DAS-VSP data first break picking method and device based on deep learning

Abstract

This disclosure relates to a method, apparatus, electronic device, and storage medium for first arrival acquisition of DAS-VSP data based on deep learning. The method includes: generating an initial region image based on the DAS-VSP data of sample shots within a target work area; acquiring the first arrival data of sample shots within the target work area and generating a predicted region image based on the first arrival data of sample shots within the target work area; training a known neural network model using the initial region image and the predicted region image; inputting the initial region image of the shot to be predicted into the trained neural network model and outputting the predicted region image of the shot to be predicted; performing trend correction and numerical correction on the first arrival position of each trace in the predicted region image of the shot to be predicted to obtain the corrected first arrival time of each trace of the shot to be acquired in the target work area. This method can achieve high-precision acquisition of low signal-to-noise ratio DAS-VSP first arrival data to meet the needs of subsequent real-time and high-precision data processing.

Description

Technical Field

[0001] This disclosure relates to the field of petroleum geophysical exploration data processing technology, and in particular to a method and apparatus for picking first arrival waves of DAS-VSP data based on deep learning, as well as electronic equipment and storage media. Background Technology

[0002] Currently, most oil and gas fields in China have entered the middle and late stages of exploration and development, and surface seismic exploration faces severe challenges in areas such as fine reservoir characterization, reservoir parameter acquisition, and improved seismic resolution. In recent years, the rapid development of DAS-VSP (Distributed Acoustic Sensing, DAS, Vertical Seismic Profile, VSP) acquisition technology has enabled the precise extraction of key parameters such as subsurface medium velocity field, anisotropy, TAR factor, formation absorption attenuation factor Q value, amplitude correction factor, and deconvolution. This is of great significance for providing high-precision reservoir parameters and improving the fine characterization of reservoirs surrounding oil and gas wells, effectively compensating for the shortcomings of surface seismic exploration.

[0003] However, the extraction of these key parameters relies on high-precision first arrivals. Currently, DAS-acquired VSP data suffers from two main problems in first arrival pickup: 1) The overall signal-to-noise ratio is relatively low. The DAS system is sensitive to stress changes in the surrounding environment of the sensing fiber. Even minor stress in the surrounding environment and weak noise within the system's internal components can generate significant noise. Actual data shows poor first arrival continuity in DAS-acquired data, with large differences in amplitude, frequency, and phase variations between adjacent channels. 2) As the well-source distance increases, the incident angle of the direct wave increases, leading to rapid amplitude response attenuation, making first arrivals almost impossible to identify effectively. Current mainstream first arrival pickup methods and commercial software are developed specifically for the first arrival characteristics of surface seismic exploration data. Their accuracy for DAS-VSP data pickup is low, requiring extensive manual interaction, resulting in high cost, low accuracy, and slow efficiency. High-efficiency, high-precision first arrival pickup technology is currently one of the key technologies restricting DAS-VSP exploration. Summary of the Invention

[0004] To solve the above-mentioned technical problems, or at least partially solve them, embodiments of this disclosure provide a method and apparatus, an electronic device, and a storage medium for picking up first arrival waves of DAS-VSP data based on deep learning.

[0005] In a first aspect, embodiments of this disclosure provide a deep learning-based method for picking first arrival waves in DAS-VSP data, the method comprising:

[0006] For the sample shot within the target work area, an initial regional image is generated in a rectangular coordinate system based on the DAS-VSP data of the sample shot, with the current position of the sample shot as the center and the shot line and seismic trace receiver line as the horizontal and vertical axes, respectively.

[0007] Acquire the initial arrival data of the sample shots in the target work area, transform the initial arrival data of the sample shots in the target work area into image data, plot it on the image of the area, and generate a prediction area image;

[0008] A known neural network model is trained using the initial region image and the predicted region image to obtain a trained neural network model;

[0009] The initial region image of the gun to be predicted is input into the trained neural network model, and the predicted region image of the gun to be predicted is output.

[0010] Trend correction and numerical correction are performed on the initial arrival position of each track in the prediction area image of the shot to be predicted, so as to obtain the initial arrival time of each track of the shot to be picked up in the target work area after correction.

[0011] In one possible implementation, the sample shot is determined through the following steps:

[0012] The preset proportion of sample shots to be extracted from all shot points in the target work area is determined based on the signal-to-noise ratio of the seismic data.

