Deep learning-based DAS-VSP data first-motion wave pickup method and device
By generating and correcting regional images of DAS-VSP data using deep learning technology, and utilizing an improved U-Net network model and linear regression model, the problem of low first arrival accuracy in DAS-VSP data was solved, achieving efficient and high-precision first arrival picking and meeting the requirements for reservoir parameter extraction.
Patent Information
- Application Number
- CN202310505916.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-06
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2043-05-06
AI Technical Summary
The first arrival acquisition of DAS-VSP data suffers from low signal-to-noise ratio and difficulty in identifying the first arrival, resulting in low acquisition accuracy and slow efficiency, which cannot meet the needs of high-precision reservoir parameter extraction.
By employing a deep learning-based approach, initial and predicted region images are generated. An improved U-Net network model and a linear regression model are used to perform trend and numerical corrections on the first arrival wave of DAS-VSP data, establish a mapping between input and output, extract data features, and achieve high-precision picking.
It improves the accuracy and efficiency of first arrival picking in DAS-VSP data, meets the requirements for high-precision reservoir parameter extraction, and reduces the cost of manual interactive picking.
Smart Images

Figure CN118918405A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the technical field of petroleum geophysical exploration data processing, and in particular to a DAS-VSP data first arrival wave picking method and device based on deep learning, an electronic device and a storage medium. Background Art
[0002] At present, most of China's oil and gas fields have entered the middle and late stages of exploration and development, and ground seismic exploration faces severe challenges in terms of fine reservoir description, reservoir parameter acquisition, and seismic resolution improvement. In recent years, the rapid development of DAS-VSP (Distributed Acoustic Sensing, DAS, distributed fiber acoustic sensing system, Vertical Seismic Profile, VSP, vertical seismic) acquisition technology can finely extract key parameters such as underground medium velocity field, anisotropy, TAR factor, formation absorption attenuation factor Q value, amplitude correction factor, deconvolution, etc., which is of great significance for providing high-precision reservoir parameters and improving the ability to finely describe the oil reservoirs around oil and gas wells, effectively making up for the shortcomings of ground seismic exploration.
[0003] However, the basis for extracting these key parameters is based on high-precision first arrivals, and the VSP data collected by DAS currently have two main problems in first arrival picking: 1) The overall signal-to-noise ratio of the data is relatively low. The DAS system is relatively sensitive to stress changes in the surrounding environment of the sensing fiber. The extremely small stress in the surrounding environment and the weak noise in the system of internal components will produce strong noise. Actual data show that the continuity of the first arrival wave of DAS data is poor, and the amplitude, frequency, and phase changes of the first arrival waves of adjacent channels are quite different. 2) As the distance from the well source increases, the incident angle of the direct wave increases, the amplitude response decays rapidly, and the first arrival wave can hardly be effectively identified. The current mainstream first arrival picking methods and commercial software are all developed for the first arrival wave characteristics of ground seismic exploration data. The first arrival picking accuracy of DAS-VSP data is low, requiring a large amount of manual interactive picking, which is costly, low-precision, and slow. High-efficiency and high-precision first arrival picking technology is one of the key technologies currently restricting DAS-VSP exploration. Summary of the invention
[0004] In order to solve the above technical problems or at least partially solve the above technical problems, embodiments of the present disclosure provide a method and device for picking first-arrival waves of DAS-VSP data based on deep learning, an electronic device, and a storage medium.
[0005] In a first aspect, an embodiment of the present disclosure provides a method for picking first arrival waves of DAS-VSP data based on deep learning, the method comprising:
[0006] For the sample shot in the target work area, the initial regional image is generated in the rectangular coordinate system based on the DAS-VSP data of the sample shot, with the current sample shot position as the center and the shot line and seismic trace receiving line as the horizontal and vertical axes;
[0007] Acquire the first arrival data of the sample shot in the target work area, transform the first arrival data of the sample shot in the target work area into image data, plot it on the regional image, and generate a predicted regional image;
[0008] Using the initial region image and the predicted region image to train a known neural network model, to obtain a trained neural network model;
[0009] Input the initial area image of the to-be-predicted gun into the trained neural network model, and output the predicted area image of the to-be-predicted gun;
[0010] The trend correction and value correction are performed on the first arrival position of each track in the prediction area image of the predicted gun to be picked up, and the corrected first arrival time of each track of the gun to be picked up in the target work area is obtained.
