CN116413786A - Seismic wave abnormal first arrival correction method and device and related equipment - Google Patents
Seismic wave abnormal first arrival correction method and device and related equipment Download PDFInfo
- Publication number
- CN116413786A CN116413786A CN202111647719.0A CN202111647719A CN116413786A CN 116413786 A CN116413786 A CN 116413786A CN 202111647719 A CN202111647719 A CN 202111647719A CN 116413786 A CN116413786 A CN 116413786A
- Authority
- CN
- China
- Prior art keywords
- arrival
- abnormal
- wave
- neural network
- seismic data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 162
- 238000000034 method Methods 0.000 title claims abstract description 108
- 238000012937 correction Methods 0.000 title claims abstract description 41
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 82
- 238000003062 neural network model Methods 0.000 claims abstract description 18
- 238000013528 artificial neural network Methods 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims description 86
- 238000005070 sampling Methods 0.000 claims description 32
- 238000010801 machine learning Methods 0.000 claims description 21
- 238000004590 computer program Methods 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 9
- 238000003860 storage Methods 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 2
- 238000004422 calculation algorithm Methods 0.000 abstract description 18
- 230000003042 antagnostic effect Effects 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 14
- 230000006870 function Effects 0.000 description 11
- 238000001514 detection method Methods 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 238000007796 conventional method Methods 0.000 description 2
- 238000005520 cutting process Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 238000012805 post-processing Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 241000282326 Felis catus Species 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/30—Analysis
- G01V1/303—Analysis for determining velocity profiles or travel times
- G01V1/305—Travel times
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/36—Effecting static or dynamic corrections on records, e.g. correcting spread; Correlating seismic signals; Eliminating effects of unwanted energy
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Remote Sensing (AREA)
- General Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Environmental & Geological Engineering (AREA)
- Geology (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geophysics (AREA)
- Theoretical Computer Science (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention discloses a seismic wave abnormal first arrival correction method, a seismic wave abnormal first arrival correction device and related equipment, wherein the method can comprise the following steps: picking up the collected single shot seismic data based on an energy ratio method to determine first arrival waves of the single shot seismic data; the first arrival wave is input into a pre-trained deep neural network model comprising a classified convolutional neural network and an opposite convolutional neural network, wherein the classified convolutional neural network outputs an abnormal first arrival wave, and the abnormal first arrival wave is input into the opposite convolutional neural network to determine a corrected first arrival wave. The invention adopts an energy ratio algorithm to carry out primary picking up first arrivals on low signal-to-noise ratio seismic data, eliminates abnormal first arrivals by using a classification network, combines the data from which the abnormal first arrivals are eliminated with manually modified first arrival data to generate an antagonistic neural network, corrects the abnormal first arrivals by using a trained generation network model, and improves the first arrival wave picking-up accuracy and the channel number.
Description
Technical Field
The invention relates to the technical field of seismic exploration data processing, in particular to a seismic wave abnormal first arrival correction method, a seismic wave abnormal first arrival correction device and related equipment.
Background
The wavefront of the seismic wave first reaches an observation point at which the particles of the medium begin to vibrate, and the recorded wave is called the first arrival wave, or the first arrival (time) of the wave. Up to now, the research and application of the first-arrival picking technology has been in the past 40 years, and many students have studied different first-arrival automatic picking methods from different angles and have made great progress. Many first-arrival wave pickup algorithms have been proposed, largely divided into conventional methods and machine learning methods. The conventional method generally detects the first arrival according to the attribute difference of the signal and the noise, and can be further divided into a classical algorithm and a combination of classical algorithms.
Classical algorithms mainly include long-short window average energy ratio (STA/LTA) (Stevenson, 1976), energy ratio method (Coppens et al, 1985), fractal dimension (Boschett et al, 1996), kurtosis criterion method (Tselentis et al, 2011;Saragiotis et al, 2002;Lois,et al.2013), correlation method (Gu Hanming et al, 1992; molyneux & Schmitt,1999; peraldi & clemente, 2006;), polarization method; further Thoma and Stewart uses grid templates for accurate analysis of first arrival times. For high signal-to-noise ratio seismic data, a single algorithm can be used for picking up first arrivals well, but a large amount of noise is often present in actually acquired data, and the first arrivals are not easy to pick up.
Some scholars transform the time domain information into other domains and pick it up again using classical methods (Mousa & Alshuhail, gu Ruisheng et al, 2015; li Juan et al, 2019;Cheng,Yao,et al,2019). Mousa and Alshuhail use t-p transforms on energy vs. seismic recordings to make the first arrival wave more pronounced before picking up. And further, as adding preprocessing to enhance the signal-to-noise ratio (Song Longlong et al, 2019), performing time difference correction (Wei Meng et al, 2017; yu Zhichao et al, 2019), adding post-processing to perform accurate pickup (root happiness et al, 2008; he Xianlong et al, 2016; liu Yuan et al, 2019). Zhu Quanjie et al (2018) added the boundary detection factor and stability factor constraint by improving the traditional sliding time window energy method to achieve accurate pick-up of the waveform arrival time of "take-off blurriness, take-off not crisp and background interference too much".