[0013] Based on the coordinates of each blast point and the wellhead coordinates within the target work area, calculate the well-source distance and azimuth of each blast point;

[0014] Based on the well-source distance and azimuth of each shot point, a sample shot of a predetermined proportion is uniformly extracted from all shot points in the target work area.

[0015] In one possible implementation, the known neural network model is an improved U-Net network model, which is obtained by improving the U-Net network model through the following steps:

[0016] The learning rate η of the U-Net network model was improved using the Adam optimization algorithm through the following expression:

[0017]

[0018] in,

[0019]

[0020] Among them, a j,k Let y be the probability value of the predicted initial arrival position of the j-th shot and the k-th trajectory. j,k Let be the probability value of the expected initial arrival position of the j-th shot and the k-th trajectory.

[0021] In one possible implementation, training a known neural network model using an initial region image and a predicted region image to obtain a trained neural network model includes:

[0022] The initial region image of each sample shot is used as input, and the predicted region image is used as output. The known neural network model is trained to obtain the trained neural network model.

[0023] In one possible implementation, the initial region image of the gun to be predicted is determined through the following steps:

[0024] First arrival time windows are set for the sample shots, and spatial interpolation techniques are used to determine the time window range for each shot to be picked up.

[0025] Using the position of the shot to be predicted as the center, and the shot line and seismic trace receiving line as the horizontal and vertical axes, an initial regional image is generated in a rectangular coordinate system based on the DAS-VSP data within the time window range.

[0026] In one possible implementation, the initial arrival position of each track in the prediction area image of the shot to be predicted is corrected for both trend and numerical values ​​to obtain the corrected initial arrival time of each track of the shot to be picked up in the target work area, including:

[0027] An initial linear regression model is constructed based on the first arrival position of each trace in the prediction region image of the shot to be predicted.

[0028] Determine the initial arrival position of the target where the deviation from the linear regression model exceeds a preset threshold;

[0029] Remove the target's initial arrival position from the initial arrival position of each trace in the prediction region image of the shot to be predicted to obtain the removed initial arrival position;

[0030] A new linear regression model is constructed based on the initial arrival position after removal;

[0031] The trend correction of the initial arrival position after removal is performed according to the new linear regression model to obtain the corrected position of the initial arrival position after removal;

[0032] The corrected position of the target initial arrival position is obtained by fitting a polynomial after removing the initial arrival position and the corresponding seismic trace number.

[0033] The target's initial arrival position and the corrected position after removal are transformed into the target's initial arrival time and the initial arrival time after removal, respectively, and used as the corrected initial arrival time of each shot to be picked up in the target work area.

[0034] In one possible implementation, after transforming the corrected positions of the target initial arrival position and the initial arrival position after removal into the target initial arrival time and the initial arrival time after removal, respectively, the method further includes:

[0035] The target's initial arrival time and the time after removal are corrected using the following expression. The corrected target's initial arrival time and the time after removal are then used as the corrected initial arrival time for each shot to be picked up in the target work area:

[0036]

[0037] Where w is the starting point of the smoothing window, h i The smoothing coefficient has a value of 1, x k Let k be the arrival time of the first path.

[0038] Secondly, embodiments of this disclosure provide a deep learning-based DAS-VSP data first arrival pickup device, comprising:

[0039] The first generation module is used to generate an initial regional image in a rectangular coordinate system based on the DAS-VSP data of the sample shot, with the current position of the sample shot as the center and the shot line and seismic trace receiving line as the horizontal and vertical axes, for the sample shot in the target work area.

[0040] The second generation module is used to acquire the initial arrival data of the sample shot in the target work area, and transform the initial arrival data of the sample shot in the target work area into image data, draw it on the image of the area, and generate a prediction area image.

[0041] The training module 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.

[0042] The prediction module is used to input the initial region image of the gun to be predicted into the trained neural network model and output the predicted region image of the gun to be predicted.

[0043] The correction module is used to perform trend correction and numerical correction on the initial arrival position of each track in the prediction area image of the shot to be predicted, so as to obtain the initial arrival time of each track of the shot to be picked up in the target work area after correction.

[0044] Thirdly, embodiments of this disclosure provide 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] When the processor executes the program stored in memory, it implements the above-mentioned deep learning-based DAS-VSP data first arrival wave picking method.