[0011] In a possible implementation, the sample gun is determined by the following steps:
[0012] Determine the preset ratio of sample shots to be drawn from all shot points in the target work area based on the signal-to-noise ratio of the seismic data;
[0013] According to the coordinates of each shot point and the wellhead coordinates in the target work area, the well source distance and azimuth of each shot point are calculated;
[0014] According to the well source distance and azimuth of each shot point, a preset proportion of sample shots are evenly extracted from all the shot points in the target work area.
[0015] In a possible implementation, the known neural network model is an improved U-Net network model, and the improved U-Net network model is obtained by improving the U-Net network model by the following steps:
[0016] Through the Adam optimization algorithm, the learning rate η of the U-Net network model is improved by the following expression:
[0017]
[0018] in,
[0019]
[0020] Among them, a j,k is the probability value of the predicted initial arrival position of the j-th shot, y j,k is the probability value of the expected first arrival position of the kth track of the jth shot.
[0021] In a possible implementation, the using the initial region image and the predicted region image to train a known neural network model to obtain a trained neural network model includes:
[0022] The initial area image of each sample shot is taken as input, the predicted area image is taken as output, and the known neural network model is trained to obtain a trained neural network model.
[0023] In a possible implementation manner, the initial area image of the to-be-predicted shot is determined by the following steps:
[0024] Set the first arrival time window for the sample shot, and use the spatial interpolation technology to determine the time window range of each shot to be picked up;
[0025] With the position of the shot to be predicted as the center, and the shot line and the seismic receiving line as the horizontal and vertical axes, an initial regional image is generated in a rectangular coordinate system according to the DAS-VSP data within the time window.
[0026] In a possible implementation, the trend correction and value correction are performed on the first arrival position of each track in the prediction area image of the to-be-predicted shot to obtain the corrected first arrival time of each track of the to-be-picked shot in the target work area, including:
[0027] An initial linear regression model is constructed according to the first arrival position of each trace in the prediction area image of the to-be-predicted shot;
[0028] Determine the first arrival position of the target whose deviation distance from the linear regression model exceeds a preset threshold;
[0029] Subtract the target first arrival position from the first arrival position of each track in the prediction area image of the to-be-predicted shot to obtain the first arrival position after subtraction;
[0030] Construct a new linear regression model based on the first arrival position after removal;
[0031] According to the new linear regression model, the trend of the first arrival position after removal is corrected to obtain the corrected position of the first arrival position after removal;
[0032] The corrected position of the target first arrival position is obtained by fitting a polynomial between the removed first arrival position and the corresponding seismic trace number;
[0033] The corrected positions of the target initial arrival position and the initial arrival position after removal are respectively transformed into the target initial arrival time and the initial arrival time after removal, which are used as the corrected initial arrival time of each shot to be picked up in the target work area.
[0034] In a possible implementation, after converting the target first arrival position and the corrected position after the removal of the first arrival position into the target first arrival time and the first arrival time after the removal of the first arrival time, the method further includes:
[0035] The target first arrival time and the first arrival time after removal are corrected by the following expression, and the corrected target first arrival time and the first arrival time after removal are used as the corrected first arrival time of each shot to be picked up in the target work area:
[0036]
[0037] Among them, w is the starting point of the smoothing window, h i is the smoothing coefficient, the value is 1, x k is the first arrival time of the kth track.
[0038] In a second aspect, an embodiment of the present disclosure provides a DAS-VSP data first arrival wave picking device based on deep learning, comprising:
[0039] The first generation module is used to generate an initial area image according to the DAS-VSP data of the sample shot in the target work area, with the position of the current sample shot as the center, the shot line and the seismic channel receiving line as the horizontal axis and the vertical axis in the rectangular coordinate system;
[0040] A second generation module is used to obtain the first arrival data of the sample shot in the target work area, and transform the first arrival data of the sample shot in the target work area into image data, and draw it on the regional image to generate a predicted regional image;
[0041] A training module, used for training a known neural network model using the initial region image and the predicted region image to obtain a trained neural network model;
[0042] A prediction module, used to input the initial area image of the to-be-predicted gun into the trained neural network model, and output the predicted area image of the to-be-predicted gun;
[0043] The correction module is used to perform trend correction and value correction on the first arrival position of each track in the prediction area image of the predicted gun, and obtain the corrected first arrival time of each track of the gun to be picked up in the target work area.
[0044] In a third aspect, an embodiment of the present disclosure provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;
[0045] Memory, used to store computer programs;
[0046] The processor is used to implement the above-mentioned DAS-VSP data first arrival wave picking method based on deep learning when executing the program stored in the memory.
[0047] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the above-mentioned deep learning-based DAS-VSP data first-arrival wave picking method is implemented.
