Three-dimensional fault identification method based on geological stress constraint and adaptive transfer learning

Abstract

The application discloses a three-dimensional fault identification method based on geological stress constraint and adaptive migration learning, comprising the following steps: determining the position, direction and tectonic movement background of the geological structure stress; generating three-dimensional seismic data and corresponding three-dimensional fault labels based on three-dimensional modeling of the geological stress constraint; decomposing the amplitude and phase of the three-dimensional seismic data, mixing and taking as the source domain data set of deep learning; constructing a three-dimensional seismic geological database of the study area, taking the constructed three-dimensional actual seismic data and three-dimensional actual fault labels as the target domain data set of deep learning; and constructing a dense connection 3D-Unet migration learning network based on domain adaptive, connecting the source domain and the target domain data, and realizing large-scale migration learning by mixing the feature distribution rules of the source domain and the target domain data. The application has the advantages of high calculation efficiency, high prediction accuracy, strong fault continuity and strong applicability.

Description

Technical Field

[0001] This invention belongs to the field of geophysical exploration, specifically relating to a three-dimensional fault identification method based on geological stress constraints and adaptive transfer learning. Background Technology

[0002] Fault identification has always been a key issue in oilfield exploration and development. While controlling hydrocarbon migration, faults also play a crucial role in lateral sealing during the formation of lithologic oil and gas reservoirs. The Bohai Bay Basin, a typical representative of terrestrial sedimentary deposits in my country, has an extremely complex fault system, further complicated by the extensive development of strike-slip faults. As exploration in the Bohai Oilfield deepens, efficiently and accurately interpreting complex three-dimensional faults is the key to unlocking its oil and gas treasures. During the exploration phase, small faults, due to their small vertical displacement, short horizontal extension, and fracturing of surrounding strata, make it difficult to determine their connection to the main fault on seismic profiles. This is especially true in areas with complex fault development, where accurate identification of micro-faults is even more challenging. Furthermore, the identification of micro-faults with strike-slip characteristics becomes even more difficult. Accurate identification of strike-slip faults, even hidden ones, is a significant factor limiting the success of exploration in the Bohai Oilfield.

[0003] Due to factors such as seismic data resolution and noise, traditional methods like conventional coherence properties are relatively vague in characterizing small-scale faults, making it difficult to meet exploration needs. Industry research on AI-based 3D fault identification is primarily based on mathematical theoretical analysis, using mathematical formulas to simulate stratigraphic undulations and constructing AI sample databases. This entire AI detection process lacks geological background and understanding, and ignores strike-slip fault characteristics. Essentially, it still simulates mathematical discontinuities, just like traditional coherence algorithms, failing to fundamentally improve fault identification accuracy and making it unsuitable for the detailed interpretation of complex faults generated by actual geological movements in oilfield exploration and development. Furthermore, the sample databases built in conventional AI fault identification processes are often overly idealized; amplitude and noise characteristics are difficult to effectively match with actual seismic data, and the feature distribution patterns of the sample databases differ significantly from those of actual seismic data. Therefore, deep network models trained on sample databases are unsuitable for actual 3D seismic data. In addition, manual fault interpretation is complex and tedious, time-consuming and labor-intensive, and the reliability of the interpretation results is influenced by the professional level and experience of the interpreters, introducing a degree of subjectivity.

[0004] Therefore, it is necessary to construct a Bohai-specific seismic geological database based on the typical seismic geological characteristics of the Bohai Oilfield, and to develop a three-dimensional fault intelligent identification method with Bohai characteristics, so as to make the interpretation of three-dimensional seismic faults in the Bohai Oilfield more efficient and accurate. Summary of the Invention

[0005] This invention is proposed to solve the problems existing in the prior art, and its purpose is to provide a three-dimensional fault identification method based on geological stress constraints and adaptive transfer learning.

