Transform-based abnormal amplitude suppression method and device

Abstract

The invention relates to the field of seismic exploration, and particularly discloses an abnormal amplitude suppression method and device based on Transform, and the method comprises the steps: carrying out the frequency division processing of training seismic shot gather data, and obtaining a plurality of data sets; screening abnormal amplitude channels from each data set according to frequency bands; extracting a training sample and a training label from the abnormal amplitude channel, sending the training sample and the training label to a Transform model for training, and establishing a Transform abnormal amplitude suppression model; and repeating seismic shot gather data training operation on the new seismic shot gather data to obtain new samples and labels, and sending the new samples and labels into the Transform abnormal amplitude suppression model to obtain trace data after abnormal amplitude suppression. According to the method, the abnormal amplitude suppression model based on Transform is formed through data conversion and establishment of a nonlinear relation between a shot data sequence and a non-abnormal amplitude channel data sequence, and abnormal amplitude suppression work can be rapidly carried out by calling the model.

Description

Technical Field

[0001] The present invention relates to the field of seismic exploration, and particularly to a method and device for suppressing abnormal amplitudes based on Transformer. Background Art

[0002] In seismic exploration, abnormally strong amplitude noise is widespread. This type of noise has strong energy, great destructiveness, a wide frequency distribution range, unstable energy, large amplitude differences, strong randomness, cannot use coherent signal enhancement processing, and has complex noise types. In order to better suppress abnormal noise, a lot of work has been done in the direction of seismic abnormal amplitude suppression by predecessors, mainly including artificial trace editing technology, frequency division suppression method, adaptive abnormal amplitude suppression method, and abnormal amplitude suppression method based on seismic wave attenuation, etc. However, these methods do not combine the data situation of the entire shot.

[0003] As one of the representative algorithms in NLP, Transformer plays an important role in natural language processing and even in the field of image processing. It can transform an input data sequence into another data sequence. For the suppression of abnormal amplitudes in seismic data, its suppression of abnormal amplitudes can be said to transform a sequence with abnormal amplitudes into a sequence without abnormal amplitudes. Therefore, it becomes possible to apply Transformer to the suppression of abnormal amplitudes.

[0004] Based on this technical background, the present invention studies a method and device for suppressing abnormal amplitudes based on Transformer. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a method and device for suppressing abnormal amplitudes based on Transformer. This method establishes a non-linear relationship between the shot data sequence and the non-abnormal amplitude trace data sequence through data conversion, thereby forming a Transformer-based abnormal amplitude suppression model. By calling this model, the suppression work of abnormal amplitudes can be quickly carried out, improving work efficiency.

[0006] To achieve the above object, the first aspect of the present invention provides a method for suppressing abnormal amplitudes based on Transformer, including:

[0007] Performing frequency division processing on the training seismic shot gather data to obtain multiple data sets;

[0008] Screening abnormal amplitude traces from each data set according to the frequency band;

[0009] Extracting training samples and training labels from the abnormal amplitude traces and sending them into the Transformer model for training to establish a Transformer abnormal amplitude suppression model;

[0010] Repeat the operation on the training seismic shot gather data for the new seismic shot gather data to obtain new samples and labels, and send the new samples and labels into the Transformer abnormal amplitude suppression model to obtain the trace data after suppressing the abnormal amplitude.

[0011] The second aspect of the present invention provides a Transformer-based abnormal amplitude suppression device, including:

[0012] A frequency division module for performing frequency division processing on the training seismic shot gather data to obtain a plurality of data sets;

[0013] A screening module for screening abnormal amplitude traces from each data set according to frequency bands;

[0014] A training module for extracting training samples and training labels from the abnormal amplitude traces and sending them into the Transformer model for training to establish a Transformer abnormal amplitude suppression model;

[0015] An amplitude suppression module for repeating the operation on the training seismic shot gather data for the new seismic shot gather data to obtain new samples and labels, and sending the new samples and labels into the Transformer abnormal amplitude suppression model to obtain the trace data after suppressing the abnormal amplitude.

[0016] The third aspect of the present invention provides an electronic device, and the electronic device includes:

[0017] A memory storing executable instructions;

[0018] A processor that runs the executable instructions in the memory to implement the Transformer-based abnormal amplitude suppression method described in the first aspect.