[0048] The above technical solution provided by the embodiment of the present disclosure has at least some or all of the following advantages compared with the prior art:
[0049] The deep learning-based DAS-VSP data first arrival wave picking method described in the embodiment of the present disclosure is as follows: for a sample shot in a target work area, an initial regional image is generated in a rectangular coordinate system according to the DAS-VSP data of the sample shot, with the position of the current sample shot as the center and the shot line and the seismic trace receiving line as the horizontal axis and the vertical axis; the first arrival data of the sample shot in the target work area are obtained, and the first arrival data of the sample shot in the target work area are converted into image data, which are plotted on the regional image to generate a predicted regional image; a known neural network model is trained using the initial regional image and the predicted regional image to obtain a trained neural network model; the initial regional image of the shot to be predicted is input into the trained neural network model Through the network model, the prediction area image of the to-be-predicted shot is output; the trend correction and value correction are performed on the first arrival position of each track in the prediction area image of the to-be-predicted shot, and the corrected first arrival time of each track of the to-be-picked shot in the target work area is obtained. The manually picked first arrival data is used as samples, and the deep network model is trained through deep learning to establish a mapping between input and output. The feature extractor composed of convolutional layers and subsampling layers is used to extract more abstract features in the data, so as to predict the first arrival time of the DAS-VSP data. Then, through the identification and correction of abnormal first arrivals, the high-precision picking of the DAS-VSP first arrival data with low signal-to-noise ratio is achieved to meet the needs of real-time and high-precision processing of subsequent data. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0052] Figure 1 The schematic diagram of the process of the DAS-VSP data first arrival wave picking method based on deep learning according to an embodiment of the present disclosure is shown;
[0053] Figure 2A schematic diagram of DAS-VSP single shot data according to an embodiment of the present disclosure is schematically shown;
[0054] Figure 3 A schematic diagram of the initial arrival prediction effect according to an embodiment of the present disclosure is schematically shown;
[0055] Figure 4 The schematic diagram of the first arrival trend correction effect according to the embodiment of the present disclosure is schematically shown;
[0056] Figure 5 The schematic diagram of the first arrival and fine correction effect according to the embodiment of the present disclosure is schematically shown;
[0057] Figure 6 A structural block diagram of a device for first-break wave picking of DAS-VSP data based on deep learning according to an embodiment of the present disclosure is schematically shown; and
[0058] Figure 7 The structural block diagram of an electronic device according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION
[0059] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.
[0060] See also Figure 1 , an embodiment of the present disclosure provides a DAS-VSP data first arrival wave picking method based on deep learning, the method comprising:
[0061] S1, for the sample shot in the target work area, with the current sample shot position as the center, the shot line and the seismic trace receiving line as the horizontal and vertical axes, generate the initial regional image according to the DAS-VSP data of the sample shot in the rectangular coordinate system.
[0062] In some embodiments, the DAS-VSP data is single-shot DAS-VSP seismic data that is excited and collected and recorded, with m channels and n sampling points per channel.
[0063] S2, obtaining the first arrival data of the sample shot in the target work area, and transforming the first arrival data of the sample shot in the target work area into image data, and plotting it on the regional image to generate a predicted regional image.
[0064] In some embodiments, the first arrival waves of the extracted sample shots are finely picked, and then the picked first arrival data are transformed into image data. Before the transformation, the display resolution, channel interval, time interval and other parameters of the seismic data are set and recorded to ensure the first arrival time positioning between the actual first arrival data and the image data.
[0065] In this embodiment, image data is used as samples for deep network model training, and deep learning technology is used to abstract and extract first-arrival features and first-arrival position information from different levels.
[0066] In this embodiment, the first arrival time of the sample shot is precisely calibrated manually and a sample label is generated to ensure the accuracy of the sample data input into the neural network model.
[0067] S3, using the initial region image and the predicted region image to train a known neural network model to obtain a trained neural network model.
[0068] S4, inputting the initial area image of the to-be-predicted gun into the trained neural network model, and outputting the predicted area image of the to-be-predicted gun.
[0069] S5, performing trend correction and value correction on the first arrival position of each track in the prediction area image of the to-be-predicted gun, and obtaining the corrected first arrival time of each track of the to-be-picked gun in the target work area.
[0070] In this embodiment, trend correction and value correction are performed on the predicted first arrival, which can meet the needs of real-time and high-precision data processing.
