial velocity model to obtain convolution features;
[0058] The seismic pre-stack shot gather data is deactivated using the first and second deactivation units in the depth domain initial velocity model to obtain deactivation characteristics.
[0059] The inactive features are weighted and mapped using the self-attention mechanism layer in the initial velocity model of the depth domain to obtain the mapped features.
[0060] The mapping features are subjected to continuous deconvolution processing to obtain the speed update amount.
[0061] Specifically, both the first convolutional unit and the second convolutional unit contain two convolutional layers and one pooling layer.
[0062] Specifically, the first deactivation unit includes two convolutional layers, one deactivation layer, and one pooling layer, and the second deactivation unit includes two convolutional layers and one deactivation layer.
[0063] In this embodiment of the invention, the self-attention mechanism layer includes three self-attention block convolutional layers.
[0064] In this embodiment of the invention, the self-attention mechanism layer is an important part of deep learning. Its core idea is to allow the model to dynamically adjust the degree of attention to each element when processing sequence data by calculating the correlation score between elements at different positions in the sequence, thereby capturing the complex dependencies within the sequence and improving the accuracy of network data feature mining.
[0065] In this embodiment of the invention, the continuous deconvolution processing of the mapped features refers to performing continuous deconvolution on the mapped features using three deconvolution units. Each deconvolution unit includes one deconvolution layer and two convolutional layers.
[0066] In this embodiment of the invention, reference is made to Figure 3 The diagram shown is a schematic representation of the network structure of a neural network model provided in this embodiment.
[0067] Specifically, the step of using the self-attention mechanism layer in the initial velocity model of the depth domain to perform weighted mapping processing on the inactive features to obtain mapped features includes:
[0068] The inactive features are mapped into a query matrix, a key matrix, and a value matrix using the self-attention mechanism layer in the deep domain initial velocity model.
[0069] The attention score is calculated using the query matrix and the key matrix, and the attention score is normalized to obtain the attention weight.
[0070] The value matrix is weighted and summed using the attention weights to obtain a weighted value matrix.
[0071] The weighted matrix is linearly transformed and mapped back to the original space to obtain the mapped features.
[0072] In this embodiment of the invention, by acquiring pre-stack seismic shot gather data for the current region, the efficiency of obtaining velocity updates is improved. Furthermore, by utilizing a pre-trained depth-domain initial velocity model based on the pre-stack seismic shot gather data, the velocity update is output, thus improving the efficiency of obtaining the updated velocity.
[0073] S2. Obtain the updated speed based on the speed update amount.
[0074] In this embodiment of the invention, obtaining the updated velocity based on the velocity update amount is achieved by adding the velocity update amount to the initial velocity model in the depth domain to obtain an accurate initial velocity model, and then using the accurate initial velocity model to update the velocity to obtain the updated velocity.
[0075] In this embodiment of the invention, by obtaining the updated velocity based on the velocity update amount, the efficiency of subsequent forward modeling of the acoustic wave equation is improved.
[0076] S3. Perform forward modeling of the acoustic wave equation on the updated velocity according to the preset forward modeling parameters to obtain the forward modeling record.
[0077] In this embodiment of the invention, the preset forward modeling parameters are set based on the parameters of the field seismic acquisition and observation system. These forward modeling parameters include, but are not limited to, shot-receiver point distribution and wavelet frequency.
[0078] In detail, the distribution of shot and receiver points includes parameters such as shot point spacing, trace spacing, shot line spacing, and receiver line spacing in seismic exploration.
[0079] In this embodiment of the invention, the forward modeling of the acoustic wave equation is a numerical simulation technique in geophysical exploration, used to simulate the propagation process of seismic waves in a medium when the structure of the underground medium is known.
[0080] The forward modeling of the acoustic wave equation, as detailed in the text, involves solving the wave equation, which describes the propagation behavior of sound waves in a medium. In the two-dimensional case, the acoustic wave equation can be expressed as a continuous partial differential equation and solved numerically through discretization (such as the finite difference method). The finite difference method is a commonly used numerical solution method in the forward modeling of the acoustic wave equation. It discretizes continuous spatial and temporal coordinates and uses the finite difference form to approximate the propagation behavior of sound waves in a homogeneous medium.
[0081] In detail, forward modeling of the acoustic wave equations requires consideration of appropriate boundary conditions to ensure the accuracy of the simulation results. These boundary conditions may include absorbing boundaries, perfectly matched layers, etc., to reduce artificial boundary reflections.
