Method and device for suppressing multiple refracted waves

Abstract

The invention discloses a method and a device for suppressing multiple refracted waves. The method comprises the following steps: acquiring actual seismic data in a to-be-processed seismic work area; carrying out preprocessing operation of nonlinear leveling on the actual seismic data; inputting the pre-processed seismic data into a pre-trained denoising model to obtain the seismic data after refraction wave suppression; and performing data recovery processing on the seismic data after refraction wave suppression to obtain de-noised seismic data.

Description

Technical Field

[0001] This article relates to the field of oil and gas geophysical exploration technology, and in particular to a method and device for suppressing multiple refracted waves. Background Technology

[0002] Given the numerous challenges faced by domestic oil and gas exploration, including complex surface conditions, intricate geological conditions, and concealed oil and gas reservoirs, recovering effective seismic data from complex raw seismic data has become a crucial foundational task for overcoming these difficulties. Multiple refraction waves are one of the main interference waves in single-shot records, severely interfering with the effective reflected waves of the target layer. Furthermore, the similar velocities of far-off gathers make them difficult to remove through post-processing, thus affecting the quality of seismic data.

[0003] Traditional methods for suppressing multiple-reflection waves primarily rely on the frequency and velocity differences between the interfering and effective waves to separate them, such as FK filtering, pull transform, and KL transform. While these methods are effective to some extent, they tend to suppress both the effective and interfering waves when they are spatially overlapping, and their processing speed is low. Recently, the FX-domain Cadzow filtering method based on random misalignment has been developed, which can extract multiple-reflection waves more completely than other traditional filtering methods while maintaining the relative horizontality of the phase axes of the multiple-reflection and refracted waves. However, it suffers from significant time consumption when processing large amounts of data.

[0004] The difficulty and low efficiency of suppressing multiple-refracted waves make it an urgent problem to develop an intelligent method for suppressing multiple-refracted waves. Summary of the Invention

[0005] This application provides a method and apparatus for suppressing multiple refracted waves. By preprocessing actual seismic data, this application utilizes a denoising model with a dual-domain, multi-scale, dual-constraint network to effectively suppress multiple refracted waves in actual seismic data, which can significantly improve denoising efficiency and accuracy.

[0006] In a first aspect, this application provides a method for suppressing multiple refracted waves, the method comprising:

[0007] Acquire actual seismic data from the seismic work area to be processed; perform nonlinear flattening preprocessing on the actual seismic data; input the preprocessed seismic data into a pre-trained denoising model to obtain seismic data after refracted wave suppression; perform data recovery processing on the seismic data after refracted wave suppression to obtain denoised seismic data.

[0008] Secondly, embodiments of the present invention also provide an apparatus for suppressing multiple refracted waves, characterized in that the apparatus includes: a memory and a processor; the memory is used to store a program for suppressing multiple refracted waves, and the processor is used to read and execute the program for suppressing multiple refracted waves, and execute the method described in any one of the above embodiments.

[0009] Thirdly, embodiments of the present invention also provide a computer-readable storage medium storing a data processing program, wherein the data processing program is executed by a processor using the method for suppressing multiple refracted waves as described in any of the above embodiments.

[0010] Compared with related technologies, this application provides a method and apparatus for suppressing multiple refracted waves. The method includes: acquiring actual seismic data from the seismic area to be processed; performing a nonlinear flattening preprocessing operation on the actual seismic data; inputting the preprocessed seismic data into a pre-trained denoising model to obtain seismic data with suppressed refracted waves; and performing data recovery processing on the seismic data with suppressed refracted waves to obtain denoised seismic data. This application, through preprocessing the actual seismic data and utilizing a dual-domain, multi-scale, dual-constraint network denoising model, effectively suppresses multiple refracted waves in the actual seismic data, significantly improving denoising efficiency and accuracy.

[0011] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. Other advantages of this application can be realized and obtained by means of the solutions described in the description and the accompanying drawings. Attached Figure Description

[0012] The accompanying drawings are used to provide an understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0013] Figure 1 This is a flowchart of a method for suppressing multiple refracted waves according to an embodiment of this application;

[0014] Figure 2 This is a schematic diagram of a device for suppressing multiple refracted waves according to an embodiment of this application;

[0015] Figure 3 This is a schematic diagram illustrating the method of picking up the theoretical refracted wave propagation travel timeline in some exemplary embodiments;

[0016] Figure 4 This is a schematic diagram illustrating the method of picking up theoretical surface wave propagation travel time lines in some exemplary embodiments;