In recent decades, machine learning algorithms have been proposed and widely used in a variety of problems, including first-arrival picking, where neural network algorithms have developed heat in recent years. Mccormack et al propose to pick up first arrivals using BNN; inputting the seismic three-component data into a 17-layer acceptance depth network model on cotyledons and the like, and outputting arrival time information; jiang Yiran et al propose to pick up first arrivals automatically using a support vector machine; hu, lianlia et al pick up first stops using a split network U-net; wu, hao et al trained the codec convolutional neural network to predict the first arrival of a microseismic event;
Any automatic pick-up algorithm does not fully meet the criteria, and requires manual intervention (sabbiene & Velis, 2010). Some students propose an abnormal first arrival rejection and correction algorithm, such as the method of the traditional method of picking up first arrivals by Zhongkeda Duan Xudong et al (Xudong Duan et al, 2018), and then inputting five seismic data into CNN at the same time, and classifying the picking results into two types: good picking or bad picking, and providing two correction strategies, wherein the first is to correct by selecting seeds and using adjacent channel cross-correlation algorithm; the second is to iteratively send the first arrival of bad movement into the convolutional neural network for detection, and bad pick-up is continuously reduced in the process. The university of labra sabione and Velis (sabione & Velis, 2010) first uses three methods of modified Coppens, entropy-based algorithm and modified fractal dimension algorithm to find the approximate first arrival location, then uses EPS algorithm to enhance the property change, then uses least squares to incorporate the common sense factors of manual correction into the auto pick algorithm, limits the first arrival to the vicinity of these lines, and finally uses simple criteria to correct or discard the wrong pick. Pengyu Yuan et al at houston university (Pengyu Yuan et al, 2019) uses a simplified U-net network, replaces cross entropy with lovasz loss, picks up first arrivals, and proposes a post-processing method of closest point pick up. Duan Xudong et al (Xudong Duan et al, 2019) 2019 have obtained advanced properties by existing methods (long-short window ratio), introduced 3 more channel properties considering spatial relationships, and optimized the model by SVM. Yaniv Hollander et al (Yaniv Hollander et al, 2019) devised a deep neural network to classify the presence or absence of first-arrival waves in an orbital sliding window. The exact location of the pick is then found within the boundaries of the identified window by calculating the energy ratio attribute and capturing its maximum. Akram et al (Akram et al, 2019) combine the traditional approach with a machine learning approach to propose optimizing the performance of the STA/LTA approach by testing different input forms and different feature functions of the three-component waveform data, including the proposed k-mean feature function.
Disclosure of Invention
With the wide application of the high-density high-efficiency acquisition technology in complex areas of oil fields, the first-arrival wave pickup work of mass data with low signal to noise ratio faces great challenges. The inventor finds that the traditional automatic first arrival wave picking method is poor in noise resistance, a large amount of man-machine interaction is needed to modify abnormal first arrivals, the picking precision and efficiency are affected by the large amount of interaction, the first arrival wave picking takes about one third of the whole processing period, and the oil and gas exploration and development process is severely restricted. Meanwhile, the inventor finds that the current automatic first-arrival wave pickup method is ideal for high signal-to-noise ratio data, but under the condition of low signal-to-noise ratio, the pickup precision is not ideal, a large amount of man-machine interaction is needed for modification, and the requirement of real-time processing of the current data cannot be met.
Therefore, for data with a low signal-to-noise ratio and a large data volume, a study on an abnormal first arrival correction method is necessary.
The present invention has been made in view of the above problems, and it is an object of the present invention to provide a seismic anomaly first arrival correction method, apparatus and related devices that overcome or at least partially solve the above problems.
In a first aspect, an embodiment of the present invention provides a method for correcting an abnormal first arrival of a seismic wave, which may include:
Picking up the collected single shot seismic data based on an energy ratio method to determine first arrival waves of the single shot seismic data;
inputting the first arrival wave into a pre-trained deep neural network model comprising a classified convolutional neural network and an opposite convolutional neural network, wherein the classified convolutional neural network outputs an abnormal first arrival wave, and inputting the abnormal first arrival wave into the opposite convolutional neural network to determine a corrected first arrival wave.
Optionally, after determining the first arrival wave of the single shot seismic data, the method further includes:
performing file format conversion on the first arrival wave of the single shot seismic data, and converting the first arrival wave data format into text format data;
and converting the text format data into image format data.
Optionally, the method further comprises: dividing the first arrival wave of the single shot seismic data according to the preset number of seismic channels and the preset number of sampling points to determine a plurality of slices.
Optionally, the picking up the collected single shot seismic data based on the energy ratio method to determine a first arrival wave of the single shot seismic data includes:
determining the position of a peak of the first arrival wave based on the position of the sampling point with the maximum energy ratio;
Determining the sampling time of the peak position of the first arrival wave as the first arrival time so as to determine the first arrival wave of the single shot seismic data;
or alternatively, the first and second heat exchangers may be,
determining the position of the sampling point with the energy ratio larger than a preset threshold value as the arrival position of the first arrival wave;
and determining the sampling time of the arrival position of the first arrival wave as the first arrival time so as to determine the first arrival wave of the single shot seismic data.
In a second aspect, an embodiment of the present invention provides a training method for a machine learning model, which may include:
acquiring a training sample set, wherein each sample in the training sample set comprises first arrival waves of single shot seismic data, abnormal first arrivals in the first arrival waves and first arrival waves after correction processing of the abnormal first arrivals;
training a deep neural network model comprising a classified convolutional neural network and an opposite convolutional neural network by using the sample degree in the training sample set; the first arrival wave of the single shot seismic data is input to the abnormal first arrival output in the classified convolution neural network, and identification parameter estimation is carried out; and inputting the abnormal first arrival into the convolutional neural network for correction parameter estimation.
Optionally, the method for judging whether the first arrival wave of the single shot seismic data is an abnormal first arrival may include:
And judging whether the first arrival wave of the seismic channel is an abnormal first arrival or not based on waveform continuous variation trend and/or waveform energy trend of the first arrival wave between adjacent channels.
Optionally, the determining whether the first arrival wave of the seismic trace is an abnormal first arrival based on the waveform continuous variation trend and/or the waveform energy trend of the first arrival wave between adjacent traces may include:
dividing the first arrival wave of the single shot seismic data according to the preset number of seismic channels and the preset number of sampling points to obtain a plurality of slices;
judging whether each first arrival wave in the slice is an abnormal first arrival or not according to the waveform continuous change trend and/or waveform energy trend of the first arrival wave between adjacent channels in the plurality of seismic channels contained in the slice and the comparison result of the preset sampling point position threshold;
and determining whether the section is the abnormal first arrival according to the comparison result of the proportion of the abnormal first arrivals of all the seismic channels contained in the section and a preset proportion threshold value.
Optionally, the process of correcting the abnormal first arrival may include the following steps:
removing the abnormal first arrival;
determining a fitting line of the first arrival wave of the single shot seismic data according to the waveform continuous variation trend and/or the waveform energy trend of the first arrival waves of the abnormal first arrival adjacent seismic channels;
And modifying the eliminated abnormal first arrival by taking the fitting line as a reference to obtain a modified first arrival wave.
Optionally, the acquiring a training sample set may include:
and screening out the single shot seismic data according to a preset shot interval based on the signal-to-noise ratio of the seismic data.
