Seismic data curvature pickup method and device, computer equipment and storage medium

Abstract

The invention provides a seismic data curvature pickup method and device, computer equipment and a storage medium, and the method comprises the steps: obtaining a pre-stack seismic data volume as a sample label data set which records the reflection waveform position information; determining seed point information of a zero offset position in the sample label data set according to the reflection waveform position information, and obtaining a seed point set based on the seed point information of the zero offset position; extracting a seismic data curvature curve from the sample label data set to obtain a curvature curve label data set; obtaining a stacked seismic data volume, extracting similar spectrum data from the stacked seismic data volume according to the residual curvature range, establishing a corresponding relation between the similar spectrum data and the seed point set, and obtaining a similar spectrum data label data set; and inputting the curvature curve label data set and the similar spectrum data label data set into a neural network for learning to obtain a neural network model for seismic data curvature pickup.

Description

Technical Field

[0001] The present invention relates to the field of seismic survey technology, and in particular to a method, device, computer equipment and storage medium for picking up seismic data curvature. Background Art

[0002] In the past decade, seismic exploration has thoroughly entered the field of processing and interpretation in the depth domain both in theory and in actual industrial processing procedures, which is particularly important for the depth domain processing of seismic data. For deep domain processing, the most important processing links are velocity modeling in the depth domain and migration imaging in the depth domain. Traditional depth domain velocity modeling is usually divided into two steps: near-surface velocity modeling and mid-deep velocity modeling. Mid-deep velocity modeling includes six steps: residual curvature calculation, inclination scanning, seed point picking, residual curvature picking, ray tracing matrix construction, and matrix solution.

[0003] The specific steps of traditional depth domain velocity modeling are to first use the initial velocity model to perform prestack depth migration to obtain the migration gathers and profiles, then pick up the residual depth difference and reflection dip angle of the imaging gathers, establish the tomographic equation, and use the iterative method to solve the discrete large sparse equations through ray tracing to obtain the velocity correction, and then update the initial velocity model, perform a new round of iterations, and obtain a high-precision velocity model. The picking of residual curvature affects the accurate estimation of the residual time difference, and the accuracy of the residual curvature picking largely determines the accuracy of the ray tracing tomographic matrix establishment, which also affects the accuracy of seismic exploration depth domain velocity modeling and the success or failure of this round of seismic exploration processing.

[0004] In the prior art, the residual curvature picking technology mainly relies on the relative relationship between the peaks and troughs of the seismic data event axes to determine the residual curvature. However, in complex exploration areas, when there are event axis intersections or the seismic data energy is weak, the residual curvature picking is not in place and the accuracy of the residual curvature picking is low. Summary of the invention

[0005] Based on this, it is necessary to provide a method, device, computer equipment and storage medium for picking curvature of seismic data in response to the above technical problems.

[0006] A method for picking curvature of seismic data, comprising:

[0007] Acquire a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records reflection waveform position information;

[0008] Determine the seed point information of the zero offset position in the sample label data set according to the reflected waveform position information, and acquire a seed point set based on the seed point information of the zero offset position;

[0009] Extracting a seismic data curvature curve from the sample label data set to obtain a curvature curve label data set;

[0010] Acquire a post-stack seismic data volume, extract similar spectrum data from the post-stack seismic data volume according to the residual curvature range, establish a corresponding relationship between the similar spectrum data and the seed point set, and obtain a similar spectrum data label data set;

[0011] The curvature curve label data set and the similar spectrum data label data set are input into a neural network for learning to obtain a neural network model for seismic data curvature picking.

[0012] In one embodiment, the step of obtaining a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records the reflection waveform position information, comprises:

[0013] Obtain pre-stack seismic data volume;

[0014] Slicing the pre-stack seismic data volume according to the number of seismic traces and the number of sampling points to obtain a sliced ​​data volume;

[0015] A seed point of zero offset in the slice data volume is obtained in the direction of the 0th seismic track along the sampling point direction, and a time-distance curve is produced by using random velocity disturbance at the seed point of non-zero offset to generate the sample label data set.

[0016] In one embodiment, the step of obtaining a pre-stack seismic data volume as a sample label data set includes:

[0017] Acquire a pre-stack seismic data volume as the sample label data set;

[0018] Preprocessing the sample label data set to obtain a processed sample label data set;

[0019] The step of extracting the seismic data curvature curve from the sample label data set to obtain the curvature curve label data set comprises:

[0020] The seismic data curvature curve is extracted from the processed sample label data set to obtain a curvature curve label data set.

[0021] In one embodiment, the step of preprocessing the sample label data set to obtain a processed sample label data set includes:

[0022] An amplitude value distribution range of a pre-stack seismic data volume is identified, and a truncation operation is performed on the pre-stack seismic data volume based on the amplitude value distribution range.

[0023] In one embodiment, the identifying of the amplitude value distribution range of the pre-stack seismic data volume, after truncating the pre-stack seismic data volume based on the amplitude value distribution range, comprises:

[0024] The truncated seismic data volume is filtered, gray-scale processed, and normalized to obtain the processed sample label data set.

[0025] In one embodiment, before the step of acquiring the post-stack seismic data volume, the step further includes:

[0026] The remaining curvature range is determined according to the work area.

[0027] A seismic data curvature picking device, comprising:

[0028] A sample label data set acquisition module is used to acquire a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records the reflection waveform position information;

[0029] A seed point set acquisition module: used to determine the seed point information of the zero offset position in the sample label data set according to the reflected waveform position information, and acquire a seed point set based on the seed point information of the zero offset position;

[0030] Curvature curve acquisition module: used to extract the seismic data curvature curve from the sample label data set to obtain the curvature curve label data set;

[0031] Similar spectrum data acquisition module: used to acquire a post-stack seismic data volume, extract similar spectrum data from the post-stack seismic data volume according to the residual curvature range, establish a corresponding relationship between the similar spectrum data and the seed point set, and obtain a similar spectrum data label data set;

[0032] Model training module: used to input the curvature curve label data set and the similar spectrum data label data set into the neural network for learning, so as to obtain a neural network model for seismic data curvature picking.