[0019] The fourth aspect of the present invention provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the Transformer-based abnormal amplitude suppression method described in the first aspect.

[0020] The beneficial effects of the present invention include:

[0021] (1) The Transformer-based abnormal amplitude suppression method proposed by the present invention establishes a non-linear relationship between the shot data sequence and the non-abnormal amplitude trace data sequence through data conversion, thereby forming a Transformer-based abnormal amplitude suppression model. By calling this model, the suppression work of abnormal amplitude can be quickly carried out, improving work efficiency.

[0022] (2) The abnormal amplitude suppression method based on Transformer proposed by the present invention introduces Transformer for training during the abnormal amplitude suppression process, and establishes a non-linear relationship between adjacent traces of the abnormal amplitude trace and all traces of the shot where this trace is located except this trace, so that when performing abnormal amplitude suppression, the situation of the entire shot can be fully considered, and during the learning and training, the weight of the influence of each trace on the result is adjusted, so that the effective suppression of abnormal amplitude is achieved under the overall consideration of the global situation.

[0023] Other features and advantages of the present invention will be described in detail in the following specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] By describing the exemplary embodiments of the present invention in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present invention will become more apparent.

[0025] Figure 1 It is a schematic flow chart of the abnormal amplitude suppression method based on Transformer proposed by the present invention.

[0026] Figure 2 It is a schematic flow chart of a specific implementation manner of the abnormal amplitude suppression method based on Transformer proposed by the present invention.

[0027] Figure 3 It is a schematic diagram of the Transformer model in a specific implementation manner of the abnormal amplitude suppression method based on Transformer proposed by the present invention. SPECIFIC IMPLEMENTATION MANNER

[0028] The following will describe the preferred embodiments of the present invention in more detail. Although the following describes the preferred embodiments of the present invention, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein.

[0029] The present invention provides an abnormal amplitude suppression method based on Transformer, as Figure 1 shown, including:

[0030] Performing frequency division processing on the training seismic shot gather data to obtain a plurality of data sets;

[0031] Screening abnormal amplitude traces from each data set according to frequency bands;

[0032] Extracting training samples and training labels from the abnormal amplitude traces and feeding them into the Transformer model for training to establish a Transformer abnormal amplitude suppression model;

[0033] Repeat the operation of training seismic shot gather data on the new seismic shot gather data to obtain new samples and labels, and send the new samples and labels into the Transformer abnormal amplitude suppression model to obtain the trace data after suppressing the abnormal amplitude.

[0034] In the present invention, through data conversion, a non-linear relationship is established between the shot data sequence and the non-abnormal amplitude trace data sequence, thereby forming an abnormal amplitude suppression model based on Transformer. By calling this model, the suppression of abnormal amplitude can be quickly carried out, improving the work efficiency.

[0035] According to the present invention, the screening method for screening abnormal amplitude traces in frequency bands is as follows:

[0036] Judge whether the ratio of the amplitude value of a certain trace in a certain data set to the amplitude value of the shot where this trace is located exceeds a specified threshold. If it exceeds the threshold, this trace is an abnormal amplitude trace.

[0037] Or the ratio of the average amplitude value of a certain shot to the overall amplitude average value exceeds a specified threshold. If it exceeds the threshold, all traces in this shot are abnormal amplitude traces.

[0038] According to the present invention, screening abnormal amplitude traces in frequency bands from each data set further includes:

[0039] Establish an index of the frequency band, shot number, and trace number where the abnormal amplitude trace is located.

[0040] According to the present invention, the training samples and training labels are two sequences;

[0041] The two sequences form a sample-label pair and are sent into the Transformer model.

[0042] Preferably, the training sample is a sequence composed of all traces in the shot where the abnormal amplitude trace is located except this trace;

[0043] The original label is the adjacent trace of the abnormal amplitude trace.

[0044] According to the present invention, the selection method for adjacent traces is as follows:

[0045] Search sequentially in the order of trace numbers. If the adjacent trace is also an abnormal amplitude trace, continue to search in the order of trace numbers. If all traces in the whole shot are abnormal amplitude traces, search sequentially in the order of shot numbers.

[0046] Preferably, the training is used to establish a non-linear relationship between the two sequences;

[0047] The training is iterative training;

[0048] The termination condition of the iterative training is that the Loss value reaches the requirement, or the number of training rounds reaches the specified number of rounds.