[0082] In this embodiment of the invention, the step of performing forward modeling of the acoustic wave equation on the updated velocity according to preset forward modeling parameters to obtain a forward modeling record includes:
[0083] A grid model is generated based on the size and resolution of the current region;
[0084] The finite difference method is used to transform the pre-defined acoustic wave equation into a discrete difference equation.
[0085] Numerical calculations are performed on each point in the medium based on the difference equation and the updated velocity to obtain the calculation results;
[0086] The wave field value of each grid in the grid model is iteratively updated based on the calculation results and the forward modeling parameters to obtain the forward modeling record.
[0087] In detail, the size and resolution of the pre-stack region are data included in the pre-stack shot gather data.
[0088] Specifically, the wave field values include both the pressure field and the velocity field.
[0089] In this embodiment of the invention, the updated velocity is subjected to forward modeling of the acoustic wave equation according to preset forward modeling parameters to obtain a forward modeling record, which improves the efficiency of subsequent iterative inversion.
[0090] S4. Based on the forward modeling records and the pre-acquired observation records, establish a L2 target loss function, and use the L2 target loss function to iteratively invert the initial velocity model in the depth domain.
[0091] In this embodiment of the invention, the pre-acquired observation records are data included in the pre-stack seismic shot gather data.
[0092] In this embodiment of the invention, establishing the L2 objective loss function based on the forward modeling records and pre-acquired observation records includes:
[0093] The L2-norm objective loss function can be expressed using the following formula:
[0094]
[0095] Where J is the loss value, n represents the number of samples in the forward modeling record, and U k (θ) represents the k-th sample in the forward modeling record. This represents the k-th sample in the observation record.
[0096] In this embodiment of the invention, the iterative inversion of the initial velocity model in the depth domain using the L2 norm objective loss function includes:
[0097] The loss value between the forward modeling record and the observation record is calculated using the L2-norm objective loss function;
[0098] After updating the weight parameters of the depth domain initial velocity model based on the loss value, the process returns to the step of outputting the velocity update amount based on the pre-trained depth domain initial velocity model and the seismic pre-stack shot gather data.
[0099] In this embodiment of the invention, after returning to the step of outputting velocity update amount based on the pre-trained depth domain initial velocity model and the pre-stack seismic shot gather data, the following steps are performed: obtaining the updated velocity based on the velocity update amount; performing acoustic wave equation forward modeling on the updated velocity based on preset forward modeling parameters to obtain forward modeling records; establishing a L2 norm target loss function based on the forward modeling records and the pre-acquired observation records; and using the L2 norm target loss function to perform iterative inversion on the depth domain initial velocity model.
[0100] In this embodiment of the invention, the weight parameters are a core component of the model learning process, which determines how the model extracts features from the input data and makes predictions.
[0101] The weight parameters, described in detail, are the coefficients of the edges connecting neurons, which determine the strength of the signal transmitted from one neuron to another. These weights are adjusted during training using optimization algorithms (such as gradient descent) to minimize the model's loss function.
[0102] In this embodiment of the invention, updating the weight parameters of the initial velocity model in the depth domain based on the loss value is achieved by using a gradient descent algorithm to update the weight parameters based on the loss value.
[0103] In detail, the gradient descent algorithm is one of the most commonly used optimization algorithms in machine learning and deep learning. Its core idea is to iteratively update the parameters and search along the direction of gradient descent of the objective function (usually the loss function) until a local minimum of the function is found.
[0104] In this embodiment of the invention, a L2-norm objective loss function is established based on the forward modeling records and pre-acquired observation records. The L2-norm objective loss function is then used to iteratively invert the initial velocity model in the depth domain, thereby improving the efficiency of establishing the depth domain velocity model.
[0105] S5. After reaching the preset number of iterations, the depth domain velocity model is confirmed. The depth domain velocity model is then used to perform depth domain velocity analysis based on the seismic data of the target area to obtain velocity data.
[0106] In this embodiment of the invention, the depth domain velocity model is a model used in seismic exploration to describe the change of velocity in the subsurface medium with depth. It is crucial for the simulation of seismic wave propagation, seismic data processing, and interpretation.