[0047] Fourthly, embodiments of this disclosure provide a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the aforementioned deep learning-based DAS-VSP data first arrival wave picking method.

[0048] Compared with the prior art, the technical solutions provided in this disclosure have at least some or all of the following advantages:

[0049] The deep learning-based DAS-VSP data first arrival acquisition method described in this embodiment generates an initial region image in a Cartesian coordinate system based on the DAS-VSP data of the sample shot, with the current position of the sample shot as the center and the shot line and seismic trace receiver line as the horizontal and vertical axes, respectively. It then acquires the first arrival data of the sample shot within the target area, transforms this data into image data, and plots it on the region image to generate a predicted region image. A known neural network model is trained using the initial region image and the predicted region image to obtain a trained neural network model. Finally, the initial region image of the shot to be predicted is input into the trained neural network model. The network model outputs the predicted area image of the shot to be predicted. The first arrival position of each trace in the predicted area image of the shot to be predicted is corrected by trend and numerical correction to obtain the corrected first arrival time of each trace of the shot to be picked in the target work area. Using manually picked first arrival data as samples, a deep network model is trained by deep learning to establish the mapping between input and output. The feature extractor composed of convolutional layers and subsampling layers extracts more abstract features from the data, thereby predicting the first arrival time of DAS-VSP data. Then, by identifying and correcting abnormal first arrivals, high-precision picking of low signal-to-noise ratio DAS-VSP first arrival data is achieved to meet the needs of real-time and high-precision processing of subsequent data. Attached Figure Description

[0050] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0051] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0052] Figure 1 This illustration schematically shows a flowchart of a deep learning-based DAS-VSP data first arrival wave picking method according to an embodiment of the present disclosure;

[0053] Figure 2A schematic diagram of single-gun data for DAS-VSP according to an embodiment of the present disclosure is shown.

[0054] Figure 3 This illustration schematically shows a preliminary prediction effect diagram according to an embodiment of the present disclosure;

[0055] Figure 4 This diagram illustrates the initial trend correction effect according to an embodiment of the present disclosure.

[0056] Figure 5 This illustration schematically shows the initial to fine-tuning effect according to an embodiment of the present disclosure;

[0057] Figure 6 A schematic diagram illustrates a structural block diagram of a deep learning-based DAS-VSP data first-arrival pickup apparatus according to an embodiment of the present disclosure; and

[0058] Figure 7 A schematic block diagram of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0060] See Figure 1 The embodiments of this disclosure provide a deep learning-based method for picking first arrival waves in DAS-VSP data, the method comprising:

[0061] S1. For the sample shot within the target work area, an initial area image is generated in a rectangular coordinate system based on the DAS-VSP data of the sample shot, with the current position of the sample shot as the center and the shot line and seismic trace receiver line as the horizontal and vertical axes, respectively.

[0062] In some embodiments, the DAS-VSP data are single-shot DAS-VSP seismic data that are excited and acquired and recorded, with m traces and n sampling points per trace.

[0063] S2, acquire the initial arrival data of the sample shot in the target work area, transform the initial arrival data of the sample shot in the target work area into image data, plot it on the image of the area, and generate the prediction area image.

[0064] In some embodiments, the first arrival waves of the extracted sample shots are finely picked up, and then the picked first arrival data is transformed into image data. Before the transformation, parameters such as the display resolution, trace interval, and time interval 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 training a deep network model, and deep learning technology is used to abstract and extract initial arrival features and initial arrival location information from different levels.

[0066] In this embodiment, the first arrival time of the sample shots is precisely calibrated manually, and sample labels are generated to ensure the accuracy of the sample data input to 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: Input the initial region image of the gun to be predicted into the trained neural network model, and output the predicted region image of the gun to be predicted.

[0069] S5 performs trend correction and numerical correction on the initial arrival position of each track in the prediction area image of the shot to be predicted, and obtains the initial arrival time of each track of the shot to be picked up in the target work area after correction.

[0070] In this embodiment, trend correction and numerical correction are performed on the predicted arrival time, which can meet the requirements of real-time and high-precision data processing.

[0071] In this embodiment, in step S1, the sample shot is determined through the following steps:

[0072] The preset proportion of sample shots to be extracted from all shot points in the target work area is determined based on the signal-to-noise ratio of the seismic data.