[0049] In the present invention, during the process of suppressing abnormal amplitudes, Transformer is introduced for training, and a non-linear relationship is established between the adjacent traces of the abnormal amplitude traces and all the traces of the shot where the trace is located except this trace. Thus, when suppressing abnormal amplitudes, the situation of the entire shot can be fully considered, and during the learning and training process, the weight of the influence of each trace on the result is adjusted. Therefore, under the overall consideration of the global situation, effective suppression of abnormal amplitudes is achieved.

[0050] The present invention will be described in more detail below through embodiments.

[0051] Embodiment 1:

[0052] As Figure 2 shown, this embodiment provides a method for suppressing abnormal amplitudes based on Transformer, and the specific steps are as follows:

[0053] Step 1: Obtain seismic shot gather data and perform frequency division processing on it; the data after frequency division is denoted as D 1 , D 2 ,..., D k ; where k is the number of frequency bands divided; for any one of the frequency bands in D 1 , D 2 ,..., D k , the data D i in it has D i = {S 1 , S 2 ,..., S p}. Among them, S 1 , S 2 ,..., S p is the data of each shot in D i ; for any one shot S 1 , S 2 ,..., S p in S i , there is S i = {T 1 , T 2 ,..., T q}, where T 1 , T 2 ,..., T q is the trace data of any one shot S i .

[0054] Step 2: For the data sets D 1 , D 2 ,..., D k after frequency division, calculate the average amplitude value of each frequency band data; denote any one shot S in the data set D i ​i The average amplitude value is A ii ; then, for the data set D i The average amplitude values of all the shots are {A i1 , A i2 ,..., A ip}; that is, for the data of one frequency band, an average amplitude value is calculated for each shot;

[0055] At this time, a double-layer loop is constructed. The outer loop loops through each frequency band, and the inner loop loops through each shot; for the data of each shot in the inner loop, its average amplitude value is read, and the relationship between the amplitude value of each trace of this shot and the average amplitude value is calculated;

[0056] ① If for any shot S i the average amplitude value is A ii , and the ratio of it to the average value of {A i1 , A i2 ,..., A ip} is greater than the specified threshold θ s , then all the traces of this shot are recorded as abnormal traces, and the traces are recorded and indexed with a triple (p n , s n , t n ), where p n is the frequency band number, s n is the shot number, and t n is the trace number;

[0057] ② If the ratio of the amplitude value of any trace in this shot to the average amplitude value of this shot is greater than the pre-set threshold θ, then this trace is recorded as an abnormal amplitude trace, and the corresponding frequency band identification number, shot number, and trace number of this trace are recorded, and the trace is recorded and indexed with a triple (p n , s n , t n ); thus, all the abnormal amplitude traces are obtained;

[0058] Step 3: For all the abnormal amplitude traces obtained in Step 2, find their adjacent traces as the data set required for constructing the Transformer algorithm;

[0059] ① When constructing the data set, the shot where the abnormal amplitude trace is located is used as the sample of this abnormal amplitude trace, and the non-abnormal amplitude trace adjacent to this abnormal amplitude trace is used as the sample label, and the two sets of data form a sample-label pair; assume that any shot set S i ={T 1 , T 2 ,..., T q}, and there is an abnormal amplitude trace T a , then the sample of this abnormal amplitude trace is: S i ={T 1,T 2 ,...,T a-1 ,T a+1 ,...,T q}, labeled as T a+1 ; By default, select the non-abnormal amplitude trace of the next trace of the abnormal amplitude trace as the label. If the abnormal amplitude trace itself is the last trace of the shot gather, retrieve forward; if the abnormal amplitude trace itself is the first trace of the shot gather, retrieve backward;

[0060] ② If the adjacent traces of the abnormal amplitude trace are also abnormal amplitude traces, then in ①, continue to iterate on both sides until finding the nearest non-abnormal amplitude trace to the abnormal amplitude trace, and thus form a sample-label pair with it; Assume any shot gather S i ={T 1 ,T 2 ,...,T q}, there is an abnormal amplitude trace T a , then the sample of this abnormal amplitude trace is: S i ={T 1 ,T 2 ,...,T a-1 ,T a+1 ,...,T q}, labeled as T a+2 ;