[0006] This invention is achieved through the following technical solution: A three-dimensional fault identification method based on geological stress constraints and adaptive transfer learning includes the following steps: S1. Based on the geological movement patterns and tectonic phases of the study area, analyze the geological stress and tectonic movement of the study area to determine the location, direction, and background of geological tectonic stress and tectonic movement in the study area. Specifically, the following steps are included: S11. Conduct a comprehensive review of the geological activity background of the study area, analyze the geological stress and tectonic movement periods of the study area, and clarify the geological tectonic movement patterns of the study area. S12. Based on the geological tectonic movement law of the work area to be studied as defined in step S11, determine the stress location and stress direction of different geological movements, and obtain the tensile stress, compressive stress or shear stress of different phases, locations and directions. S2. By extracting a large number of two-dimensional seismic profiles with the characteristics of the study area and typical fault features during the exploration and development of the oilfield in the study area, and combining the location and direction of geological structural stress, a three-dimensional model based on the geological stress constraints of the study area is realized, generating a large number of three-dimensional seismic data and corresponding three-dimensional fault labels. Specifically, the following steps are included: S21. Organize the massive amount of two-dimensional seismic profiles with the characteristics of the study area and typical fault features extracted during the exploration and development of the oilfield in the study area, and expand the corresponding seismic data and fault interpretation data into three dimensions through dimensional extension to obtain a three-dimensional model. The three-dimensional expansion of two-dimensional data through dimensional extension is a mature existing technology in this field. In simple terms, it is achieved by "copying and pasting" a two-dimensional cross-section along the third dimension, thereby expanding the dimension and laying the foundation for subsequent three-dimensional spatial deformation. S22. By applying the stress location and stress direction of different geological movements determined in step S12 to the three-dimensional model obtained by dimensional extension in step S21, the three-dimensional model is subjected to three-dimensional spatial folding deformation under geological stress constraints, thereby realizing three-dimensional modeling based on the stress constraints of the geological structure of the work area under study. This process generates massive amounts of three-dimensional seismic data and response three-dimensional fault labels. S3. By performing amplitude and phase decomposition on the 3D seismic data used in the actual exploration and development of the oilfield in the research area, and by aliasing the amplitude and phase information respectively, domain transfer based on amplitude and phase aliasing data is achieved. The amplitude and noise features in the actual seismic data are introduced into the 3D seismic data generated by the 3D modeling in step S2. This (the 3D seismic data with noise from the actual seismic data introduced after amplitude and phase aliasing between the 3D modeling data generated in step S2 and the actual seismic data) and the corresponding 3D fault labels are used as the source domain dataset for deep learning. Specifically, the following steps are included: S31. Perform a Fourier transform on the three-dimensional seismic data constructed in step S2 using the Fourier transform formula to obtain the corresponding amplitude information. and phase information ; Amplitude information The calculation formula is: In the above formula: This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Indicates amplitude information; Phase information The calculation formula is: In the formula: This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Represents the arctangent function; For phase information; Amplitude information and phase information In the calculation formula: In the above formula: This represents the Fourier transform operation; The three-dimensional seismic data constructed in step S2; S32. Using the Fourier transform formula, perform a Fourier transform on the actual 3D seismic data used in the oilfield exploration and development process of the study area to obtain the corresponding amplitude information. and phase information ; Amplitude information The calculation formula is: In the above formula: This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Indicates amplitude information; Phase information The calculation formula is: In the above formula: Represents the arctangent function; This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Represents phase information; Amplitude information and phase information In the calculation formula: In the above formula: This represents the Fourier transform operation; The three-dimensional seismic data constructed in step S2; S33. By aliasing the amplitude information A1 and A2, and the phase information F1 and F2 obtained in steps S31 and S32, data domain migration based on amplitude and phase aliasing is achieved. While ensuring the construction information of the three-dimensional data obtained from the three-dimensional modeling based on the geological stress constraints of the study area in step S22, amplitude and noise information from the actual three-dimensional seismic data of the oilfield in the study area are introduced. The obtained three-dimensional seismic data and the corresponding three-dimensional fault label data are used as the source domain dataset for deep learning.

[0007] The amplitude and phase aliasing data domain migration technology in this invention innovatively introduces Fourier transform into seismic data processing. By leveraging the principle that amplitude spectrum can reflect noise characteristics and phase spectrum can reflect tectonic characteristics, it adds actual seismic data noise to the massive modeling data to reduce the feature differences between actual data and modeling data.