[0107] In detail, the shape of the seismic wavelet in depth-domain migration data is influenced by the formation velocity. The maximum peak value of the depth-domain wavelet coincides with the formation reflection interface, indicating that the depth-domain velocity model plays an important role in identifying the shape of the seismic wavelet and the reflection interface.
[0108] In this embodiment of the invention, the step of using the depth domain velocity model to perform depth domain velocity analysis based on seismic data of the target area to obtain velocity data involves using the depth domain velocity model to transform the seismic data from the time domain to the depth domain to obtain depth domain data, and finally calculating the velocity at different depth layers based on the depth domain data. Finally, the velocity inversion is performed on the calculation results to obtain the velocity data.
[0109] In this embodiment of the invention, after a preset number of iterations is reached, a depth domain velocity model is confirmed. By using the depth domain velocity model to perform depth domain velocity analysis based on seismic data of the target area, velocity data is obtained, which improves the efficiency of establishing the depth domain velocity model and improves the efficiency and accuracy of performing depth domain velocity analysis on seismic data.
[0110] In this embodiment of the invention, reference is made to Figure 2 The diagram shown is a flowchart of a method for establishing a depth domain velocity model in this embodiment.
[0111] In this embodiment of the invention, a method for establishing a deep domain velocity model is proposed. By introducing a network architecture with a self-attention mechanism, seismic shot gather data is used as input, and the network output is the velocity update quantity. Next, the initial velocity model established by traditional methods for the data of this area is added to the velocity update quantity output by the network to obtain a velocity model, which is then input into the wave equation forward modeling kernel for acoustic wave equation forward modeling. The obtained forward modeling record and observation data are used to establish a L2 target loss function for iterative updates. This establishes a chain relationship between the neural network and the wave equation forward modeling kernel, realizing an intelligent wave equation velocity modeling technology based on neural network constraints, which improves the accuracy and efficiency of velocity modeling.
[0112] Example 2
[0113] Based on the above embodiments, this embodiment provides an application example.
[0114] This invention addresses the problem of high-precision velocity modeling in the depth domain of seismic data and combines a specific process of full-waveform inversion technology. It fully leverages the advantages of deep neural networks in global information mining and nonlinear operators, and establishes a network architecture based on a deep learning framework that incorporates a self-attention mechanism. This achieves the mapping from seismic data to velocity updates. Furthermore, it establishes a chain relationship based on the DeepWave forward modeling operator in the Python open-source package, fully utilizing the automatic differentiation mechanism of neural networks to obtain velocity updates based on neural network operators. In essence, this realizes the regularization constraint of velocity by the neural network.
[0115] The technical process adopted in this invention is as follows: Figure 1As shown, the overall technology comprises two parts. The first part is the neural network velocity update part: a network architecture incorporating a self-attention mechanism is established, using seismic shot gather data as input and the network output as the velocity update quantity. The second part is the wave equation forward modeling part. This part employs an algorithm where an initial velocity model established based on traditional methods for the data in this area is added to the velocity update quantity output from the first part of the network to obtain a velocity model, which is then used as input to the forward modeling operator for acoustic wave equation forward modeling. Furthermore, the obtained forward modeling record and observation data are used to establish a L2-norm objective loss function for iterative updates, while simultaneously updating the neural network weights. Ultimately, intelligent wave equation velocity modeling based on neural network constraints is achieved. The network structure used in this invention is as follows: Figure 2 As shown, the overall structure adopts the UNET network, and the accuracy of network feature mining is improved by introducing a self-attention mechanism.
[0116] The basic process of this algorithm can be summarized in the following steps:
[0117] (1) Algorithm Flow Construction and Data Preparation: Based on Figure 1 The algorithm flow shown completes the construction of the neural network and the entire algorithm chain relationship. According to the algorithm requirements of this invention, data preparation is carried out: ① seismic pre-stack shot gather data obtained from actual field observations in the current work area; ② the initial velocity model in the depth domain obtained by conventional algorithms in the current work area.
[0118] (2) Parameter settings: ① Set the number of neural network iterations; ② Set the forward modeling parameters according to the parameters of the field seismic acquisition and observation system, including shot-receiver point distribution, wavelet frequency, etc.
[0119] (3) Iterative inversion: Based on the above algorithm flow and parameter settings, iterative inversion is performed. Since the main role of the neural network in the algorithm process is essentially the regularization constraint of speed, the training process of the neural network is actually an iterative inversion process, and the final output is the inversion result.