[0073] Based on the coordinates of each blast point and the wellhead coordinates within the target work area, calculate the well-source distance and azimuth of each blast point;

[0074] Based on the well-source distance and azimuth of each shot point, a sample shot of a predetermined proportion is uniformly extracted from all shot points in the target work area.

[0075] In some embodiments, if the number of shots acquired across the entire area is S, then h% of the sample shots are extracted based on the signal-to-noise ratio of the seismic data. Simultaneously, based on the shot point coordinates and wellhead coordinates, the well-to-source distance and azimuth angle for each shot are calculated. The well-to-source distance for each azimuth is then averaged for extraction. That is, if a total of g sample shots are extracted, then... If angles and ρ meters are extracted, then each azimuth and each interval will be sampled. The cannons are used to ensure that the sample cannons are evenly distributed and can cover the initial arrival features of the entire area.

[0076] In some embodiments, h% of the sample shots are extracted based on the signal-to-noise ratio (SNR) of the seismic data. If the SNR is greater than 2, then h is 5, that is, 5% of the sample shots are extracted; if the SNR is less than 1, then h is 10, that is, 10% of the sample shots are extracted; generally, h is 8, that is, 8% of the sample shots are extracted.

[0077] In some embodiments, average extraction is performed based on the well-source distance in each azimuth, at intervals of angles. The spacing ρ is set based on the signal-to-noise ratio of the data; the lower the signal-to-noise ratio, the smaller the angle. The smaller the value of the spacing ρ, the more uniform the distribution of the sample shots. Generally, the angle... The value is 20, and the spacing ρ is 500.

[0078] In this embodiment, seismic data is sampled according to well-source distance and azimuth to ensure uniform distribution of sample shots and to 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, which is obtained by improving the U-Net network model through the following steps:

[0080] The learning rate η of the U-Net network model was improved using the Adam optimization algorithm through the following expression:

[0081]

[0082] in,

[0083]

[0084] Among them, a j,k Let y be the probability value of the predicted initial arrival position of the j-th shot and the k-th trajectory. j,k Let be the probability value of the expected initial arrival position of the j-th shot and the k-th trajectory.

[0085] In some embodiments, the value of ε is set differently: for seismic data with a high signal-to-noise ratio, ε is set to a larger value; for seismic data with a low signal-to-noise ratio, ε is set to a smaller value; generally, ε is set to 0.6 by default.

[0086] In some embodiments, the learning rate η is given a large learning rate in the early stage of training in order to speed up the training efficiency. As the number of training times increases, the matching rate between the predicted initial arrival position and the actual position increases, that is, T continuously increases. If T is greater than a given threshold, the learning rate is gradually reduced to improve the model accuracy and thus improve the initial arrival detection accuracy.

[0087] In this embodiment, step S3, which involves training a known neural network model using the initial region image and the predicted region image to obtain a trained neural network model, includes:

[0088] The initial region image of each sample shot is used as input, and the predicted region image is used as output. The known neural network model is trained to obtain the trained neural network model.

[0089] In this embodiment, in step S4, the initial region image of the shot to be predicted is determined through the following steps:

[0090] First arrival time windows are set for the sample shots, and spatial interpolation techniques are used to determine the time window range for each shot to be picked up.

[0091] Using the position of the shot to be predicted as the center, and the shot line and seismic trace receiving line as the horizontal and vertical axes, an initial regional image is generated in a rectangular coordinate system based on the DAS-VSP data within the time window range.

[0092] In some embodiments, the seismic gather data is divided into N regions in a Cartesian coordinate system, with the sample shot location as the center point, the shot line as the vertical axis, and the seismic trace receiver line as the horizontal axis. Seismic traces within each region are sorted in ascending order of shot-receiver distance. A first-arrival time window is set for each region. Based on the first-arrival time window for each region of the sample shot, spatial interpolation is used to determine the time window range for each first-arrival of the shot to be picked up in the work area. Specifically, the seismic gather data for each shot is divided into N regions in a Cartesian coordinate system. N can be an integer such as 1, 2, 4, or 8 (i.e., a power of 2 greater than or equal to 0). A larger value for N results in a smaller first-arrival time window range and higher first-arrival picking accuracy. The default is to divide the data into one region.

[0093] In some embodiments, an initial arrival time window is set for each region. 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, step S5 involves performing trend correction and numerical correction on the initial arrival position of each track in the prediction area image of the shot to be predicted, to obtain the corrected initial arrival time of each track of the shot to be picked up in the target work area, including:

[0095] An initial linear regression model is constructed based on the first arrival position of each trace in the prediction region image of the shot to be predicted.