In a third aspect, an embodiment of the present invention provides a method for identifying an abnormal first arrival of a seismic wave, which may include:
picking up the collected single shot seismic data based on an energy ratio method to determine first arrival waves of the single shot seismic data;
inputting the first arrival wave into a pre-trained convolutional neural network model to identify abnormal first arrivals in the first arrival wave of the single shot seismic data.
Optionally, the convolutional neural network model is pre-trained by:
acquiring a training sample set, wherein each sample in the training sample set comprises a first arrival wave of single shot seismic data and an abnormal first arrival in the first arrival wave;
and training the convolutional neural network model by using the sample degree in the training sample set, wherein the first arrival wave of the single shot seismic data is input to the abnormal first arrival output in the convolutional neural network, and identification parameter estimation is performed.
In a fourth aspect, an embodiment of the present invention provides a seismic wave abnormal first arrival correction apparatus, which may include:
the pickup module is used for picking up the collected single shot seismic data based on an energy ratio method so as to determine first arrival waves of the single shot seismic data;
the determining module is used for inputting the first arrival wave into a pre-trained deep neural network model comprising a classified convolutional neural network and an opposite convolutional neural network, wherein the classified convolutional neural network outputs an abnormal first arrival wave, and the abnormal first arrival wave is input into the opposite convolutional neural network to determine the corrected first arrival wave.
In a fifth aspect, an embodiment of the present invention provides a training apparatus for a machine learning model, which may include:
the acquisition module is used for acquiring a training sample set, wherein each sample in the training sample set comprises first arrival waves of single shot seismic data, abnormal first arrival waves in the first arrival waves and first arrival waves after correction processing of the abnormal first arrival waves;
the training module is used for training a deep neural network model comprising a classified convolutional neural network and an anti-convolutional neural network by using the sample degree in the training sample set; the first arrival wave of the single shot seismic data is input to the abnormal first arrival output in the classified convolution neural network, and identification parameter estimation is carried out; and inputting the abnormal first arrival into the convolutional neural network for correction parameter estimation.
In a sixth aspect, an embodiment of the present invention provides a device for identifying an abnormal first arrival of a seismic wave, which may include:
the pickup module is used for picking up the collected single shot seismic data based on an energy ratio method so as to determine first arrival waves of the single shot seismic data;
the identification module is used for inputting the first arrival wave into a pre-trained convolutional neural network model so as to identify abnormal first arrival in the first arrival wave of the single shot seismic data.
In a seventh aspect, an embodiment of the present invention provides a computer readable storage medium, on which a computer program is stored, the program, when executed by a processor, implementing the method for correcting seismic wave anomalies first arrivals as described in the first aspect, or implementing the method for training a machine learning model as described in the second aspect, or implementing the method for identifying seismic wave anomalies first arrivals as described in the third aspect.
In an eighth aspect, an embodiment of the present invention provides a computer device, including a memory, a processor and a computer program stored in the memory and capable of running on the processor, where the processor implements the method for correcting seismic wave anomalies first arrivals according to the first aspect, the training method for a machine learning model according to the second aspect, or the method for identifying seismic wave anomalies first arrivals according to the third aspect when executing the program.
The technical scheme provided by the embodiment of the invention has the beneficial effects that at least:
the embodiment of the invention provides a seismic wave abnormal first arrival correction method, a seismic wave abnormal first arrival correction device and related equipment, wherein the method can comprise the following steps: picking up the collected single shot seismic data based on an energy ratio method to determine first arrival waves of the single shot seismic data; the first arrival wave is input into a pre-trained deep neural network model comprising a classified convolutional neural network and an opposite convolutional neural network, wherein the classified convolutional neural network outputs an abnormal first arrival wave, and the abnormal first arrival wave is input into the opposite convolutional neural network to determine a corrected first arrival wave. The embodiment of the invention adopts an energy ratio algorithm to carry out primary picking up of the seismic data with low signal to noise ratio, eliminates abnormal first arrivals by using a classification network, combines the data with the manually modified first arrival data to generate an antagonistic neural network, corrects the abnormal first arrivals by using a trained generation network model, and improves the accuracy and the number of channels of the first arrival wave picking up.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description and drawings.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention. In the drawings:
FIG. 1 is a flow chart of a training method of a machine learning model provided in embodiment 1 of the present invention;
FIG. 2 is an example of single shot seismic data provided in embodiment 1 of the present invention;
FIG. 3 is an example of a designed first arrival time window provided in embodiment 1 of the present invention;
FIG. 4 is a specific flowchart of the implementation of step S11 provided in embodiment 1 of the present invention;
FIG. 5 is a schematic view of a first-arrival slice of a determination abnormality provided in example 1 of the present invention;
FIG. 6 is a flowchart showing the step S44 provided in embodiment 1 of the present invention;
FIG. 7 is a schematic diagram of the embodiment 1 of the present invention after eliminating abnormal first arrival;
FIG. 8 is a schematic diagram of the modified first arrival provided in example 1 of the present invention;
FIG. 9 is a schematic diagram of a classified convolutional neural network provided in example 1 of the present invention;
FIG. 10 is a schematic diagram of an antagonistic convolutional neural network provided in example 1 of the present invention;
Fig. 11 is a schematic diagram of a generator G network provided in embodiment 1 of the present invention;
fig. 12 is a diagram of the architecture of the arbiter D network provided in embodiment 1 of the present invention;
FIG. 13 is a schematic structural view of a training device for machine learning model provided in embodiment 1 of the present invention;
FIG. 14 is a flowchart of a seismic wave anomaly first arrival correction method provided in embodiment 2 of the present invention;
FIG. 15 is a schematic structural diagram of a seismic wave anomaly first arrival correction device according to embodiment 2 of the present invention;
FIG. 16 is a flow chart of the method for identifying seismic anomalies in first arrival provided in embodiment 3 of the present invention;
fig. 17 is a schematic structural diagram of a seismic wave abnormal first arrival identification apparatus provided in embodiment 3 of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
Example 1
In embodiment 1 of the present invention, a training method of a machine learning model is provided, and referring to fig. 1, the method may include the following steps:
step S11, a training sample set is obtained, and each sample in the training sample set comprises first arrival waves of single shot seismic data, abnormal first arrival waves in the first arrival waves and first arrival waves after correction processing is carried out on the abnormal first arrival waves.
