Methods and systems for seismic imaging using s-wave velocity models and machine learning

Abstract

A method can include obtaining a P-wave velocity model (476) and velocity ratio data (411) for a geological region of interest. The method can also include generating an initial S-wave velocity model (420) for the geological region of interest based on the P-wave velocity model (476) and the velocity ratio data (411). The method can also include determining a plurality of velocity boundaries within the initial S-wave velocity model (420) using a trained model. The method can also include updating the initial S-wave velocity model (420) using the velocity boundaries, automatically selected cross-correlation lag values based on a plurality of seismic gather, and migration velocity analysis to produce an updated S-wave velocity model (470). The method also includes generating a combined velocity model for the geological region of interest using the updated S-wave velocity model (470) and the P-wave velocity model (476).

Description

BACKGROUND

[0001] By performing a full wavefield inversion, a migration algorithm can convert time-based seismic data into a depth representation of the subsurface. In particular, a migration algorithm can use a velocity model representing different velocity values of matter within the subsurface to determine image data from data in the data domain. However, migration algorithms can be computationally intensive processes due to the number of computations required. SUMMARY

[0002] This Summary is provided to introduce a selection of concepts that are further described below in the. This Summary is neither intended to identify key or essential features of the claimed subject matter nor is it used to determine the scope of the claimed subject matter.

[0003] Generally, in one aspect, embodiments relate to a method including obtaining, by a computer processor, P-wave velocity model and velocity ratio data for a geological region of interest. The method also includes generating, by the computer processor and based on the P-wave velocity model and the velocity ratio data, an initial S-wave velocity model for the geological region of interest. The method also includes determining, by the computer processor, using a trained model, a plurality of velocity boundaries within the initial S-wave velocity model. The method also includes updating, by the computer processor, using the velocity boundaries, automatically selected cross-correlation lag values based on a plurality of seismic migration gathers, and migration velocity analysis, the initial S-wave velocity model to produce an updated S-wave velocity model. The method also includes generating, by the computer processor, using the updated S-wave velocity model and the P-wave velocity model, a combined velocity model for the geological region of interest.

[0004] Generally, in one aspect, embodiments relate to a non-transitory computer readable medium storing instructions executable by a computer processor. The instructions obtain P-wave velocity model and velocity ratio data for a geological region of interest. The instructions also generate, based on the P-wave velocity model and the velocity ratio data, an initial S-wave velocity model for the geological region of interest. The instructions also determine, using a trained model, a plurality of velocity boundaries within the initial S-wave velocity model. The instructions also update, using the velocity boundaries, automatically selected cross-correlation lag values based on a plurality of seismic migration gathers, and migration velocity analysis, the initial S-wave velocity model to produce an updated S-wave velocity model. The instructions also generate, using the updated S-wave velocity model and the P-wave velocity model, a combined velocity model for the geological region of interest.

[0005] In general, in one aspect, embodiments relate to a system including a seismic survey system including a seismic source and a plurality of seismic receivers. The system also includes a seismic interpreter including a computer processor. The seismic interpreter is coupled to the seismic survey system. The seismic interpreter obtains a P-wave velocity model and velocity ratio data for a geological region of interest. The seismic interpreter generates an initial S-wave velocity model for the geological region of interest based on the P-wave velocity model and the velocity ratio data. The seismic interpreter determines a plurality of velocity boundaries within the initial S-wave velocity model using a trained model. The seismic interpreter updates the initial S-wave velocity model using the velocity boundaries, automatically selected cross-correlation lag values based on a plurality of seismic common-image gathers, and migration velocity analysis to produce an updated S-wave velocity model. The seismic interpreter generates a combined velocity model for the geological region of interest using the updated S-wave velocity model and the P-wave velocity model.

[0006] Other aspects of the disclosure will become apparent from the following description and appended claims. BRIEF DESCRIPTION OF DRAWINGS

[0007] Specific embodiments of the disclosed technology will now be described in detail with reference to the figures. Like reference numbers are used to indicate like elements in the figures.

[0008] Figure 1 and Figure 2 A system is shown in accordance with one or more embodiments.

[0009] Figure 3 A flowchart is shown in accordance with one or more embodiments.

[0010] Figure 4 , Figure 5 , Figure 6 and Figure 7 An example is shown in accordance with one or more embodiments.

[0011] Figure 8 An example is shown in accordance with one or more embodiments.

[0012] Figure 9 A computing system is shown in accordance with one or more embodiments. DETAILED DESCRIPTION

[0013] In the following detailed description of embodiments of the disclosure, numerous specific details are set forth in order to provide a more thorough understanding of the disclosure. However, it will be apparent to one of ordinary skill in the art having the benefit of this disclosure that the disclosure can be practiced without these specific details. In other instances, well-known features have not been described in detail to avoid unnecessarily complicating the description.

[0014] Throughout the application, ordinal numbers (e.g., first, second, third, etc.) may be used as adjectives for elements (i.e., any noun in this application). Unless explicitly disclosed, such as by using the terms “before,” “after,” “single,” and other such terms, the use of ordinal numbers does not imply or create any particular order of elements, nor does it limit any element to a single element. Rather, the use of ordinal numbers is intended to distinguish between elements. As an example, a first element is distinct from a second element, and a first element may contain more than one element and be placed after (or before) the second element in the order of elements.

[0015] In general, embodiments of this disclosure include systems and methods for automating the updating of S-wave velocity models. Where the P-wave corresponds to the first arrival seismic wave in seismic surveys, the S-wave may correspond to a secondary wave or shear wave following the P-wave. In particular, some embodiments include obtaining a set of migration gathers to identify optimal cross-correlation lag values ​​for the final S-wave velocity model. For example, this set of migration gathers can be performed using migration velocity analysis such as reverse time migration. Using the selected cross-correlation time lag values, the S-wave velocity boundaries within the initial S-wave velocity model can then be predicted using a trained model. The initial S-wave velocity model can be determined based on a pre-existing P-wave velocity model and velocity ratio data. Thus, by focusing on building the S-wave velocity model through machine learning, computational costs can be reduced using time-migrated gathers while also addressing errors in the final S-wave velocity model.

[0016] Some embodiments may include a trained model, which could be a machine learning model for determining velocity boundaries within a geological region of interest (e.g., a specific subsurface stratum). This trained model can use migrated depth images based on seismic data to classify S-wave velocity boundaries for fault photography updates. Seismic imaging of complex basins (such as the Red Sea) is challenging and may require supplementary information from converted S-waves. Where P-wave imaging is frequently used for hydrocarbon prospecting, some embodiments use combined imaging (e.g., images with corresponding P-wave and S-wave values) to obtain this supplementary information. Combined imaging techniques can avoid various problems encountered by some migration algorithms when the P-wave velocity model is complex (e.g., with sharp boundaries). Furthermore, combined depth images can highlight information that is obscured when analyzing P-wave images individually.

[0017] Turning Figure 1 , Figure 1 A schematic diagram according to one or more embodiments is shown. Figure 1 As shown, Figure 1The diagram illustrates a seismic survey system 100 and multiple formation paths of pressure waves (also referred to as seismic waves). The seismic survey system 100 includes a source 122 that includes functionality for generating pressure waves that penetrate a subsurface layer 124, such as a reflected wave 136, a latent wave A142, or a latent wave B146. The pressure waves generated by the source 122 can penetrate the subsurface layer 124 along several paths according to a particle velocity V1 for detection at multiple seismic receivers 126 along a profile line. Similarly, particle velocity can refer to various velocity types, such as the two types of particle motion caused by seismic waves: the velocity of the first arrival wave (P-wave) and the different velocities of the second arrival wave (S-wave) through a specific medium. The source 122 can be a seismic vibrator, such as a seismic vibrator using controlled-source technology, an air gun in marine seismic surveys, explosives, etc. The seismic receivers 126 can include detectors, hydrophones, accelerometers, and other sensing devices. Similarly, the seismic receiver 126 may include single-component and / or multi-component sensors for measuring pressure waves on multiple spatial axes.

[0018] like Figure 1 As shown, the seismic source 122 generates an air wave 128, formed from a portion of the emitted seismic energy, which travels over the Earth's surface 130 to the seismic receiver 126. The seismic source 122 can also emit surface waves 132 traveling along the Earth's surface 130. The velocity of the surface waves 132 (also known as Rayleigh waves or roll waves) can correspond to a particle velocity that is generally slower than that of secondary waves. Although Figure 1 The seismic survey shown is a two-dimensional survey along a seismic profile in the longitudinal direction, but other embodiments, such as three-dimensional surveys, are also conceivable.

[0019] Furthermore, subsurface layer 124 has a particle velocity V1, while subsurface layer 140 has a particle velocity V2. In other words, different subsurface layers can correspond to different particle velocity values. Specifically, particle velocity can refer to the speed at which a pressure wave travels through a medium; for example, a latent wave B 146 travels through subsurface layer 124 forming a curved ray path 148. Particle velocity may depend on the density and elasticity of a particular medium, as well as various wave properties, such as the frequency at which the pressure wave is emitted. When the particle velocities differ between two subsurface layers, this seismic wave impedance mismatch can lead to seismic reflections of the pressure wave. For example, Figure 1 The diagram illustrates a pressure wave propagating downwards from the source 122 to the subsurface interface 138, which, in response to seismic reflection, becomes an upward-propagating reflected wave 136. The source 122 can also generate a direct wave 144 that travels directly from the source 122 through the subsurface layer 124 to the seismic receiver 126 at a particle velocity V1.

[0020] The pressure wave is redirected and refracted. Source 122 can also generate a refracted wave (i.e., a latent wave A 142), which is refracted at the subsurface interface 138 and travels a distance along the interface 138 (e.g., ...). Figure 1 As shown), the refracted pressure wave travels upwards to the seismic receiver 126. Thus, the refracted pressure wave may include latent waves (e.g., latent wave A 142, latent wave B 146) that can be analyzed to map subsurface layers 124, 140. For example, a latent wave can be a refracted wave that continuously refracts throughout the Earth's subsurface. Therefore, latent waves can be generated where the particle velocity gradually increases with depth. Similarly, unlike reflected seismic energy, the vertex of a latent wave can be offset from the common midpoint (CMP). However, for analytical purposes, the vertex of the latent wave can be considered as the common midpoint of the refracted energy. Thus, the vertex can be used as a basis for organizing and classifying seismic survey datasets.

[0021] Furthermore, when analyzing seismic data acquired using the seismic survey system 100, rays can be used to approximate seismic wave propagation. For example, reflected waves (e.g., reflected wave 136) and latent waves (e.g., latent waves 142, 146) can be scattered at the subsurface interface 138. For example, in Figure 1 In this context, the latent wave B 146 can present a wide-angle ray path similar to a reflected wave, enabling the mapping of the subsurface. For example, using latent waves, a velocity model for the underlying subsurface can be generated, describing the particle velocities in different regions within different subsurface layers. An initial velocity model can be generated by simulating the velocity structure of the subsurface medium using seismic data inversion (often referred to as seismic inversion). In seismic inversion, the velocity model is iteratively updated until the velocity model and the seismic data have a minimum mismatch; for example, the solution of the velocity model converges to a globally optimal value that meets predetermined criteria.

[0022] Regarding velocity models, they can map various subsurface layers based on particle velocities in different sub-regions (e.g., P-wave velocities, S-wave velocities, and various anisotropic effects within the sub-regions). For example, a velocity model can be used, along with the arrival times and directions of P-waves and S-waves, to locate seismic events. Anisotropic effects can correspond to subsurface properties that cause pressure waves to exhibit direction dependence. Thus, seismic anisotropy can correspond to multiple parameters in geophysics involving variations in wave velocity based on propagation direction. One or more anisotropic algorithms can be performed to determine anisotropic effects, such as anisotropic ray tracing localization algorithms or algorithms using inclined shaft sonic logging, vertical seismic profiles (VSPs), and core measurements. Similarly, a velocity model can include multiple velocity boundaries that define areas of rock type variation, such as interfaces between different subsurface layers. In some embodiments, the velocity model is updated using one or more fault photography updates to adjust the velocity boundaries within the velocity model.

[0023] Turning Figure 2 , Figure 2 A system according to one or more embodiments is shown. Figure 2 As shown, a seismic volume 290 comprises multiple seismic traces (e.g., seismic trace 250) acquired by multiple seismic receivers (e.g., seismic receiver 226) positioned on the Earth's surface 230. More specifically, the seismic volume 290 can be a three-dimensional cubic dataset of seismic traces. Individual cubic units within the seismic volume 290 may be referred to as voxels or volume pixels (e.g., voxel 260). In particular, different portions of the seismic traces may correspond to various depth points within the Earth's volume. To generate the seismic volume 290, a three-dimensional array of seismic receivers 226 is positioned along the Earth's surface 230, and seismic data is acquired in response to multiple pressure waves emitted from a source. Within voxel 260, statistics can be calculated on the first arrival data assigned to a specific voxel to determine the multimodal distribution of wave travel times and derive travel time estimates associated with azimuth sectors (e.g., based on mean, median, mode, standard deviation, kurtosis, and other appropriate statistical accuracy metrics). First arrival data can describe the initial arrival of refracted or latent waves generated by a specific source signal at seismic receiver 226.