[0096] Determine the initial arrival position of the target where the deviation from the linear regression model exceeds a preset threshold;

[0097] Remove the target's initial arrival position from the initial arrival position of each trace in the prediction region image of the shot to be predicted to obtain the removed initial arrival position;

[0098] A new linear regression model is constructed based on the initial arrival position after removal;

[0099] The trend correction of the initial arrival position after removal is performed according to the new linear regression model to obtain the corrected position of the initial arrival position after removal;

[0100] The corrected position of the target initial arrival position is obtained by fitting a polynomial after removing the initial arrival position and the corresponding seismic trace number.

[0101] The target's initial arrival position and the corrected position after removal are transformed into the target's initial arrival time and the initial arrival time after removal, respectively, and used as the corrected initial arrival time of each shot to be picked up in the target work area.

[0102] In some embodiments, the initial arrival time of the i-th pickup for each shot is t. i For each shot, the first arrivals of N seismic traces are collected. An expression for the objective function J(β) is established, and the regression coefficient β for minimizing the objective function is calculated, thereby establishing an initial linear regression model.

[0103] J(β)=∑(τ i -t i β) 2 +∑λ i β 2

[0104] Where J(β) is the objective function, t i For the initial arrival time of each course, τ i Let λ be the initial arrival time for each linear regression. i This is a penalty term for the objective function, whose value is selected based on the variance and bias used to balance the model, thereby improving the accuracy of the regression coefficient β calculation.

[0105] In some embodiments, deviations affecting the linear regression model are deleted according to the following expression:

[0106]

[0107] in, Let t be the i-th main diagonal element of the statistical matrix H, and p be the independent variable t. i The number of elements, where N is the sample size of the modeling dataset. If the diagonal elements satisfy the above expression, then it represents the i-th initial arrival anomaly.

[0108] In some embodiments, because linear regression models are susceptible to extreme values, statistical methods are needed to test for deviations from the observed samples. If outlier data is found during the modeling process, the dataset is rectified. The initial linear regression model is solved based on the derived regression coefficients, and the linear regression model is expressed as follows:

[0109]

[0110] Where μ is the linear regression equation, t is the initial 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 deleting the initial arrival of anomalies, a new linear regression model is obtained for each shot using the initial linear regression model.

[0112] In some embodiments, if there is seismic data without preliminary first arrival results and a linear regression model cannot be established, a linear regression model can be obtained by interpolating an existing model using the inverse distance weighting method, thus obtaining the first arrival trend line for each seismic data.

[0113] In this embodiment, after transforming the corrected positions of the initial target location and the initial location after removal into the initial target location time and the initial location after removal, respectively, the method further includes:

[0114] The target's initial arrival time and the time after removal are corrected using the following expression. The corrected target's initial arrival time and the time after removal are then used as the corrected initial arrival time for each shot to be picked up in the target work area:

[0115]

[0116] Where w is the starting point of the smoothing window, h i The smoothing coefficient has a value of 1, x k Let L be the arrival time of the kth channel, and L be the time window length.

[0117] In some embodiments, abnormal first arrivals with large deviations are removed, and the channel numbers with predicted first arrivals for each shot are counted. Assuming the minimum channel number is Φ and the maximum channel number is Ψ, the unpredicted seismic traces from Φ to Ψ can be located by fitting a polynomial to the first arrival time and channel number. At the same time, the Savitzky-Gola algorithm is used to smooth the fitted first arrivals to obtain a refined first arrival time. The calculation expression is as follows:

[0118]

[0119] Where w is the starting point of the smoothing window, h i The smoothing coefficient has a value of 1, x k Let L be the arrival time of the kth channel, and L be the time window length.

[0120] This disclosure extracts sample shots based on 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 shots; trains a high-precision deep network model using an improved U-Net; uses the trained deep network model to predict the data to be predicted to obtain the first arrival position of each trace; corrects the predicted first arrival trend using a multiple linear regression model; and combines the Savitzky-Gola algorithm to refine the first arrival, resulting in a significant improvement in both picking accuracy and efficiency.