[0071] In this embodiment, in step S1, the sample gun is determined by the following steps:
[0072] Determine the preset ratio of sample shots to be drawn from all shot points in the target work area based on the signal-to-noise ratio of the seismic data;
[0073] According to the coordinates of each shot point and the wellhead coordinates in the target work area, the well source distance and azimuth of each shot point are calculated;
[0074] According to the well source distance and azimuth of each shot point, a preset proportion of sample shots are evenly extracted from all the shot points in the target work area.
[0075] In some embodiments, the number of shots collected in the whole area is S, then h% of sample shots are extracted according to the signal-to-noise ratio of the seismic data, and the well source distance and azimuth of each shot are calculated according to the shot point coordinates and the wellhead coordinates, and the average extraction is performed according to the well source distance in each azimuth, that is, if a total of g sample shots are extracted, the sample is averaged according to the well source distance every azimuth. Angle and ρ meters are used for extraction, then each azimuth and each interval are extracted guns, so as to ensure that the sample guns are evenly distributed and can cover the first arrival characteristics of the entire area.
[0076] In some embodiments, h% of sample shots are extracted based on the signal-to-noise ratio of the seismic data. If the signal-to-noise ratio is greater than 2, h is taken as 5, i.e., 5% of the sample shots are extracted; if the data signal-to-noise ratio is less than 1, h is taken as 10, i.e., 10% of the sample shots are extracted; generally, h is taken as 8, i.e., 8% of the sample shots are extracted.
[0077] In some embodiments, the average extraction is performed based on the well source distance in each azimuth, with an interval angle of The angle and spacing ρ are set according to the data signal-to-noise ratio. The lower the signal-to-noise ratio, the greater the angle. The smaller the value of and spacing ρ, the more evenly the sample guns are distributed. In general, the angle The value is 20 and the spacing ρ is 500.
[0078] In this embodiment, sample shots are extracted from seismic data according to the well source distance and azimuth to ensure that the sample shots are evenly distributed and can cover the first arrival characteristics of the entire area.
[0079] In this embodiment, in step S3, the known neural network model is an improved U-Net network model, and the improved U-Net network model is obtained by improving the U-Net network model by the following steps:
[0080] Through the Adam optimization algorithm, the learning rate η of the U-Net network model is improved by the following expression:
[0081]
[0082] in,
[0083]
[0084] Among them, a j,k is the probability value of the predicted initial arrival position of the j-th shot, y j,k is the probability value of the expected first arrival position of the kth track of the jth shot.
[0085] In some embodiments, the value of ε is, for seismic data with a high signal-to-noise ratio, a larger value of ε is taken; for seismic data with a low signal-to-noise ratio, a smaller value of ε is taken; generally, the default value of ε is 0.6.
[0086] In some embodiments, the learning rate η is given a large learning rate in the early stage of training to speed up the training efficiency. As the number of training times increases, the consistency rate between the predicted initial arrival position and the actual position increases, that is, T continues to increase. If T is greater than a given threshold, the learning rate is gradually reduced to improve the model accuracy, thereby improving the initial arrival detection accuracy.
[0087] In this embodiment, in step S3, the use of the initial region image and the predicted region image to train a known neural network model to obtain a trained neural network model includes:
[0088] The initial area image of each sample shot is taken as input, the predicted area image is taken as output, and the known neural network model is trained to obtain a trained neural network model.
[0089] In this embodiment, in step S4, the initial area image of the to-be-predicted shot is determined by the following steps:
[0090] Set the first arrival time window for the sample shot, and use the spatial interpolation technology to determine the time window range of each shot to be picked up;
[0091] With the position of the shot to be predicted as the center, and the shot line and the seismic receiving line as the horizontal and vertical axes, an initial regional image is generated in a rectangular coordinate system according to the DAS-VSP data within the time window.
[0092] In some embodiments, the position of the sample shot is taken as the center point, the shot line is taken as the vertical axis, and the seismic trace receiving line is taken as the horizontal axis. The seismic trace data is divided into N regional sets in a rectangular coordinate system. The seismic traces in each region are sorted from small to large according to the shot offset. A first arrival time window is set in each region. According to the first arrival wave time window of each region of the sample shot, a spatial interpolation technique is used to determine the time window range of each first arrival wave of the shot to be picked up in the work area. The seismic traces of each shot are divided into N regional sets in a rectangular coordinate system. N can be an integer such as 1, 2, 4, 8, that is, 2 is taken to the power of an integer greater than or equal to 0. The larger the value of N, the smaller the first arrival time window range, the higher the first arrival accuracy, and the default is divided into 1 region.
[0093] In some embodiments, a first arrival time window is set in each area. If the near-surface structure is relatively horizontal, the time window length is set to 200 sample points; if the near-surface structure is complex, the time window length is set to 600 sample points; generally, the time window length is set to 400 sample points.
[0094] In this embodiment, in step S5, the trend correction and value correction are performed on the first arrival position of each track in the prediction area image of the to-be-predicted shot to obtain the corrected first arrival time of each track of the to-be-picked shot in the target work area, including:
[0095] An initial linear regression model is constructed according to the first arrival position of each trace in the prediction area image of the to-be-predicted shot;
[0096] Determine the first arrival position of the target whose deviation distance from the linear regression model exceeds a preset threshold;
[0097] Subtract the target first arrival position from the first arrival position of each track in the prediction area image of the to-be-predicted shot to obtain the first arrival position after subtraction;
[0098] Construct a new linear regression model based on the first arrival position after removal;
[0099] According to the new linear regression model, the trend of the first arrival position after removal is corrected to obtain the corrected position of the first arrival position after removal;
[0100] The corrected position of the target first arrival position is obtained by fitting a polynomial between the removed first arrival position and the corresponding seismic trace number;
[0101] The corrected positions of the target initial arrival position and the initial arrival position after removal are respectively transformed into the target initial arrival time and the initial arrival time after removal, which are used as the corrected initial arrival time of each shot to be picked up in the target work area.
[0102] In some embodiments, let the first arrival time of the i-th pick of each shot be t i , pick up the first arrival of N seismic traces for each shot, establish the expression of the objective function J(β), calculate the regression coefficient β of the minimum value of the objective function, and thus establish the initial linear regression model:
[0103] J(β)=∑(τ i -t i β) 2 +∑λ i β 2
[0104] Among them, J(β) is the objective function, t i is the first arrival time of each track, τ i is the first arrival time of each linear regression, λ i It is the penalty term of the objective function, which is selected according to the variance and bias used to balance the model and improve the calculation accuracy of the regression coefficient β.
[0105] In some embodiments, the outlier points that affect the linear regression model are deleted according to the following expression:
[0106]
[0107] in, is the i-th main diagonal element of the statistical matrix H, and p is the independent variable t i , N is the sample size of the modeling data set. If the diagonal element satisfies the above expression, it represents the i-th first arrival anomaly.
[0108] In some embodiments, since the linear regression model is easily affected by extreme values, it is necessary to use statistical methods to test the observed samples for deviation points. If abnormal data is found during the modeling process, the data set is rectified. The initial linear regression model is solved based on the derived regression coefficients. The linear regression model is expressed as follows:
[0109]
[0110] Among them, μ is the linear regression equation, t is the first arrival time, α is the disturbance term, which is small and can be ignored, and H is the statistical matrix.
[0111] In some embodiments, after the abnormal initial arrivals are deleted, the initial linear regression model is used again to obtain a new linear regression model for each shot.
[0112] In some embodiments, if there is seismic data without preliminary first arrival results and a linear regression model cannot be established, an inverse distance weighted method can be used to interpolate an existing model to obtain a linear regression model to obtain a first arrival trend line for each seismic data.
[0113] In this embodiment, after converting the target first arrival position and the corrected position of the removed first arrival position into the target first arrival time and the removed first arrival time respectively, the method further includes:
[0114] The target first arrival time and the first arrival time after removal are corrected by the following expression, and the corrected target first arrival time and the first arrival time after removal are used as the corrected first arrival time of each shot to be picked up in the target work area:
[0115]
[0116] Among them, w is the starting point of the smoothing window, h i is the smoothing coefficient, the value is 1, x k is the first arrival time of the kth track, and L is the time window length.
[0117] In some embodiments, the abnormal first arrival with large deviation is deleted, and the channel number with predicted first arrival result for each shot is counted. Assuming that the minimum channel number is Φ and the maximum channel number is Ψ, the unpredicted seismic traces from Φ to Ψ can be fitted with a polynomial by the first arrival time and the channel number to obtain the corresponding position. At the same time, the Savitzky-Gola algorithm is used to smooth the fitted first arrival to obtain the refined first arrival time. The calculation expression is as follows:
[0118]
[0119] Among them, w is the starting point of the smoothing window, h i is the smoothing coefficient, the value is 1, x k is the first arrival time of the kth track, and L is the time window length.