[0024] Seismic data can refer to raw time-domain data acquired from seismic surveys (e.g., the acquired seismic data may generate seismic bodies 290). However, seismic data can also refer to data acquired at different time intervals, such as in the case of repeated seismic surveys to obtain time-shifted data. Seismic data can also refer to various seismic properties derived in response to processing the acquired seismic data. Furthermore, in some contexts, seismic data can also refer to depth data or image data. Similarly, seismic data can also refer to processed data (e.g., using seismic inversion operations to generate velocity models of subsurface strata) or migrated seismic images of rock layers within the Earth's surface. Seismic data can also be preprocessed data, such as time-domain data arranged within a two-dimensional shot gather.

[0025] Furthermore, seismic data can include multiple spatial coordinates, such as (x, y) coordinates for each shot point and (x, y) coordinates for each receiver. Thus, seismic data can be grouped into common shot point gathers or common receiver point gathers. In some embodiments, seismic data is grouped based on a common domain, such as a common midpoint (i.e., Xmidpoint = (Xshot + Xrec) / 2, where Xshot corresponds to the shot point location and Xrec corresponds to the seismic receiver location) and a common shot-receiver offset (i.e., Xoffset = Xshot - Xrec).

[0026] In some embodiments, seismic data is processed to generate one or more seismic images. For example, a process called migration can be used to perform seismic imaging. In some embodiments, migration can transform a pre-processed shot gather from the data domain to the image domain corresponding to the depth data. In the data domain, seismic events in the shot gather can represent subsurface seismic events recorded in field surveys. In the image domain, seismic events in the migrated shot gather can represent subsurface geological interfaces. Similarly, various types of migration algorithms can be used for seismic imaging. For example, one type of migration algorithm corresponds to inverse time migration. In inverse time migration, seismic gathers can be analyzed by: 1) performing forward modeling of the seismic wavefield starting from a synthetic source wavelet and velocity model via mathematical modeling; 2) performing backpropagation of the seismic data via mathematical modeling using the same velocity model; 3) performing cross-correlation of the seismic wavefield based on the results of the forward modeling and backpropagation; and 4) applying imaging conditions during the cross-correlation to generate a seismic image at each time step. Under the basic assumption that the source wavefield represents the downlink wavefield and the receiver wavefield represents the uplink wavefield, the imaging conditions can be determined by estimating the cross-correlation between the source and receiver wavefields to determine how to form the actual image. For example, in Kirchhoff and beamforming methods, imaging conditions may include the sum of contributions from the input data channels after they have been partially expanded along the individual isochrones (e.g., using the principles of constructive and destructive interference to form an image).

[0027] Continuing with seismic imaging, it can be near the end of the seismic data workflow before the seismic interpreter performs analysis. The seismic interpreter can then derive an understanding of the subsurface geology from one or more final migrated images. To confirm whether a particular seismic data workflow accurately simulates the subsurface, normal time-of-motion (NMO) stacks can be generated, which consist of multiple NMO gathers with amplitudes sampled from a common centroid (CMP). In particular, NMO correction can be based on a seismic imaging approximation that calculates reflection travel times. However, in cases of complex subsurface geology with large inhomogeneities in particle velocities, or when seismic surveys are not acquired on a horizontal plane, NMO stacking results may fail to indicate accurate subsurface geology. Ocean-bottom-node surveys and seismic surveys on rough-topographic land are examples of how NMO stacking results may fail to describe subsurface geology.

[0028] Although Figure 2 The diagram generally shows seismic traces with zero source-receiver offset, but these traces can be stacked, migrated, and / or used to generate attribute volumes derived from the underlying traces. For example, an attribute volume could be a dataset of seismic volumes subjected to one or more processing techniques, such as amplitude-versus-offse (AVO) processing. In AVO processing, seismic data can be classified based on variations in reflection amplitude caused by the presence of hydrocarbon accumulation in the subsurface strata. Using AVO, the seismic properties of subsurface interfaces can be determined based on the dependence of detected seismic reflection amplitude on the incident angle of seismic energy. This AVO processing can determine the normal incidence coefficient and / or gradient components of seismic reflections. Similarly, seismic data can also be processed based on the vertices of pressure waves. In particular, vertices can be used as data gather points to classify the first arrival of seismic data records or traces into multiple source-receiver offset cells based on survey dimensional data (e.g., the xy position of seismic receiver 226 on the Earth's surface 230). These elements can include different numbers of channels and / or different coordinate dimensions.

[0029] Turning to seismic interpreter 261, seismic interpreter 261 may include hardware and / or software having functionality for storing seismic body 290, well logs, core sample data, and other data for seismic data processing, well data processing, training operations, and other corresponding data processes. In some embodiments, seismic interpreter 261 may include similar features to those described below. Figure 9The computer system 602 described in the corresponding description. While a seismic interpreter may refer to one or more computer systems used to perform seismic data processing, it may also refer to a human analyst who performs seismic data processing in conjunction with a computer. Although seismic interpreter 261 is shown at a seismic survey site, in some embodiments, seismic interpreter 261 may be located away from the seismic survey site.

[0030] Continuing with the discussion of seismic interpreter 261, seismic interpreter 261 may include hardware and / or software having the capability to generate one or more machine learning models 270 for analyzing seismic data and one or more subsurface strata. For example, seismic interpreter 261 may use and / or process seismic data and other types of data to generate and / or update one or more machine learning models 270 and / or one or more velocity models. Therefore, different types of machine learning models can be trained, such as convolutional neural networks, deep neural networks, recurrent neural networks, support vector machines, decision trees, inductive learning models, deductive learning models, supervised learning models, unsupervised learning models, reinforcement learning models, etc. Supervised learning algorithms may include linear regression algorithms, nearest neighbor algorithms, decision trees, etc. In some embodiments, two or more different types of machine learning models are integrated into a single machine learning architecture; for example, the machine learning model may include decision trees and neural networks. In some embodiments, seismic interpreter 261 may generate augmented or synthetic data to produce a large amount of interpretable data for training a particular model.

[0031] Regarding neural networks, for example, a neural network may include one or more hidden layers, where each hidden layer comprises one or more neurons. A neuron may be a modeling node or object that roughly mimics neurons in the human brain. Specifically, a neuron can combine data inputs with a set of coefficients (i.e., a set of network weights and biases used to adjust the data inputs). These network weights and biases can amplify or decrease the value of a particular data input, thereby assigning importance to multiple data inputs used for the task being modeled. Through machine learning, a neural network can determine which data inputs should receive higher priority when determining one or more specified outputs of the neural network. Similarly, these weighted data inputs can be summed such that the sum is passed through the activation function of the neuron to other hidden layers within the neural network. In this way, the activation function can determine whether and to what extent the output of a neuron progresses to other neurons, where the output can be weighted again for use as input to the next hidden layer.