[0121] The following are applied to Figure 2 The deep learning-based first arrival picking method for DAS-VSP single-shot data shown includes:

[0122] 1) Excite and acquire / record DAS-VSP single-shot seismic data, such as Figure 2 As shown, the single-gun receiver has 430 channels, a sampling interval of 4 milliseconds, and 500 sampling points per channel.

[0123] 2) Sample Shot Extraction: Seismic data was sampled based on well-source distance and azimuth. A total of 32,895 shots were collected across the entire area. Based on the signal-to-noise ratio, 1,600 shots were extracted, using a grid of 20° and 500 meters, resulting in an average of 5 shots per grid to ensure even distribution and coverage of first-arrival characteristics across the entire area. Simultaneously, a time window was set for the sample shots, with a window length of 400 sample points. Spatial interpolation techniques were used to determine the time window range for each shot to be extracted.

[0124] 3) Sample Data Generation: First, the extracted sample shots are manually and meticulously calibrated. Then, the first arrival data is extracted into density images. Before extraction, parameters such as the display resolution, trace interval, and time interval of the seismic data must be set and recorded to ensure the first arrival time is accurately located between the first arrival data and the images. The extracted density images serve as input labels for training the deep network model, using deep learning techniques to abstract and extract first arrival features and location information from different levels.

[0125] 4) The learning rate η of the deep network U-Net was improved by combining the Adam optimization algorithm:

[0126]

[0127] in,

[0128]

[0129] Among them, a j,k Let y be the probability value of the predicted initial arrival position of the j-th shot and the k-th trajectory. j,k Let be the probability value of the expected initial arrival position of the j-th shot and the k-th trajectory.

[0130] 5) The improved U-Net network is used for model training to obtain a deep network model;

[0131] 6) Using step 2), extract the data within the first arrival time window of the shot to be predicted. Simultaneously, using steps 4) and 5), perform prediction on the data using the trained deep network model. The prediction results are as follows: Figure 3 As shown, the overall prediction accuracy is high, but due to interference noise, the prediction of the first arrival anomaly at channel numbers 20, 40, 232, 258, 294, 347, 367, 402, and 422 did not pick up the accurate first arrival position.

[0132] 7) Utilize initial trend correction techniques to... Figure 3 The predicted first arrival is revised to depict the overall trend of the first arrival in the earthquake data, such as... Figure 4 As shown, the revised initial arrival trend is reasonable.

[0133] 8) To meet the requirements of high-precision data processing, fine correction is needed for the initial arrival. Fine correction includes, for example... Figure 5 As shown in the figure, the overall picking accuracy is high after correction, and the continuity of the initial arrival is significantly improved, which can meet the needs of extracting key parameters.

[0134] See Figure 6 The embodiments of this disclosure provide a deep learning-based DAS-VSP data first arrival pickup device, comprising:

[0135] The first generation module 11 is used to generate an initial regional image in a rectangular coordinate system based on the DAS-VSP data of the sample shot, with the current position of the sample shot as the center and the shot line and seismic trace receiving line as the horizontal and vertical axes, for the sample shot in the target work area.

[0136] The second generation module 12 is used to acquire the initial arrival data of the sample shot in the target work area, and transform the initial arrival data of the sample shot in the target work area into image data, draw it on the image of the area, and generate a prediction area image.

[0137] 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 is used to input the initial region image of the gun to be predicted into the trained neural network model and output the predicted region image of the gun to be predicted.

[0139] The correction module 15 is used to perform trend correction and numerical correction on the initial arrival position of each track in the prediction area image of the shot to be predicted, so as to obtain the initial arrival time of each track of the shot to be picked up in the target work area after correction.

[0140] The first-arrival pickup device for DAS-VSP data based on deep learning disclosed herein first extracts sample shots from seismic data according to well-source distance and azimuth, sets first-arrival time windows for the sample shots, and uses spatial interpolation to determine the time window range for each shot to be picked up; secondly, it performs precise first-arrival calibration on the sample shots and generates sample labels to ensure the accuracy of the input sample data; thirdly, it uses an improved U-Net network for model training to obtain a deep network model; then, it uses the obtained deep network model to predict the data to be picked up; finally, it performs trend and fine-tuning on the predicted first arrivals to meet the requirements of real-time and high-precision data processing.

[0141] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0142] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0143] In the above embodiments, any plurality 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 into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality 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 hardware circuitry, 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-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. 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, which can perform corresponding functions when the computer program module is run.