[0033] In one embodiment, the sample label data set acquisition module includes:

[0034] A pre-stack seismic data volume acquisition module is used to acquire pre-stack seismic data volumes;

[0035] A pre-stack seismic data volume slicing module is used to slice the pre-stack seismic data volume according to the number of seismic traces and the number of sampling points to obtain a sliced ​​data volume;

[0036] The sample label data set generation module is used to obtain the seed point of zero offset in the slice data volume in the direction of the 0th seismic track along the sampling point direction, use random velocity perturbation to make time-distance curves at the seed point of non-zero offset, and generate the sample label data set.

[0037] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the following steps when executing the computer program:

[0038] Acquire a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records reflection waveform position information;

[0039] Determine the seed point information of the zero offset position in the sample label data set according to the reflected waveform position information, and acquire a seed point set based on the seed point information of the zero offset position;

[0040] Extracting a seismic data curvature curve from the sample label data set to obtain a curvature curve label data set;

[0041] Acquire a post-stack seismic data volume, extract similar spectrum data from the post-stack seismic data volume according to the residual curvature range, establish a corresponding relationship between the similar spectrum data and the seed point set, and obtain a similar spectrum data label data set;

[0042] The curvature curve label data set and the similar spectrum data label data set are input into a neural network for learning to obtain a neural network model for seismic data curvature picking.

[0043] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0044] Acquire a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records reflection waveform position information;

[0045] Determine the seed point information of the zero offset position in the sample label data set according to the reflected waveform position information, and acquire a seed point set based on the seed point information of the zero offset position;

[0046] Extracting a seismic data curvature curve from the sample label data set to obtain a curvature curve label data set;

[0047] Acquire a post-stack seismic data volume, extract similar spectrum data from the post-stack seismic data volume according to the residual curvature range, establish a corresponding relationship between the similar spectrum data and the seed point set, and obtain a similar spectrum data label data set;

[0048] The curvature curve label data set and the similar spectrum data label data set are input into a neural network for learning to obtain a neural network model for seismic data curvature picking.

[0049] The above-mentioned seismic data curvature picking method, device, computer equipment and storage medium obtain a curvature curve label set by extracting the seismic data curvature curve from the sample label data set, and extracting similar spectrum data from the stacked seismic data body according to the remaining curvature range, thereby establishing a corresponding relationship between the similar spectrum data and the seed point set, and then inputting the curvature curve label data set and the similar spectrum data label data set into the neural network for learning, thereby obtaining a neural network model capable of seismic data curvature picking. When training the neural network model, after a large amount of data calculation, the neural network model is gradually optimized so that the neural network model can pick up the curvature of the seismic data more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic diagram of a flow chart of a method for picking curvature of seismic data in one embodiment;

[0051] Figure 2 A schematic diagram of the structure of a seismic data curvature picking device in one embodiment;

[0052] Figure 3 is a diagram of the internal structure of a computer device in one embodiment;

[0053] Figure 4 A schematic diagram of a flow chart of a method for picking curvature of seismic data in another embodiment;

[0054] Figure 5 A schematic diagram of a flow chart of a method for picking curvature of seismic data in another embodiment;

[0055] Figure 6 An example diagram of a seismic data volume sample set in an embodiment;

[0056] Figure 7 An example diagram of a preprocessed seismic data sample set in one embodiment;

[0057] Figure 8 is similarity coefficient spectrum data obtained based on seismic data in one embodiment;

[0058] Fig. 9 An example diagram of a seismic data label set in one embodiment;

[0059] Fig.10 is a schematic diagram of a neural network model in one embodiment;

[0060] Fig.11 A result diagram of a neural network model output in one embodiment;

[0061] Fig.12 It is a schematic diagram of the time-distance curve shape in one embodiment. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0063] Embodiment 1

[0064] In this embodiment, Figure 1 As shown, a method for picking curvature of seismic data is provided, which comprises:

[0065] Step 110 : Acquire a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records the reflection waveform position information.

[0066] In this embodiment, a seismic data volume is obtained as sample data, and the seismic data volume is a data volume of a pre-stack gather. The pre-stack seismic data volume contains a record of a reflection waveform. In this embodiment, the reflection waveform position corresponding to the reflection waveform is marked, and the marked seismic data volume is used as label data, thereby obtaining a sample label data set.

[0067] In the sample label data set, the sample data of the earthquake data volume is as follows Figure 6 As shown, the label data is as follows Fig. 9 shown.

[0068] Step 120 : determining seed point information of a zero offset position in the sample label data set according to the reflected waveform position information, and acquiring a seed point set based on the seed point information of the zero offset position.

[0069] In this embodiment, the position of the first reflected waveform is the seed point of the zero offset position. In this step, the seed point information of the zero offset position in the sample label data set is determined according to the reflected waveform position information, thereby determining the position information of the first seed point, and based on the position information of the first seed point, the position information of other seed points can be obtained, thereby obtaining various sub-points in the sample label data set to obtain a seed point set.

[0070] Step 130: extracting the seismic data curvature curve from the sample label data set to obtain a curvature curve label data set.

[0071] In this embodiment, the curvature curve label data set includes the location information of the seed point and the residual curvatures corresponding to various sub-points.

[0072] Step 140, obtaining a post-stack seismic data volume, extracting similar spectrum data from the post-stack seismic data volume according to the residual curvature range, establishing a corresponding relationship between the similar spectrum data and the seed point set, and obtaining a similar spectrum data label data set.

[0073] In this embodiment, the similar spectrum data includes the residual curvature of the seed point, and the correspondence between the similar spectrum data and the seed point set is established, so that the relationship between the position information of the seed point and the corresponding residual curvature can be obtained. Different from step 130, the residual curvature of step 140 is screened and extracted according to the residual curvature range, and the correspondence between the seed point position and the residual curvature can be obtained more specifically according to the residual curvature range.

[0074] Step 150: Input the curvature curve label data set and the similar spectrum data label data set into a neural network for learning, so as to obtain a neural network model for seismic data curvature picking.

[0075] In this embodiment, the curvature curve label data set and the similar spectrum data label data set are used as training sets and input into the neural network, the neural network is trained, the parameters of the neural network are determined, and the trained neural network is obtained. The trained neural network can be used to pick up the remaining curvature.