[0017] Figure 5 This is a schematic diagram of seismic data after nonlinear flattening in some exemplary embodiments;

[0018] Figure 6 This is a schematic diagram of a dual-domain, multi-scale, dual-constraint network structure in some exemplary embodiments;

[0019] Figure 7 This is a schematic diagram of raw seismic data in some exemplary embodiments;

[0020] Figure 8 This is a schematic diagram showing the use of FK filtering in some exemplary embodiments;

[0021] Figure 9 This is a schematic diagram showing the use of Cadzow filtering in some exemplary embodiments;

[0022] Figure 10 This is an example of using data after suppressing multiple refracted waves in some exemplary embodiments. Detailed Implementation

[0023] This application describes several embodiments, but these descriptions are exemplary and not restrictive, and it will be apparent to those skilled in the art that many more embodiments and implementations are possible within the scope of the embodiments described herein. Although many possible combinations of features are shown in the drawings and discussed in the detailed description, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with, or may replace, any feature or element of any other embodiment.

[0024] This application includes and contemplates combinations of features and elements known to those skilled in the art. The embodiments, features, and elements disclosed in this application may also be combined with any conventional features or elements to form a unique inventive scheme as defined by the claims. Any feature or element of any embodiment may also be combined with features or elements from other inventive schemes to form another unique inventive scheme as defined by the claims. Therefore, it should be understood that any feature shown and / or discussed in this application may be implemented individually or in any suitable combination. Therefore, the embodiments are not limited except by the limitations imposed by the appended claims and their equivalents. Furthermore, various modifications and changes may be made within the scope of the appended claims.

[0025] Furthermore, in describing representative embodiments, the specification may have presented methods and / or processes as a specific sequence of steps. However, the method or process should not be limited to the specific order of steps described herein, to the extent that it does not depend on such a specific order. As will be understood by those skilled in the art, other sequences of steps are also possible. Therefore, the specific order of steps set forth in the specification should not be construed as a limitation of the claims. Moreover, the claims concerning the method and / or process should not be limited to the steps performed in the written order, and those skilled in the art will readily understand that these orders can be varied and still remain within the spirit and scope of the embodiments of this application.

[0026] This invention provides a method for suppressing multiple refracted waves, such as... Figure 1 As shown, the method includes steps S100-S130:

[0027] S100: Obtain actual seismic data in the seismic work area to be processed;

[0028] S110: Perform a nonlinear flattening preprocessing operation on the actual seismic data;

[0029] S120: Input the preprocessed seismic data into the pre-trained denoising model to obtain the seismic data after refracted wave suppression;

[0030] S130: Perform data recovery processing on the seismic data after the refracted wave suppression to obtain denoised seismic data.

[0031] In one exemplary embodiment, the preprocessing operation of nonlinear flattening of the actual seismic data includes: travel time stretching correction or inter-trace cross-correlation stretching correction; depending on the actual seismic data, the nonlinear flattening preprocessing operation can be selected, and there are three processing methods:

[0032] First method: Only perform timeline stretching correction;

[0033] The second method involves only performing inter-track cross-correlation stretching correction.

[0034] The third approach involves using both the running time line stretching correction and the inter-track correlation stretching correction. In this third case, the running time line stretching correction must be performed first, followed by the inter-track correlation stretching correction.

[0035] In this embodiment, the process of travel time line stretching correction is as follows: calculate the travel time line stretching correction amount for each actual seismic data, and perform stretching correction on the actual seismic data for that channel based on the travel time line stretching correction amount.

[0036] The process of inter-trace cross-correlation stretching correction is as follows: calculate the inter-trace cross-correlation stretching correction amount for each actual seismic data trace, and perform stretching correction on the actual seismic data of that trace based on the inter-trace cross-correlation stretching correction amount.

[0037] In one exemplary embodiment, the formula for calculating the travel time stretching correction for each actual seismic data trace is as follows:

[0038]

[0039] Where i represents the i-th seismic trace. This represents the stretching correction amount for the i-th track. This represents the theoretical travel time of the refracted wave in the i-th channel. This represents the minimum theoretical travel time of refracted waves in all seismic traces.

[0040] In one exemplary embodiment, calculating the inter-trace cross-correlation stretching correction for each actual seismic data trace includes:

[0041] Calculate the inter-trace cross-correlation value between each actual seismic data trace and adjacent traces; where the inter-trace cross-correlation value is:

[0042]

[0043] Among them, R i,i+1 [k] represents the inter-trace cross-correlation value between the i-th and (i+1)-th seismic traces, d i [n] represents the nth sampling point in the i-th seismic data, d i+1 [n+k] represents the (n+k)th sampling point in the (i+1)th seismic trace, where k is the time delay and n represents the time sampling sequence number of the seismic trace.