[0144] See Figure 7 The electronic device provided in the embodiments of this 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 is used to store computer programs;

[0146] When processor 1110 executes the program stored in memory 1130, it implements the following deep learning-based DAS-VSP data first arrival wave picking method:

[0147] For the sample shot within the target work area, an initial regional image is generated in a rectangular coordinate system based on the DAS-VSP data of the sample shot, with the current position of the sample shot as the center and the shot line and seismic trace receiver line as the horizontal and vertical axes, respectively.

[0148] Acquire the initial arrival data of the sample shots in the target work area, transform the initial arrival data of the sample shots in the target work area into image data, plot it on the image of the area, and generate a prediction area image;

[0149] A known neural network model is trained using the initial region image and the predicted region image to obtain a trained neural network model;

[0150] The initial region image of the gun to be predicted is input into the trained neural network model, and the predicted region image of the gun to be predicted is output.

[0151] Trend correction and numerical correction are performed on the initial arrival position of each track in the prediction area image of the shot to be predicted, so as to obtain the initial arrival time of each track of the shot to be picked up in the target work area after correction.

[0152] The first arrival picking method for DAS-VSP data disclosed herein involves extracting sample shots based on well-source distance and azimuth, setting first arrival time windows, determining the time window range for each shot to be picked using spatial interpolation, accurately calibrating the first arrivals of the sample shots, generating sample labels, training a deep network model using an improved U-Net network, predicting the data to be picked, and performing trend and fine-tuning on the predicted first arrivals. This method can achieve high-precision picking of low signal-to-noise ratio DAS-VSP first arrival data to meet the needs of subsequent real-time and high-precision data processing.

[0153] The aforementioned communication bus 1140 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 1140 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus.

[0154] The communication interface 1120 is used for communication between the above-mentioned electronic device and other devices.

[0155] The memory 1130 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory 1130 may also be at least one storage device located remotely from the aforementioned processor 1110.

[0156] The processor 1110 mentioned above 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 gate or transistor logic devices, or discrete hardware components.

[0157] Embodiments of this disclosure also provide a computer-readable storage medium. The computer-readable storage medium stores a computer program that, when executed by a processor, implements the deep learning-based DAS-VSP data first-arrival acquisition method as described above.

[0158] The computer-readable storage medium may be included in the device / apparatus described in the above embodiments; or it may exist independently and not assembled into the device / apparatus. The computer-readable storage medium carries one or more programs, which, when executed, implement the deep learning-based DAS-VSP data first arrival wave picking method according to the embodiments of this disclosure.

[0159] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0160] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0161] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily 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 this disclosure. Therefore, this disclosure is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for first arrival picking of DAS-VSP data based on deep learning, characterized in that, The method includes: For the sample shot within the target work area, an initial regional image is generated in a rectangular coordinate system based on the DAS-VSP data of the sample shot, with the current position of the sample shot as the center and the shot line and seismic trace receiver line as the horizontal and vertical axes, respectively. Acquire the initial arrival data of the sample shot in the target work area, transform the initial arrival data of the sample shot in the target work area into image data, plot it on the initial area image, and generate the prediction area image; A known neural network model is trained using the initial region image and the predicted region image to obtain a trained neural network model; The initial region image of the gun to be predicted is input into the trained neural network model, and the predicted region image of the gun to be predicted is output. Trend correction and numerical correction are performed on the initial arrival position of each track in the prediction area image of the shot to be predicted, and the initial arrival time of each track of the shot to be picked up in the target work area after correction is obtained. The initial arrival position of each track in the prediction area image of the shot to be predicted is corrected by trend and numerical values ​​to obtain the corrected initial arrival time of each track of the shot to be picked up in the target work area, including: An initial linear regression model is constructed based on the first arrival position of each trace in the prediction region image of the shot to be predicted. Determine the initial arrival position of the target where the deviation from the linear regression model exceeds a preset threshold; Remove the target's initial arrival position from the initial arrival position of each trace in the prediction region image of the shot to be predicted to obtain the removed initial arrival position; A new linear regression model is constructed based on the initial arrival position after removal; The trend correction of the initial arrival position after removal is performed according to the new linear regression model to obtain the corrected position of the initial arrival position after removal; The corrected position of the target initial arrival position is obtained by fitting a polynomial after removing the initial arrival position and the corresponding seismic trace number. The target's initial arrival position and the corrected position after removal are transformed into the target's initial arrival time and the initial arrival time after removal, respectively. The target's initial arrival time and the initial arrival time after removal are corrected by the following expression. The corrected target's initial arrival time and the initial arrival time after removal are used as the corrected initial arrival time of each shot to be picked up in the target work area. in, As the starting point of the smoothing time window, This is the smoothing coefficient, with a value of 1. Let L be the arrival time of the kth channel, and L be the time window length.

2. The method of claim 1, wherein, The sample gun was determined through the following steps: The preset proportion of sample shots to be extracted from all shot points in the target work area is determined based on the signal-to-noise ratio of the seismic data. Based on the coordinates of each blast point and the wellhead coordinates within the target work area, calculate the well-source distance and azimuth of each blast point; Based on the well-source distance and azimuth of each shot point, a sample shot of a predetermined proportion is uniformly extracted from all shot points in the target work area.

3. The method of claim 1, wherein, The known neural network model is an improved U-Net network model, which is obtained by improving the U-Net network model through the following steps: The learning rate of the U-Net network model is calculated using the Adam optimization algorithm and the following expression. Improvements have been made: in, in, For the first Cannon The probability value of predicting the initial arrival position. For the first Cannon The probability value of the expected initial position.

4. The method according to claim 1, characterized in that, The process of training a known neural network model using an initial region image and a predicted region image to obtain a trained neural network model includes: The initial region image of each sample shot is used as input, and the predicted region image is used as output. The known neural network model is trained to obtain the trained neural network model.

5. The method of claim 1, wherein, The initial region image of the gun to be predicted is determined through the following steps: First arrival time windows are set for the sample shots, and spatial interpolation techniques are used to determine the time window range for each shot to be picked up. Using the position of the shot to be predicted as the center, and the shot line and seismic trace receiving line as the horizontal and vertical axes, an initial regional image is generated in a rectangular coordinate system based on the DAS-VSP data within the time window range.

6. A deep learning-based DAS-VSP data first break picking device, characterized in that, include: The first generation module is used to generate an initial regional image in a rectangular coordinate system based on the DAS-VSP data of the sample shot, with the current position of the sample shot as the center and the shot line and seismic trace receiving line as the horizontal and vertical axes, for the sample shot in the target work area. The second generation module is used to acquire the initial arrival data of the sample shot in the target work area, and transform the initial arrival data of the sample shot in the target work area into image data, draw it on the initial area image, and generate the prediction area image. The training module 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. The prediction module is used to input the initial region image of the gun to be predicted into the trained neural network model and output the predicted region image of the gun to be predicted. The correction module is used to perform trend correction and numerical correction on the initial arrival position of each track in the prediction area image of the shot to be predicted, so as to obtain the initial arrival time of each track of the shot to be picked up in the target work area after correction. The initial arrival position of each track in the prediction area image of the shot to be predicted is corrected by trend and numerical values ​​to obtain the corrected initial arrival time of each track of the shot to be picked up in the target work area, including: An initial linear regression model is constructed based on the initial arrival position of each trace in the prediction region image of the shot to be predicted. Determine the initial position of the target where the deviation from the linear regression model exceeds a preset threshold; Remove the target's initial arrival position from the initial arrival position of each trace in the prediction region image of the shot to be predicted to obtain the initial arrival position after removal; A new linear regression model is constructed based on the initial arrival position after removal; The trend correction of the initial arrival position after removal is performed according to the new linear regression model to obtain the corrected position of the initial arrival position after removal; The corrected position of the target initial arrival position is obtained by fitting a polynomial after removing the initial arrival position and the corresponding seismic trace number. The target's initial arrival position and the corrected position after removal are transformed into the target's initial arrival time and the initial arrival time after removal, respectively. The target's initial arrival time and the initial arrival time after removal are corrected by the following expression. The corrected target's initial arrival time and the initial arrival time after removal are used as the corrected initial arrival time of each shot to be picked up in the target work area. in, As the starting point of the smoothing time window, This is the smoothing coefficient, with a value of 1. Let L be the arrival time of the kth channel, and L be the time window length.

7. An electronic device, comprising: 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 through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the deep learning-based DAS-VSP data first arrival wave picking method as described in any one of claims 1-5.

8. A computer readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the deep learning-based DAS-VSP data first arrival wave picking method as described in any one of claims 1-5.

Images