Multi-scale full-waveform inversion method for node ultra-large offset data

Through the multi-scale full-waveform inversion method of node ultra-large offset data, the large-scale model and initial wave information are used to solve the local extreme value problem caused by the lack of medium and low-frequency data in the full-waveform inversion, and the inversion of high-precision speed model is realized performance reduces costs and improves efficiency.

CN118938295APending Publication Date: 2024-11-12CHINA PETROLEUM & CHEMICAL CORP +1
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202310515803.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-09
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The full waveform inversion method is prone to fall into local extreme values ​​when the low-frequency data is missing, resulting in unstable inversion results and low accuracy. It is especially difficult to obtain high-precision initial models in terrestrial low signal-to-noise ratio seismic data.

Method used

The multi-scale full waveform inversion method of node ultra-large offset data is adopted, and forward simulation is performed by large-scale theoretical model, reasonable multi-scale inversion frequency band is designed, objective function is constructed to improve wavelet inversion accuracy, and diagonal Hessian matrix is ​​used to pre-exist. Conditional gradient algorithm and smoothing filtering technology, frequency domain wave sub-estimation, multi-scale frequency division inversion strategy and Gaussian beam offset imaging technology, combining low-frequency wave field information and first-to-date wave information to avoid local extreme value and constraint solution characteristics, ensuring The robustness of the inversion process.

Benefits of technology

A stable and high-precision speed model is achieved in the absence of low-frequency components, reducing the dependence of multi-scale full-waveform inversion on low-frequency data, improving the reliability and cost-effectiveness of inversion results, and is suitable for land node acquisition. Extra large offset data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118938295A_ABST
    Figure CN118938295A_ABST
Patent Text Reader

Abstract

The invention provides a node ultra-large offset data multi-scale full-waveform inversion method, and the method comprises the steps: 1, carrying out the forward modeling of a large-scale theoretical model, and obtaining a seismic response; 2, designing a reasonable multi-scale inversion frequency band; 3, constructing and adding an objective function to improve the wavelet inversion precision; step 4, constructing a steady waveform inversion process; step 5, calculating to obtain a time window function corresponding to each channel of prediction data; and step 6, calculating a pre-stack migration imaging gather and a post-stack migration imaging gather. According to the node ultra-large offset data multi-scale full-waveform inversion method, the requirement of multi-scale full-waveform inversion on low-frequency data in field acquisition data is reduced while a high-precision inversion result is obtained, the cost is greatly reduced, the benefit is improved, and a stable and high-precision speed model can be obtained; the cost is greatly reduced and the benefit is improved while the accurate inversion of the shallow, medium and deep layers and low, medium and high wave number sections is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of oil and gas exploration seismic data processing, and in particular to a multi-scale full waveform inversion method for node ultra-large offset data. Background Art

[0002] Full waveform inversion is an effective method for reconstructing high-resolution model parameters of underground media based on the wave equation, which describes the propagation process of seismic waves in underground media. It uses the kinematic and dynamic information of the seismic wave field to minimize the residual of the predicted wave field and the actual wave field. It has the potential to reveal structural details and lithology under complex geological backgrounds. Compared with traditional velocity modeling methods such as migration analysis methods and travel time tomography, full waveform inversion can obtain the medium and high frequency components of the velocity field. The inversion results can meet the requirements of high-precision complex structural seismic imaging and reservoir prediction for velocity parameter accuracy. The full waveform inversion method is a gradient-based optimization iterative inversion method that has a serious dependence on the initial model. However, it is difficult to obtain a high-precision velocity model in practical applications. Therefore, it is difficult to ensure the successful application of full waveform inversion, especially in my country's low signal-to-noise ratio seismic data on land. In addition, land seismic data has now turned to full-node acquisition, with wireless node detectors replacing conventional cable detectors, breaking through the limitations of traditional acquisition channel capabilities. As long as the number of detectors is sufficient, they can be arbitrarily deployed, and the data of ultra-large offset distances has increased significantly.