The single shot seismic data refer to seismic data acquired when one excitation point is excited. The first arrival wave refers to a time when the wavefront of a seismic wave reaches a certain seismic observation point (detection point) in seismic exploration, and at the observation point, the moment when the wave detector detects particle vibration is called as a first arrival time of the wave, which is abbreviated as a first arrival.
Referring to fig. 1, which is an example of single shot seismic data provided in the embodiment of the invention, when a first-shot wave is picked up, a person skilled in the art designs a time window of the first-shot wave to limit the first-shot wave within a local range, so as to improve the accuracy of picking up, and referring to the first-shot wave time window design example shown in fig. 3, the first-shot wave of the single shot seismic data is picked up within the local range more accurately.
Step S12, training a deep neural network model comprising a classified convolutional neural network and an opposite convolutional neural network by using the sample degree in the training sample set; the first arrival wave of the single shot seismic data is input to the abnormal first arrival output by the classified convolution neural network, and identification parameter estimation is carried out; the abnormal first arrival is input into the convolutional neural network for correction parameter estimation.
According to the training method of the machine learning model, provided by the embodiment of the invention, based on the first arrival wave of the single shot seismic data, the abnormal first arrival in the first arrival wave and the samples of the first arrival wave after the correction processing is carried out, the deep neural network model comprising the classified convolutional neural network and the opposite convolutional neural network is trained, and the identification parameters and the correction parameters are determined. The neural network model provided by the embodiment of the invention can accurately predict the types of first arrivals (normal first arrivals and abnormal first arrivals), improves the accuracy of identifying the abnormal first arrivals, and resists the generation of the corrected first arrivals by the generator network in the convolutional neural network and the discrimination of the first arrivals by the identifier network to judge the true and false, and iterates until the identifier network considers the first arrivals generated by the generator network to be the true first arrivals. The correction of the abnormal first arrival of the low signal to noise ratio is effectively realized, and the picking accuracy and the track number are obviously improved.
In the step S11 of the embodiment of the present invention, referring to fig. 4, the process of obtaining the training sample set mainly includes the following steps:
step S41, obtaining seismic data. The above-mentioned seismic data in this step are seismic data that are excited and acquired in the field and that may include seismic data acquired after a plurality of shots are excited.
Step S42, single shot seismic data are screened out according to preset shot intervals based on the signal-to-noise ratio of the seismic data.
In the step, a training sample set is selected, namely a certain number of seed cannons are selected as training data, and the selected standard is: according to the shot position sequencing, the number of training shots is selected according to a certain proportion according to the signal-to-noise ratio of the seismic data, wherein for the seismic data with high signal-to-noise ratio (the signal-to-noise ratio threshold is 1.5), one shot is generally selected every 200 shots, and for the seismic data with low signal-to-noise ratio, one shot is generally selected every 100 shots.
And step S43, picking up the collected single shot seismic data based on an energy ratio method to determine first arrival waves of the single shot seismic data.
Implementation of this step may include various ways, such as thresholding and maximum-value methods, in particular:
determining the position of a peak of the first arrival wave based on the position of the sampling point with the maximum energy ratio;
determining the sampling time of the peak position of the first arrival wave as the first arrival time so as to determine the first arrival wave of the single shot seismic data;
or alternatively, the first and second heat exchangers may be,
determining the position of the sampling point with the energy ratio larger than a preset threshold value as the arrival position of the first arrival wave;
and determining the sampling time of the arrival position of the first arrival wave as the first arrival time so as to determine the first arrival wave of the single shot seismic data.
Step S44, judging whether the first arrival wave of the seismic channel is an abnormal first arrival or not based on the waveform continuous variation trend and/or waveform energy trend of the first arrival wave between adjacent channels.
In general, the first arrival of seismic waves should be continuously variable, i.e., the waveform of a first arrival wave of single shot seismic data should have a continuous trend and an energy trend of waveform continuity. If the waveform of the first arrival wave is disturbed or energy is abnormal, it is highly likely that the first arrival wave belongs to an abnormal first arrival.
In the step, it is determined which part of first-arrival waves are abnormal first-arrival waves in the first-arrival waves of the single shot seismic data, namely which first-arrival wave is the abnormal first-arrival wave, or which first-arrival waves are the abnormal first-arrival waves. For example, as shown in fig. 3 and 5, the waveform continuity of the first arrival waves of 1068747 ～ 1068757, 1068853 ～ 1068877, and 1068907 ～ 1068922 is deteriorated, the trend is completely changed, and the energy trend is also changed, and the partial first arrival is an abnormal first arrival. The abnormal first arrival is used as an input parameter of a machine learning model so that the model learns how to identify the abnormal first arrival.
Of course, before executing this step, the original seismic data needs to be dimension-reduced and converted into a two-dimensional matrix, stored in a text format, and then converted into an image format, so as to facilitate machine learning.
Referring to fig. 6, the implementation of this step may include the following procedure:
step S441, dividing first arrival waves of single shot seismic data according to a preset seismic channel number and a preset sampling point number to obtain a plurality of slices.
For example, in this step, the number of seismic traces of the single shot seismic data is N, and the number of sampling points per trace is M. The first arrival of the single shot seismic data is divided into n multiplied by m slices, namely the number of seismic channels in each slice data is n, and the number of sampling points in each channel is m, so that the first arrival of the single shot seismic data can be extracted by the slices in the step.
It should be noted that, in the embodiment of the present invention, the number n of the seismic traces and the number m of sampling points of each trace of the slice may be adjusted according to actual needs, which is not limited in particular.
Step 442, judging whether each first arrival wave in the slice is an abnormal first arrival according to the comparison result of the waveform continuous variation trend and/or waveform energy trend of the first arrival wave between adjacent channels in the plurality of seismic channels contained in the slice and the preset sampling point position threshold.
The sampling point position threshold in the embodiment of the invention may be 2 sampling points, for example, the difference between the first arrival wave acquired by the seismic trace and the adjacent trace is 4 sampling points, and the first arrival wave acquired by the seismic trace is an abnormal first arrival.