[0120] Example 3
[0121] like Figure 4 The diagram shown is a functional block diagram of a depth domain velocity model establishment device provided in this embodiment.
[0122] The depth domain velocity model building device 100 described in this invention can be installed in an electronic device. Depending on the functions implemented, the depth domain velocity model building device 100 may include a data acquisition module 101, a velocity update module 102, an iterative inversion module 103, and a data analysis module 104. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.
[0123] In this embodiment, the functions of each module / unit are as follows:
[0124] The data acquisition module 101 is used to acquire pre-stack seismic shot gather data for the current region;
[0125] The velocity update module 102 is used to output a velocity update amount based on the pre-trained depth domain initial velocity model and the pre-stack seismic shot gather data, and to obtain the updated velocity based on the velocity update amount.
[0126] The iterative inversion module 103 is used to perform forward modeling of the acoustic equation on the updated velocity according to the preset forward modeling parameters to obtain forward modeling records, establish a L2 target loss function based on the forward modeling records and the pre-acquired observation records, and use the L2 target loss function to perform iterative inversion of the initial velocity model in the depth domain.
[0127] The data analysis module 104 is used to confirm the depth domain velocity model after reaching a preset number of iterations, and to perform depth domain velocity analysis based on the seismic data of the target area using the depth domain velocity model to obtain velocity data.
[0128] Example 4
[0129] Based on the above embodiments, this embodiment 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 perform the following steps:
[0130] Acquire pre-stack seismic shot gather data for the current region, and output velocity update values based on the pre-stack seismic shot gather data using a pre-trained depth domain initial velocity model.
[0131] The updated speed is obtained based on the speed update amount;
[0132] The updated velocity is subjected to forward modeling of the acoustic wave equation based on preset forward modeling parameters to obtain a forward modeling record;
[0133] Based on the forward modeling records and the pre-acquired observation records, a L2 target loss function is established, and the L2 target loss function is used to iteratively invert the initial velocity model in the depth domain.
[0134] After reaching the preset number of iterations, the depth domain velocity model is confirmed. The depth domain velocity model is then used to perform depth domain velocity analysis based on the seismic data of the target area to obtain velocity data.
[0135] In some embodiments of this example, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that the computer program, when executed by a processor, performs the following steps:
[0136] Acquire pre-stack seismic shot gather data for the current region, and output velocity update values based on the pre-stack seismic shot gather data using a pre-trained depth domain initial velocity model.
[0137] The updated speed is obtained based on the speed update amount;
[0138] The updated velocity is subjected to forward modeling of the acoustic wave equation based on preset forward modeling parameters to obtain a forward modeling record;
[0139] Based on the forward modeling records and the pre-acquired observation records, a L2 target loss function is established, and the L2 target loss function is used to iteratively invert the initial velocity model in the depth domain.
[0140] After reaching the preset number of iterations, the depth domain velocity model is confirmed. The depth domain velocity model is then used to perform depth domain velocity analysis based on the seismic data of the target area to obtain velocity data.
[0141] In some embodiments of this example, a computer program product is provided, including a computer program, characterized in that the computer program, when executed by a processor, implements the steps of the method described in the above embodiments.
[0142] The processor may include, but is not limited to, one or more processors or microprocessors. Each processor may be implemented as an Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), controller, microcontroller, microprocessor, or other electronic component, for executing the methods in the above embodiments.
[0143] Computer-readable storage media can be implemented by any type of volatile or non-volatile storage device or a combination thereof. Computer-readable storage media may include, but are not limited to, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, and computer storage media (e.g., hard disks, floppy disks, solid-state drives, removable disks, Blu-ray discs, etc.).
[0144] Computer-readable storage media may also store at least one computer-executable program, such as computer-readable instructions. Computer-readable storage media include, but are not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Computer-readable storage media may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, a non-transitory computer-readable storage medium may be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions stored on the computer-readable storage medium, the various methods described above can be performed.
[0145] In addition, the computer device may include (but is not limited to) a data bus, an input / output (I / O) bus, a display, and input / output devices (e.g., keyboard, mouse, speakers, etc.).
[0146] The processor can communicate with external devices via the I / O bus through wired or wireless networks.
[0147] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product / computer program product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.