[0076] The above-mentioned seismic data curvature picking method obtains a curvature curve label set by extracting the seismic data curvature curve from the sample label data set, and extracts similar spectrum data from the stacked seismic data body according to the remaining curvature range, thereby establishing a corresponding relationship between the similar spectrum data and the seed point set, and then inputs the curvature curve label data set and the similar spectrum data label data set into the neural network for learning, thereby obtaining a neural network model capable of seismic data curvature picking. When training the neural network model, after a large amount of data calculation, the neural network model is gradually optimized, so that the neural network model can pick up the curvature of the seismic data more accurately.

[0077] In one embodiment, the step of obtaining a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records the reflection waveform position information, includes: obtaining a pre-stack seismic data volume; slicing the pre-stack seismic data volume according to the number of seismic traces and the number of sampling points to obtain a sliced ​​data volume; obtaining a seed point of zero offset in the sliced ​​data volume in the direction of the 0th seismic trace along the sampling point direction, and using random velocity perturbations at the seed points of non-zero offsets to create a time-distance curve to generate the sample label data set.

[0078] In this embodiment, the number of seismic traces in the seismic data volume is N, and the number of sampling points is M. The seismic data volume is sliced ​​according to the number of N seismic traces and the number of M sampling points to obtain a sliced ​​data volume. The number of zero-offset seed points is obtained in the direction of the 0th seismic trace along the sampling point direction, and the time-distance curve is produced by random velocity perturbation at non-zero offset to generate a label data set.

[0079] In this embodiment, for the seismic reflection waveform sample data, the Ricker wavelet is convolved with the label data to construct a sample label data set, which is calculated by the following formula (1):

[0080] T (n, m) = F (m, v, n) n∈[0, N-1], v∈[-25%, 25%], m∈[0, M-1] (1)

[0081] Where T is the function expression of a curvature curve in the label data, and F is the rule for generating the curve, that is, the label data set is generated by perturbing the time-distance curve starting from the seed points n=0 to N-1 with a velocity perturbation range of -25% to 25%. Formula (1) is the calculation formula for the seismic exploration time-distance curve. T(n,m) refers to the two-dimensional graph of all time-distance curves on the entire data volume corresponding to the size of the pre-stack seismic data slice. Each curve will produce different curvature changes according to the perturbation of v. The specific calculation is as follows:

[0082]

[0083] Where T(n,m) is a two-dimensional graph of all time-distance curves on the entire data volume corresponding to the size of the pre-stack seismic data slice. Each time-distance curve will produce different curvature changes according to the disturbance of v. x is the index of n, and n corresponds to the offset X. offset , h is the index of m, and m corresponds to the number of seed points. The two-dimensional graph of the time-distance curve is as follows Fig.12 As shown, there are m seed points, and the time-distance curves generated by different V are as follows Fig.12 shown.

[0084] In one embodiment, the step of obtaining a pre-stack seismic data volume as a sample label data set includes: obtaining a pre-stack seismic data volume as the sample label data set; preprocessing the sample label data set to obtain a processed sample label data set; the step of extracting a seismic data curvature curve from the sample label data set to obtain a curvature curve label data set includes: extracting a seismic data curvature curve from the processed sample label data set to obtain a curvature curve label data set. The preprocessed sample label data set is as follows: Figure 7 shown.

[0085] In one embodiment, the step of preprocessing the sample label data set to obtain a processed sample label data set includes: identifying the amplitude value distribution range of the pre-stack seismic data volume, and truncating the pre-stack seismic data volume based on the amplitude value distribution range; filtering, grayscale processing, and normalization operations on the truncated seismic data volume to obtain the processed sample label data set.

[0086] In this embodiment, the pre-stack seismic data volume is identified, the specific amplitude value distribution range of the earthquake number is determined, and the pre-stack seismic data volume is truncated by the following equations (2) and (3):

[0087] D_1(n,m)=Clip(D(n,m)) n∈[0,N-1], m∈[0,M-1] (2)

[0088] Clip(D(n,m))=if(D(n,m)>0.90*Max(D(n,m)))then D(n,m)

[0089] =0.90*Max(D(n,m)) (3)

[0090] Wherein, D(n, m) is seismic data, which is a slice of pre-stack seismic data gather, n is the trace number index, and m is the sampling index.

[0091] The truncated seismic data volume is filtered, gray-scale processed, and normalized using the following formula (4):

[0092] D_2(n,m)=SplitNorm(StructSmooth((D_1(n,m))))

[0093] n∈[0, N-1], m∈[0, M-1] (4)

[0094] Among them, SplitNorm adopts the piecewise normalization calculation along the sampling direction, and StructSmooth adopts the structural filtering technology to perform filtering and smoothing operations on the discontinuous seismic data event axes along the event axis direction.

[0095] In one embodiment, the step of obtaining the post-stack seismic data volume further includes: determining the residual curvature range according to the work area. In this embodiment, in order to screen out more residual curvatures that meet the conditions of the work area, the residual curvature range is determined according to the requirements of the work area, so as to extract corresponding similar spectrum data.

[0096] In one embodiment, the steps of obtaining a post-stack seismic data volume, extracting similar spectrum data from the post-stack seismic data volume according to a residual curvature range, establishing a corresponding relationship between the similar spectrum data and the seed point set, and obtaining a similar spectrum data label data set include: scanning the post-stack seismic data volume according to the residual curvature range, calculating a similarity coefficient spectrum according to the following formula (5), and further extracting seismic features:

[0097] RD(g,m)=Semblance(D_2(n,m)) n∈[0,N-1], m∈[0,M-1], g∈[g1,g2] (5)

[0098] Among them, Semblance (D_2 (n, m)) is the calculation method of seismic data similarity spectrum, g is the residual curvature scanning range;

[0099] The similarity spectrum is calculated using formula (6):

[0100]

[0101] Among them, T zx Indicates the offset is x halfoffset The remaining delay calculated at z0 is the time corresponding to zero bias, and γ is the residual curvature range.