[0044] The inter-track cross-correlation tensile correction amount is determined based on the calculated inter-track cross-correlation value;

[0045] Wherein, the inter-channel cross-correlation tensile correction amount is:

[0046]

[0047] In the above formula, This represents the cross-correlation stretching correction amount for the i-th channel. Indicates when cross-correlation R i,i+1 The time delay corresponding to reaching the maximum value.

[0048] In one exemplary embodiment, after the preprocessing operation of nonlinear flattening on the actual seismic data, the method further includes:

[0049] The actual seismic data is subjected to seismic trace randomization processing, wherein the seismic trace randomization processing includes:

[0050] Using earthquake arrangement as the basic calculation unit, obtain the sequence number of all traces in each arrangement;

[0051] Randomize the track numbers;

[0052] The seismic traces are rearranged according to their disordered trace numbers to obtain randomized seismic trace data.

[0053] In one exemplary embodiment, the denoising model includes:

[0054] FX domain channel network, shared network, signal branch network and noise branch network.

[0055] In one exemplary embodiment, the denoising model employs a dual-constraint loss function:

[0056]

[0057] Where l(Θ) is the double-constraint loss function, K is the total number of samples in the dataset, k is the sample index, and Net(d k ;Θ s Net(d) represents the data output after passing through the signal branching network. k ;Θ n ) represents the data output through the noisy branch network, d k s is the training sample for the original seismic data. k For noise-removed seismic data training samples (or signal data training samples), n k Training samples for noisy data, This represents the square of the Frobenius norm, where μ is the balance coefficient, μ∈[0,1], and the default value is 0.5.

[0058] In one exemplary embodiment, the step of inputting the preprocessed seismic data into a pre-trained denoising model to obtain seismic data after refracted wave suppression includes:

[0059] The preprocessed seismic data is input into a pre-trained denoising model, and the output is... and

[0060] According to the output and Determine the data after noise suppression for:

[0061]

[0062] In the above formula, This is the data after noise suppression. The seismic signal data output by the network. λ represents the noisy data output by the network, which is an adjustable parameter with a default value of 0.5; d represents the noisy seismic data.

[0063] In one exemplary embodiment, the step of performing data recovery processing on the seismic data after the refracted wave suppression to obtain denoised seismic data includes:

[0064] Step 1: Obtain the track number in the randomization process;

[0065] The second step is to match the track number in the randomization process with the original track number.

[0066] The third step is to rearrange the seismic traces of the refracted wave suppressed seismic data according to the original sequence number to obtain the denoised seismic data.

[0067] In one exemplary embodiment, the step of performing data recovery processing on the seismic data after the refracted wave suppression to obtain denoised seismic data further includes:

[0068] Based on the travel time stretching correction for each actual seismic data, all sampling points of the seismic data are shifted downwards for reverse stretching correction according to the travel time stretching correction.

[0069] Based on the inter-trace cross-correlation stretching correction of each actual seismic data, all sampling points of the seismic data are moved up and down according to the absolute value of the inter-trace cross-correlation stretching correction to perform reverse stretching correction.

[0070] If the inter-track cross-correlation tensile correction is negative, it will be shifted downwards; if the inter-track cross-correlation tensile correction is positive, it will be shifted upwards.

[0071] The intelligent multiple refraction wave suppression method implemented in this embodiment has the following technical effects:

[0072] Through the entire process of constructing a training dataset of multiple refracted waves, designing and training a dual-domain, multi-scale, dual-constraint network, and applying the network (noise suppression), and by employing key technologies such as nonlinear flattening correction and seismic trace randomization, the generalization and denoising performance of the network are greatly improved. Compared with conventional techniques, this technique can significantly improve denoising efficiency and denoising accuracy.

[0073] Secondly, embodiments of the present invention also provide a device for suppressing multiple refracted waves, such as... Figure 2 As shown, the device includes a memory 200 and a processor 210; the memory is used to store a program for suppressing multiple refracted waves, and the processor is used to read and execute the program for suppressing multiple refracted waves, and to execute the method described in any of the above embodiments.

[0074] Thirdly, embodiments of the present invention also provide a computer-readable storage medium storing a data processing program, wherein the data processing program is executed by a processor using the method for suppressing multiple refracted waves as described in any of the above embodiments.