Step S443, determining whether the slice is an abnormal first arrival according to the comparison result of the proportion of the abnormal first arrivals of all the seismic channels contained in the slice and a preset proportion threshold.
In the embodiment of the invention, each slice can be assigned with labels of 0 and 1, wherein 0 is a normal first arrival, and 1 is an abnormal first arrival. The preset proportion threshold value in the embodiment of the invention can be 90%, if the normal first arrival in all the seismic channels in the slice is greater than 90%, the slice is the normal first arrival, otherwise, the slice is the abnormal first arrival, and the first arrival in the slice needs to be corrected.
It should be noted that, in the embodiment of the present invention, the position threshold of the sampling point and the preset proportion threshold may be adjusted according to actual needs, which is not limited in particular. The embodiment of the invention uses the image data to train the machine learning model, because the image data can intuitively describe the characteristics of the seismic data and the accuracy of each first arrival, so as to better establish the relation between the experimental result and the seismic data and further adjust the parameters to observe the experimental change.
And S45, eliminating the abnormal first arrival. Referring to fig. 7, the single shot seismic data after the first arrival of the anomaly is removed is shown.
And step S46, determining a fitting line of the first arrival wave of the single shot seismic data according to the waveform continuous variation trend and/or the waveform energy trend of the first arrival waves of the abnormal first arrival adjacent seismic channels.
In this step, a fitting curve can be made for all the waveforms of the normal first arrival, and the abnormal first arrival is repaired by taking the fitting curve as a reference.
And step S47, modifying the rejected abnormal first arrival by taking the fitting line as a reference to obtain a modified first arrival wave.
In the step S12, training the deep neural network model including the classified convolutional neural network and the anti-convolutional neural network by using the sample degree in the training sample set; the first arrival wave of the single shot seismic data is input to the abnormal first arrival output by the classified convolution neural network, and identification parameter estimation is carried out; the abnormal first arrival is input into the convolutional neural network for correction parameter estimation.
Specifically, training sample data is firstly input into a classification convolutional neural network for classification, abnormal first arrival is identified, in the process, recognition parameters are machine-learned, the classification convolutional neural network in the embodiment of the invention can be an H neural network, the learning rate, the number of samples for one training and the number of training rounds are set, and training set images are input into the classification network H according to batches at random. And in the training process, the model is stored once every same round number, meanwhile, the loss function and the classification accuracy of the verification set are calculated, and the classification accuracy is gradually improved along with gradual convergence of the loss function, so that the parameter model capable of predicting the first arrival category is obtained. Referring to fig. 9, the network may include four convolutional layers, four max-pooling layers, and three fully-connected layers. To avoid overfitting, a random culling dropout function is used. The network outputs a 1 x 2 vector: a, b, if a > b, respectively predicting the category as 0; and conversely, the prediction is 1. The four convolution layers of the network H respectively adopt two convolution kernels of 5 multiplied by 5, two convolution kernels of 3 multiplied by 3, and the pooled kernels are 2 multiplied by 2, and the embodiment of the invention does not limit the architecture of the network specifically. Considering that the information in the seismic image is less complex than in real images (such as cat and dog vehicle buildings, etc.), a simpler network is adopted to improve the calculation efficiency, and the function of the network is to identify whether m first arrivals in the input image are abnormal first arrivals. The training model can recognize that the abnormal first arrival is related to the training set used in the training stage, each image in the training set is assigned with a label of 0 or 1, the label is respectively an index of the normal or abnormal first arrival image category, and the output image category is one of the two categories.
The learning rate is related to the signal-to-noise ratio of the first arrival data, and the lower the signal-to-noise ratio is, the smaller the learning rate is, and the default is set to be 0.000001; of course, the number of samples of one training is related to the graphics card of the computer, and the higher the graphics card performance is, the more the number of samples of one training is; the training round number is related to the signal-to-noise ratio of the first arrival data, the lower the signal-to-noise ratio is, the more training times are, and the default training times are 600.
And scanning the first arrival which is picked up by the energy ratio method without training in the sliding window by using the trained model, automatically detecting the abnormal first arrival, and eliminating the abnormal first arrival. In the process of rejecting, the machine can automatically recognize according to the threshold value and reject, or reject after confirmation by the user.
After the abnormal first arrival is removed, part of channels in each shot of seismic images lack first arrival information. To automatically supplement first arrival information faster, an anti-convolution neural network-SDGAN is generated through construction. The network structure is shown with reference to fig. 10, and the convolutional neural network includes a generator G network and a discriminator D network. The generator G network structure is shown with reference to fig. 11, and is actually a network for semantic segmentation, and includes 11 convolution layers and 3 maximum pooling layers, where conv 1-9 adopts a convolution kernel of 5×5, conv10 and 11 respectively adopt convolution kernels of 15×15 and 1×1, and the pooling kernel size is 2×2. And the data after the abnormal first arrival is removed is used as input data of the generator G network. The identifier D network may use the abnormal first arrival detection network CNN, or may reconstruct the identifier network to select better, and in the embodiment of the present invention, a markov identifier is constructed, as in fig. 12, which is different from a general classifier in that the output layer is not two or more nodes, but is an n×n matrix, and then the average value of the output matrix is determined as True or False.
And eliminating the seismic data image of the abnormal first arrival as input, and taking the first arrival image picked up in the corresponding training sample set as a label. The G network generates fake first arrival images which are as close to the artificial first arrival in the training sample set as possible according to the input image data. The real first arrival image after manual picking and the fake first arrival image generated by the G network are respectively spliced with the corresponding seismic data images, the true and fake images are judged according to the batch of random input to the D network, and iterative training is carried out until the D network considers that the generated first arrival is the real first arrival. And automatically correcting the seismic channels with the abnormal first arrival removed by using the trained generation model to obtain a pickup result.
As shown in fig. 8, the result of correcting the abnormal first arrival by using the generator G network can be seen that the removed abnormal first arrival can be effectively corrected basically, and the first arrival wave with extremely low signal to noise ratio can not be effectively corrected, mainly because the waveforms and energy of adjacent first arrival wavelets have larger mutation, and the adjacent first arrival constraint can not be used for correcting.