[0102] Among them, the similarity spectrum calculated based on the seismic data volume is as follows Figure 8 shown.

[0103] In this embodiment, the residual curvature scanning range of equation (5) is [-0.5, 0.5].

[0104] In this embodiment, the seed point is selected by selecting the vertical depth position of the point with the maximum energy picked up on the similarity spectrum, so as to construct a corresponding relationship between the similarity spectrum data and the seed point set.

[0105] In one embodiment, the neural network is based on a convolutional neural network autoencoder as the main body, and a 29-layer deep neural network model is built based on tensorflow 2.0. The built deep neural network model is as follows Fig.10 shown.

[0106] In this embodiment, considering that the most important feature of the pre-stack seismic data body or the post-stack seismic data body as a whole is its spatial position information, an autoencoder based on a convolutional neural network is used as the main body of this technology, and a 29-layer deep neural network model is built based on tensorflow 2.0. The deep neural network model includes seven convolution upsampling blocks and seven convolution downsampling blocks, each of which consists of two convolution layers, a BN layer, a downsampling / upsampling layer, and a LeakyRelu activation function layer. The convolution kernels used in the two convolution layers are 7×1 and 1×7, so as to obtain a larger spatial receptive field and reduce the amount of calculation.

[0107] In one embodiment, the curvature curve label data set and the similar spectrum data label data set are input into a neural network for learning, and the steps of obtaining a neural network model for seismic data curvature picking include: using the curvature curve label data set and the similar spectrum data label data set to establish a training set, and performing distributed training on the neural network; wherein, the error metric of the neural network uses the RMSE error functional based on L2 regularization as the objective function of the neural network, the initialization method of each parameter of the convolution layer in the neural network adopts Xavier initialization, and the L2 regularization coefficient of the convolution layer parameters is set to 0.01, and the optimization algorithm adopts the Adam algorithm.

[0108] In this embodiment, the Adam algorithm is a first-order optimization algorithm that can replace the traditional stochastic gradient descent process. It designs independent adaptive learning rates for different parameters by calculating the first-order moment estimate and the second-order moment estimate of the gradient, and then iteratively updates the neural network weights. It has better convergence than the traditional stochastic gradient descent algorithm.

[0109] In this embodiment, the RMSE functional expression is shown in formula (7):

[0110]

[0111] Where E(x) is the objective function of the neural network, x is the model parameter of the neural network, m is the number of selected mini-batches, and y i (x) is the actual label data, is the predicted data, and γ is the regularization parameter. Further, the regularization parameter γ is 0.01.

[0112] In this embodiment, after obtaining the neural network model for seismic data curvature picking, the parameters of the trained neural network are saved to the local disk and deployed to a dedicated cluster for seismic data curvature picking to form a reliable seismic data curvature picking technology.

[0113] It should be understood that although Figure 1The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0114] Embodiment 2

[0115] In this embodiment, Figure 4 As shown, a method for picking curvature of seismic data is provided, comprising:

[0116] Step 310, establish a seismic data volume as sample data, in which there are a large number of reflection waveform records, and at the same time establish a reflection waveform position data volume that corresponds accurately to these reflection waveforms one by one as label data, thereby establishing this sample label data set.

[0117] In this embodiment, a sample label data set is established in which the seismic curvature data corresponds to the actual curvature position one by one: the following method is specifically adopted: first, the seismic data with N seismic channels and M sampling points are sliced, and the number of zero offset seed points is established along the sampling point direction in the 0th seismic channel direction, and the time-distance curve produced by random velocity perturbation is used at non-zero offset to generate a label data set. For the seismic reflection waveform sample data, the Ricker wavelet is convolved with the label data to construct the sample label data set. The specific formula is shown in formula (1):

[0118] T (n, m) = F (m, v, n) n∈[0, N-1], v∈[-25%, 25%], m∈[0, M-1] (1)

[0119] Wherein, T is a function expression of a curvature curve existing in the label data, and F is the rule for generating the curve in the present invention, i.e., using the time-distance curve starting from the seed points n=0 to n=N-1 with a velocity disturbance range of -25% to 25% to perform disturbance. Formula (1) is the calculation formula for the time-distance curve of seismic exploration, and T(n,m) refers to the two-dimensional graph of all time-distance curves on the entire data volume corresponding to the size of the pre-stack seismic data slice. Each curve will produce different curvature changes according to the disturbance of v. The specific calculation is as follows:

[0120]

[0121] Where T(n,m) is a two-dimensional graph of all time-distance curves on the entire data volume corresponding to the size of the pre-stack seismic data slice. Each time-distance curve will produce different curvature changes according to the disturbance of v. x is the index of n, and n corresponds to the offset X. offset , h is the index of m, and m corresponds to the number of seed points. The two-dimensional graph of the time-distance curve is as follows Fig.12 As shown, there are m seed points, and the time-distance curves generated by different V are as follows Fig.12 shown.

[0122] Step 320: pre-process the established sample label data set.

[0123] In this embodiment, the seismic data volume is preprocessed to a sampling standard and a numerical value that can be input into deep learning, and the seed point information of the zero offset position is extracted.

[0124] In this embodiment, the seismic data volume is preprocessed as follows: first, the seismic data volume data is input, the specific amplitude value distribution range of the earthquake number is determined, and the input seismic data is truncated according to the following formula:

[0125] D_1(n,m)=Clip(D(n,m))n∈[0,N-1],m∈[0,M-1]

[0126] Clip(D(n,m))=if(D(n,m)>0.90*Max(D(n,m)))then D(n,m)

[0127] =0.90*Max(D(n,m))

[0128] Wherein, D(n, m) is seismic data, which is a slice of pre-stack seismic data gather, n is the trace number index, and m is the sampling index.

[0129] The truncated seismic data is filtered, gray-scaled, and normalized using the following formula:

[0130] D_2(n,m)=SplitNorm(StructSmooth((D_1(n,m))))

[0131] n∈[0,N-1],m∈[0,M-1]

[0132] Among them, SplitNorm adopts the piecewise normalization calculation along the sampling direction, and StructSmooth adopts the structural filtering technology to perform filtering and smoothing operations on the discontinuous seismic data event axes along the event axis direction.