[0075] Example 1

[0076] This example demonstrates the training process of a denoising model using a dual-domain, multi-scale, dual-constraint network. The specific training process is as follows:

[0077] Step 1: Construct a training dataset for a denoising model based on a dual-domain, multi-scale, dual-constraint network.

[0078] The denoising model is based on a dual-domain, multi-scale, dual-constraint network. The training dataset for the denoising model includes simulated data, object model data, and actual historical processed data of suppressed refracted waves (referred to as actual data). The data is organized in the form of "noisy data - noisy data" pairs. The specific construction steps are as follows:

[0079] Step 1: Data Preprocessing

[0080] Step 1.1 Nonlinear tensile correction

[0081] Step 1.1.1 Calculate the theoretical travel time of refracted wave propagation

[0082] Based on the given seismic data, two points (tr1, ti1) and (tr2, ti2) are obtained along the first refracted wave, where tr is the seismic recording time and ti is the trace number. For example... Figure 3 As shown, Figure 3 The first data point is located at the leftmost "cross" position (tr1, ti1); Figure 3 The second data point is located at the leftmost "cross" position (tr2, ti2).

[0083] Based on the characteristics of seismic data records, the coordinates of the excitation point and receiver point of a trace can be retrieved in the trace header by the trace number. Here, the coordinates of the excitation point are denoted as (sx1, sy1) and (sx2, sy2), and the coordinates of the receiver point are denoted as (rx1, ry1) and (rx2, ry2).

[0084] Calculate the offsets x1 and x2 using the following formulas:

[0085]

[0086] Where (sx,sy) and (rx,ry) represent the coordinates of the excitation point and the receiver point, respectively.

[0087] Calculate time t0 and velocity v based on (x1, ti1) and (x2, ti2) respectively, using the following formulas:

[0088]

[0089] Where x and t i x1 represents the offset and seismic recording time corresponding to a certain point, x2 represents the offset of the first data point, x1 represents the offset of the second data point, ti1 represents the seismic recording time of the first data point, and ti2 represents the seismic recording time of the second data point.

[0090] Obtain the coordinates of the excitation and receiving points of all channels, and then obtain the offset distance of the channel. Calculate the time of t0 and velocity v calculated in the above steps and the time point of all channels using formula (2). The line connecting the time points of all channels is called the theoretical refracted wave propagation travel time line.

[0091] Step 1.1.2 Calculate the stretching correction amount of the travel time line.

[0092] Calculate the travel line stretch correction for track i using the following formula:

[0093]

[0094] Where i represents the i-th seismic trace. This represents the stretching correction amount for the i-th track. This represents the theoretical travel time of the refracted wave in the i-th channel. This represents the minimum theoretical travel time of refracted waves in all seismic traces. Figure 3 The highest point of the black line is represented in the middle.

[0095] Step 1.1.3 Calculate the inter-channel cross-correlation.

[0096] Due to the complexity of surface conditions in field seismic acquisitions, relying on travel time stretching corrections can only achieve trend corrections. Corrections for some minor jitter require calculation of cross-correlation corrections.

[0097] Arranged by earthquakes, such as Figure 3 As shown and Figure 4 As shown, a "mountain" shaped data is a permutation. Figure 3 and Figure 4 There are a total of 4 complete permutations. Using a "mountain"-shaped data point as the basic calculation unit, the seismic trace with the smallest theoretical refracted wave travel time in the permutation is taken as the standard trace, which is labeled as 0. The traces to the left are numbered -1, -2, ..., and the traces to the right are numbered 1, 2, 3, ... . The cross-correlation between trace i and trace i+1 is then expressed as:

[0098]

[0099] Among them, R i,i+1 [k] is the cross-correlation value between channel i and channel i+1, d i [n] represents the nth sampling point in the i-th seismic data, d i+1 [n+k] represents the (n+k)th sampling point in the (i+1)th seismic trace, where k is the time delay and n represents the time sampling sequence number of the seismic trace.

[0100] Step 1.1.4 Calculate the inter-channel cross-correlation tensile correction amount

[0101] Calculate the inter-pass cross-correlation tensile correction amount using the following formula:

[0102]

[0103] in, This represents the cross-correlation stretching correction amount for the i-th channel. Indicates when cross-correlation R i,i+1 The time delay corresponding to reaching the maximum value.

[0104] Step 1.1.5 Calculate the theoretical surface wave travel time.

[0105] The theoretical surface wave travel time is calculated to shield surface waves and prevent them from being suppressed by multiple refracted waves.