According to the embodiment of the invention, the classification network model is constructed according to the first arrival picked up by the energy ratio method and the data after the abnormal first arrival is manually removed, so that the model for predicting the first arrival category is obtained, the accuracy rate of identifying the abnormal first arrival is improved, the countermeasures network model is generated by combining the data with the abnormal first arrival removed and the manually modified first arrival data, the correction of the abnormal first arrival with low signal to noise ratio is effectively realized, and the picking accuracy and the number of channels are obviously improved.
Further, the core of the invention is that a first arrival wave time window is designed for the collected seismic data, the first arrival wave is automatically picked up by using an energy ratio method, and a part of data is selected to be used as a sample label by manually eliminating abnormal first arrival; secondly, cutting the obtained energy ratio first arrival image and the abnormal first arrival image by n paths as a unit into n multiplied by m subgraphs respectively, and distributing 0 and 1 labels for each subgraph; and constructing a classification network again, setting parameters such as learning rate, the number of samples for one training, the number of training rounds and the like, and simultaneously calculating a verification set loss function and classification accuracy to obtain a parameter model capable of predicting first arrival categories. Scanning the first arrival of the energy ratio which does not participate in training in the sliding window by using the trained model, automatically detecting the abnormal first arrival, and eliminating the abnormal first arrival; then, the data from which the abnormal first arrival is removed and the manually modified first arrival data are utilized to generate a G network of the first arrival and a D network for distinguishing the true and false of the first arrival in a combined way; and finally, automatically generating the seismic channels with the abnormal first arrival removed by using a trained generation model.
The method mainly comprises the steps of designing a first arrival wave time window for collected seismic data, automatically picking up the first arrival wave by using a conventional energy ratio algorithm, selecting a part of data, manually removing abnormal first arrival as a sample label, cutting the obtained energy ratio first arrival image and the removed abnormal first arrival image into n multiplied m subgraphs respectively by n times of a unit, distributing 0 label and 1 label to each subgraph, constructing a classification network to obtain a parameter model capable of predicting the first arrival category, automatically detecting the abnormal first arrival which does not participate in training by using a trained model, generating a true G network and a true D network for distinguishing the first arrival by using the data of removing the abnormal first arrival and the manually modified first arrival data, and automatically correcting the seismic channels of removing the abnormal first arrival by using the trained generation model.
Based on the same inventive concept, the embodiment of the invention further provides a training device of a machine learning model, and referring to fig. 13, the device may include: the working principle of the acquisition module 131 and the training module 132 is as follows:
the obtaining module 131 is configured to obtain a training sample set, where each sample in the training sample set includes a first arrival wave of single shot seismic data, an abnormal first arrival in the first arrival wave, and a first arrival wave after correction processing is performed on the abnormal first arrival.
The training module 132 is configured to train a deep neural network model including a classified convolutional neural network and an opposite convolutional neural network with a sample degree in a training sample set; the first arrival wave of the single shot seismic data is input to the abnormal first arrival output by the classified convolution neural network, and identification parameter estimation is carried out; the abnormal first arrival is input into the convolutional neural network for correction parameter estimation.
Based on the same inventive concept, there is also provided in an embodiment of the present invention a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements a training method of a machine learning model as described above.
Based on the same inventive concept, the embodiment of the invention also provides a computer device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the training method of the machine learning model when executing the program.
The specific manner in which the apparatus, medium and associated devices in the above embodiments are described in detail in connection with the embodiments of the method will not be described in detail herein.
Example 2
The embodiment of the invention provides a seismic wave abnormal first arrival correction method, which is shown by referring to fig. 14, and comprises the following steps:
and step S141, picking up the collected single shot seismic data based on an energy ratio method to determine first arrival waves of the single shot seismic data.
Specifically, determining the position of the peak of the first arrival wave based on the position of the sampling point with the maximum energy ratio; and determining the sampling time of the peak position of the first arrival wave as the first arrival time so as to determine the first arrival wave of the single shot seismic data.
Or determining the position of the sampling point with the energy ratio larger than a preset threshold value as the arrival position of the first arrival wave; and determining the sampling time of the arrival position of the first arrival wave as the first arrival time so as to determine the first arrival wave of the single shot seismic data.
Step S142, inputting the first arrival wave into a pre-trained deep neural network model comprising a classified convolutional neural network and an opposite convolutional neural network, wherein the classified convolutional neural network outputs an abnormal first arrival wave, and inputting the abnormal first arrival wave into the opposite convolutional neural network to determine a corrected first arrival wave.
The invention aims to provide a method for carrying out preliminary picking up first arrivals on low signal-to-noise ratio seismic data by adopting an energy ratio algorithm, eliminating abnormal first arrivals by utilizing a classification network, generating an antagonistic neural network by utilizing the combination of the data with the abnormal first arrivals eliminated and manually modified first arrival data, correcting the abnormal first arrivals by utilizing a trained generation network model and improving the picking accuracy and the channel number of first arrival waves.
It should be noted that, in the embodiment of the present invention, the pre-trained deep neural network model including the classified convolutional neural network and the opposite convolutional neural network may be obtained by training in the manner of embodiment 1 or may be obtained by training in other manners, which is not limited in particular in the embodiment of the present invention.
In an alternative embodiment, after determining the first arrival wave of the single shot seismic data, the method may further include:
performing file format conversion on first arrival waves of single shot seismic data, and converting the first arrival wave data format into text format data; the text format data is converted into image format data.
In another alternative embodiment, the method may further include: dividing first arrival waves of single shot seismic data according to the preset number of seismic channels and the preset number of sampling points to determine a plurality of slices.
Based on the same inventive concept, the embodiment of the invention also provides a seismic wave abnormal first arrival correction device, which can include, as shown in fig. 15: the working principle of the pick-up module 151 and the determining module 152 is as follows:
the pickup module 151 is configured to pick up the collected single shot seismic data based on an energy ratio method, so as to determine a first arrival wave of the single shot seismic data;
the determining module 152 is configured to input the first arrival wave to a pre-trained deep neural network model including a classified convolutional neural network and an opposite convolutional neural network, where the classified convolutional neural network outputs an abnormal first arrival wave, and input the abnormal first arrival wave to the opposite convolutional neural network to determine a corrected first arrival wave.