[0120] The present invention extracts sample shots according to the signal-to-noise ratio, well source distance and azimuth of seismic data; determines the time window range of the shot to be predicted through the first arrival time window of the sample shot; uses an improved U-Net to train a high-precision deep network model, and uses the trained deep network model to predict the data to be predicted to obtain the first arrival position of each trace; uses a multivariate linear regression model to correct the predicted first arrival wave trend, and combines the Savitzky-Gola algorithm to perform fine correction on the first arrival, so that the picking accuracy and efficiency are greatly improved.
[0121] The following are applied to Figure 2 The DAS-VSP data first arrival wave picking method based on deep learning for the DAS-VSP single shot data shown includes:
[0122] 1) Stimulate and collect DAS-VSP single-shot seismic data, such as Figure 2 As shown, the number of receiving channels for this single shot is 430, the sampling interval is 4 milliseconds, and the number of sampling points per channel is 500.
[0123] 2) Sample shot extraction: Sample shots were extracted from seismic data according to well source distance and azimuth. A total of 32,895 shots were collected in the whole area. According to the data signal-to-noise ratio, 1,600 shots were extracted. The samples were extracted at 20° and 500-meter grids, with an average of 5 shots per grid, ensuring that the sample shots were evenly distributed and could cover the first arrival characteristics of the whole area. At the same time, the sample shots were set with a time window of 400 sample points, and the time window range of each shot to be picked up was determined using spatial interpolation technology.
[0124] 3) Sample data generation: First, the sample shots are manually calibrated, and then the first arrival data is extracted as a density image. Before extraction, the display resolution, channel interval, time interval and other parameters of the seismic data must be set and recorded to ensure the first arrival time positioning between the first arrival data and the image. The extracted density image is used as the input label for deep network model training, and deep learning technology is used to abstract and extract the first arrival features and location information from different levels.
[0125] 4) Combined with the Adam optimization algorithm, the learning rate η of the deep network U-Net is improved:
[0126]
[0127] in,
[0128]
[0129] Among them, a j,k is the probability value of the predicted initial arrival position of the j-th shot, y j,k is the probability value of the expected first arrival position of the kth track of the jth shot.
[0130] 5) Use the improved U-Net network to train the model and obtain a deep network model;
[0131] 6) Using step 2), extract the data within the time window of the first arrival of the cannon to be predicted, and using steps 4) and 5), use the trained deep network model to predict the data to be predicted. The prediction results are as follows: Figure 3 As shown, it can be seen that the overall prediction accuracy is high, but affected by interference noise, the predicted first arrival anomalies at channel numbers 20, 40, 232, 258, 294, 347, 367, 402, and 422 are not accurate. The first arrival position is not picked up.
[0132] 7) Using the initial trend correction technique, Figure 3 The predicted first arrivals are corrected to describe the overall trend of the first arrivals of earthquake data, such as Figure 4 As shown, it can be seen that the revised first arrival trend is reasonable.
[0133] 8) In order to meet the needs of high-precision data processing, it is necessary to make fine corrections to the first arrivals. Figure 5 As shown, it can be seen that after correction, the overall picking accuracy is higher, the first arrival continuity is significantly improved, and it can meet the needs of processing key parameter extraction.
[0134] See also Figure 6 The embodiment of the present disclosure provides a DAS-VSP data first arrival wave picking device based on deep learning, comprising:
[0135] The first generation module 11 is used to generate an initial area image according to the DAS-VSP data of the sample shot in the target work area, with the position of the current sample shot as the center, the shot line and the seismic trace receiving line as the horizontal axis and the vertical axis in a rectangular coordinate system;
[0136] The second generation module 12 is used to obtain the first arrival data of the sample shot in the target work area, and transform the first arrival data of the sample shot in the target work area into image data, and draw it on the regional image to generate a predicted regional image;
[0137] A training module 13 is used to train a known neural network model using the initial region image and the predicted region image to obtain a trained neural network model;
[0138] Prediction module 14, used for inputting the initial area image of the to-be-predicted gun into the trained neural network model, and outputting the predicted area image of the to-be-predicted gun;
[0139] The correction module 15 is used to perform trend correction and value correction on the first arrival position of each track in the prediction area image of the to-be-predicted gun, and obtain the corrected first arrival time of each track of the to-be-picked gun in the target work area.
[0140] The DAS-VSP data first arrival wave picking device based on deep learning disclosed in the present invention first extracts sample shots from seismic data according to the well source distance and azimuth, sets the first arrival wave time window for the sample shots, and uses spatial interpolation to determine the time window range of each shot to be picked up; secondly, the sample shots are accurately calibrated for the first arrival, and sample labels are generated to ensure the accuracy of the input sample data; the improved U-Net network is used for model training again to obtain a deep network model; then the obtained deep network model is used to predict the data to be picked up; finally, the predicted first arrival is trended and finely corrected to meet the needs of real-time and high-precision data processing.