[0106] like Figure 4 As shown, based on the given seismic data, two points are obtained along the first surface wave. The remaining steps are the same as in step 1.1.1. Based on the given seismic data, two points (tr1, ti1) and (tr2, ti2) are obtained along the surface wave, where tr is the seismic recording time and ti is the trace number. According to the characteristics of seismic data recording, in the standard segy file, the coordinates of the excitation point and receiver point of a trace can be retrieved from the trace header using the trace number. Here, the coordinates of the excitation point are denoted as (sx1, sy1) and (sx2, sy2), and the coordinates of the receiver point are denoted as (rx1, ry1) and (rx2, ry2).

[0107] Calculate the offsets x1 and x2 using the following formulas:

[0108]

[0109] Where (sx,sy) and (rx,ry) represent the coordinates of the excitation point and the receiver point, respectively.

[0110] Calculate time t0 and velocity v based on (x1, ti1) and (x2, ti2) respectively, using the following formulas:

[0111]

[0112] Where x and t i x1 represents the offset and seismic recording time corresponding to a certain point, x2 represents the offset of the first data point, x1 represents the offset of the second data point, ti1 represents the seismic recording time of the first data point, and ti2 represents the seismic recording time of the second data point.

[0113] Obtain the coordinates of the excitation and receiving points of all channels, and then obtain the offset distance of the channel. Calculate the time of t0 and velocity v calculated in the above steps and the time point of all channels using formula (2). The line connecting the time points of all channels is called the theoretical surface wave propagation travel time line.

[0114] Step 1.1.6 Shielding Surface Wave Region

[0115] Set the region below the theoretical surface wave propagation travel time line to zero.

[0116] Step 1.1.7 Perform nonlinear stretching correction on the seismic data.

[0117] In this step, firstly, based on the travel time stretching correction calculated in step 1.1.2, stretching correction is performed on the seismic data one channel at a time. The correction method is to move all sampling points of the seismic data of that channel upwards, and the upward movement amount is the travel time stretching correction amount.

[0118] Secondly, based on the inter-trace cross-correlation stretching correction calculated in step 1.1.4, stretching correction is performed on the seismic data for each trace. The correction method involves moving all sampling points of the seismic data for that trace vertically, with the moving distance equal to the absolute value of the inter-trace cross-correlation stretching correction. If the inter-trace cross-correlation stretching correction is negative, the points are moved upwards; otherwise, they are moved downwards. Note: If the surface is relatively flat (such as in desert areas), inter-trace cross-correlation stretching correction can be omitted to save time. The corrected effect is as follows: Figure 5 The seismic data shown is after nonlinear flattening.

[0119] Step 1.2 Seismic Trace Randomization Processing

[0120] After nonlinear stretching correction in step 1.1, the refracted waves are close to horizontal. Further randomization processing of the seismic traces is then performed, with the specific steps as follows:

[0121] ① Arranged by earthquake ( Figure 3 In the text, a "mountain" shaped data is arranged as one permutation. Figure 3 (A total of 4 complete permutations) are used as basic calculation units to obtain the sequence numbers of all channels in the permutation.

[0122] ② Randomly shuffle the track numbers;

[0123] ③ Rearrange the seismic traces according to the new serial numbers;

[0124] Since the refracted waves are close to horizontal after multiple refractions, they remain horizontal after randomization. However, other seismic signals become random pulses in the trace direction, which increases the difference in signal-to-noise characteristics and reduces the difficulty of network signal-to-noise separation.

[0125] Step 2: Create the training dataset

[0126] ① Select data

[0127] After performing all the data preprocessing steps in step 1 on all data pairs (noisy data and denoised data) from all data sources, ensure that the sampling length and total number of channels of the two data pairs are completely consistent. According to the given dataset data patch size (pt, px), randomly set j and m, select channels j to j+px-1 from the seismic data, and then randomly select sampling points m to m+pt-1 from the above channel range to form a matrix of size pt×px. The data patch selected in the noisy data is denoted as the noisy sample patch d, and the data patch selected in the denoised data is denoted as the signal sample patch s.

[0128] ② Calculate the noise sample size n according to the following formula:

[0129] d = s + n

[0130] ③ A set of data containing a set of {d; s, n} is called a sample set; d is noisy seismic data, s is signal data, and n is noise data.

[0131] ④ Given the number of samples n, the default value is 30000; repeat steps ①-③ until the number of samples meets n.

[0132] ⑤ Given a partition ratio 'a', for example, a = 0.2, the dataset is partitioned into 80% as the training set and 20% as the validation set.