[0008] S4. By transforming the massive amount of three-dimensional fault interpretation schemes accumulated in the exploration and development process of the research area into three dimensions, the conversion of points, lines and surfaces in space is realized. Combined with the massive amount of actual oilfield seismic data of the research area, a three-dimensional seismic geological database of the Bohai Sea is constructed. The constructed three-dimensional actual seismic data and three-dimensional actual fault labels are used as the target domain dataset for deep learning. Specifically, the following steps are included: S41. Compile the actual three-dimensional seismic and fault interpretation data corresponding to different oilfields or structures during the exploration and development of the work area to be studied. S42. Convert a large number of discrete fault interpretation points in the fault interpretation results data into standard three-dimensional continuous cross-section label data by reconstructing three-dimensional cross-sections. The conversion formula is: In the above formula: This represents the operation of finding a line between two points. This represents bilinear interpolation. Indicates different fault numbers. Indicates the numbering of different fault points on the same fault line; Indicates the number is Numbering on the fault The fault interpretation point of the breakpoint; S43. Standardize the actual three-dimensional seismic data of different oil fields or structures to eliminate the influence of amplitude differences under different seismic tectonic backgrounds. The formula for standardization is: In the above formula: This represents standardized 3D actual seismic data; Represents actual three-dimensional seismic data; This represents the operation of finding the minimum value; This represents the operation to find the maximum value; S44. The standardized 3D actual seismic data and 3D continuous cross-section label data are mapped one by one, and then the data is augmented by mirroring, scaling and rotation, and then used as the target domain dataset for deep learning. The mirror formula is: The scaling formula is: The rotation formula is: In the above formula: This represents the scaling factor in the x-direction. This represents the scaling factor in the y-direction; Indicates the scaling factor in the z-direction; express Number of directional sampling points; express Number of directional sampling points; Indicates the number of vertical sampling points; This represents the angle of clockwise rotation around the origin in the xOy plane; This represents the spatial x-coordinate before the data augmentation transformation; This represents the spatial y-coordinate before the data augmentation transformation; This represents the spatial z-coordinate before the data augmentation transformation; S5. Construct a densely connected 3D-Unet transfer learning network based on domain adversarial adaptation to link source domain and target domain data. Achieve large-scale transfer learning by confusing the feature distribution patterns of source and target domain data to improve the accuracy of automatic three-dimensional fault identification of new actual seismic data. Specifically, the following steps are included: S51. Based on the traditional U-net network structure, the source domain samples are used as grayscale image inputs. The input grayscale image is converted into R, G, and B components using the RGB conversion method. The R, G, and B components are used as multi-channel input data. The input of different channels is processed by two convolutional neural networks and one pooling layer to complete one downsampling process, and dense connections are formed in different downsampling processes. Then, the downsampling results obtained from multiple channels are fused together and an upsampling convolution is performed. The convolution result is then fused with the downsampling result of the previous layer and two normalized convolutions are performed to complete one upsampling process, thus constructing a 3D-Unet network structure based on dense connections. S52. Based on the densely connected 3D-Unet network structure constructed in step S51, target domain sample input is added to the input end, and a binary classification domain discrimination network is constructed through source domain samples and target domain samples. The adversarial domain adaptive network is added to the densely connected 3D-Unet network through the gradient reversal layer to construct a densely connected 3D-Unet network structure based on adversarial domain adaptive transfer learning. Adversarial domain adaptive networks are used to enhance the confusion between source domain samples and target domain samples, so as to extract feature information without distinguishing whether the sample comes from the source domain or the target domain. The densely connected 3D-Unet network is used to extract feature information for learning and to achieve the function of predicting three-dimensional faults.

[0009] S53. The source domain and target domain sample datasets obtained in steps S3 and S4 are respectively used as grayscale inputs. Based on the RGB conversion of the grayscale, the input three-dimensional seismic data is converted into R, G, and B components as input data. At the same time, by using the data feature distribution patterns of different domain sources, a large-scale transfer learning is performed on the three-dimensional fault feature information in the source domain samples using a large number of actual three-dimensional fault interpretation results, so as to improve the accuracy of automatic three-dimensional fault identification of actual seismic data.

[0010] S54. Through the training and learning process described above, a deep learning network model can be obtained. Based on this deep network model, the three-dimensional seismic data to be predicted only needs to be input into the deep network model to automatically obtain a high-precision intelligent identification result of the three-dimensional fault.