Based on the same inventive concept, the embodiment of the invention also provides a computer readable storage medium, on which a computer program is stored, which when being executed by a processor, implements the seismic wave abnormal first arrival correction method.
Based on the same inventive concept, the embodiment of the invention also provides a computer device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the seismic wave abnormal first arrival correction method when executing the program.
The specific manner in which the apparatus, medium and associated devices in the above embodiments are described in detail in connection with the embodiments of the method will not be described in detail herein.
Example 3
Based on the same inventive concept, the embodiment of the invention also provides a seismic wave abnormal first arrival identification method, and referring to fig. 16, the method can comprise the following steps:
step 161, picking up the collected single shot seismic data based on an energy ratio method to determine a first arrival wave of the single shot seismic data.
Step S162, inputting the first arrival wave into a pre-trained convolutional neural network model to identify abnormal first arrivals in the first arrival wave of the single shot seismic data.
In an alternative embodiment, the convolutional neural network model is pre-trained by:
acquiring a training sample set, wherein each sample in the training sample set comprises a first arrival wave of single shot seismic data and an abnormal first arrival in the first arrival wave;
and training the convolutional neural network model by using the sample degree in the training sample set, wherein the first arrival wave of the single shot seismic data is input to the abnormal first arrival output in the convolutional neural network, and identification parameter estimation is performed.
Based on the same inventive concept, the embodiment of the invention also provides a device for identifying abnormal first arrival of seismic waves, which can comprise: the pick-up module 171 and the identification module 172 operate on the following principles:
the pickup module 171 is configured to pick up the collected single shot seismic data based on an energy ratio method, so as to determine a first arrival wave of the single shot seismic data;
the identification module 172 is configured to input the first arrival wave into a pre-trained convolutional neural network model to identify an abnormal first arrival in the first arrival wave of the single shot seismic data.
Based on the same inventive concept, the embodiment of the invention also provides a computer readable storage medium, on which a computer program is stored, which when being executed by a processor, realizes the method for identifying abnormal first arrival of seismic waves.
Based on the same inventive concept, the embodiment of the invention also provides a computer device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the method for identifying the abnormal first arrival of the seismic wave when executing the program.
The specific manner in which the apparatus, medium and associated devices in the above embodiments are described in detail in connection with the embodiments of the method will not be described in detail herein.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, magnetic disk storage, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
Claims (16)
1. The method for correcting the abnormal first arrival of the seismic wave is characterized by comprising the following steps of:
picking up the collected single shot seismic data based on an energy ratio method to determine first arrival waves of the single shot seismic data;
inputting the first arrival wave into a pre-trained deep neural network model comprising a classified convolutional neural network and an opposite convolutional neural network, wherein the classified convolutional neural network outputs an abnormal first arrival wave, and inputting the abnormal first arrival wave into the opposite convolutional neural network to determine a corrected first arrival wave.
2. The method of claim 1, wherein after determining the first arrival wave of the single shot seismic data, further comprising:
performing file format conversion on the first arrival wave of the single shot seismic data, and converting the first arrival wave data format into text format data;
and converting the text format data into image format data.
3. The method as recited in claim 2, further comprising: dividing the first arrival wave of the single shot seismic data according to the preset number of seismic channels and the preset number of sampling points to determine a plurality of slices.
4. A method according to any one of claims 1 to 3, wherein the picking up the acquired single shot seismic data based on the energy ratio method to determine a first arrival wave of the single shot seismic data comprises:
Determining the position of a peak of the first arrival wave based on the position of the sampling point with the maximum energy ratio;
determining the sampling time of the peak position of the first arrival wave as the first arrival time so as to determine the first arrival wave of the single shot seismic data;
or alternatively, the first and second heat exchangers may be,
determining the position of the sampling point with the energy ratio larger than a preset threshold value as the arrival position of the first arrival wave;
and determining the sampling time of the arrival position of the first arrival wave as the first arrival time so as to determine the first arrival wave of the single shot seismic data.
5. A method of training a machine learning model, comprising:
acquiring a training sample set, wherein each sample in the training sample set comprises first arrival waves of single shot seismic data, abnormal first arrivals in the first arrival waves and first arrival waves after correction processing of the abnormal first arrivals;
training a deep neural network model comprising a classified convolutional neural network and an opposite convolutional neural network by using the sample degree in the training sample set; the first arrival wave of the single shot seismic data is input to the abnormal first arrival output in the classified convolution neural network, and identification parameter estimation is carried out; and inputting the abnormal first arrival into the convolutional neural network for correction parameter estimation.
6. The method of claim 5, wherein the method for determining whether the first arrival wave of the single shot seismic data is an abnormal first arrival comprises:
and judging whether the first arrival wave of the seismic channel is an abnormal first arrival or not based on waveform continuous variation trend and/or waveform energy trend of the first arrival wave between adjacent channels.
7. The method of claim 6, wherein determining whether the first arrival wave of the seismic trace is an abnormal first arrival based on waveform continuity variation trend and/or waveform energy trend of the first arrival wave between adjacent traces, comprises:
dividing the first arrival wave of the single shot seismic data according to the preset number of seismic channels and the preset number of sampling points to obtain a plurality of slices;
judging whether each first arrival wave in the slice is an abnormal first arrival or not according to the waveform continuous change trend and/or waveform energy trend of the first arrival wave between adjacent channels in the plurality of seismic channels contained in the slice and the comparison result of the preset sampling point position threshold;
and determining whether the section is the abnormal first arrival according to the comparison result of the proportion of the abnormal first arrivals of all the seismic channels contained in the section and a preset proportion threshold value.
8. The method of claim 6, wherein the process of correcting the abnormal first arrival comprises the steps of:
removing the abnormal first arrival;
determining a fitting line of the first arrival wave of the single shot seismic data according to the waveform continuous variation trend and/or the waveform energy trend of the first arrival waves of the abnormal first arrival adjacent seismic channels;
and modifying the eliminated abnormal first arrival by taking the fitting line as a reference to obtain a modified first arrival wave.
9. The method of any one of claims 5-8, wherein the obtaining a training sample set comprises:
and screening out the single shot seismic data according to a preset shot interval based on the signal-to-noise ratio of the seismic data.