[0133] Step 3: Establish a dual-domain, multi-scale, dual-constraint network model

[0134] Step 3.1 Model Architecture

[0135] like Figure 6 As shown, the frequency spatial domain multi-scale dual-constraint network consists of an FX domain channel network, a shared network, a signal branch network, and a noise branch network.

[0136] The FX (frequency space) domain channel network consists of FXT (FX transform layer), FM (feature mapping layer), GAM attention mechanism module and IFXT (inverse FX transform layer);

[0137] The shared network employs a multi-scale network structure, consisting of four feature extraction (FE) layers and four feature reconstruction (FM) layers. To avoid feature loss, skip connections are added between feature maps of the same scale.

[0138] The signal branch network and the noise branch network have the same structure, consisting of 4 feature mapping layers and 1 output layer;

[0139] Among them, the FX (frequency space) domain channel network and the shared network are connected in series, and the signal branch network and the noise branch network are connected in parallel and then connected in series with the above network.

[0140] The FE feature extraction layer includes convolutional layers and downsampling layers. The convolutional layers consist of five groups of convolutions with 128 channels and a filter size of 3. Each group of convolutions contains the ReLU activation function. The downsampling layer consists of a group of convolutional layers with 128 channels, a filter size of 3, and a stride of 2.

[0141] The FM feature reconstruction layer includes an upsampling layer and a convolutional layer. The convolutional layer consists of five groups of convolutional layers with 128 channels and a filter size of 3. Each group of convolutional layers contains a ReLU activation function. The upsampling layer consists of a group of deconvolutional layers with 128 channels, a filter size of 3, and a stride of 2.

[0142] The output layer consists of a set of convolutional layers with 128 input channels, 2 output channels, and a filter size of 3.

[0143] Step 3.2 Model Parameters

[0144] Set a double-constraint loss function:

[0145]

[0146] Where l(Θ) is the double-constraint loss function, K is the total number of samples in the dataset, k is the sample index, and Net(d k ;Θ s Net(d) represents the data output after passing through the signal branching network. k ;Θ n ) represents the data output through the noisy branch network, d k s is the training sample for the original seismic data. k For noise-removed seismic data training samples (or signal data training samples), n k Training samples for noisy data, This represents the square of the Frobenius norm, where μ is the balance coefficient, μ∈[0,1], and the default value is 0.5.

[0147] Step 4: Randomly initialize the parameters of the dual-domain, multi-scale, dual-constraint network to obtain the initial model.

[0148] Step 5: Train a dual-domain, multi-scale, dual-constraint network model

[0149] The training set of the earthquake data is input into the initial model of the dual-domain multi-scale dual-constraint network to obtain the data of the output layer. The loss function shown in Equation (7) is calculated, and it is determined whether the current iteration number meets the maximum training number (the default value is 50). If not, the network parameters of the current network are adjusted by the backpropagation algorithm until the maximum training number is reached, and the network model that has reached the training number is determined as the dual-domain multi-scale dual-constraint network model. The backpropagation algorithm of the dual-domain multi-scale dual-constraint network is the process of obtaining the optimal network parameters by minimizing the loss function shown in Equation (7). The minimization process of the objective function (Equation (7)) can be achieved by the Adam optimization algorithm.

[0150] Step 6: Test the trained model using test samples.

[0151] The test samples of the seismic data are input into the denoising model in each iteration of step 5. The loss function shown in formula (7) and the signal-to-noise ratio snr shown in formula (8) are calculated. The performance of the denoising model can be evaluated based on the obtained signal-to-noise ratio snr.

[0152] in,

[0153] In the above formula, K1 is the size of the test set. This refers to the denoising result after inputting the k-th sample in the sample set into the network, i.e. λ is an adjustment parameter, as above. The meanings of other symbols are the same as in Formula 7.

[0154] Example 2

[0155] This example uses actual seismic data collected in a basin (such as...). Figure 7 The high energy of multiple refracted waves causes severe interference, making it impossible to accurately image shallow layers and affecting further processing and interpretation of seismic data. The specific steps for denoising using the seismic data denoising model based on a dual-domain, multi-scale, dual-constraint network constructed in this application are as follows:

[0156] The first step is to perform preprocessing operations on the actual seismic data, including nonlinear stretching correction and seismic trace randomization.

[0157] The nonlinear stretching correction process specifically includes:

[0158] Step 1. Calculate the theoretical travel time of refracted wave propagation.

[0159] Step 2. Calculate the stretching correction amount for the running time line.