[0011] The beneficial effects of this invention are: This invention provides a three-dimensional fault identification method based on geological stress constraints and adaptive transfer learning. It innovatively proposes a three-dimensional modeling method based on geological stress constraints, incorporating the geological structural characteristics of the Bohai Sea into the construction process of an artificial intelligence big data database. This avoids the drawbacks of conventional mathematical modeling methods that lack geological background and understanding. It fully combines the complex fault characteristics unique to the Bohai Sea with deep learning processes, significantly enhancing fault continuity while making the planar combination characteristics of faults more consistent with geological laws. Through amplitude and phase decomposition and aliasing, it innovatively achieves domain transfer based on amplitude and phase aliasing data, introducing amplitude and noise characteristics from actual Bohai Sea seismic data into the three-dimensional modeling data. This effectively solves the problem of significant differences in feature distribution patterns between training sample data and data to be predicted, significantly improving the accuracy of three-dimensional fault identification in actual work areas. Furthermore, it innovatively constructs the first three-dimensional seismic geological big data database for the Bohai Oilfield and initiates the first research on strike-slip fault identification based on artificial intelligence in the Bohai Sea, significantly improving the identification accuracy of strike-slip faults. This invention has advantages such as high computational efficiency, high prediction accuracy, strong fault continuity, and strong applicability. Attached Figure Description

[0012] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of a two-dimensional seismic profile with typical fault features in the Bohai Sea, as described in an embodiment of the present invention. Figure 3 This is a schematic diagram of the data volume and its faults obtained by dimensional topology based on two-dimensional seismic profiles in an embodiment of the present invention. Figure 4 This is a schematic diagram of the three-dimensional data volume and its fault label volume obtained by three-dimensional modeling based on geological stress constraints in an embodiment of the present invention; Figure 5 The three-dimensional modeling data and its corresponding amplitude and phase spectra in this embodiment of the invention are shown (a is the modeling data; b is the amplitude spectrum; c is the phase spectrum). Figure 6 The following are three-dimensional data of the Bohai Sea and their corresponding amplitude and phase spectra in the embodiments of the present invention (a is the actual data; b is the amplitude spectrum; c is the phase spectrum). Figure 7 This embodiment of the invention presents three-dimensional data based on amplitude and phase spectrum aliasing and its corresponding amplitude and phase spectra (a is the aliased data; b is the amplitude spectrum; c is the phase spectrum). Figure 8 A schematic diagram of reconstructed three-dimensional cross-sections from discrete actual fault interpretation points in an example of the present invention (a is a discrete actual fault interpretation point; b is a two-dimensional fault; c is a three-dimensional cross-section). Figure 9 This is a schematic diagram of the actual three-dimensional seismic data volume and its fault label volume in an embodiment of the present invention; Figure 10 This is a schematic diagram of a slice of the tomographic detection result obtained by a conventional coherence algorithm in an embodiment of the present invention; Figure 11 This is a schematic diagram of the slice of intelligent tomographic recognition results in an embodiment of the present invention.

[0013] For those skilled in the art, other related figures can be obtained from the above figures without any creative effort. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0015] Example 1 A three-dimensional fault identification method based on geological stress constraints and adaptive transfer learning, the technical process of which is as follows: Figure 1 As shown, it includes the following steps: S1. Based on the geological movement patterns and tectonic phases of the Bohai Sea, analyze the geological stress and tectonic movements of the Bohai Bay Basin to determine the location, direction, and background of geological tectonic stress in the Bohai Bay Basin. Specifically, the following steps are included: S11. Conduct a comprehensive review of the geological activity background of the research area in the Bohai Bay Basin, analyze the geological stress and tectonic movement periods of the research area in the Bohai Bay Basin, and clarify the geological tectonic movement patterns of the research area in the Bohai Bay Basin. S12. Based on the macroscopic geological movement patterns of the research area in the Bohai Bay Basin, determine the stress location and stress direction of different geological movements, and obtain the tensile stress, compressive stress or shear stress of different stages, locations and directions; the macroscopic geological movement of this target area occurs in the early strata, that is, the deepest part generates compressive stress approximately parallel to the northeast direction.