[0141] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.
[0142] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can refer to the partial description of the method embodiments. The device embodiments described above are only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without paying any creative work.
[0143] In the above embodiment, any multiple of the first generation module 11, the second generation module 12, the training module 13, the prediction module 14 and the correction module 15 can be combined in one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. At least one of the first generation module 11, the second generation module 12, the training module 13, the prediction module 14 and the correction module 15 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in any appropriate combination of any of them. Alternatively, at least one of the first generation module 11, the second generation module 12, the training module 13, the prediction module 14 and the correction module 15 can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be executed.
[0144] See also Figure 7 , an electronic device provided by an embodiment of the present disclosure includes a processor 1110, a communication interface 1120, a memory 1130 and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1140;
[0145] Memory 1130, used for storing computer programs;
[0146] The processor 1110 is configured to implement the following DAS-VSP data first arrival wave picking method based on deep learning when executing the program stored in the memory 1130:
[0147] For the sample shot in the target work area, the initial regional image is generated in the rectangular coordinate system based on the DAS-VSP data of the sample shot, with the current sample shot position as the center and the shot line and seismic trace receiving line as the horizontal and vertical axes;
[0148] Acquire the first arrival data of the sample shot in the target work area, transform the first arrival data of the sample shot in the target work area into image data, plot it on the regional image, and generate a predicted regional image;
[0149] Using the initial region image and the predicted region image to train a known neural network model, to obtain a trained neural network model;
[0150] Input the initial area image of the to-be-predicted gun into the trained neural network model, and output the predicted area image of the to-be-predicted gun;
[0151] The trend correction and value correction are performed on the first arrival position of each track in the prediction area image of the predicted gun to be picked up, and the corrected first arrival time of each track of the gun to be picked up in the target work area is obtained.
[0152] The DAS-VSP data first arrival wave picking method disclosed in the present invention extracts sample shots according to the well source distance and azimuth, sets the first arrival wave time window, uses spatial interpolation to determine the time window range of each shot to be picked, accurately calibrates the first arrival of the sample shots, generates sample labels, uses the improved U-Net network training to obtain a deep network model, predicts the data to be picked, and performs trend and fine correction on the predicted first arrival. It can achieve high-precision picking of low signal-to-noise ratio DAS-VSP first arrival data to meet the needs of real-time and high-precision processing of subsequent data.
[0153] The communication bus 1140 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus 1140 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.
[0154] The communication interface 1120 is used for communication between the above electronic device and other devices.
[0155] The memory 1130 may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory 1130 may also be at least one storage device located away from the processor 1110.
[0156] The above-mentioned processor 1110 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.
[0157] The embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for first-break picking of DAS-VSP data based on deep learning as described above is implemented.
[0158] The computer-readable storage medium may be included in the device / apparatus described in the above embodiment; or it may exist independently without being assembled into the device / apparatus. The above computer-readable storage medium carries one or more programs. When the above one or more programs are executed, the DAS-VSP data first arrival wave picking method based on deep learning according to the embodiment of the present disclosure is implemented.
[0159] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.
[0160] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0161] The foregoing is merely a specific embodiment of the present disclosure, which enables those skilled in the art to understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features applied herein.
Claims
1. A method for first arrival wave picking of DAS-VSP data based on deep learning, It is characterized in that The method comprises: For the sample shot in the target work area, the initial regional image is generated in the rectangular coordinate system based on the DAS-VSP data of the sample shot, with the current sample shot position as the center and the shot line and seismic trace receiving line as the horizontal and vertical axes; Acquire the first arrival data of the sample shot in the target work area, transform the first arrival data of the sample shot in the target work area into image data, plot it on the regional image, and generate a predicted regional image; Using the initial region image and the predicted region image to train a known neural network model, to obtain a trained neural network model; Input the initial area image of the to-be-predicted gun into the trained neural network model, and output the predicted area image of the to-be-predicted gun; The trend correction and value correction are performed on the first arrival position of each track in the prediction area image of the predicted gun to be picked up, and the corrected first arrival time of each track of the gun to be picked up in the target work area is obtained.
2. The method according to claim 1, It is characterized in that The sample gun is determined by the following steps: Determine the preset ratio of sample shots to be drawn from all shot points in the target work area based on the signal-to-noise ratio of the seismic data; According to the coordinates of each shot point and the wellhead coordinates in the target work area, the well source distance and azimuth of each shot point are calculated; According to the well source distance and azimuth of each shot point, a preset proportion of sample shots are evenly extracted from all the shot points in the target work area.