[0061] ③ If all the traces of a shot of data are abnormal amplitude traces, then select the trace data of the trace with the smaller trace number of the adjacent shot as the label trace data of the abnormal amplitude trace, thus forming a set of sample-label pairs;

[0062] It should be noted that for the generation of each sample-label pair, the trace data of the abnormal amplitude trace needs to be removed from the shot data where the sample is located;

[0063] Through the implementation of ①②③, a suitable sample-label set can be equipped for all abnormal amplitude traces; The reason for selecting this method to construct the data set is mainly considered as follows;

[0064] Select this method to construct the data set. Use all the data of the shot where the abnormal amplitude is located as the sample value, and use the non-abnormal amplitude traces in this shot or the non-abnormal amplitude traces of the adjacent shot as the label for training; In this way, the nonlinear relationship between the overall shot data of the shot where the abnormal trace is located and the non-abnormal amplitude trace can be learned, and then this nonlinear relationship is applied to the relationship between the shot data of the abnormal amplitude trace and the abnormal amplitude trace, so as to recalculate the trace data of the abnormal amplitude trace through the model constructed by this nonlinear relationship, thus realizing the generation of new abnormal amplitude traces, that is, the abnormal amplitude traces that need to be attenuated;

[0065] During the construction of the dataset, it was noted that the first-shot data was used as the sample value for a single trace. This approach can establish a non-linear relationship between the entire shot and a single trace of data, enabling the model to utilize the data from the entire shot when performing abnormal amplitude attenuation. As a result, the abnormal amplitude attenuation takes into account the overall situation of the entire shot, providing a more comprehensive consideration.

[0066] Step 4: Use the dataset constructed in Step 3 for the Transformer model set in this step. In terms of the encoder:

[0067] ① Convert the shot gather data into an n×1 matrix. That is, the original shot gather data is a t×q matrix, where t is the trace length in a single-shot data and q is the number of traces in the shot gather. Then n = t * q.

[0068] ② After converting the data into a one-dimensional sequence, the data can be used as the input for the Transformer algorithm. The data undergoes normalization, multi-head attention mechanism, residual addition, layer normalization, feed-forward neural network, and residual addition in the encoder and finally outputs the data processed by the encoder.

[0069] In terms of the decoder:

[0070] Compared with the decoder, the encoder has an additional module with a multi-head mechanism. The attention of this module needs to use the data output from the encoder to calculate the weight matrix in the encoder. Other processes are the same as those of the decoder. Finally, the decoder outputs probability values through the cross-entropy loss function and iterates repeatedly until the loss function reaches the set threshold or the number of training rounds meets the corresponding requirements.

[0071] Step 5: After being trained in Step 4, a Transformer model for suppressing abnormal amplitude can be formed. By invoking this model, it can be used for abnormal amplitude attenuation.

[0072] Step 6: After completing the above five steps, the training of the abnormal suppression model is completed. When suppressing the abnormal amplitude of new data, it is necessary to use the steps from Step 1 to Step 3 to construct the data, then invoke the model trained in Step 4, and then the trace data after amplitude suppression can be obtained. Thus, the work of abnormal amplitude suppression is completed.

[0073] The Transformer algorithm model used in this embodiment is as Figure 3 shown. Using the method of this embodiment, the rapid suppression of abnormal amplitude is achieved, improving work efficiency.

[0074] Embodiment 2:

[0075] This embodiment provides a method for suppressing abnormal amplitudes based on Transformer, as follows Figure 1 shown, including:

[0076] Performing frequency division processing on the training seismic shot gather data to obtain multiple data sets;

[0077] Screening abnormal amplitude channels from each data set according to frequency bands;

[0078] Extracting training samples and training labels from the abnormal amplitude channels and sending them into the Transformer model for training to establish a Transformer abnormal amplitude suppression model;

[0079] Repeating the operations on the training seismic shot gather data for the new seismic shot gather data to obtain new samples and labels, and sending the new samples and labels into the Transformer abnormal amplitude suppression model to obtain the channel data after suppressing the abnormal amplitudes;

[0080] The screening method for screening abnormal amplitude channels according to frequency bands is as follows:

[0081] Judging whether the ratio of the amplitude value of a certain channel in a certain data set to the amplitude value of the shot where this channel is located exceeds a specified threshold. If it exceeds the threshold, this channel is an abnormal amplitude channel.