[0133] Step 330, output the seismic data volume, determine the remaining curvature range according to the work area requirements, extract the seismic data similarity spectrum data, and establish a data set in which the similarity spectrum corresponds to the data set of the seed point one by one.

[0134] In this embodiment, for the input seismic data volume, a suitable residual curvature range is selected according to the work area for scanning, and the similarity coefficient spectrum is calculated to further extract the seismic characteristics, and the following formula is used for calculation:

[0135] RD(g,m)=Semblance(D_2(n,m))n∈[0,N-1], m∈[0,M-1], g∈[g1,g2]

[0136] Wherein, Semblance(D_2(n,m)) is the calculation method of seismic data similarity spectrum, g is the residual curvature scanning range, and the sample label data set range is [-0.5, 0.5]. The present invention uses the following formula to calculate the similarity spectrum:

[0137]

[0138] Among them, T zx Indicates the offset is x halfoffset The remaining delay calculated at z0 is the time corresponding to zero bias, and γ is the residual curvature range.

[0139] In this embodiment, the seed point is selected by selecting the vertical depth position of the point with the maximum energy picked up on the similarity spectrum, and a one-to-one correspondence data set between the similarity spectrum and the seed point is constructed.

[0140] Step 340, construct a neural network model based on deep learning.

[0141] In this embodiment, considering that the seismic data body as a whole has the most important feature of the spatial position information of the seismic data body, an autoencoder based on a convolutional neural network is used as the main body, and a 29-layer deep neural network model is built based on tensorflow 2.0. The deep neural network model includes seven convolution upsampling blocks and seven convolution downsampling blocks, each of which consists of two convolution layers, a BN layer, a downsampling / upsampling layer, and a LeakyRelu activation function layer. The two convolution layers use convolution kernels of 7×1 and 1×7 to obtain a larger spatial receptive field and reduce the amount of calculation.

[0142] Step 350, using the training set established in step 320 and step 330, perform distributed training on the neural network model built in step 340.

[0143] In this embodiment, the error metric of the training of the neural network model is based on the L2 regularized RMSE error functional as the objective function of the neural network model, the initialization method of each parameter of the convolution layer adopts Xavier initialization, and the L2 regularization with a coefficient of 0.01 is set in the convolution layer parameters. The optimization algorithm of the neural network model adopts the Adam algorithm. The Adam algorithm is a first-order optimization algorithm that can replace the traditional stochastic gradient descent process. It designs independent adaptive learning rates for different parameters by calculating the first-order moment estimate and the second-order moment estimate of the gradient, and then iteratively updates the weights of the neural network model, which has better convergence than the traditional stochastic gradient descent algorithm.

[0144] In this embodiment, the RMSE functional expression is as follows:

[0145]

[0146] Where E(x) is the objective function, x is the neural network model parameter, m is the number of selected mini-batches, and y i (x) is the actual label data, is the predicted data, γ is the regularization parameter, and the regularization parameter γ is 0.01.

[0147] Step 360, save the parameters of the trained network model to the local disk and deploy it to a dedicated cluster for seismic data curvature picking to form a reliable seismic data curvature picking technology.

[0148] The key to this application is the training of the neural network model and the saving and loading of the neural network model. The present invention uses GPU computing resources to train the neural network model according to the requirements of step 350. After the training is completed, the parameters of the neural network model are saved locally, and then the model can be uploaded to the processing cluster, and the seismic data to be processed can be inferred using CPU resources, completing the promotion and application of the technology of the present invention.

[0149] Embodiment 3

[0150] In this embodiment, Figure 5 As shown, a method for picking up curvature of seismic data is provided, including: generating a label data set according to the number of seed points in the sampling direction and random velocity disturbance, convolving the label data set using Ricker wavelet to establish a sample data set, processing the sample data set in a preprocessing manner, establishing a sample label data set of seismic data curvature curve, calculating a similarity coefficient spectrum using residual curvatures in different ranges, establishing a similarity spectrum and a sample label data set of seed points, building a convolutional neural network and training it, training to obtain deep learning model parameters and arranging clusters. Input the test data into the deep learning model and output the prediction results. Fig.11As shown in the figure, the processing results of the deep learning model on the actual seismic data body show that the present invention can accurately pick up the position of the coinciding axis of the seismic data body and can meet the requirements of deep domain modeling.

[0151] The curvature picking method of seismic data in this application is of great significance to the processing of deep-domain seismic data by establishing an accurate deep-domain velocity model. The picking of residual curvature of seismic data in the deep-domain velocity modeling link is a major problem in this link. The traditional curvature picking technology based on peaks and troughs cannot adapt to exploration areas with complex structures. This application uses big data and artificial intelligence technologies based on image recognition that have gradually been put into practical use in the industry in recent years. By constructing a residual curvature picking system based on deep learning, it can achieve accurate extraction of the position of the phase axis on the seismic data body, thereby providing accuracy guarantee for the subsequent links of deep-domain velocity modeling of seismic exploration, and solving the problem of inaccurate deep-domain seismic velocity modeling in areas with complex surface and underground geological structures such as piedmont belts and alluvial fans.

[0152] Embodiment 4

[0153] In this embodiment, Figure 2 As shown, a seismic data curvature picking device is provided, comprising:

[0154] The sample label data set acquisition module 210 is used to acquire a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records the reflection waveform position information;

[0155] A seed point set acquisition module 220 is used to determine the seed point information of the zero offset position in the sample label data set according to the reflected waveform position information, and acquire a seed point set based on the seed point information of the zero offset position;

[0156] The curvature curve acquisition module 230 is used to extract the curvature curve of the seismic data from the sample label data set to obtain the curvature curve label data set;

[0157] Similar spectrum data acquisition module 240: used to acquire a post-stack seismic data volume, extract similar spectrum data from the post-stack seismic data volume according to the residual curvature range, establish a corresponding relationship between the similar spectrum data and the seed point set, and obtain a similar spectrum data label data set;

[0158] Model training module 250: used to input the curvature curve label data set and the similar spectrum data label data set into a neural network for learning, so as to obtain a neural network model for seismic data curvature picking.

[0159] The sample label data set acquisition module 210 is used to acquire a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records the reflection waveform position information.

[0160] In this embodiment, a seismic data volume is obtained as sample data, and the seismic data volume is a data volume of a pre-stack gather. The pre-stack seismic data volume contains a record of a reflection waveform. In this embodiment, the reflection waveform position corresponding to the reflection waveform is marked, and the marked seismic data volume is used as label data, thereby obtaining a sample label data set.

[0161] The seed point set acquisition module 220 is used to determine the seed point information of the zero offset position in the sample label data set according to the reflected waveform position information, and acquire a seed point set based on the seed point information of the zero offset position.

[0162] In this embodiment, the position of the first reflected waveform is the seed point of the zero offset position. In this step, the seed point information of the zero offset position in the sample label data set is determined according to the reflected waveform position information, thereby determining the position information of the first seed point, and based on the position information of the first seed point, the position information of other seed points can be obtained, thereby obtaining various sub-points in the sample label data set to obtain a seed point set.

[0163] The curvature curve acquisition module 230 is used to extract the seismic data curvature curve from the sample label data set to obtain a curvature curve label data set.

[0164] In this embodiment, the curvature curve label data set includes the location information of the seed point and the residual curvatures corresponding to various sub-points.

[0165] The similar spectrum data acquisition module 240 is used to acquire the post-stack seismic data volume, extract similar spectrum data from the post-stack seismic data volume according to the residual curvature range, establish a corresponding relationship between the similar spectrum data and the seed point set, and obtain a similar spectrum data label data set.

[0166] In this embodiment, the similar spectrum data includes the residual curvature of the seed point. By establishing the correspondence between the similar spectrum data and the seed point set, the relationship between the position information of the seed point and the corresponding residual curvature can be obtained. Different from the curvature curve acquisition module 230, the residual curvature of the similar spectrum data acquisition module 240 is screened and extracted according to the residual curvature range, and the correspondence between the seed point position and the residual curvature can be obtained more specifically according to the residual curvature range.

[0167] The model training module 250 is used to input the curvature curve label data set and the similar spectrum data label data set into a neural network for learning, so as to obtain a neural network model for seismic data curvature picking.

[0168] In this embodiment, the curvature curve label data set and the similar spectrum data label data set are used as training sets and input into the neural network, the neural network is trained, the parameters of the neural network are determined, and the trained neural network is obtained. The trained neural network can be used to pick up the remaining curvature.

[0169] In one embodiment, the sample label data set acquisition module 210 includes:

[0170] A pre-stack seismic data volume acquisition module 211 is used to acquire a pre-stack seismic data volume;

[0171] A pre-stack seismic data volume slicing module 212 is used to slice the pre-stack seismic data volume according to the number of seismic traces and the number of sampling points to obtain a sliced ​​data volume;

[0172] The sample label data set generation module 213 is used to obtain the seed point of zero offset in the slice data volume along the sampling point direction in the 0th seismic trace direction, and use random velocity perturbation to make time-distance curves at the seed point of non-zero offset to generate the sample label data set.

[0173] In this embodiment, the number of seismic traces in the seismic data volume is N, and the number of sampling points is M. The seismic data volume is sliced ​​according to the number of N seismic traces and the number of M sampling points to obtain a sliced ​​data volume. The number of zero-offset seed points is obtained in the direction of the 0th seismic trace along the sampling point direction, and the time-distance curve is produced by random velocity perturbation at non-zero offset to generate a label data set.

[0174] In this embodiment, for the seismic reflection waveform sample data, the Ricker wavelet is convolved with the label data to construct a sample label data set, which is calculated by the following formula (1):

[0175] T(n,m)=F(m,v,n) n∈[0,N-1], v∈[-25%, 25%], m∈[0,M-1] (1)

[0176] Among them, T is the function expression of a curvature curve existing in the label data, and F is the rule for generating the curve, that is, the label data set is generated by perturbing the time-distance curve starting from the seed points n=0 to N-1 with a speed perturbation range of -25% to 25%.

[0177] In one embodiment, the sample label data set acquisition module 210 is used to acquire a pre-stack seismic data volume as the sample label data set; preprocess the sample label data set to obtain a processed sample label data set; the step of extracting a seismic data curvature curve from the sample label data set to obtain a curvature curve label data set includes: extracting a seismic data curvature curve from the processed sample label data set to obtain a curvature curve label data set.

[0178] In one embodiment, the sample label data set acquisition module 210 is used to identify the amplitude value distribution range of the pre-stack seismic data body, and truncate the pre-stack seismic data body based on the amplitude value distribution range; filter, grayscale process, and normalize the truncated seismic data body to obtain the processed sample label data set.

[0179] In this embodiment, the pre-stack seismic data volume is identified, the specific amplitude value distribution range of the earthquake number is determined, and the pre-stack seismic data volume is truncated by the following equations (2) and (3):

[0180] D_1(n,m)=Clip(D(n,m)) n∈[0,N-1], m∈[0,M-1] (2)

[0181] Clip(D(n,m))=if(D(n,m)>0.90*Max(D(n,m)))then D(n,m)

[0182] =0.90*Max(D(n,m)) (3)

[0183] Wherein, D(n, m) is seismic data, which is a slice of pre-stack seismic data gather, n is the trace number index, and m is the sampling index.

[0184] The truncated seismic data volume is filtered, gray-scale processed, and normalized using the following formula (4):

[0185] D_2(n,m)=SplitNorm(StructSmooth((D_1(n,m))))

[0186] n∈[0, N-1], m∈[0, M-1] (4)

[0187] Among them, SplitNorm adopts the piecewise normalization calculation along the sampling direction, and StructSmooth adopts the structural filtering technology to perform filtering and smoothing operations on the discontinuous seismic data event axes along the event axis direction.

[0188] In one embodiment, the similar spectrum data acquisition module is further used to determine the remaining curvature range according to the work area. In this embodiment, in order to screen out more remaining curvatures that meet the conditions of the work area, the remaining curvature range is determined according to the work area requirements, so as to extract the corresponding similar spectrum data.

[0189] In one embodiment, the similarity spectrum data acquisition module 240 is used to scan the stacked seismic data volume according to the remaining curvature range, calculate the similarity coefficient spectrum according to the following formula (5), and further extract seismic features:

[0190] RD(g,m)=Semblance(D_2(n,m)) n∈[0,N-1], m∈[0,M-1], g∈[g1,g2] (5)

[0191] Among them, Semblance (D_2 (n, m)) is the calculation method of seismic data similarity spectrum, g is the residual curvature scanning range;

[0192] The similarity spectrum is calculated using formula (6):

[0193]

[0194] Among them, T zx Indicates the offset is x halfoffset The remaining delay calculated at z0 is the time corresponding to zero bias, and γ is the residual curvature range.

[0195] In this embodiment, the residual curvature scanning range of equation (5) is [-0.5, 0.5].

[0196] In this embodiment, the seed point is selected by selecting the vertical depth position of the point with the maximum energy picked up on the similarity spectrum, so as to construct a corresponding relationship between the similarity spectrum data and the seed point set.

[0197] In one embodiment, the neural network is based on a convolutional neural network autoencoder as the main body, and a 29-layer deep neural network model is built based on tensorflow 2.0.

[0198] In this embodiment, considering that the most important feature of the pre-stack seismic data body or the post-stack seismic data body as a whole is its spatial position information, an autoencoder based on a convolutional neural network is used as the main body of this technology, and a 29-layer deep neural network model is built based on tensorflow 2.0. The deep neural network model includes seven convolution upsampling blocks and seven convolution downsampling blocks, each of which consists of two convolution layers, a BN layer, a downsampling / upsampling layer, and a LeakyRelu activation function layer. The convolution kernels used in the two convolution layers are 7×1 and 1×7, so as to obtain a larger spatial receptive field and reduce the amount of calculation.

[0199] In one embodiment, the model training module 250 is used to establish a training set using a curvature curve label data set and a similar spectrum data label data set to perform distributed training on the neural network; wherein, the error metric of the neural network uses the RMSE error functional based on L2 regularization as the objective function of the neural network, and the initialization method of each parameter of the convolution layer in the neural network adopts Xavier initialization. For the parameters of the convolution layer, the L2 regularization coefficient is set to 0.01, and the optimization algorithm adopts the Adam algorithm.

[0200] In this embodiment, the Adam algorithm is a first-order optimization algorithm that can replace the traditional stochastic gradient descent process. It designs independent adaptive learning rates for different parameters by calculating the first-order moment estimate and the second-order moment estimate of the gradient, and then iteratively updates the neural network weights. It has better convergence than the traditional stochastic gradient descent algorithm.

[0201] In this embodiment, the RMSE functional expression is shown in formula (7):

[0202]

[0203] Where E(x) is the objective function of the neural network, x is the model parameter of the neural network, m is the number of selected mini-batches, and y i (x) is the actual label data, is the predicted data, and γ is the regularization parameter. Further, the regularization parameter γ is 0.01.

[0204] In this embodiment, after obtaining the neural network model for seismic data curvature picking, the parameters of the trained neural network are saved to the local disk and deployed to a dedicated cluster for seismic data curvature picking to form a reliable seismic data curvature picking technology.

[0205] The specific definition of the seismic data curvature picking device can refer to the definition of the seismic data curvature picking method in the above text, which will not be repeated here. Each unit in the above seismic data curvature picking device can be implemented in whole or in part by software, hardware and a combination thereof. The above units can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above units.

[0206] Embodiment 5

[0207] In this embodiment, a computer device is provided. Its internal structure diagram can be shown as follows: Figure 3As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program, and the non-volatile storage medium is deployed with a database, which is used to store pre-stack seismic data bodies and post-stack seismic data bodies. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with other computer devices deployed with application software. When the computer program is executed by the processor, a method for picking up curvature of seismic data is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse.

[0208] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0209] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0210] Step 110 : Acquire a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records the reflection waveform position information.

[0211] In this embodiment, a seismic data volume is obtained as sample data, and the seismic data volume is a data volume of a pre-stack gather. The pre-stack seismic data volume contains a record of a reflection waveform. In this embodiment, the reflection waveform position corresponding to the reflection waveform is marked, and the marked seismic data volume is used as label data, thereby obtaining a sample label data set.

[0212] Step 120 : determining seed point information of a zero offset position in the sample label data set according to the reflected waveform position information, and acquiring a seed point set based on the seed point information of the zero offset position.

[0213] In this embodiment, the position of the first reflected waveform is the seed point of the zero offset position. In this step, the seed point information of the zero offset position in the sample label data set is determined according to the reflected waveform position information, thereby determining the position information of the first seed point, and based on the position information of the first seed point, the position information of other seed points can be obtained, thereby obtaining various sub-points in the sample label data set to obtain a seed point set.

[0214] Step 130: extracting the seismic data curvature curve from the sample label data set to obtain a curvature curve label data set.

[0215] In this embodiment, the curvature curve label data set includes the location information of the seed point and the residual curvatures corresponding to various sub-points.

[0216] Step 140, obtaining a post-stack seismic data volume, extracting similar spectrum data from the post-stack seismic data volume according to the residual curvature range, establishing a corresponding relationship between the similar spectrum data and the seed point set, and obtaining a similar spectrum data label data set.

[0217] In this embodiment, the similar spectrum data includes the residual curvature of the seed point, and the correspondence between the similar spectrum data and the seed point set is established, so that the relationship between the position information of the seed point and the corresponding residual curvature can be obtained. Different from step 130, the residual curvature of step 140 is screened and extracted according to the residual curvature range, and the correspondence between the seed point position and the residual curvature can be obtained more specifically according to the residual curvature range.

[0218] Step 150: Input the curvature curve label data set and the similar spectrum data label data set into a neural network for learning, so as to obtain a neural network model for seismic data curvature picking.

[0219] In this embodiment, the curvature curve label data set and the similar spectrum data label data set are used as training sets and input into the neural network, the neural network is trained, the parameters of the neural network are determined, and the trained neural network is obtained. The trained neural network can be used to pick up the remaining curvature.

[0220] Embodiment 6

[0221] In this embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0222] Step 110 : Acquire a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records the reflection waveform position information.

[0223] In this embodiment, a seismic data volume is obtained as sample data, and the seismic data volume is a data volume of a pre-stack gather. The pre-stack seismic data volume contains a record of a reflection waveform. In this embodiment, the reflection waveform position corresponding to the reflection waveform is marked, and the marked seismic data volume is used as label data, thereby obtaining a sample label data set.

[0224] Step 120 : determining seed point information of a zero offset position in the sample label data set according to the reflected waveform position information, and acquiring a seed point set based on the seed point information of the zero offset position.

[0225] In this embodiment, the position of the first reflected waveform is the seed point of the zero offset position. In this step, the seed point information of the zero offset position in the sample label data set is determined according to the reflected waveform position information, thereby determining the position information of the first seed point, and based on the position information of the first seed point, the position information of other seed points can be obtained, thereby obtaining various sub-points in the sample label data set to obtain a seed point set.

[0226] Step 130: extracting the seismic data curvature curve from the sample label data set to obtain a curvature curve label data set.

[0227] In this embodiment, the curvature curve label data set includes the location information of the seed point and the residual curvatures corresponding to various sub-points.

[0228] Step 140, obtaining a post-stack seismic data volume, extracting similar spectrum data from the post-stack seismic data volume according to the residual curvature range, establishing a corresponding relationship between the similar spectrum data and the seed point set, and obtaining a similar spectrum data label data set.

[0229] In this embodiment, the similar spectrum data includes the residual curvature of the seed point, and the correspondence between the similar spectrum data and the seed point set is established, so that the relationship between the position information of the seed point and the corresponding residual curvature can be obtained. Different from step 130, the residual curvature of step 140 is screened and extracted according to the residual curvature range, and the correspondence between the seed point position and the residual curvature can be obtained more specifically according to the residual curvature range.

[0230] Step 150: Input the curvature curve label data set and the similar spectrum data label data set into a neural network for learning, so as to obtain a neural network model for seismic data curvature picking.

[0231] In this embodiment, the curvature curve label data set and the similar spectrum data label data set are used as training sets and input into the neural network, the neural network is trained, the parameters of the neural network are determined, and the trained neural network is obtained. The trained neural network can be used to pick up the remaining curvature.

[0232] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0233] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0234] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for picking curvature of seismic data, characterized in that: include: Acquire a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records reflection waveform position information; Determine the seed point information of the zero offset position in the sample label data set according to the reflected waveform position information, and acquire a seed point set based on the seed point information of the zero offset position; Extracting a seismic data curvature curve from the sample label data set to obtain a curvature curve label data set; Acquire a post-stack seismic data volume, extract similar spectrum data from the post-stack seismic data volume according to the residual curvature range, establish a corresponding relationship between the similar spectrum data and the seed point set, and obtain a similar spectrum data label data set; The curvature curve label data set and the similar spectrum data label data set are input into a neural network for learning to obtain a neural network model for seismic data curvature picking.

2. The method according to claim 1, characterized in that The step of obtaining a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records the reflection waveform position information, comprises: Obtain pre-stack seismic data volume; Slicing the pre-stack seismic data volume according to the number of seismic traces and the number of sampling points to obtain a sliced ​​data volume; A seed point of zero offset in the slice data volume is obtained in the direction of the 0th seismic track along the sampling point direction, and a time-distance curve is produced by using random velocity disturbance at the seed point of non-zero offset to generate the sample label data set.

3. The method according to claim 1, characterized in that The step of obtaining a pre-stack seismic data volume as a sample label data set comprises: Acquire a pre-stack seismic data volume as the sample label data set; Preprocessing the sample label data set to obtain a processed sample label data set; The step of extracting the seismic data curvature curve from the sample label data set to obtain the curvature curve label data set comprises: The seismic data curvature curve is extracted from the processed sample label data set to obtain a curvature curve label data set.

4. The method according to claim 3, characterized in that The step of preprocessing the sample label data set to obtain a processed sample label data set includes: An amplitude value distribution range of a pre-stack seismic data volume is identified, and a truncation operation is performed on the pre-stack seismic data volume based on the amplitude value distribution range.

5. The method according to claim 4, characterized in that The identifying of the amplitude value distribution range of the pre-stack seismic data volume and performing a truncation operation on the pre-stack seismic data volume based on the amplitude value distribution range comprises: The truncated seismic data volume is filtered, gray-scale processed, and normalized to obtain the processed sample label data set.

6. The method according to claim 1, characterized in that The step of obtaining the post-stack seismic data volume also includes: The remaining curvature range is determined according to the work area.

7. A seismic data curvature picking device, characterized in that: include: A sample label data set acquisition module is used to acquire a pre-stack seismic data volume as a sample label data set, wherein the sample label data set records the reflection waveform position information; A seed point set acquisition module: used to determine the seed point information of the zero offset position in the sample label data set according to the reflected waveform position information, and acquire a seed point set based on the seed point information of the zero offset position; Curvature curve acquisition module: used to extract the seismic data curvature curve from the sample label data set to obtain the curvature curve label data set; Similar spectrum data acquisition module: used to acquire a post-stack seismic data volume, extract similar spectrum data from the post-stack seismic data volume according to the residual curvature range, establish a corresponding relationship between the similar spectrum data and the seed point set, and obtain a similar spectrum data label data set; Model training module: used to input the curvature curve label data set and the similar spectrum data label data set into the neural network for learning, so as to obtain a neural network model for seismic data curvature picking.

8. The device according to claim 7, characterized in that The sample label data set acquisition module includes: A pre-stack seismic data volume acquisition module is used to acquire pre-stack seismic data volumes; A pre-stack seismic data volume slicing module is used to slice the pre-stack seismic data volume according to the number of seismic traces and the number of sampling points to obtain a sliced ​​data volume; The sample label data set generation module is used to obtain the seed point of zero offset in the slice data volume in the direction of the 0th seismic track along the sampling point direction, use random velocity perturbation to make time-distance curves at the seed point of non-zero offset, and generate the sample label data set.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images