[0003] Full waveform inversion is a highly nonlinear inverse problem, and the accuracy of the inversion results is highly dependent on the initial model. In practical applications, it is difficult to obtain a high-precision initial model, and frequency jumps are prone to occur during data fitting, causing the full waveform inversion to fall into a local extreme value solution. A common solution is to adopt a multi-scale full waveform inversion strategy, which decomposes the problem into different scales to ensure the stable convergence of the inversion process. The frequency multi-scale method is to decompose the frequency according to the problem and divide the frequency into appropriate frequency groups, that is, to use the inversion result of the previous scale as the initial model for the next scale inversion. By inverting from low frequency to high frequency, the low, medium, and high wavenumber components of the model can be obtained respectively, reducing the possibility of falling into a local extreme value, and effectively solving the problem of inversion non-convergence caused by the initial value being far from the true solution. However, conventional seismic data often lack low-frequency information and rarely contain low-frequency data below 5Hz, while the price of high-performance low-frequency geophones is generally dozens of times that of conventional geophones. Therefore, the frequency-band multi-scale full waveform inversion method often cannot be directly applied to conventionally acquired data with missing low frequencies, which greatly limits its scope of application.

[0004] In the Chinese patent application with application number: CN201710791533.X, a method and device for seismic full waveform inversion are involved. In one embodiment, the seismic full waveform inversion method includes: obtaining seismic data for full waveform inversion from the memory of the electronic terminal or receiving data uploaded by the user in real time; filtering and preprocessing the seismic data; establishing a regularization model based on the preprocessed seismic data; iteratively solving the regularization model based on the fast gradient algorithm, and inverting to obtain a velocity model of the seismic full waveform.

[0005] In the Chinese patent application with application number: CN201410373265.6, an efficient full waveform inversion method in the time domain is involved, which belongs to the field of high-efficiency, low-storage, and high-precision velocity modeling of petroleum seismic exploration data. The first step is to use three random schemes: source polarity random, source position random, and source number random, to randomly form super-shot sets of different frequency bands for the original observed seismic records; the second step is to use CUDA-based multi-GPU parallel computing technology to accelerate the high-order staggered grid finite difference forward simulation of the first-order acoustic wave equation system and the elastic wave equation system in the time domain, and obtain the GPU simulation ifreq frequency band super-shot set; in the forward simulation, CPU and GPU collaborative technology is used, and multiple GPU cores are used to simulate different source propagations, and the boundary data of the source wave field and all wave fields at the last moment are stored in the computer memory.

[0006] In the Chinese patent application with application number: CN202210357986.2, a reflection waveform inversion method and system based on deep learning of convolutional neural network is involved, including: preprocessing the original seismic data to obtain an initial velocity model; a blockization step: creating a velocity model base by blockizing the initial velocity model into spatially correlated and non-connected feature sets; generating a training set based on the velocity model base; training a preset convolutional neural network based on the training set to obtain a trained convolutional neural network; using the initial velocity model and the initial reverse time migration image as input, using the trained convolutional neural network for prediction to obtain a predicted velocity model; judging whether the error between the seismic data simulated by the predicted velocity model forward model and the original seismic data is less than a preset threshold; if not, using the predicted velocity model as a new initial velocity model, and returning to execute the blockization step. This invention alleviates the technical problem of inaccurate predicted velocity models existing in the prior art.

[0007] In the Chinese patent application with application number: CN201410206451.0, a method for extracting the full waveform inversion gradient operator based on the excitation main energy optimization algorithm is involved, which is to establish a full waveform inversion parameter model; use conventional pre-stacked single shot data as observation data to establish a forward and inversion observation system; place the source at the surface shot point, and use the optimized finite difference algorithm to calculate the forward wave field generated by the source; determine the maximum value of the source wave field at each point underground and its corresponding time during the forward propagation of the source wave field, and save the wave field of each point underground within the excitation main energy time window; then use the optimized finite difference simulation algorithm to calculate the wave field residual reverse time propagation wave field, judge whether the wave field residual reverse time propagation wave field at different times of each point underground is within the excitation main energy time window of the point, and obtain the full waveform inversion gradient operator based on the excitation main energy optimization algorithm. This invention greatly reduces I / O operations and improves calculation efficiency while ensuring accuracy.