[0082] Or judging whether the ratio of the average amplitude value of a certain shot to the overall average amplitude value exceeds a specified threshold. If it exceeds the threshold, all channels within this shot are abnormal amplitude channels;

[0083] Screening abnormal amplitude channels from each data set according to frequency bands further includes:

[0084] Establishing indexes of the frequency band, shot number, and channel number where the abnormal amplitude channels are located;

[0085] The training samples and training labels are two sequences;

[0086] The two sequences form a sample-label pair and are sent into the Transformer model;

[0087] The training sample is a sequence composed of all channels in the shot where the abnormal amplitude channel is located except this channel;

[0088] The original label is the adjacent channel of the abnormal amplitude channel;

[0089] The selection method for adjacent channels is as follows:

[0090] Searching in order according to the channel number size. If the adjacent channel is also an abnormal amplitude channel, continue searching in order according to the channel number size. If all channels in the shot are abnormal amplitude channels, search in order according to the shot number size;

[0091] Training is used to establish the non-linear relationship between the two sequences.

[0092] The training is iterative training;

[0093] The termination condition of the iterative training is that the Loss value reaches the requirement or the number of training rounds reaches the specified number of rounds.

[0094] Embodiment 3:

[0095] This embodiment provides an abnormal amplitude suppression device based on Transformer, including:

[0096] A frequency division module for performing frequency division processing on the training seismic shot gather data to obtain multiple data sets;

[0097] A screening module for screening abnormal amplitude channels from each data set according to the frequency band;

[0098] A training module for extracting training samples and training labels from the abnormal amplitude channels and sending them into the Transformer model for training to establish a Transformer abnormal amplitude suppression model;

[0099] An amplitude suppression module for repeating the operations on the training seismic shot gather data for the new seismic shot gather data to obtain new samples and labels, and sending the new samples and labels into the Transformer abnormal amplitude suppression model to obtain the channel data after suppressing the abnormal amplitude;

[0100] The screening method for screening abnormal amplitude channels according to the frequency band is as follows:

[0101] Judge whether the ratio of the amplitude value of a certain channel in a certain data set to the amplitude value of the shot where this channel is located exceeds the specified threshold. If it exceeds the threshold, this channel is an abnormal amplitude channel.

[0102] Or whether the ratio of the average amplitude value of a certain shot to the overall amplitude average value exceeds the specified threshold. If it exceeds the threshold, all channels in this shot are abnormal amplitude channels;

[0103] Screening abnormal amplitude channels from each data set according to the frequency band further includes:

[0104] Establishing indexes of the frequency band, shot number, and channel number where the abnormal amplitude channel is located.

[0105] According to the present invention, the training samples and training labels are two sequences;

[0106] The two sequences form a sample-label pair and are sent into the Transformer model;

[0107] The training sample is a sequence composed of all channels in the shot where the abnormal amplitude channel is located except this channel;

[0108] The original label is the adjacent channel of the abnormal amplitude channel;

[0109] The selection method for adjacent channels is as follows:

[0110] Search sequentially in the order of channel numbers. If the adjacent channel is also an abnormal amplitude channel, continue to search in the order of channel numbers. If all channels in a whole shot are abnormal amplitude channels, search sequentially in the order of shot numbers;

[0111] Training is used to establish a non-linear relationship between two sequences;

[0112] The training is iterative training;

[0113] The termination condition of the iterative training is that the Loss value meets the requirements or the number of training rounds reaches the specified number of rounds.

[0114] Embodiment 4:

[0115] An embodiment of the present invention provides an electronic device including a memory and a processor,

[0116] The memory stores executable instructions;

[0117] The processor runs the executable instructions in the memory to implement the abnormal amplitude suppression method based on Transformer.

[0118] This memory is used to store non-temporary computer-readable instructions. Specifically, the memory may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0119] The processor may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In an embodiment of the present invention, the processor is used to run the computer-readable instructions stored in the memory.

[0120] Those skilled in the art should understand that, in order to solve the technical problem of how to obtain good user experience effects, this embodiment may also include well-known structures such as communication buses, interfaces, etc., and these well-known structures should also be included in the protection scope of the present invention.