[0032] In some embodiments, multiple types of machine learning algorithms (e.g., machine learning algorithm 271) can be used to train the model, such as the backpropagation algorithm. In the backpropagation algorithm, the gradient for each hidden layer of the neural network is computed backwards from the layer closest to the output layer to the layer closest to the input layer. Thus, the gradient can be computed using the transpose of the weights of the corresponding hidden layer based on an error function (also known as a "loss function"). The error function can be based on multiple criteria, such as a mean squared error function, a similarity function, etc., where the error function can be used as a feedback mechanism for adjusting the weights in the electronic model. An example of a backpropagation algorithm is the Levenberg-Marqardt algorithm. In some embodiments, multiple epochs are used to train the machine learning model. For example, an epoch can be an iteration of the model through a portion or all of the training dataset. Thus, a single machine learning epoch can correspond to a specific batch of training data, where the training data is divided into multiple batches for multiple epochs. Therefore, the machine learning model can be trained iteratively using epochs until the model reaches a predetermined level of prediction accuracy. Therefore, better training of the model can lead to better predictions by the trained model.

[0033] In some embodiments, a training dataset comprising manually picked data, augmented data, and / or synthetic data is used to train a machine learning model. For example, data augmentation may include performing multiple processes (e.g., logcropping or adding noise) on the acquired manually picked boundary data to generate augmented boundary data. In particular, data augmentation can introduce multiple machine learning algorithms into uncommon problems to increase the ability of a trained model to predict boundary data (e.g., velocity boundaries). Similarly, data augmentation can be performed to generate an expanded dataset sufficient to train the model. For example, a data augmentation process can alter a normal seismic dataset to produce a different or more complex seismic dataset. With such augmented boundary data, the artificial intelligence model can be made unaffected by multiple anomalies in the velocity boundary picking.

[0034] In some embodiments, the data augmentation process may include random modification, where the migrated seismic data is modified data from the original state to the modified state. Data augmentation may also include smoothing operations to remove data spikes within manually picked boundary data, such as by resampling or smoothing the data. In another embodiment, data augmentation may include intrusion operations, where random values ​​are added to manually picked boundary data in a specific area, depending on the type of augmentation factor. In another embodiment, data augmentation may include random noise operations that add varying amounts of noise to different depths of the migrated seismic image. Such added noise may increase the complexity of velocity boundary picking and thus increase boundary data. In another embodiment, data augmentation includes a cut operation that randomly removes velocity boundary data from the original data. For example, data augmentation may be performed to simulate migrated seismic data and velocity boundaries of different lithofacies, differences in layer thickness within strata, and other geological scenarios. Other data augmentation operations may include random zeroing operations that randomly assign zero values, random shift operations that shift the range of data values ​​within the data, masking data, or rotating the data. While some data augmentation operations are described as random, the data augmentation process may include pseudo-random processes tailored to specific criteria. In some embodiments, for example, data augmentation operations may be a function of a particular geologist's requirement to manipulate manually picked boundary data.

[0035] Turning Figure 3 , Figure 3 A flowchart according to one or more embodiments is shown. Specifically, Figure 3 A general method for generating a combined velocity model and / or generating seismic images using that combined velocity model is described. Figure 3 One or more boxes in the structure can be formed by, for example Figure 1 and Figure 2 This is performed by one or more components described herein (e.g., seismic interpreter 261). Although Figure 3 The boxes in the document are presented and described in sequence, but those skilled in the art will understand that some or all of these boxes may be executed in a different order, may be combined or omitted, and may be executed in parallel. Furthermore, these boxes may be executed actively or passively.

[0036] In block 300, according to one or more embodiments, velocity ratio data and / or seismic data about a geological area of ​​interest are obtained. In some embodiments, for example, the velocity ratio data describes the ratio of P-wave velocity to S-wave velocity (e.g., V0.05) at a specific location in the geological area of ​​interest. P / V S(Value). For example, the velocity ratio can depend on various geological properties, such as porosity, degree of consolidation, clay content, pressure differential, pore geometry, and other geological factors. Thus, for different rock types, the velocity ratio can correspond to a constant value or a function of values. Similarly, velocity ratio data can be determined using well logging samples, core samples, and / or seismic data. Seismic data can be similar to the above regarding... Figure 1 and Figure 2 The earthquake data described.

[0037] In block 310, a P-wave velocity model is obtained for a geological region of interest, according to one or more embodiments. For example, the P-wave velocity model may describe the particle velocities of P-waves at different locations within the subsurface. In some embodiments, the P-wave velocity model is known from previous seismic data processing. Similarly, the P-wave velocity model can be determined using one or more seismic inversion operations and / or one or more migration algorithms. In some embodiments, it is assumed that the background S-wave velocity has no effect on the source-side dynamics of the P-wave velocity model.

[0038] In box 320, according to one or more embodiments, an initial S-wave velocity model is generated for the geological region of interest and based on a P-wave velocity model and velocity ratio data. The initial S-wave velocity model (also known as a shear wave velocity model) can be generated by simulating the velocity structure of the subsurface medium using a form of seismic data inversion (commonly referred to as seismic inversion). After obtaining the P-wave velocity model via seismic inversion, for example, the P-wave velocity values ​​can be converted to S-wave velocity values ​​using velocity ratio data. During the seismic inversion operation, the velocity model can be iteratively improved until it is consistent with the seismic data and / or velocity ratio data, for example, the solution of the velocity model converges to a globally optimal value satisfying a certain criterion.

[0039] In box 330, according to one or more embodiments, multiple seismic migration gathers with different cross-correlation hysteresis values ​​are generated based on migration velocity analysis and an initial S-wave velocity model. For example, time-delayed migration gathers can be generated based on an initial S-wave velocity model (e.g., the initial S-wave velocity model from box 320). A gather can include a collection of input traces formed from seismic data according to some trace head parameters. The migration gather can be the result of migration velocity analysis applied to the traces within the gather. Before achieving a complete fault photogrammetric inversion of the initial S-wave velocity model, migration gathers can be generated, for example, using inverse time migration (RTM) or phase shift plus interpolation (PSPI) methods with multiple cross-correlation hysteresis values. Therefore, multiple volumetric stacked images can be generated within the migration gathers for different cross-correlation hysteresis values.

[0040] In some embodiments, cross-correlation imaging conditions are used to generate multiple non-zero hysteresis co-imaging point gathers (CIGs), which are then combined to form a co-imaging cube (CIC). Slicing the CIC at different cross-correlation hysteresis values ​​can generate a series of CIGs. Therefore, when an incorrect velocity model is used in the migration algorithm, flattening events may occur in the CIGs at cross-correlation hysteresis values ​​other than zero hysteresis.

[0041] In some embodiments, a smoothing function is used to smooth the offset gather. For example, the offset gather can be generated at each surface location and spatially smoothed along a local slope within a sliding (i.e., moving) window. The highest amplitude coherent energy can be selected on the smoothed gather. Due to the "smoothing," the highest amplitude can be coherent or continuous between adjacent locations within the corresponding gather. Smoothing can also reduce noise from the gather.

[0042] In box 340, according to one or more embodiments, multiple seismic migration gathers are used and cross-correlation hysteresis values ​​are automatically selected based on predetermined criteria. For example, the cross-correlation hysteresis values ​​may correspond to the correlation shift Δτ (i.e., tau value) in the cross-correlation equation. More specifically, the correlation shift Δτ can be used in multiple migration velocity analysis techniques to update the initial velocity model or generate migration gathers.

[0043] In some embodiments, for example, the seismic interpreter can automatically select cross-correlation hysteresis values ​​by analyzing the maximum stacked response across multiple migration sets. Specifically, the maximum stacked response can describe the highest degree of continuity and amplitude coherence energy between different migration sets based on different cross-correlation hysteresis values. Thus, the predetermined criterion can be the highest amplitude coherence energy within the migration set. Similarly, the predetermined criterion can also correspond to a predetermined degree of coherence energy within the migration set.

[0044] In some embodiments, the seismic interpreter analyzes the CIG to determine the cross-correlation hysteresis values ​​at the focal depth where the event in the CIG is flattest. Thus, the seismic interpreter can simulate the Green's function by seeding sources at the focal depth using one-way wave equation tomography. Specifically, the seismic interpreter can shift the modeled wavefield using the corresponding cross-correlation hysteresis values. This migration velocity analysis process can be repeated for events at different lateral and vertical locations in the CIG. The result of the migration velocity analysis can be a set of velocity modeling data whose wavefield approximates the wavefield of the most accurate velocity model from the actual subsurface that has been used to model these migration gathers.

[0045] In block 350, according to one or more embodiments, a trained model is used to determine multiple S-wave velocity boundaries within an initial S-wave velocity model. In some embodiments, for example, a machine learning model is trained based on a training dataset that includes manually picked boundary data and augmented boundary data. Therefore, one or more training operations can be performed to generate a trained model from the machine learning model.

[0046] In some embodiments, the seismic interpreter uses a trained model to determine velocity boundaries of one or more subsurface layers or phases. For example, the trained model can be a convolutional neural network (e.g., the U-net model), trained using machine learning algorithms. More specifically, the U-net model can have a deep neural network architecture that includes functions for classifying and / or segmenting images. In a deep neural network, layers of neurons can be trained on a predetermined list of features based on the output of previous network layers. Thus, as data progresses through the deep neural network, more complex features can be identified within the data by neurons in later layers. Similarly, U-net or other types of convolutional neural networks can include multiple convolutional layers, pooling layers, fully connected layers, and / or normalization layers to produce specific types of output. Therefore, convolutional and pooling functions can be activation functions within the convolutional neural network. For more information on the U-net model, see below. Figure 8 And the corresponding description.

[0047] After one or more training operations, the trained model can take a migrated S-wave image as input and produce S-wave velocity boundaries as output (e.g., in a velocity boundary map). These velocity boundaries can correspond to multiple layers of a specific facies within the subsurface. Therefore, the velocity boundaries can be used for fault photography updates of the initial S-wave velocity model. Thus, velocity boundaries can be selected from the optimal stacked images based on automatically selected cross-correlation hysteresis values. In some embodiments, for example, the top and bottom velocity boundaries of salt bodies within the formation are automatically determined. For example, the seismic interpreter can use an automated selection process to isolate salt bodies from non-salt bodies. Specifically, this automated selection process can use artificial intelligence to identify these salt velocity boundaries for subsalt imaging. In contrast, velocity boundary picking can be performed manually by geologists over days or months. With the automated selection process, the seismic interpreter can reduce the amount of time required to generate optimal seismic images and thus increase resources available for other areas. For example, by reducing the amount of time spent determining velocity boundaries, better seismic images can be obtained. These better seismic images may increase the chances of discovering oil and gas generated in complex environments, such as the Red Sea, which is considered one of the most complex geological basins in the world. Therefore, the automatic selection process can determine the boundaries of complex subsurface structures, thick salt deposits, and / or rugged seafloor topography (which may prove problematic in generating accurate subsurface seismic images).

[0048] In some embodiments, a trained model is validated by determining whether it predicts S-wave velocity boundaries at a predetermined level of accuracy. For example, a test dataset not used in the training operation can be used to validate the trained model. For example, if the validation falls below an accuracy threshold, the trained model can continue training until a satisfactory output (i.e., a satisfactory prediction of S-wave velocity boundaries) is obtained.

[0049] In block 360, according to one or more embodiments, an initial S-wave velocity model is updated using multiple velocity boundaries, automatically selected cross-correlation hysteresis values, and migration velocity analysis. For example, S-wave velocity boundaries from block 350 and automatically selected cross-correlation hysteresis values ​​can be input into migration velocity analysis to update the initial S-wave velocity model. In some embodiments, migration velocity analysis is based on wave equation traveltime tomography, which can update the velocity model in the presence of multiple velocity errors and complex environments. For example, in one or more embodiments, migration velocity analysis is one-way wave equation tomography. In other embodiments, other migration functions, such as ray tracing, or two-way wave equation tomography can be used to update the velocity model.

[0050] In some embodiments, the updated S-wave velocity model is validated. For example, the S-wave velocity model can be updated until the velocity values ​​converge to a predetermined criterion (e.g., the change between update iterations is below a predetermined threshold). Similarly, the seismic interpreter can use velocity ratio data, seismic data, and / or the P-wave model to determine whether the updated S-wave velocity model has achieved a global optimum.

[0051] In box 370, according to one or more embodiments, an updated S-wave velocity model and a P-wave velocity model are used to generate a combined velocity model for a geological region of interest. In some embodiments, the combined velocity model includes both a P-wave velocity model and an S-wave velocity model. For example, the combined velocity model may be a model of the Earth describing the geological region of interest. Where generating a combined velocity model using only conventional P-wave imaging techniques can be challenging, dual P-wave and S-wave imaging can produce an accurate combined velocity model. Therefore, seismic images based on the combined velocity model can provide more information that might be obscured from P-wave images.

[0052] In some embodiments, the combined velocity model is used for one or more shear wave analyses of a geological area of ​​interest. For example, site-specific shear wave velocities can be used to determine the classification of potential well sites (e.g., regarding structural support and different types of well plans). Therefore, the combined velocity model can be used for hydrocarbon exploration and well construction.

[0053] In box 380, according to one or more embodiments, a combined velocity model is used to generate a combined image of the geological region of interest. For example, the combined image may be a PS-image showing P-wave velocity boundaries and S-wave velocity boundaries, as well as other seismic attribute information. In some embodiments, the geological region of interest corresponds to an area of ​​subsurface strata that may be desired for further analysis (e.g., for anticipated drilling operations or reservoir modeling). Therefore, the combined image can provide spatial and depth illustrations of subsurface strata for various practical applications (e.g., predicting hydrocarbon deposition, predicting wellbore paths for geological guidance, etc.).

[0054] Regarding PS-images, they can provide supplementary information missing from conventional P-wave images of the geological area of ​​interest. Therefore, PS-images can improve image accuracy beneath gas-bearing areas, which is challenging with standalone P-images. Similarly, PS-images may offer better resolution for imaging near-surface structures. In some embodiments, PS-images are used to determine fracture density and fracture orientation analyses, for example, for well enhancement operations or hydraulic fracturing operations.

[0055] As mentioned above Figure 3As illustrated, some embodiments use machine learning, automated velocity picking, and velocity ratio data to construct subsurface S-wave velocity models for complex geological environments from acquired seismic survey data. From this S-wave velocity model, combined velocity models can be generated for exploration in complex geological areas (e.g., salt bodies). Therefore, various techniques can reduce the high drilling costs associated with drilling through such complex geological structures. Consequently, some embodiments provide an automated workflow to overcome the complexities associated with various geological environments used for hydrocarbon exploration and production.

[0056] Go to Figure 4 , Figure 5 , Figure 6 and Figure 7 , Figure 4 , Figure 5 , Figure 6 and Figure 7 Examples are provided for updating the S-wave velocity model by automatically selecting cross-correlation hysteresis values ​​and using the trained model to determine the S-wave velocity boundaries. The examples below are for illustrative purposes only and are not intended to limit the scope of the disclosed techniques.

[0057] exist Figure 4 In the process, the seismic interpreter obtains velocity ratio data A 411 and P-wave velocity data B 412. Using the S-wave inversion function 415, the velocity ratio data A 411 and P-wave velocity data B 412 are used to generate an initial S-wave velocity model E 420, which includes S-wave velocity data E 422. The seismic interpreter uses the migration gather generation function 421, which performs the RTM or PSPI method, to generate multiple migration gathers (i.e., migration gather A 432, migration gather B 433, and migration gather C 434). The seismic interpreter can use an automatic selection process (i.e., the cross-correlation hysteresis selection function 431) on migration gathers 432, 433, and 434 to determine the final cross-correlation hysteresis value 435. The final cross-correlation hysteresis value 435 can correspond to the cross-correlation hysteresis value that produces the highest coherent amplitude among migration gathers 432, 433, and 434, i.e., as a predetermined standard.

[0058] Go to Figure 5 The initial S-wave velocity model E 420 is updated using the offset velocity analysis function 461. Here, in the one-way wave equation offset velocity analysis function 461, the value from... Figure 4The final cross-correlation hysteresis value 435 and the S-wave velocity boundary B436 are used to perform one-way wave equation fault photography on the initial S-wave velocity model E420. The output of the migration velocity analysis function 461 is an updated S-wave velocity model E470, which includes updated S-wave velocity data G472. For example, the seismic interpreter can use an automatic velocity boundary selection function (not shown) with a trained model (not shown) to determine a velocity boundary map from one or more migration seismic images. In particular, the velocity boundary map may include multiple velocity boundaries corresponding to multiple salt bodies deposited in the subsurface region shown in the migration seismic images. These velocity boundaries can be used to generate the updated S-wave velocity model E470. Therefore, the seismic interpreter then uses a combined image generation function 475 with the updated S-wave velocity model E470 and the P-wave velocity model E476 to generate a PS-image B485.

[0059] Turning Figure 6 and Figure 7 The seismic interpreter uses velocity models 420, 470, and 476 as input to the combined image generation function 475. Figure 6 In the input velocity model X 491, P-wave velocity model E 476 and initial S-wave velocity model E420 are included. Using these velocity models 476 and 420, the seismic interpreter generates PS-image A 483. Figure 7 In the input velocity model Y 492, P-wave velocity model E 476 and the updated S-wave velocity model E 470 are included. Using these velocity models 476 and 470, the seismic interpreter generates PS-image B 485. Based on a comparison of PS-image A 483 and PS-image B 485, PS-image B 485 provides a better representation of the underlying subsurface strata.

[0060] Turning Figure 8 , Figure 8 An example of generating a U-Net model to predict S-wave velocity boundaries in subsurface strata is provided. The example below is for illustrative purposes only and is not intended to limit the scope of the disclosed techniques.

[0061] exist Figure 8 In this study, the supervised learning algorithm Q570 is used to train the U-Net model X551 to predict velocity boundaries within migrated seismic images (e.g., migrated S-wave image X550). Specifically, the U-Net model X551 includes five hidden layers: three convolutional layers (i.e., convolutional layer A581, convolutional layer C583, and convolutional layer E585), a downscaling layer B582, and an upscaling layer D584. Furthermore, the U-Net model X551 receives an image (i.e., migrated S-wave image X550) as input for both predicting S-wave velocity boundaries and training.

[0062] Continue to refer to Figure 8 The U-Net model X 551 comprises a contraction path (left) and an expansion path (right). In the contraction path, the U-Net model X 551 corresponds to a convolutional network architecture, where multiple modified linear units (not shown) and max-pooling operations (not shown) produce downsampled descriptions (e.g., feature maps) of predetermined features within the offset S-wave image X 550. This increases the number of feature channels during the contraction path. In the expansion path, the feature map is upsampled, which reduces the number of feature channels. At the final layer (i.e., convolutional layer E 585), the resulting feature map is associated with a desired number of categories describing the S-wave velocity boundaries within the S-wave velocity boundary map Y 556.

[0063] The embodiments can be implemented on a computer system. Figure 9 This is a block diagram of a computer system for providing computing functions associated with algorithms, methods, functions, procedures, flows, and programs as described in this disclosure. The computer 602 shown is intended to encompass any computing device, such as a high-performance computing (HPC) device, server, desktop computer, laptop / notebook computer, wireless data port, smartphone, personal data assistant (PDA), tablet computing device, one or more processors within these devices, or any other suitable processing device, including physical or virtual instances (or both) of the computing device. Additionally, computer 602 may include a computer comprising: an input device, such as a keypad, keyboard, touchscreen, or other device capable of accepting user information; and an output device that transmits information associated with the operation of computer 602, including digital data, visual or audio information (or a combination of information); or a GUI.

[0064] Computer 602 may function as a client, network component, server, database, or other persistent storage, or any other component (or combination of roles) in a computer system for performing the subjects described in this disclosure. The illustrated computer 602 is communicatively coupled to network 630 or the cloud. In some specific implementations, one or more components of computer 602 may be configured to operate within an environment including a cloud-based environment, a local environment, a global environment, or other environments (or combinations thereof).

[0065] At a higher level, computer 602 is an electronic computing device capable of receiving, transmitting, processing, storing, or managing data and information associated with the described subject. Depending on some specific implementations, computer 602 may also include, or be communicatively coupled to, application servers, email servers, web servers, cache servers, streaming media data servers, business intelligence (BI) servers, or other servers (or combinations thereof).

[0066] Computer 602 may receive requests via network 630 or cloud from client applications (e.g., executing on another computer 602) and respond to the requests by processing the received requests in a suitable software application. Additionally, requests may also be sent to computer 602 from internal users (e.g., from a command console or via other suitable access methods), external or third parties, other automated applications, and any other suitable entity, individual, system, or computer.

[0067] Each component of computer 602 can communicate using system bus 603. In some specific implementations, any or all components (hardware or software, or a combination of hardware and software) of computer 602 can interact with each other or with interface 604 (or a combination of both) on system bus 603 using application programming interface (API) 612 or service layer 613 (or a combination of API 612 and service layer 613). API 612 may include descriptions of routines, data structures, and object classes. API 612 may be independent of or dependent on a computer language and refers to a complete interface, a single function, or even a set of APIs. Service layer 613 provides software services to computer 602 or other components (whether shown or not) communicatively coupled to computer 602. The functionality of computer 602 may be accessible to all service consumers using the service layer. Software services (such as those provided by service layer 613) provide reusable, defined business functionality through defined interfaces. For example, the interface may be software written in JAVA, C++, or other suitable languages ​​that provide data in Extensible Markup Language (XML) format or other suitable formats. Although shown as an integrated component of computer 602, alternative concrete implementations may be shown as API 612 or service layer 613 as a separate component relative to or communicatively coupled to other components of computer 602 (whether shown or not). Furthermore, any part or all of API 612 or service layer 613 may be implemented as a submodule or sub-module of another software module, enterprise application, or hardware module without departing from the scope of this disclosure.

[0068] Computer 602 includes interface 604. Although in Figure 9While shown as a single interface 604, two or more interfaces 604 may be used depending on specific needs, expectations, or a particular implementation of computer 602. Interface 604 is used by computer 602 to communicate with other systems in a distributed environment connected to network 630. Generally, interface 604 includes logic coded in software or hardware (or a combination of software and hardware) and operable to communicate with network 630 or the cloud. More specifically, interface 604 may include software supporting one or more communication protocols associated with the communication, enabling network 630 or the hardware of the interface to communicate physical signals both inside and outside the illustrated computer 602.

[0069] Computer 602 includes at least one computer processor 605. Although in Figure 9 The computer processor 605 is shown as a single computer processor 605, but two or more processors may be used depending on specific needs, expectations, or a particular implementation of the computer 602. Generally, the computer processor 605 executes instructions and manipulates data to perform the operations of the computer 602 and any algorithms, methods, functions, procedures, flows, and programs as described in this disclosure.

[0070] Computer 602 also includes memory 606, which stores data for computer 602 or other components (or a combination of both) that can be connected to network 630. For example, memory 606 may be a database storing data consistent with this disclosure. Although in Figure 9 The memory 606 is shown as a single memory 606, but two or more memories may be used depending on specific needs, expectations, or the specific implementation of the computer 602 and the described functionality. Although the memory 606 is shown as an integrated component of the computer 602, in alternative specific implementations, the memory 606 may be external to the computer 602.

[0071] Application 607 is an algorithmic software engine that provides functionality (particularly with respect to the functionality described in this disclosure) for a specific need, expectation, or specific implementation of computer 602. For example, application 607 can function as one or more components, modules, applications, etc. Furthermore, although shown as a single application 607, application 607 can be implemented as multiple applications 607 on computer 602. Additionally, although shown as integrated with computer 602, in alternative specific implementations, application 607 may be located external to computer 602.

[0072] Any number of computers 602 may exist, either associated with or outside the computer system containing computer 602, wherein each computer 602 communicates on network 630. Furthermore, the terms "client," "user," and other suitable terms may be used interchangeably where appropriate without departing from the scope of this disclosure. Moreover, this disclosure envisions a plurality of users using one computer 602, or a single user using multiple computers 602.

[0073] In some embodiments, computer 602 is implemented as part of a cloud computing system. For example, the cloud computing system may include one or more remote servers and various other cloud components, such as cloud storage units and edge servers. In particular, the cloud computing system can perform one or more computing operations without direct, active management by user devices or local computer systems. Thus, the cloud computing system can have different functions distributed across multiple locations from a central server, which can be executed using one or more Internet connections. More specifically, the cloud computing system can operate according to one or more service models, such as Infrastructure as a Service (IaaS), Platform as a Service (PaaS), Software as a Service (SaaS), Mobile Backend as a Service (MBaaS), Artificial Intelligence as a Service (AIaaS), Serverless Computing, and / or Function as a Service (FaaS).

[0074] Although only a few exemplary embodiments have been described in detail above, those skilled in the art will readily understand that many modifications can be made to the exemplary embodiments without substantially departing from the invention. Therefore, all such modifications are intended to be included within the scope of this disclosure as defined by the appended claims.

[0075] Although this disclosure has been described with respect to a limited number of embodiments, those skilled in the art will recognize, with benefit from this disclosure, that other embodiments can be devised without departing from the scope of the invention disclosed herein. Therefore, the scope of this disclosure should be defined only by the appended claims.

Claims

1. A method comprising: The P-wave velocity model (476) and velocity ratio data (411) for the geological area of ​​interest are obtained by the computer processor (605); An initial S-wave velocity model (420) for the geological region of interest is generated by the computer processor (605) based on the P-wave velocity model (476) and the velocity ratio data (411). The computer processor (605) uses a trained model to determine multiple velocity boundaries within the initial S-wave velocity model (420); The computer processor (605) updates the initial S-wave velocity model (420) using the multiple velocity boundaries, automatically selected cross-correlation hysteresis values ​​based on multiple seismic migration gathers, and migration velocity analysis to generate an updated S-wave velocity model (470); and The computer processor (605) uses the updated S-wave velocity model (470) and the P-wave velocity model (476) to generate a combined velocity model for the geological region of interest.

2. The method according to claim 1, further comprising: The combined velocity model is used to generate a combined image of the geological region of interest. The combined image describes multiple different S-wave velocities and multiple different P-wave velocities in the geological region of interest.

3. The method according to any one of claims 1 and 2, further comprising: Based on the migration velocity analysis and the initial S-wave velocity model (420), multiple seismic migration gathers with different cross-correlation hysteresis values ​​are generated. and The computer processor (605) automatically selects a predetermined cross-correlation hysteresis value using the plurality of seismic migration gathers and based on predetermined criteria. The automatically selected cross-correlation lag value corresponds to the predetermined cross-correlation lag value.

4. The method according to claim 3, in, The different cross-correlation hysteresis values ​​are based on the travel-time inversion process performed using ray tracing, one-way wave equation tomography, or two-way wave equation tomography.

5. The method according to any one of claims 1 to 4, in, The trained model is a convolutional neural network, which includes multiple convolutional layers (585), at least one downscaling layer (582), and at least one upscaling layer (584). The trained model receives the migrated seismic image (550) as input, and The trained model generates a velocity boundary map as output.

6. The method according to any one of claims 1 to 5, in, The trained model is a machine learning model, which is trained using a training dataset that includes manually picked boundary data and augmented boundary data. The trained model is trained using multiple machine learning cycles, and Each of the plurality of machine learning cycles uses a portion of the training dataset to train a machine learning model to produce the trained model.

7. The method according to any one of claims 1 to 6, in, The migration velocity analysis is a single-pass wave equation tomography operation.

8. The method according to any one of claims 1 to 7, in, The velocity ratio data (411) describes a predetermined ratio between the P-wave value and the S-wave value at a predetermined location in the geological region of interest.

9. The method according to any one of claims 1 to 8, further comprising: Seismic data about the geological area of ​​interest is obtained using a seismic survey system (100); and The P-wave velocity model (476) is generated using the earthquake data and earthquake inversion operation.

10. The method according to any one of claims 1 to 9, further comprising: Obtain first manually picked boundary data describing the first plurality of velocity boundaries within the first migrated seismic image (550); Enhanced boundary data describing a second plurality of velocity boundaries within a second migrated seismic image (550) is obtained, wherein the enhanced boundary data is generated from second manually picked boundary data using enhancement operations; The trained model is generated using multiple machine learning cycles and training data including the first manually picked boundary data and the enhanced boundary data; and The trained model and migrated seismic data are used to determine one or more S-wave velocity boundaries.

11. The method of claim 10, further comprising: Obtain boundary data picked by a third person manually. During the plurality of machine learning cycles, the trained model is updated based on a comparison between at least a portion of the third manually picked boundary data and the predicted boundary data generated by the trained model.

12. The method according to claim 10, in, The multiple machine learning cycles are training iterations for the trained model, and Each of the plurality of machine learning cycles uses a predetermined portion of the training data to train the trained model.

13. The method according to claim 10, in, The enhancement operation is selected from the group consisting of the following operations: rotation operation, displacement operation, cutting operation, intrusion operation, and deformation operation.

14. A non-transitory computer-readable medium storing instructions executable by a computer processor (605), the instructions comprising the following functions: Obtain P-wave velocity models (476) and velocity ratio data (411) for the geological region of interest; Based on the P-wave velocity model (476) and the velocity ratio data (411), an initial S-wave velocity model (420) for the geological region of interest is generated. The trained model is used to determine multiple velocity boundaries within the initial S-wave velocity model (420); The initial S-wave velocity model (420) is updated using the multiple velocity boundaries, automatically selected cross-correlation hysteresis values ​​based on multiple seismic migration gathers, and migration velocity analysis to produce an updated S-wave velocity model (470); and The updated S-wave velocity model (470) and the P-wave velocity model (476) are used to generate a combined velocity model for the geological region of interest.

15. The non-transitory computer-readable medium according to claim 14, wherein, The instructions also include the following functions: The combined velocity model is used to generate a combined image of the geological region of interest. The combined image describes multiple different S-wave velocities and multiple different P-wave velocities in the geological region of interest.

16. The non-transitory computer-readable medium according to any one of claims 14 and 15, wherein, The instructions also include the following functions: Based on the migration velocity analysis and the initial S-wave velocity model (420), multiple seismic migration gathers with different cross-correlation hysteresis values ​​are generated; and The computer processor (605) automatically selects a predetermined cross-correlation hysteresis value using the plurality of seismic migration gathers and based on predetermined criteria. The automatically selected cross-correlation lag value corresponds to the predetermined cross-correlation lag value.

17. The non-transitory computer-readable medium according to any one of claims 14 to 16, in, The trained model is a machine learning model, which is trained using a training dataset that includes manually picked boundary data and augmented boundary data. The trained model is trained using multiple machine learning cycles, and Each of the plurality of machine learning cycles uses a portion of the training dataset to train a machine learning model to produce the trained model.

18. A system comprising: A seismic survey system (100) comprising a seismic source (122) and multiple seismic receivers (226); as well as A seismic interpreter (261) including a computer processor (605), wherein the seismic interpreter (261) is coupled to the seismic survey system (100), and the seismic interpreter (261) includes the following functions: Obtain P-wave velocity models (476) and velocity ratio data (411) for the geological region of interest; Based on the P-wave velocity model (476) and the velocity ratio data (411), an initial S-wave velocity model (420) for the geological region of interest is generated. The trained model is used to determine multiple velocity boundaries within the initial S-wave velocity model (420); The initial S-wave velocity model (420) is updated using the multiple velocity boundaries, automatically selected cross-correlation hysteresis values ​​based on multiple seismic migration gathers, and migration velocity analysis to produce an updated S-wave velocity model (470); and The updated S-wave velocity model (470) and the P-wave velocity model (476) are used to generate a combined velocity model for the geological region of interest.

19. The system according to claim 18, wherein, The seismic interpreter (261) also includes the following functions: The combined velocity model is used to generate a combined image of the geological region of interest. The combined image describes multiple different S-wave velocities and multiple different P-wave velocities in the geological region of interest.

20. The system according to any one of claims 18 and 19, in, The trained model is a convolutional neural network, which includes multiple convolutional layers (585), at least one downscaling layer (582), and at least one upscaling layer (584). The trained model receives the migrated seismic image (550) as input, and The trained model generates a velocity boundary map as output.

Images