[0008] The above existing technologies are all significantly different from the present invention and fail to solve the technical problem we want to solve. For this reason, we have invented a new multi-scale full waveform inversion method for node ultra-large offset data. Summary of the invention

[0009] The purpose of the present invention is to provide a multi-scale full-waveform inversion method for node ultra-large offset data, which mainly solves the frequency skipping problem caused by the lack of low-frequency data in the full-waveform inversion method, can avoid the inversion algorithm from falling into local extreme values, and obtain accurate and reliable inversion results.

[0010] The object of the present invention can be achieved by the following technical measures: a multi-scale full-waveform inversion method for node ultra-large offset data, the multi-scale full-waveform inversion method for node ultra-large offset data comprising:

[0011] Step 1: forward modeling the large-scale theoretical model to obtain the seismic response;

[0012] Step 2, design a reasonable multi-scale inversion frequency band;

[0013] Step 3, construct and add an objective function to improve the accuracy of wavelet inversion;

[0014] Step 4, construct a robust waveform inversion process;

[0015] Step 5, calculate and obtain the time window function corresponding to each prediction data;

[0016] Step 6: Calculate pre-stack and post-stack migration imaging gathers.

[0017] The purpose of the present invention can also be achieved by the following technical measures:

[0018] In step 1, the seismic response characteristics and spectral features of small offset and large offset are analyzed through forward simulation of the large-scale model.

[0019] In step 2, the seismic response is divided into components of different frequencies to provide a data basis for multi-scale full waveform inversion.

[0020] In step 2, different frequency bands of large offset seismic data are analyzed to obtain the lowest effective seismic data frequency and design several reasonable multi-scale inversion frequency bands.

[0021] The multi-scale full-waveform inversion method for ultra-large offset data of the node also includes, after step 2, using a diagonal Hessian matrix preconditioned gradient algorithm combined with a smoothing filter technique to optimize the conventional gradient. The Hessian operator contains curvature information that can clearly predict the caustics phenomenon in the gradient operator and artifacts caused by double scattering.

[0022] After step 2, the frequency domain wavelet estimation method is adopted and applied to the full waveform inversion process; since the wavelet inversion in the frequency domain can be expressed as a linear optimization problem, an initial wavelet is first given, and the seismic wave field is obtained by forward modeling in the time domain through the finite difference method, and then compared and analyzed with the observed data in the frequency domain, and the predicted updated seismic wavelet is inverted.

[0023] In step 3, in order to alleviate the ill-posedness and constrained solution characteristics of the inversion, an objective function with Tikhonov regularization constraint term is constructed to improve the accuracy of wavelet inversion.

[0024] In step 4, a multi-scale frequency division inversion strategy is used to achieve step-by-step inversion of low wavenumber, medium wavenumber, and high wavenumber, and to construct a robust waveform inversion process.

[0025] In step 4, the large-scale velocity structure is first inverted using low-frequency wave field information, and then the frequency components are gradually increased, and the velocity model obtained by the low-frequency information is used as the initial model for the next high-frequency inversion to invert a more detailed velocity structure.

[0026] In step 5, a reasonable data waveform fitting window is designed to prevent the full waveform inversion from falling into a local extreme value and causing a cycle jump phenomenon.

[0027] In step 5, the time shift between each prediction data and the corresponding observation data is calculated, and each prediction data is moved toward the corresponding observation data to generate intermediate transition data, thereby obtaining a time window function corresponding to each prediction data.

[0028] In step 6, the pre-stack and post-stack migration imaging gathers are calculated using the reflected waves based on the Gaussian beam migration technique.

[0029] In step 6, the reflected waves are migrated using the Gaussian beam migration imaging method to obtain pre-stack migration imaging gathers and stacked gathers.

[0030] The multi-scale full-waveform inversion method for node ultra-large offset data also includes, after step 6, determining the correctness of the velocity model.

[0031] When judging the correctness of the velocity model, the gather inclination prediction technology is used to determine whether the gather is flattened, and the correctness of the velocity model is judged to ensure that the model is updated in the correct direction.

[0032] The present invention mainly solves the problem that the inversion results are easily trapped in local extreme values ​​due to the lack of low-frequency components in multi-scale full waveform inversion in exploration geophysics. The present invention designs a stable and practical multi-scale full waveform inversion implementation method for node ultra-large offset data to solve its current application technical difficulties. The seismic records collected at large offsets of all nodes contain obvious first arrival waves (deep refraction waves, rotation waves), which can improve the technical difficulty that the multi-scale full waveform inversion algorithm cannot stably obtain inversion results due to the lack of low-frequency data by updating the medium and low wavenumber components of the deep layer in the velocity model, and finally obtain a high-precision velocity model. The first arrival wave information in the large offset data provides technical support for the successful application of multi-scale full waveform inversion. In the case of the lack of low-frequency components in the seismic data, this invention patent uses the first arrival wave information of the large offset data to automatically compensate for the low wavenumber information in the early inversion results of the multi-scale full waveform, so that the multi-scale full waveform inversion is successfully implemented. For land node acquisition, ultra-large offset data is easy to obtain. This method reduces the requirements of multi-scale full waveform inversion for low-frequency data in field acquisition data while obtaining high-precision inversion results, greatly reducing costs and improving benefits. The present invention has advantages that other methods do not have, and its specific advantages and characteristics are reflected in the following aspects:

[0033] 1. Reliability of the method. This method uses the rich first arrival wave information in the large offset data to invert the low-wavenumber velocity structure in the deep strata, providing a good initial velocity model for subsequent inversion. The subsequent large offset node acquisition and multi-scale full waveform inversion joint strategy can obtain a stable and high-precision velocity model.

[0034] Second, the method is cost-effective. For land node acquisition, ultra-large offset data is easy to obtain. This method can achieve accurate inversion of velocity fields in shallow, medium and deep layers and low, medium and high wavenumber segments while significantly reducing costs and improving benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A schematic diagram of a real velocity field of a Marmous i model in a specific embodiment of the present invention;

[0036] Figure 2 A schematic diagram of the initial velocity field of the Marmous i model input for multi-scale full waveform inversion in a specific embodiment of the present invention;

[0037] Figure 3 It is a schematic diagram of forward shot records of three frequency bands of the Marmous i small offset model in a specific embodiment of the present invention;

[0038] Figure 4 A schematic diagram of the multi-scale full-waveform velocity inversion results of the Marmous i small offset model obtained using the patent of the present invention;

[0039] Figure 5 It is a schematic diagram of forward shot records of three frequency bands of the offset model in Marmous i in a specific embodiment of the present invention;

[0040] Figure 6 A schematic diagram of the multi-scale full-waveform velocity inversion results of the Marmous i medium offset model obtained using the patent of the present invention;

[0041] Figure 7 It is a schematic diagram of forward shot records of three frequency bands of the Marmous i large offset model in a specific embodiment of the present invention;

[0042] Figure 8 A schematic diagram of the multi-scale full-waveform velocity inversion results of the Marmous i large offset model obtained using the patent of the present invention;

[0043] Fig. 9 It is a schematic diagram of 5 Hz seismic forward modeling data in a specific embodiment of the present invention;

[0044] Fig.10 It is a schematic diagram of 7 Hz seismic forward modeling data in a specific embodiment of the present invention;

[0045] Fig.11 is a schematic diagram of an initial velocity model in a specific embodiment of the present invention;

[0046] Fig.12 It is a schematic diagram of the full waveform inversion result of 5 Hz frequency band in a specific embodiment of the present invention;

[0047] Fig.13 It is a schematic diagram of the full waveform inversion result of 7 Hz frequency band in a specific embodiment of the present invention;

[0048] Fig.14 It is a schematic diagram of the least square migration imaging result of the velocity imaging seismic section in a specific embodiment of the present invention;

[0049] Fig.15A schematic diagram of a least squares migration imaging result of DACIG in a specific embodiment of the present invention;

[0050] Fig.16 The present invention is a flowchart of a specific embodiment of the multi-scale full-waveform inversion method for node ultra-large offset data. DETAILED DESCRIPTION

[0051] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0052] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations and / or combinations thereof.

[0053] In order to solve the problem of missing low-frequency information faced by full waveform inversion, the present invention first uses the rich first-arrival wave information in the large offset data to invert the low-wavenumber velocity structure in the deep strata, providing a good initial velocity model for subsequent inversion. Then, the "step-by-step inversion" strategy of the multi-scale inversion idea is borrowed, and finally the large offset node acquisition and multi-scale inversion joint strategy are used for inversion to obtain a stable and high-precision velocity model. This invention patent uses the first-arrival wave and reflection wave information at the same time, which is of great significance for the full-band inversion of low, medium and high wavenumbers. Figure 1 As shown, Figure 1 The flowchart of the multi-scale full waveform inversion method for node ultra-large offset data of the present invention is as follows. The multi-scale full waveform inversion method for node ultra-large offset data includes the following steps:

[0054] 101) Forward simulation of large-scale theoretical models was performed to obtain seismic responses.

[0055] 102) Divide the seismic response into components of different frequencies to provide a data basis for multi-scale full waveform inversion.

[0056] 103) In order to alleviate the ill-posedness and constrained solution characteristics of the inversion, an objective function with Tikhonov regularization constraint term is constructed to improve the accuracy of wavelet inversion.

[0057] 104) First, use the low-frequency wave field information to invert the large-scale velocity structure, then gradually increase the frequency component, and use the velocity model obtained from the low-frequency information as the initial model for the next high-frequency inversion to invert a more detailed velocity structure.

[0058] 105) Design a reasonable data waveform fitting window to avoid the full waveform inversion falling into local extreme values ​​and causing frequency skipping.

[0059] 106) The pre-stack and post-stack migration imaging gathers are calculated using the Gaussian beam migration technique using reflected waves.

[0060] 107) Determine the correctness of the velocity model.

[0061] The following are several specific embodiments of the present invention.

[0062] Example 1

[0063] In a specific embodiment 1 of the present invention, the multi-scale full waveform inversion method for node ultra-large offset data includes the following steps:

[0064] 1) Through forward simulation of large-scale models, the seismic response characteristics and spectral features of small offset and large offset are analyzed.

[0065] 2) Analyze different frequency bands of large-offset seismic data, obtain the lowest effective seismic data frequency, and design several reasonable multi-scale inversion frequency bands.

[0066] 3) The conventional gradient is optimized by using the diagonal Hessian matrix preconditioned gradient algorithm combined with smoothing filtering technology. The Hessian operator contains curvature information that can clearly predict the caustics phenomenon in the gradient operator and the artifacts caused by double scattering. It effectively avoids the problem that the gradient operator does not scale with increasing depth due to the limited frequency band of seismic records, non-uniform coverage of the wave field, and double scattering in the traditional gradient calculation formula, and effectively improves the depth and accuracy of deep velocity updates.

[0067] 4) A frequency domain wavelet estimation method is used and applied to the full waveform inversion process. Since the wavelet inversion in the frequency domain can be expressed as a linear optimization problem, we first give an initial wavelet, obtain the seismic wave field by finite difference method in the time domain, and compare and analyze it with the observed data in the frequency domain to invert the predicted updated seismic wavelet.

[0068] 5) In order to alleviate the inversion ill-posedness and constrained solution characteristics, the Tikhonov regularization term is added to the objective function to constrain the inversion process to improve the accuracy of wavelet inversion.

[0069] 6) Use a multi-scale frequency division inversion strategy to achieve step-by-step inversion of low wavenumber, medium wavenumber and high wavenumber, and build a robust waveform inversion process.

[0070] 7) Calculate the time shift between each predicted data and the corresponding observed data, move each predicted data to the corresponding observed data, generate intermediate transition data, and obtain the time window function corresponding to each predicted data. By reasonably designing the data waveform fitting window, the cycle jump problem of full waveform inversion can be avoided and local extreme values ​​can be prevented.

[0071] 8) The Gaussian beam migration imaging method is used to migrate the reflected waves to obtain pre-stack migration imaging gathers and stacked gathers.

[0072] 9) Use gather inclination prediction technology to determine whether the gather is flattened, determine the correctness of the velocity model, and ensure that the model is updated in the correct direction.

[0073] Example 2

[0074] The method provided by the present invention is applied to the international standard Marmous i model velocity inversion modeling, and good inversion results are obtained. Figure 1 As shown, the initial model of full waveform inversion input is as follows Figure 2 shown. Figure 3 The forward shot records of 3Hz, 5Hz and 7Hz frequency bands of the 5km small offset model selected from the left side of Marmous i are shown in Figure 1. Figure 4 The inversion results of different frequency bands of the small offset model show that the basic structure of the model has been accurately restored. However, careful observation reveals that the restored structure is basically the reflection layer of the high wave number part. Figure 5 The forward shot records of the 3Hz, 5Hz and 7Hz frequency bands of the 10km mid-offset model selected from the left side of Marmous i. It can be seen that the shot records already contain obvious first arrival waves. This part of the waveform is of great significance for recovering the medium and low wavenumber velocity structure, especially the deep part. Figure 6 These are the inversion results of data in different frequency bands of the medium-offset model. It can be seen that the blocky structure with medium wavenumber in the shallow layer can be basically restored, but the blocky structure with low wavenumber in the deep layer is not accurately restored. Figure 7 The forward shot records of the 3Hz, 5Hz and 7Hz frequency bands of the 18km large offset model selected from the left side of Marmous i. It can be seen that the first arrival wave in the shot record is more obvious, the waveform earlier than the direct wave is richer, and it carries more deep velocity information. Figure 8 The inversion results of different frequency bands of the large offset model show that the main structures are well characterized by using 3Hz seismic data, and the high-wave number details are gradually restored when the high-frequency energy is increased.

[0075] Example 3

[0076] The method provided by the present invention is applied to the actual data velocity inversion modeling, and good inversion results are obtained. The initial model of the full waveform inversion input is as follows Fig.11 shown. Figure 9-10 Forward shot records and actual observation data in the 5Hz and 7Hz frequency bands. The first arrival waves in the shot records are obvious, and the rotation waves have rich waveform characteristics and carry deep velocity information.

[0077] Fig. 9 The lower picture is the 5Hz actual observation gun record; the upper picture is the 5Hz forward simulation gun record.

[0078] Fig.10 The lower picture is the 7Hz actual observation gun record; the upper picture is the 7Hz forward simulation gun record.

[0079] Figure 12-13 The inversion results of data in different frequency bands show that the overall outline of the depression structure is well portrayed in the inversion results of the 0-5Hz frequency band, and the medium wave number information is gradually restored. A low-velocity abnormal layer appears in the middle of the depression, which may be a favorable reservoir. Full waveform inversion has achieved good application results in the actual work area and can provide a high-precision velocity model for subsequent precise seismic reservoir interpretation. The high-frequency components of the shallow layer are significantly improved in the 0-7Hz inversion results, and the frequency jump problem is effectively avoided.

[0080] First, we use the tomographic velocity model and the velocity model updated by full waveform inversion as the migration velocity model to carry out least squares Gaussian beam migration imaging, and then obtain the pre-stack migration imaging gathers. The gather inclination prediction technology is used to determine whether the gathers are flattened and the correctness of the velocity model is determined. From the migration imaging sections corresponding to the two velocities, it can be seen that the continuity and energy of the reflection interface in the imaging section after the full waveform inversion velocity update are enhanced, the imaging noise is reduced, and the seismic imaging resolution is significantly improved. Fig.14 shown. Fig.14 In the figure below, the tomographic velocity imaging seismic profile is shown; the upper figure is the full waveform inversion velocity imaging seismic profile. And after the velocity update, the curved event axis in the dip domain gather is flattened to a certain extent, such as Fig.15 As shown, the upper figure is the DACIG based on the tomographic velocity model, and the lower figure is the DACIG based on the full waveform inversion result, which indirectly proves the accuracy of the full waveform inversion velocity model update.

[0081] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions recorded in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0082] Except for the technical features described in the specification, all other technical features are known technologies to those skilled in the art.

Claims

1. A multi-scale full waveform inversion method for node ultra-large offset data, characterized in that: The multi-scale full waveform inversion method for the node ultra-large offset data includes: Step 1: forward modeling the large-scale theoretical model to obtain the seismic response; Step 2, design a reasonable multi-scale inversion frequency band; Step 3, construct and add an objective function to improve the accuracy of wavelet inversion; Step 4, construct a robust waveform inversion process; Step 5, calculate and obtain the time window function corresponding to each prediction data; Step 6: Calculate pre-stack and post-stack migration imaging gathers.

2. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 1 is characterized in that: In step 1, the seismic response characteristics and spectral features of small offset and large offset are analyzed through forward simulation of the large-scale model.

3. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 1 is characterized in that: In step 2, the seismic response is divided into components of different frequencies to provide a data basis for multi-scale full waveform inversion.

4. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 3 is characterized in that: In step 2, different frequency bands of large offset seismic data are analyzed to obtain the lowest effective seismic data frequency and design several reasonable multi-scale inversion frequency bands.

5. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 1 is characterized in that: The multi-scale full-waveform inversion method for ultra-large offset data of the node also includes, after step 2, using a diagonal Hessian matrix preconditioned gradient algorithm combined with a smoothing filter technique to optimize the conventional gradient. The Hessian operator contains curvature information that can clearly predict the caustics phenomenon in the gradient operator and artifacts caused by double scattering.

6. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 1, characterized in that: The multi-scale full-waveform inversion method for ultra-large offset data of the node also includes, after step 2, using a frequency domain wavelet estimation method and applying it to the full waveform inversion process; since the wavelet inversion in the frequency domain can be expressed as a linear optimization problem, an initial wavelet is first given, and the seismic wave field is obtained by forward modeling in the time domain through the finite difference method, and compared and analyzed with the observed data in the frequency domain, and the predicted updated seismic wavelet is inverted.

7. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 1 is characterized in that: In step 3, in order to alleviate the ill-posedness and constrained solution characteristics of the inversion, an objective function with Tikhonov regularization constraint term is constructed to improve the accuracy of wavelet inversion.

8. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 1, characterized in that: In step 4, a multi-scale frequency division inversion strategy is used to achieve step-by-step inversion of low wavenumber, medium wavenumber, and high wavenumber, and to construct a robust waveform inversion process.

9. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 8, characterized in that: In step 4, the large-scale velocity structure is first inverted using low-frequency wave field information, and then the frequency components are gradually increased, and the velocity model obtained by the low-frequency information is used as the initial model for the next high-frequency inversion to invert a more detailed velocity structure.

10. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 1, characterized in that: In step 5, a reasonable data waveform fitting window is designed to prevent the full waveform inversion from falling into a local extreme value and causing a cycle jump phenomenon.

11. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 10, characterized in that: In step 5, the time shift between each prediction data and the corresponding observation data is calculated, and each prediction data is moved toward the corresponding observation data to generate intermediate transition data, thereby obtaining a time window function corresponding to each prediction data.

12. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 1, characterized in that: In step 6, the pre-stack and post-stack migration imaging gathers are calculated using the reflected waves based on the Gaussian beam migration technique.

13. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 12, characterized in that: In step 6, the reflected waves are migrated using the Gaussian beam migration imaging method to obtain pre-stack migration imaging gathers and stacked gathers.

14. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 1, characterized in that: The multi-scale full-waveform inversion method for node ultra-large offset data also includes, after step 6, determining the correctness of the velocity model.

15. The multi-scale full waveform inversion method for node ultra-long offset data according to claim 14, characterized in that: When judging the correctness of the velocity model, the gather inclination prediction technology is used to determine whether the gather is flattened, and the correctness of the velocity model is judged to ensure that the model is updated in the correct direction.

Citation Information

Cited By

  • Submersible wave tomography static correction method based on low wave number constraint

    CN117420602A