[0121] For a detailed description of this embodiment, reference may be made to the corresponding descriptions in the foregoing embodiments, and details will not be repeated here.

[0122] Embodiment 5:

[0123] An embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements an abnormal amplitude suppression method based on Transformer.

[0124] The computer-readable storage medium according to the embodiment of the present invention stores non-temporary computer-readable instructions. When the non-temporary computer-readable instructions are run by a processor, all or part of the steps of the methods of the various embodiments of the present invention described above are executed.

[0125] The above computer-readable storage medium includes but is not limited to: optical storage media (such as CD-ROM and DVD), magneto-optical storage media (such as MO), magnetic storage media (such as magnetic tapes or external hard drives), media with built-in rewritable non-volatile memory (such as memory cards), and media with built-in ROM (such as ROM cartridges).

[0126] The abnormal amplitude suppression method based on Transformer proposed by the embodiment of the present invention establishes a non-linear relationship between the shot data sequence and the non-abnormal amplitude trace data sequence through data conversion, thereby forming an abnormal amplitude suppression model based on Transformer. By calling this model, the suppression of abnormal amplitude can be quickly carried out, improving the work efficiency.

[0127] The various embodiments of the present invention have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A method for suppressing abnormal amplitudes based on Transformer, characterized in that, it includes: Performing frequency division processing on the training seismic shot gather data to obtain multiple data sets; Screening abnormal amplitude traces from each data set according to frequency bands; Extracting training samples and training labels from the abnormal amplitude traces and feeding them into the Transformer model for training to establish a Transformer abnormal amplitude suppression model; Repeating the operations on the training seismic shot gather data for the new seismic shot gather data to obtain new samples and labels, and feeding the new samples and labels into the Transformer abnormal amplitude suppression model to obtain the trace data after suppressing the abnormal amplitudes.

2. The method according to claim 1, characterized in that, The screening method for screening abnormal amplitude traces according to frequency bands is: Judging whether the ratio of the amplitude value of a certain trace in a certain data set to the amplitude value of the shot where this trace is located exceeds a specified threshold. If it exceeds the threshold, this trace is an abnormal amplitude trace, or the ratio of the average amplitude value of a certain shot to the overall amplitude average value exceeds a specified threshold. If it exceeds the threshold, all traces within this shot are abnormal amplitude traces.

3. The method according to claim 2, characterized in that, Screening abnormal amplitude traces from each data set according to frequency bands further includes: Establishing indexes of the frequency band, shot number, and trace number where the abnormal amplitude trace is located.

4. The method according to claim 1, characterized in that, The training samples and training labels are two sequences; The two sequences form a sample-label pair and are fed into the Transformer model.

5. The method according to claim 4, characterized in that, The training sample is a sequence composed of all traces in the shot where the abnormal amplitude trace is located except this trace; The original label is the adjacent trace of the abnormal amplitude trace.

6. The method according to claim 5, characterized in that, The selection method for the adjacent trace is: Searching in sequence according to the trace number size. If the adjacent trace is also an abnormal amplitude trace, continue searching according to the trace number size. If all traces in the shot are abnormal amplitude traces, search in sequence according to the shot number size.

7. The method according to claim 5, characterized in that, The training is used to establish a non-linear relationship between the two sequences; The training is iterative training; The termination condition for the iterative training is that the Loss value reaches the requirement, or the number of training rounds reaches the specified number of rounds.

8. A device for suppressing abnormal amplitudes based on Transformer, characterized in that, it includes: A frequency division module, used for performing frequency division processing on the training seismic shot gather data to obtain multiple data sets; A screening module, used for screening abnormal amplitude traces from each data set according to frequency bands; A training module, used for extracting training samples and training labels from the abnormal amplitude traces and feeding them into the Transformer model for training to establish a Transformer abnormal amplitude suppression model; An amplitude suppression module, which is used to repeat the operations on the training seismic shot gather data for the new seismic shot gather data to obtain new samples and labels, and send the new samples and labels into the Transformer abnormal amplitude suppression model to obtain the trace data after suppressing the abnormal amplitude.

9. An electronic device, characterized in that, the electronic device includes: a memory storing executable instructions; a processor that runs the executable instructions in the memory to implement the Transformer-based abnormal amplitude suppression method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the Transformer-based abnormal amplitude suppression method according to any one of claims 1-7.

Images