3. The method according to claim 1, It is characterized in that The known neural network model is an improved U-Net network model, and the improved U-Net network model is obtained by improving the U-Net network model by the following steps: Through the Adam optimization algorithm, the learning rate η of the U-Net network model is improved by the following expression: in, Among them, a j,k is the probability value of the predicted initial arrival position of the j-th shot, y j,k is the probability value of the expected first arrival position of the kth track of the jth shot.
4. The method according to claim 1, It is characterized in that The method of using the initial region image and the predicted region image to train a known neural network model to obtain a trained neural network model includes: The initial area image of each sample shot is taken as input, the predicted area image is taken as output, and the known neural network model is trained to obtain a trained neural network model.
5. The method according to claim 1, It is characterized in that The initial area image of the to-be-predicted shot is determined by the following steps: Set the first arrival time window for the sample shot, and use the spatial interpolation technology to determine the time window range of each shot to be picked up; With the position of the shot to be predicted as the center, and the shot line and the seismic receiving line as the horizontal and vertical axes, an initial regional image is generated in a rectangular coordinate system according to the DAS-VSP data within the time window.
6. The method according to claim 1, It is characterized in that The method of performing trend correction and value correction on the first arrival position of each track in the prediction area image of the to-be-predicted shot to obtain the corrected first arrival time of each track of the to-be-picked shot in the target work area includes: An initial linear regression model is constructed according to the first arrival position of each trace in the prediction area image of the to-be-predicted shot; Determine the first arrival position of the target whose deviation distance from the linear regression model exceeds a preset threshold; Subtract the target first arrival position from the first arrival position of each track in the prediction area image of the to-be-predicted shot to obtain the first arrival position after subtraction; Construct a new linear regression model based on the first arrival position after removal; According to the new linear regression model, the trend of the first arrival position after removal is corrected to obtain the corrected position of the first arrival position after removal; The corrected position of the target first arrival position is obtained by fitting a polynomial between the removed first arrival position and the corresponding seismic trace number; The corrected positions of the target initial arrival position and the initial arrival position after removal are respectively transformed into the target initial arrival time and the initial arrival time after removal, which are used as the corrected initial arrival time of each shot to be picked up in the target work area.
7. The method according to claim 6, It is characterized in that After converting the target first arrival position and the corrected position of the removed first arrival position into the target first arrival time and the removed first arrival time respectively, the method further comprises: The target first arrival time and the first arrival time after removal are corrected by the following expression, and the corrected target first arrival time and the first arrival time after removal are used as the corrected first arrival time of each shot to be picked up in the target work area: Among them, w is the starting point of the smoothing window, h i is the smoothing coefficient, the value is 1, x k is the first arrival time of the kth track.
8. A DAS-VSP data first arrival wave picking device based on deep learning, It is characterized in that include: The first generation module is used to generate an initial area image according to the DAS-VSP data of the sample shot in the target work area, with the position of the current sample shot as the center, the shot line and the seismic channel receiving line as the horizontal axis and the vertical axis in the rectangular coordinate system; A second generation module is used to obtain the first arrival data of the sample shot in the target work area, and transform the first arrival data of the sample shot in the target work area into image data, and draw it on the regional image to generate a predicted regional image; A training module, used for training a known neural network model using the initial region image and the predicted region image to obtain a trained neural network model; A prediction module, used to input the initial area image of the to-be-predicted gun into the trained neural network model, and output the predicted area image of the to-be-predicted gun; The correction module is used to perform trend correction and value correction on the first arrival position of each track in the prediction area image of the predicted gun, and obtain the corrected first arrival time of each track of the gun to be picked up in the target work area.
9. An electronic device, It is characterized in that It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; The processor is used to implement the DAS-VSP data first arrival wave picking method based on deep learning as described in any one of claims 1 to 7 when executing the program stored in the memory.
10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the deep learning-based DAS-VSP data first-break picking method according to any one of claims 1 to 7 is implemented.
Citation Information
Patent Citations
Intelligent first arrival screening chromatographic inversion method and system based on deep learning
CN110308484A
Preliminary wave time difference correction method and device of VSP seismic data
CN110568499A
Time shifting VSP data acquisition system and method based on distributed optical fiber acoustic wave sensing
CN112130195A
Abnormal first arrival wave identification method and device
CN112305593A
First arrival pickup method and device based on deep learning network
CN112464725A