[0160] Step 3. Calculate the inter-channel cross-correlation.

[0161] Step 4. Calculate the inter-track cross-correlation tensile correction.

[0162] Step 5. Calculate the theoretical surface wave travel time.

[0163] Step 6. Shielding the surface wave region

[0164] Step 7. Perform nonlinear stretching correction on the seismic data.

[0165] The process of randomizing seismic traces is as follows:

[0166] ① Arranged by earthquake ( Figure 3 In the text, a "mountain" shaped data is arranged as one permutation. Figure 3 (A total of 4 complete permutations) are used as basic calculation units to obtain the sequence numbers of all channels in the permutation.

[0167] ② Randomly shuffle the track numbers;

[0168] ③ Rearrange the seismic traces according to the new serial numbers;

[0169] Since the refracted waves are close to horizontal after multiple refractions, they remain horizontal after randomization. However, other seismic signals become random pulses in the trace direction, which increases the difference in signal-to-noise characteristics and reduces the difficulty of network signal-to-noise separation.

[0170] The second step is to input the preprocessed seismic data into a denoising model using a dual-domain, multi-scale, dual-constraint network to obtain the output of the denoising model. and And obtain the noise-suppressed data. for:

[0171]

[0172] Step 3: Data Recovery

[0173] In this step, the noise-suppressed data obtained in the second step is subjected to derandomization and anti-nonlinear flattening to obtain the final noise-suppressed seismic data.

[0174] The steps for derandomization are as follows:

[0175] ① Record the randomly shuffled path numbers, and then record and match them with the original path numbers;

[0176] ② Rearrange the seismic traces according to their original serial numbers.

[0177] The processing steps for anti-nonlinear flattening are as follows:

[0178] ① Based on the travel time stretching correction, reverse stretching correction is performed on the seismic data for each channel. The correction method is to shift all sampling points of the seismic data for that channel downwards, and the amount of downward shift is the travel time stretching correction.

[0179] ② Based on the inter-track cross-correlation stretching correction, perform reverse stretching correction on the seismic data for each track. The correction method is to move all sampling points of the seismic data for that track up or down by a distance equal to the absolute value of the inter-track cross-correlation stretching correction. If the inter-track cross-correlation stretching correction is negative, move it down; otherwise, move it up. Figure 7 This is the original single-shot record. Figure 8 This is a schematic diagram of single-shot seismic data after applying FK filtering. Figure 9 This is a schematic diagram of single-shot seismic data after applying Cadzow filtering. Figure 10 This is an illustration of data obtained after suppressing multiple refracted waves. From Figure 8 , 9 The single-shot seismic data after filtering using three different methods (e.g., 10) clearly demonstrate that the denoising model using the dataset constructed in this invention and the dual-domain multi-scale dual-constraint network, after training, can effectively identify and suppress multiple refracted waves (e.g., ...). Figure 10 Shallow seismic signals were recovered, and the suppression effect was superior to the FK filtering method. Figure 8 ) and Cadzow filtering method ( Figure 9 Actual data tests show that, compared with the FK filtering method, the multiple refraction wave suppression method of this application can improve the denoising accuracy by 50%; compared with the Cadzow filtering technique, the multiple refraction wave suppression method of this application improves the processing efficiency by nearly 10 times in the model application stage while maintaining the same denoising accuracy.

[0180] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

Claims

1. A method of suppressing multiple refracted waves, characterized by, The method comprises: acquiring actual seismic data in a seismic work area to be processed; performing a nonlinear flattening preprocessing operation on the actual seismic data; inputting the preprocessed seismic data into a pre-trained denoising model to obtain seismic data after refraction wave suppression; performing data recovery processing on the seismic data after refraction wave suppression to obtain denoised seismic data.

2. The method for suppressing multiple refraction waves according to claim 1, wherein the nonlinear flattening preprocessing operation comprises traveltime line stretching correction or intertrace cross-correlation stretching correction. The process of the traveltime line stretching correction is as follows: calculating the traveltime line stretching correction amount of each trace of the actual seismic data, and performing stretching correction on the actual seismic data according to the traveltime line stretching correction amount. The process of the intertrace cross-correlation stretching correction is as follows: calculating the intertrace cross-correlation stretching correction amount of each trace of the actual seismic data, and performing stretching correction on the actual seismic data according to the intertrace cross-correlation stretching correction amount.

3. The method for suppressing multiple refraction waves according to claim 2, wherein the calculation formula of the traveltime line stretching correction amount is as follows:

4. The method for suppressing multiple refraction waves according to claim 2, wherein the calculation of the intertrace cross-correlation stretching correction amount of each trace of the actual seismic data comprises: calculating the intertrace cross-correlation value of each trace of the actual seismic data and an adjacent trace; where i represents the i-th seismic trace, represents the stretch correction amount of the traveltime curve of the i-th trace, represents the theoretical refracted wave propagation traveltime curve of the i-th trace, represents the minimum value of the theoretical refracted wave propagation traveltime curve in all seismic traces. determining the intertrace cross-correlation stretching correction amount according to the calculated intertrace cross-correlation value. The intertrace cross-correlation value is as follows: The intertrace cross-correlation stretching correction amount is as follows:

5. The method for suppressing multiple refraction waves according to claim 1, wherein after the nonlinear flattening preprocessing operation on the actual seismic data, the method further comprises: performing seismic trace randomization processing on the actual seismic data. In the above formula, R i,i+1 [k] is the cross-correlation value between the i-th trace and the i+1-th trace of seismic data, d i [n] represents the n-th sampling point in the i-th trace of seismic data, d i+1 [n+k] represents the n+k-th sampling point in the i+1-th trace of seismic data, k is the time delay, and n represents the time sampling serial number of the seismic trace. The process of the seismic trace randomization processing is as follows: In the above formula, denotes the cross-correlation stretch correction amount of the i-th channel, denotes the cross-correlation R i,i+1 the time delay corresponding to the maximum value. taking a seismic array as a calculation unit, acquiring all trace numbers in each seismic array; performing random sorting processing on all trace numbers; in each seismic array, rearranging seismic traces according to the trace numbers after random sorting to obtain seismic trace data after randomization processing.

6. The method for suppressing multiple refraction waves according to claim 1, wherein the denoising model comprises: an FX domain channel network, a shared network, a signal branch network, and a noise branch network.

7. The method for suppressing multiple refraction waves according to claim 6, wherein the denoising model adopts a double-constraint loss function:

8. The method for suppressing multiple refraction waves according to claim 1, wherein the inputting of the preprocessed seismic data into the pre-trained denoising model to obtain seismic data after refraction wave suppression comprises:

9. The method for suppressing multiple refraction waves according to claim 5, wherein the data recovery processing on the seismic data after refraction wave suppression to obtain denoised seismic data comprises: acquiring the trace numbers after randomization processing; matching the trace numbers after randomization processing and the original trace numbers; rearranging the seismic data after refraction wave suppression according to the original trace numbers to obtain denoised seismic data. ​ Where l(Θ) is the double-constraint loss function, K is the total number of samples in the dataset, k is the sample index, and Net(d k ;Θ s Net(d) represents the data output by the signal branch network. k ;Θ n ) represents the data output by the noisy branch network, d k s is the training sample for the original seismic data. k For training samples of noise-removed seismic data, n k Training samples for noisy data, This represents the square of the Frobenius norm, where μ is the balance coefficient, μ∈[0,1], and the default value is 0.

5. ​ ​ inputting the preprocessed seismic data into the pre-trained denoising model to output seismic signal data and noise data According to the output seismic signal data and noise data determining noise-suppressed seismic data The noise-suppressed seismic data is: In the above formula, is the seismic data after noise suppression, is the seismic signal data output by the network, is the noise data output by the network, and λ is an adjustment parameter and d is seismic data. ​ ​ ​ ​ ​ 10. The method for suppressing multiple refracted waves according to claim 2, characterized in that, The step of performing data recovery processing on the seismic data after the refracted wave suppression to obtain denoised seismic data further includes: Based on the travel time stretching correction for each actual seismic data, reverse stretching correction is performed on all sampling points of the seismic data for each track according to the travel time stretching correction. Based on the inter-trace cross-correlation stretching correction of each actual seismic data, reverse stretching correction is performed on all sampling points of the seismic data for each trace according to the absolute value of the inter-trace cross-correlation stretching correction. If the inter-track cross-correlation tensile correction is negative, then shift downwards; If the inter-channel cross-correlation stretching correction is positive, then it will shift upwards.

11. Apparatus for suppressing multiple refracted waves, characterized by The device includes a memory and a processor; the memory is used to store a program for suppressing multiple refracted waves, and the processor is used to read and execute the program for suppressing multiple refracted waves, performing the method according to any one of claims 1-10.

12. A computer-readable storage medium storing a data processing program, the data processing program being executed by a processor as the method for suppressing multiple refracted waves according to any one of claims 1-10.

Images