10. The method for identifying the abnormal first arrival of the seismic wave is characterized by comprising the following steps of:
picking up the collected single shot seismic data based on an energy ratio method to determine first arrival waves of the single shot seismic data;
inputting the first arrival wave into a pre-trained convolutional neural network model to identify abnormal first arrivals in the first arrival wave of the single shot seismic data.
11. The method of claim 10, wherein the convolutional neural network model is pre-trained by:
Acquiring a training sample set, wherein each sample in the training sample set comprises a first arrival wave of single shot seismic data and an abnormal first arrival in the first arrival wave;
and training the convolutional neural network model by using the sample degree in the training sample set, wherein the first arrival wave of the single shot seismic data is input to the abnormal first arrival output in the convolutional neural network, and identification parameter estimation is performed.
12. An abnormal first arrival correction device for seismic waves, comprising:
the pickup module is used for picking up the collected single shot seismic data based on an energy ratio method so as to determine first arrival waves of the single shot seismic data;
the determining module is used for inputting the first arrival wave into a pre-trained deep neural network model comprising a classified convolutional neural network and an opposite convolutional neural network, wherein the classified convolutional neural network outputs an abnormal first arrival wave, and the abnormal first arrival wave is input into the opposite convolutional neural network to determine the corrected first arrival wave.
13. A training apparatus for a machine learning model, comprising:
the acquisition module is used for acquiring a training sample set, wherein each sample in the training sample set comprises first arrival waves of single shot seismic data, abnormal first arrival waves in the first arrival waves and first arrival waves after correction processing of the abnormal first arrival waves;
The training module is used for training a deep neural network model comprising a classified convolutional neural network and an anti-convolutional neural network by using the sample degree in the training sample set; the first arrival wave of the single shot seismic data is input to the abnormal first arrival output in the classified convolution neural network, and identification parameter estimation is carried out; and inputting the abnormal first arrival into the convolutional neural network for correction parameter estimation.
14. An abnormal first arrival identification device for seismic waves, comprising:
the pickup module is used for picking up the collected single shot seismic data based on an energy ratio method so as to determine first arrival waves of the single shot seismic data;
the identification module is used for inputting the first arrival wave into a pre-trained convolutional neural network model so as to identify abnormal first arrival in the first arrival wave of the single shot seismic data.
15. A computer-readable storage medium having stored thereon a computer program, characterized in that the program, when executed by a processor, implements the seismic wave anomaly first arrival correction method according to any one of claims 1 to 4, or implements the training method of the machine learning model according to any one of claims 5 to 9, or implements the seismic wave anomaly first arrival identification method according to claim 10 or 11.
16. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of seismic anomaly first-arrival correction as claimed in any one of claims 1 to 4, or the method of training the machine learning model as claimed in any one of claims 5 to 9, or the method of seismic anomaly first-arrival identification as claimed in claim 10 or 11, when the program is executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111647719.0A CN116413786A (en) | 2021-12-30 | 2021-12-30 | Seismic wave abnormal first arrival correction method and device and related equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111647719.0A CN116413786A (en) | 2021-12-30 | 2021-12-30 | Seismic wave abnormal first arrival correction method and device and related equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116413786A true CN116413786A (en) | 2023-07-11 |
Family
ID=87049827
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111647719.0A Pending CN116413786A (en) | 2021-12-30 | 2021-12-30 | Seismic wave abnormal first arrival correction method and device and related equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116413786A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117647837A (en) * | 2024-01-26 | 2024-03-05 | 东北石油大学三亚海洋油气研究院 | First arrival pickup method, system and computer equipment for seismic data |
-
2021
- 2021-12-30 CN CN202111647719.0A patent/CN116413786A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117647837A (en) * | 2024-01-26 | 2024-03-05 | 东北石油大学三亚海洋油气研究院 | First arrival pickup method, system and computer equipment for seismic data |
| CN117647837B (en) * | 2024-01-26 | 2024-04-09 | 东北石油大学三亚海洋油气研究院 | First arrival pickup method, system and computer equipment for seismic data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111611847B (en) | Video motion detection method based on scale attention hole convolution network | |
| CN109598287B (en) | Appearance flaw detection method for resisting network sample generation based on deep convolution generation | |
| CN108564026B (en) | Network construction method and system for thyroid tumor cytology smear image classification | |
| CN103927510B (en) | Pattern recognition device and image-recognizing method | |
| CN109948647A (en) | A kind of electrocardiogram classification method and system based on depth residual error network | |
| CN109829467A (en) | Image labeling method, electronic device and non-transient computer-readable storage medium | |
| CN107292330A (en) | A kind of iterative label Noise Identification algorithm based on supervised learning and semi-supervised learning double-point information | |
| CN108564123B (en) | Thyroid tumor cytology smear image classification method and device | |
| Prasetyo et al. | A comparison of yolo and mask r-cnn for segmenting head and tail of fish | |
| CN105574820A (en) | Deep learning-based adaptive ultrasound image enhancement method | |
| CN103324937A (en) | Method and device for labeling targets | |
| CN111507371A (en) | Method and apparatus | |
| Huang et al. | Qualitynet: Segmentation quality evaluation with deep convolutional networks | |
| CN111381275A (en) | First arrival picking method and device for seismic data | |
| CN115953666B (en) | Substation site progress identification method based on improved Mask-RCNN | |
| Polikar et al. | An incremental learning algorithm with confidence estimation for automated identification of NDE signals | |
| CN105956570A (en) | Lip characteristic and deep learning based smiling face recognition method | |
| CN109271546A (en) | The foundation of image retrieval Feature Selection Model, Database and search method | |
| CN104778466A (en) | Detection method combining various context clues for image focus region | |
| CN112560596A (en) | Radar interference category identification method and system | |
| CN116413786A (en) | Seismic wave abnormal first arrival correction method and device and related equipment | |
| Li et al. | ET-YOLOv5s: toward deep identification of students’ in-class behaviors | |
| CN115309860A (en) | False news detection method based on pseudo twin network | |
| CN115165366A (en) | Variable working condition fault diagnosis method and system for rotary machine | |
| CN117197591A (en) | Data classification method based on machine learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |