Artificial intelligence first arrival pickup method based on data optimization and related equipment

Abstract

The invention provides an artificial intelligence first arrival pickup method based on data optimization and related equipment, first arrival pickup is performed on first arrival data through a pre-trained artificial intelligence first arrival pickup model, the pickup speed is high, the pickup accuracy is high, and the pickup efficiency is improved through continuous pickup. Accuracy of the model can be gradually increased by collecting enough data. Meanwhile, the invention also provides an artificial intelligence first arrival pickup model training method. Optimization processing is carried out by utilizing the advantages of strong main lobe energy and weak side lobe of the Yu wavelet, so that the first arrival wavelet is more focused while the background noise is suppressed, which is a single-channel processing measure; on the basis of a refracted wave propagation theory, a virtual seismic source is solved by using adjacent channels, the channel is reconstructed, superposition is performed, and the signal-to-noise ratio of first arrival data is enhanced, which is a multi-channel processing measure; when the first-arrival reconstruction is carried out, the preliminary static correction value is applied to reduce the inter-channel dislocation, the transverse direction of the first-arrival wave group is smoother, and the reconstruction effect can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas seismic exploration, and in particular to an artificial intelligence first arrival picking method based on data optimization and related equipment. Background Art

[0002] In the field of oil and gas seismic exploration, the existing first-arrival picking algorithms can be divided into three categories. The first is based on single-channel amplitude mutation, such as the energy ratio method. This is a fully automatic first-arrival picking method with high efficiency, but it is only applicable to data with weak background noise and strong signal energy; the first is based on waveform similarity between channels, such as the cross-correlation method of interactive picking. This type of method does not require high signal-to-noise ratio, but requires human-computer interaction to pick seed channels, which is inefficient and cannot meet production needs for the current data volume of tens of thousands of channels per shot; the third is artificial intelligence methods, which train the neural network model based on the given labeled shot as a seed, and then automatically pick up other non-seed shots, while taking into account the characteristics of amplitude attributes and waveform similarity between channels. Compared with the first two methods, the efficiency is improved, but the current artificial intelligence technology is implemented on raw seismic data. For data with low signal-to-noise ratio and more complex waveforms, the picking effect cannot meet production needs.

[0003] With the advancement of exploration technology, onshore seismic acquisition uses a large number of controllable vibrators and node instrument single-point detectors, which have weak energy and strong background noise. Social and economic development, such as roads, railways, canals, oil wells, pipelines, wind power and villages, are often distributed in the work area, generating a lot of strong energy external interference. The received seismic data has a low signal-to-noise ratio and a messy wave field, resulting in poor effect and low efficiency of the current first-arrival picking technology, which seriously affects the cycle of seismic data processing and imaging quality. Summary of the invention

[0004] The purpose of the present invention is to provide an artificial intelligence first-arrival picking method and related equipment based on data optimization, so as to overcome the problem that artificial intelligence technology in the prior art is implemented on raw seismic data, and the picking effect cannot meet production needs for data with low signal-to-noise ratio and complex waveform.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] An artificial intelligence first arrival picking method based on data optimization, comprising:

[0007] Acquire original seismic data, correct the original seismic data to obtain first arrival data, optimize the first arrival data to obtain optimized first arrival data;

[0008] Inputting the optimized first arrival data into a pre-trained artificial intelligence first arrival picking model to obtain the first arrival time;

[0009] Among them, the pre-trained artificial intelligence first-arrival picking model is an artificial intelligence first-arrival picking model obtained by training with multiple pre-processed first-arrival data.

[0010] Preferably, the artificial intelligence first arrival picking model training method comprises:

[0011] Obtaining original seismic data, performing linear correction on the original seismic data, and obtaining first arrival data;

[0012] Optimize and reshape the first arrival data to obtain processed first arrival data;

[0013] Automatically pick up the processed first arrival data to obtain the first arrival time;

[0014] The high-frequency static correction is estimated and applied to the first-arrival data after the initial processing to obtain the first-arrival data after the preliminary static correction is applied;

[0015] Reconstructing the first arrival data after applying the preliminary static correction to obtain reconstructed first arrival data;

[0016] Extract and reconstruct the label cannons in the first arrival data and perform manual picking;

[0017] Based on the first arrival time of the picked label gun, the artificial intelligence model is trained until convergence and stability are achieved to obtain the artificial intelligence first arrival picking model.

[0018] Preferably, the first arrival data linear correction includes:

[0019] Select the label gun, interactively determine the linear correction parameters and perform linear correction.

[0020] Preferably, the linearly corrected first arrival data is optimized and shaped, specifically:

[0021] Perform Fourier transform on the corrected first arrival data to obtain the spectrum of each first arrival data;

[0022] According to the spectrum of the first arrival signal, the spectrum is shaped by giving the start and end frequencies of the Yu wavelet optimization;

[0023] The shaped spectrum is subjected to inverse Fourier transform to obtain the seismic trace in the time domain, thus completing the optimization of the first arrival data.

[0024] Preferably, the energy ratio method is used to automatically pick up the processed first arrival data.

[0025] Preferably, the high-frequency static correction amount estimation and application of the first arrival data after picking up are specifically as follows:

[0026] The high-frequency static correction is calculated by using the first-arrival residual static correction estimation method. Then, this set of preliminary static correction is applied to the first-arrival data to obtain the first-arrival data after static correction is applied, which is used to eliminate inter-channel jitter.

[0027] Preferably, the reconstruction of the application initial arrival data is specifically as follows:

[0028] Based on the refraction wave propagation theory, the virtual source is constructed using cross-correlation, and then the convolution algorithm is used to reconstruct the first arrival data.

[0029] An artificial intelligence first arrival picking system based on data optimization, comprising:

[0030] Data acquisition module: acquire original seismic data, correct the original seismic data to obtain first arrival data, optimize the first arrival data to obtain optimized first arrival data;

[0031] First arrival picking module: input the optimized first arrival data into a pre-trained artificial intelligence first arrival picking model to obtain the first arrival time;

[0032] Among them, the pre-trained artificial intelligence first-arrival picking model is an artificial intelligence first-arrival picking model obtained by training with multiple pre-processed first-arrival data.

[0033] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of an artificial intelligence first-arrival picking method based on data optimization are implemented.

[0034] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of an artificial intelligence first-arrival picking method based on data optimization.

[0035] Compared with the prior art, the present invention has the following beneficial effects: the present invention provides an artificial intelligence first-arrival picking method based on data optimization, which performs first-arrival picking on first-arrival data through a pre-trained artificial intelligence first-arrival picking model, with fast picking speed and high picking accuracy, and through continuous picking, sufficient data is collected to gradually increase the accuracy of the model.

[0036] Furthermore, a number of first-arrival preprocessing measures are used, including linear correction, first-arrival optimization, first-arrival reconstruction and preliminary static correction. The first-arrival data after preprocessing have a high signal-to-noise ratio, strong lateral continuity and high wave field recognizability, and the picking effect and efficiency are greatly improved. The artificial intelligence first-arrival picking model is then trained to improve the accuracy of model training. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart of an artificial intelligence first arrival picking method based on data optimization of the present invention;

[0038] Figure 2 This is a block diagram of an artificial intelligence first arrival picking system based on data optimization of the present invention;

[0039] Figure 3 This is a flow chart of an artificial intelligence first arrival picking model training method of the present invention;

[0040] Figure 4 A flowchart of artificial intelligence first arrival picking based on data optimization provided by an embodiment of the present invention;

[0041] Figure 5 A schematic diagram of an interactive linear correction frame provided by an embodiment of the present invention;

[0042] Figure 6 Schematic diagrams of three types of wavelets (upper row) and spectra (lower row) provided for the embodiments of the present invention (left: bandpass wavelet; middle: Ricker wavelet; right: Yu wavelet);

[0043] Figure 7 A schematic diagram of an example of application of Yu's wavelet optimization in suppressing vibrator side lobes provided by an embodiment of the present invention (left: before optimization; right: after optimization);

[0044] Figure 8 A schematic diagram of an example of application of Yu's wavelet optimization in improving the first-arrival signal-to-noise ratio provided by an embodiment of the present invention (left: before optimization; right: after optimization);

[0045] Fig. 9 A schematic diagram of obtaining a virtual seismic source by cross-correlation of refracted waves provided in an embodiment of the present invention;

[0046] Fig.10 A schematic diagram of convolution reconstruction of the first arrival wave field provided in an embodiment of the present invention;

[0047] Fig.11 A schematic diagram of an example of the application effect of first-break reconstruction provided by an embodiment of the present invention, (upper left: original data; upper right: after Yu wavelet optimization; lower: after optimization + reconstruction);

[0048] Fig.12 A schematic diagram of original seismic data of mountainous areas provided by an embodiment of the present invention;

[0049] Fig.13 A schematic diagram of first arrival data after linear correction provided by one embodiment of the present invention;

[0050] Fig.14 A schematic diagram of first arrival data after first arrival optimization is provided for one embodiment of the present invention;

[0051] Fig.15 A schematic diagram of first arrival data after optimization and application of preliminary static correction provided by one embodiment of the present invention;

[0052] Fig.16 A schematic diagram of first arrival picking results using conventional artificial intelligence technology provided for one embodiment of the present invention;

[0053] Fig.17 A schematic diagram of an artificial intelligence first arrival picking effect provided by an embodiment of the present invention;

[0054] Fig.18 Original seismic data of mountain data provided by another embodiment of the present invention;

[0055] Fig.19 A schematic diagram of first arrival data after linear correction is provided for another embodiment of the present invention;

[0056] Fig. 20 A schematic diagram of optimized and reconstructed first arrival data provided by yet another embodiment of the present invention;

[0057] Fig.21 A schematic diagram of first arrival data after applying preliminary static correction is provided for another embodiment of the present invention;

[0058] Fig. 22 A schematic diagram of a conventional artificial intelligence first arrival picking effect provided for yet another embodiment of the present invention;

[0059] Fig.23 A schematic diagram of an artificial intelligence first-arrival picking result provided by yet another embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0061] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0063] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0064] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0065] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0066] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0067] like Figure 1 As shown, the present invention provides an artificial intelligence first arrival picking method based on data optimization, comprising:

[0068] S101 acquires original seismic data, corrects the original seismic data to obtain first arrival data, and optimizes the first arrival data to obtain optimized first arrival data;

[0069] S102 inputs the optimized first arrival data into a pre-trained artificial intelligence first arrival picking model to obtain the first arrival time;

[0070] Among them, the pre-trained artificial intelligence first-arrival picking model is an artificial intelligence first-arrival picking model obtained by training with multiple pre-processed first-arrival data.

[0071] like Figure 3As shown, the artificial intelligence first arrival picking model training method includes:

[0072] S301 acquires original seismic data, performs linear correction on the original seismic data, and obtains first arrival data;

[0073] S302 optimizes and reshapes the first arrival data to obtain processed first arrival data;

[0074] S303 automatically picks up the processed first arrival data to obtain the first arrival time;

[0075] S304 estimates and applies high-frequency static corrections to the processed first arrival data to obtain first arrival data after applying preliminary static corrections;

[0076] S305 reconstructs the first arrival data after applying the preliminary static correction amount to obtain reconstructed first arrival data;

[0077] S306 extracts the label guns in the reconstructed first arrival data and performs manual picking;

[0078] S307 performs artificial intelligence model training based on the first arrival time of the picked label gun until convergence is stable, and obtains the artificial intelligence first arrival picking model.

[0079] Linear correction of first arrival data, including: selecting label shots, interactively determining linear correction parameters and performing linear correction.

[0080] The first arrival data after linear correction is optimized and shaped, specifically:

[0081] Perform Fourier transform on the corrected first arrival data to obtain the spectrum of each first arrival data;

[0082] According to the spectrum of the first arrival signal, the spectrum is shaped by giving the start and end frequencies of the Yu wavelet optimization;

[0083] The shaped spectrum is subjected to inverse Fourier transform to obtain the seismic trace in the time domain, thus completing the optimization of the first arrival data.

[0084] The energy ratio method is used to automatically pick the processed first arrival data.

[0085] The estimation and application of high-frequency static correction for first arrival data are as follows:

[0086] The high-frequency static correction is calculated by using the first-arrival residual static correction estimation method. Then, this set of preliminary static correction is applied to the first-arrival data to obtain the first-arrival data after static correction is applied, which is used to eliminate inter-channel jitter.

[0087] The reconstruction of the application's initial data is as follows:

[0088] Based on the refraction wave propagation theory, the virtual source is constructed using cross-correlation, and then the convolution algorithm is used to reconstruct the first arrival data.

[0089] like Figure 2 As shown, the present invention provides an artificial intelligence first-arrival picking system based on data optimization, comprising:

[0090] Data acquisition module: obtain original seismic data, perform linear correction on the original seismic data to obtain first arrival data, optimize the first arrival data to obtain optimized first arrival data;

[0091] First arrival picking module: input the optimized first arrival data into a pre-trained artificial intelligence first arrival picking model to obtain the first arrival time;

[0092] Among them, the pre-trained artificial intelligence first-arrival picking model is an artificial intelligence first-arrival picking model obtained by training with multiple pre-processed first-arrival data.

[0093] Example:

[0094] like Figure 4 The present invention provides an artificial intelligence first arrival picking model training method, comprising the following steps:

[0095] S1: Linear correction of first arrival data, including selecting seed shots, interactively determining linear correction parameters and performing linear correction;

[0096] S2: Perform Yu wavelet optimization and shaping processing on the first arrival data after linear correction to obtain the processed first arrival data;

[0097] S3: Automatically pick up the processed first arrival data to obtain the first arrival time;

[0098] S4: Estimating and applying high-frequency static corrections to the first arrival time, and obtaining first arrival data after applying preliminary static corrections;

[0099] S5: reconstructing the first arrival data after applying the preliminary static correction amount to obtain reconstructed first arrival data;

[0100] S6: Extract the label cannons in the reconstructed first arrival data and perform manual picking;

[0101] S7: Based on the first arrival time of the picked label gun, convolutional neural network artificial intelligence first arrival picking is performed.

[0102] In one embodiment, the linear correction of the first arrival data includes:

[0103] Select seed shots to determine the linear correction parameters. The distribution of seed shots can vary with the surface conditions. In places with stable terrain, the seed shots are sparsely distributed. In places with drastic terrain changes, they need to be denser in order to better control the linear correction apparent velocity.

[0104] Load and display the seed guns and interactively determine the linear correction parameters;

[0105] The linear correction parameters are spatially interpolated to obtain the linear correction parameters of each shot, and linear correction that varies with space is performed to generate first arrival data.

[0106] In one embodiment, the processing of the initial data by the Yu wavelet optimization and shaping process includes:

[0107] Perform Fourier transform on the intercepted first arrival data to obtain the spectrum of each first arrival data;

[0108] According to the bandwidth of the first arrival signal, the spectrum is shaped by giving the start and end frequencies of the Yu wavelet optimization;

[0109] The shaped spectrum is subjected to inverse Fourier transform to obtain the seismic trace in the time domain, thus completing the optimization of the first arrival data.

[0110] In one embodiment, the high-frequency static correction amount is estimated and applied based on the first arrival time automatically picked up by the aforementioned energy ratio method, and the high-frequency static correction amount is calculated using the first arrival residual static correction amount estimation method. This set of static correction amounts is applied to the first arrival data to eliminate inter-channel jitter.

[0111] In one embodiment, the first arrival data reconstruction is based on the refraction wave propagation theory, using cross-correlation to construct a virtual earthquake source, and then using a convolution algorithm to reconstruct the seismic trace;

[0112] Among them, reconstruction is a multi-channel operation, which uses the adjacent channel's first arrival data to improve the signal-to-noise ratio of the channel. The specific steps are as follows:

[0113] The virtual earthquake source is obtained by cross-correlating the adjacent channel first arrival data.

[0114] The virtual source and the adjacent channel are used for convolution calculation to reconstruct the local channel;

[0115] Superimpose with the original channel to improve the signal-to-noise ratio.

[0116] In one embodiment, after the above preprocessing in step 6, the amount of first arrival data is reduced, the signal-to-noise ratio is obtained, and the time difference between channels is improved and eliminated. Similarly, seed shots are selected to determine the linear correction parameters, and label shots are selected at a certain distance and fine manual picking is performed.

[0117] In one embodiment, the convolutional neural network artificial intelligence first arrival picking in step 7 includes: performing artificial intelligence model training based on finely picked label guns until convergence and stability;

[0118] Using the trained model, AI first arrival picking is performed on all non-labeled guns.

[0119] In the above technical scheme, an artificial intelligence first-arrival picking system based on data optimization provided by the present invention has the following beneficial effects: Yu wavelet first-arrival optimization and virtual earthquake source first-arrival reconstruction are used to improve data quality, first through single-channel processing measures: utilizing the advantages of strong main lobe energy and weak side lobes of Yu wavelet, while suppressing background noise, making the first-arrival wavelet more focused; then through multi-channel processing technology, based on the refraction wave propagation theory, using adjacent channels to obtain virtual earthquake sources and reconstruct the current channel, superimposing, and enhancing the signal-to-noise ratio of the first-arrival data; applying a preliminary static correction can reduce the inter-channel offset, make the first-arrival wave group more gentle in the lateral direction, and increase the recognizability of the first-arrival data.

[0120] The above system is different from the industry's common artificial intelligence picking on raw data. The pre-processed first-arrival data has a high signal-to-noise ratio, strong lateral continuity and high wave field recognizability, which greatly improves the picking effect and efficiency.

[0121] Specifically, the main processing process of intelligent first arrival picking is described as follows: interactively determine the linear correction frame ( Figure 5 ); Three single-channel wavelets are plotted, including: left: bandpass wavelet; middle: Ricker wavelet; right: Yu wavelet ( Figure 6 ); Yu wavelet optimization is used to optimize the sidelobe suppression of vibrator, where: a: before optimization; b: after optimization ( Figure 7 ); Yu's wavelet optimization is used to optimize the first arrival signal-to-noise ratio, where: A: before optimization; B: after optimization ( Figure 8 );Determine the virtual source by cross-correlation of refracted waves ( Fig. 9 ); Convolution reconstruction of the first arrival wave field ( Fig.10 ); form the final initial reconstruction application effect diagram ( Fig.11 ), where the upper left figure is the original first arrival data, the upper right figure is the effect after first arrival optimization, and the lower figure is the effect after optimization + reconstruction.

[0122] Furthermore, the above system is mainly aimed at the increasing number of channels in onshore seismic exploration in complex exploration areas in my country, with each shot exceeding 20,000 channels. Due to the use of controllable seismic sources and node detectors, the signal-to-noise ratio is low, the wave field is messy, and the conventional first-arrival picking method is difficult to meet production needs. The present invention greatly improves the effect and efficiency of first-arrival picking, and has important economic benefits for accelerating the seismic data processing cycle and ensuring the final imaging quality and processing effect.

[0123] In the above scheme, in order to address the challenges of low signal-to-noise ratio of first-arrival data and strong external interference, the present invention adopts Yu wavelet first-arrival optimization and virtual source first-arrival reconstruction to improve data quality, and selects single-channel processing measures: taking advantage of the strong energy of the main lobe and weak side lobes of Yu wavelet, while suppressing background noise, the first-arrival wavelet is made more focused; then through multi-channel processing technology, based on the theory of refracted wave propagation, adjacent channels are used to obtain virtual sources and reconstruct the current channel, and superimposed to enhance the signal-to-noise ratio of the first-arrival data; applying preliminary static correction can reduce the inter-channel offset, make the first-arrival wave group more horizontally flat, and increase the recognizability of the first-arrival data. The present invention is different from the industry's common artificial intelligence picking on raw data. The first-arrival data after preprocessing has a high signal-to-noise ratio, strong lateral continuity, and high wave field recognizability, and the picking effect and efficiency are greatly improved.

[0124] As a further technical solution provided by the present invention: first arrival data linear correction includes:

[0125] Select seed shots to determine the linear correction parameters. The distribution of seed shots can vary with the surface conditions. In places with stable terrain, the seed shots are sparsely distributed. In places with drastic terrain changes, they need to be denser in order to better control the linear correction apparent velocity.

[0126] Load and display the seed guns and interactively determine the linear correction parameters ( Figure 5 );

[0127] The linear correction parameters are spatially interpolated to obtain the linear correction parameters of each shot, and linear correction that varies with space is performed to generate first arrival data.

[0128] Furthermore, the processing of the initial data by Yu's wavelet optimization and shaping includes:

[0129] The three collected single-channel wavelets: bandpass wavelet, Ricker wavelet and Yu wavelet are synthesized into Yu wavelet by Ricker wavelets with different frequency components;

[0130] Perform Fourier transform on the intercepted first arrival data to obtain the spectrum of each first arrival data;

[0131] According to the bandwidth of the first arrival signal, the spectrum is shaped by giving the start and end frequencies of the Yu wavelet optimization;

[0132] The shaped spectrum is subjected to inverse Fourier transform to obtain the seismic trace in the time domain, thus completing the optimization of the first arrival data.

[0133] Furthermore, the energy ratio method in the above system automatically picks the first arrival. The energy ratio method is a method of automatically picking according to amplitude mutation. This method is suitable for data with weak background noise energy and strong signal energy. Therefore, this method is only suitable for seismic traces with very high signal-to-noise ratio.

[0134] Furthermore, the high-frequency static correction is calculated and applied based on the first arrival time automatically picked up by the aforementioned energy ratio method, and the first arrival residual static correction estimation method is used to calculate the high-frequency static correction. This set of static correction is applied to the first arrival data to eliminate inter-channel jitter.

[0135] As another technical solution further provided by the present invention, first arrival data reconstruction (based on virtual source technology) is based on the refraction wave propagation theory, using cross-correlation to construct virtual sources, and then using a convolution algorithm to reconstruct seismic traces;

[0136] Among them, reconstruction is a multi-channel operation, which uses the adjacent channel's first arrival data to improve the signal-to-noise ratio of the channel. The specific steps are as follows:

[0137] The virtual earthquake source is obtained by cross-correlating the adjacent channel first arrival data.

[0138] The virtual source and the adjacent channel are used for convolution calculation to reconstruct the local channel;

[0139] Superimpose with the original channel to improve the signal-to-noise ratio.

[0140] Furthermore, extracting label shots and performing fine manual picking in step 6 means that after the above preprocessing, the amount of first arrival data is reduced, the signal-to-noise ratio is obtained, and the time difference between channels is improved and eliminated, similar to selecting seed shots to determine linear correction parameters, selecting label shots at a certain distance and performing fine manual picking.

[0141] Furthermore, in step 7, artificial intelligence first arrival picking using a convolutional neural network is performed, including:

[0142] Based on the finely picked label cannon, the artificial intelligence model is trained until convergence and stability;

[0143] Using the trained model, AI first arrival picking is performed on all non-labeled guns.

[0144] The above system first performs a series of preprocessing on the seismic data to enhance the wave field recognizability and signal-to-noise ratio, including linear correction, first arrival optimization, first arrival reconstruction and application of preliminary static correction, and then performs artificial intelligence picking based on convolutional neural networks.

[0145] Based on the detailed description of the application of the system in the above embodiment, the present invention provides an actual detection application for the above system:

[0146] The original seismic data collected from a mountain area, Fig.12 ;

[0147] Based on the linearly corrected first arrival data in step one, Fig.13 ;

[0148] Based on the first arrival data after the first arrival optimization in step 2, Fig.14 ;

[0149] Based on the first arrival data after optimization + applying preliminary static correction in steps 3 to 5, Fig.15 ;

[0150] As can be seen from the logo in the figure, after preprocessing Fig.15 The signal-to-noise ratio, lateral continuity and wave field identifiability of the conventional technology are better than those of the conventional technology. Fig.12 In terms of, higher.

[0151] Further, based on one embodiment, the collected data is compared with the conventional picking technology and the picking system provided by the present invention. The conventional artificial intelligence technology first arrival picking result, Fig.16 ; The invention is based on the artificial intelligence first arrival picking effect of data optimization, Fig.17 .

[0152] Based on the detailed description of the application of the system in the above embodiment, the present invention provides another practical detection application for the above system:

[0153] The original seismic data collected from a mountain area, Fig.18 ;

[0154] Based on the linearly corrected first arrival data in step one, Fig.19 ;

[0155] Based on the first arrival data after the first arrival optimization in step 2, Fig. 20 ;

[0156] Based on the first arrival data after optimization + applying preliminary static correction in steps 3 to 5, Fig.21 ;

[0157] As can be seen from the logo in the figure, after preprocessing Fig.21 The signal-to-noise ratio, lateral continuity and wave field identifiability of the conventional technology are better than those of the conventional technology. Fig.18 In terms of, higher.

[0158] Further, based on one embodiment, the collected data is compared with the conventional picking technology and the picking system provided by the present invention. The conventional artificial intelligence technology first arrival picking result, Fig. 22 ; The invention is based on the artificial intelligence first arrival picking effect of data optimization, Fig.23 .

[0159] Although the embodiments of the present invention are described above in conjunction with the accompanying drawings, the present invention is not limited to the above-mentioned specific embodiments and application fields, and the above-mentioned specific embodiments are only illustrative and instructive, rather than restrictive. Under the guidance of the specification, a person skilled in the art can also make many forms without departing from the scope of protection of the claims of the present invention, all of which belong to the scope of protection of the present invention.

Claims

1. An artificial intelligence first arrival picking method based on data optimization, characterized in that: include: Acquire original seismic data, correct the original seismic data to obtain first arrival data, optimize the first arrival data to obtain optimized first arrival data; Inputting the optimized first arrival data into a pre-trained artificial intelligence first arrival picking model to obtain the first arrival time; Among them, the pre-trained artificial intelligence first-arrival picking model is an artificial intelligence first-arrival picking model obtained by training with multiple pre-processed first-arrival data.

2. The artificial intelligence first arrival picking method according to claim 1, characterized in that: The artificial intelligence first arrival picking model training method includes: Obtaining original seismic data, performing linear correction on the original seismic data, and obtaining first arrival data; Optimize and reshape the first arrival data to obtain processed first arrival data; Automatically pick up the processed first arrival data to obtain the first arrival time; The high-frequency static correction is estimated and applied to the first-arrival data after the initial processing to obtain the first-arrival data after the preliminary static correction is applied; Reconstructing the first arrival data after applying the preliminary static correction to obtain reconstructed first arrival data; Extract and reconstruct the label cannons in the first arrival data and perform manual picking; Based on the first arrival time of the picked label gun, the artificial intelligence model is trained until convergence and stability are achieved to obtain the artificial intelligence first arrival picking model.

3. The artificial intelligence first arrival picking method according to claim 2, characterized in that: The first arrival data linear correction comprises: Select the label gun, interactively determine the linear correction parameters and perform linear correction.

4. The artificial intelligence first arrival picking method according to claim 2, characterized in that: The first arrival data after linear correction is optimized and shaped, specifically: Perform Fourier transform on the corrected first arrival data to obtain the spectrum of each first arrival data; According to the spectrum of the first arrival signal, the spectrum is shaped by giving the start and end frequencies of the Yu wavelet optimization; The shaped spectrum is subjected to inverse Fourier transform to obtain the seismic trace in the time domain, thus completing the optimization of the first arrival data.

5. The artificial intelligence first arrival picking method according to claim 2, characterized in that: The energy ratio method is used to automatically pick the processed first arrival data.

6. The artificial intelligence first arrival picking method according to claim 2, characterized in that: The estimation and application of high-frequency static correction for the first arrival data after picking are as follows: The high-frequency static correction is calculated by using the first-arrival residual static correction estimation method. Then, this set of preliminary static correction is applied to the first-arrival data to obtain the first-arrival data after static correction is applied, which is used to eliminate inter-channel jitter.

7. The artificial intelligence first arrival picking method according to claim 2, characterized in that: The reconstruction of the application initial data is as follows: Based on the refraction wave propagation theory, the virtual source is constructed using cross-correlation, and then the convolution algorithm is used to reconstruct the first arrival data.

8. An artificial intelligence first arrival picking system based on data optimization, characterized in that: include: Data acquisition module: acquire original seismic data, correct the original seismic data to obtain first arrival data, optimize the first arrival data to obtain optimized first arrival data; First arrival picking module: input the optimized first arrival data into a pre-trained artificial intelligence first arrival picking model to obtain the first arrival time; Among them, the pre-trained artificial intelligence first-arrival picking model is an artificial intelligence first-arrival picking model obtained by training with multiple pre-processed first-arrival data.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of an artificial intelligence first-arrival picking method based on data optimization are implemented as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of an artificial intelligence first-arrival picking method based on data optimization are implemented as described in any one of claims 1-7.

Images