[0148] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0149] It should be noted that, in this disclosure, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element limited by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0150] While the embodiments disclosed herein are as described above, the foregoing content is merely for the purpose of facilitating understanding of this disclosure and is not intended to limit this disclosure. Any person skilled in the art to which this disclosure pertains may make any modifications and changes in form and detail of the implementation without departing from the spirit and scope of this disclosure; however, the scope of patent protection of this disclosure shall still be determined by the scope defined in the appended claims.
Claims
1. A method for establishing a depth-domain velocity model, characterized in that, include: Acquire pre-stack seismic shot gather data for the current region, and output velocity update values based on the pre-stack seismic shot gather data using a pre-trained depth domain initial velocity model. The updated speed is obtained based on the speed update amount; The updated velocity is subjected to forward modeling of the acoustic wave equation based on preset forward modeling parameters to obtain a forward modeling record; Based on the forward modeling records and the pre-acquired observation records, a L2 target loss function is established, and the L2 target loss function is used to iteratively invert the initial velocity model in the depth domain. After reaching the preset number of iterations, the depth domain velocity model is confirmed. The depth domain velocity model is then used to perform depth domain velocity analysis based on the seismic data of the target area to obtain velocity data.
2. The method according to claim 1, characterized in that, The method of using a pre-trained depth-domain initial velocity model to output velocity updates based on the pre-stack seismic shot gather data includes: The seismic pre-stack shot gather data is continuously convolved using the first and second convolution units in the depth domain initial velocity model to obtain convolution features; The seismic pre-stack shot gather data is deactivated using the first and second deactivation units in the depth domain initial velocity model to obtain deactivation characteristics. The inactive features are weighted and mapped using the self-attention mechanism layer in the initial velocity model of the depth domain to obtain the mapped features. The mapping features are subjected to continuous deconvolution processing to obtain the speed update amount.
3. The method according to claim 2, characterized in that, The step of using the self-attention mechanism layer in the initial velocity model of the depth domain to perform weighted mapping processing on the inactive features to obtain mapped features includes: The inactive features are mapped into a query matrix, a key matrix, and a value matrix using the self-attention mechanism layer in the deep domain initial velocity model. The attention score is calculated using the query matrix and the key matrix, and the attention score is normalized to obtain the attention weight. The value matrix is weighted and summed using the attention weights to obtain a weighted value matrix. The weighted matrix is linearly transformed and mapped back to the original space to obtain the mapped features.
4. The method according to claim 1, characterized in that, The step of performing forward modeling of the updated velocity using preset forward modeling parameters to obtain a forward modeling record includes: A grid model is generated based on the size and resolution of the current region; The finite difference method is used to transform the pre-defined acoustic wave equation into a discrete difference equation. Numerical calculations are performed on each point in the medium based on the difference equation and the updated velocity to obtain the calculation results; The wave field value of each grid in the grid model is iteratively updated based on the calculation results and the forward modeling parameters to obtain the forward modeling record.
5. The method according to claim 1, characterized in that, The step of establishing a L2 target loss function based on the forward modeling records and pre-acquired observation records includes: The L2-norm objective loss function can be expressed using the following formula: Where J is the loss value, n represents the number of samples in the forward modeling record, and U k (θ) represents the k-th sample in the forward modeling record. This represents the k-th sample in the observation record.
6. The method according to claim 5, characterized in that, The iterative inversion of the initial velocity model in the depth domain using the L2 norm objective loss function includes: The loss value between the forward modeling record and the observation record is calculated using the L2-norm objective loss function; After updating the weight parameters of the depth domain initial velocity model based on the loss value, the process returns to the step of outputting the velocity update amount based on the pre-trained depth domain initial velocity model and the seismic pre-stack shot gather data.
7. A device for establishing a depth-domain velocity model, characterized in that, include: The data acquisition module is used to acquire pre-stack shot gather data for the current region. The velocity update module is used to output a velocity update amount based on the pre-stack seismic shot gather data using a pre-trained depth domain initial velocity model, and to obtain the updated velocity based on the velocity update amount. The iterative inversion module is used to perform forward modeling of the acoustic wave equation on the updated velocity according to preset forward modeling parameters to obtain forward modeling records. Based on the forward modeling records and pre-acquired observation records, a L2 target loss function is established, and the L2 target loss function is used to perform iterative inversion of the initial velocity model in the depth domain. The data analysis module is used to confirm the depth domain velocity model after reaching a preset number of iterations, and then use the depth domain velocity model to perform depth domain velocity analysis based on the seismic data of the target area to obtain velocity data.
8. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 6.