[0016] S2. By extracting a large amount of two-dimensional seismic profiles with typical fault features characteristic of the Bohai Sea during the exploration and development of the Bohai oilfield, and combining the location and direction of geological structural stress, a three-dimensional model based on the geological stress constraint of the Bohai Sea is realized, generating a large amount of three-dimensional seismic data and corresponding three-dimensional fault labels. Specifically, the following steps are included: S21. Compile and organize the massive amount of two-dimensional seismic profiles with typical fault characteristics that are characteristic of the Bohai Sea and interpreted during the exploration and development of the Bohai Oilfield, such as... Figure 2 As shown, its corresponding seismic data and fault interpretation data By extending two-dimensional data to three dimensions, a three-dimensional data volume is obtained. and label body ; in, and These represent the data volume after dimensional extension and the tomographic label volume, respectively. Representing the three dimensions of space respectively spatial range, and These represent the seismic data and fault interpretation data corresponding to the two-dimensional profile, respectively.

[0017] S22. Apply the approximately northeast-oriented compressive stress generated at the deepest part of the model, as determined in step S12, to the three-dimensional model obtained by dimensional extension in step S21. and In, such as Figure 3 As shown, the three-dimensional model can be subjected to geological stress constraints for three-dimensional spatial fold deformation S1, thus realizing three-dimensional modeling based on the stress constraints of the Bohai Sea geological structure. This process can generate massive amounts of three-dimensional seismic data. and response 3D tomographic labels ,like Figure 4 As shown.

[0018] in, These represent the x, y, and z coordinates of the 3D data volume and the label volume constructed in step S21, respectively. These represent the x, y, and z coordinates of the three-dimensional velocity body after the addition of wrinkle deformation; Indicates the maximum fold. Indicates the location where the wrinkles occur. The type of fold that indicates the direction of stress, whether it is an anticline or a syncline; Indicates about The two-dimensional Gaussian function, where Indicates the maximum amplitude of random folds. Indicates the radius of random folds. and The x and z coordinates represent the center of the fold.

[0019] S3. By performing amplitude and phase decomposition on the three-dimensional seismic data used in the actual exploration and development of the Bohai Oilfield, and by aliasing the amplitude and phase information respectively, domain transfer based on amplitude and phase aliasing data is achieved. The amplitude and noise features in the actual seismic data are introduced into the three-dimensional seismic data generated by the three-dimensional modeling in step S2, and the data and the corresponding three-dimensional fault labels are used as the source domain dataset for deep learning. Specifically, the following steps are included: S31. Use the Fourier transform formula to process the three-dimensional seismic data constructed in step S2. Perform a Fourier transform to obtain the corresponding amplitude information. and phase information ,like Figure 5 As shown; Amplitude information The calculation formula is: In the above formula: This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Indicates amplitude information; Phase information The calculation formula is: In the formula: This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Represents the arctangent function; For phase information; Amplitude information and phase information In the calculation formula: In the above formula: This represents the Fourier transform operation; The three-dimensional seismic data constructed in step S2; S32. Using Fourier transform formulas to analyze actual 3D seismic data used in the exploration and development process of the Bohai Oilfield. Perform a Fourier transform to obtain the corresponding amplitude information. and phase information ,like Figure 6 As shown; Amplitude information The calculation formula is: In the above formula: This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Indicates amplitude information; Phase information The calculation formula is: In the above formula: Represents the arctangent function; This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Represents phase information; Amplitude information and phase information In the calculation formula: In the above formula: This represents the Fourier transform operation; Actual 3D seismic data used in the exploration and development process of the Bohai Oilfield; S33. By aliasing the amplitude information A1 and A2, and the phase information F1 and F2 obtained in steps S31 and S32, data domain migration based on amplitude and phase aliasing is achieved, such as... Figure 7 As shown, while ensuring the structural information of the 3D data obtained from the 3D modeling based on the geological stress constraints of the Bohai Sea in step S22, amplitude and noise information from the actual 3D seismic data of the oilfield are introduced, and the obtained 3D seismic data and the corresponding 3D fault label data are used as the source domain dataset for deep learning; In the formula: Represents the imaginary unit. This represents the inverse Fourier transform.

[0020] S4. By transforming the massive three-dimensional fault interpretation schemes accumulated in the nearly 40-year exploration and development process of Bohai Oilfield into three dimensions, the conversion of points, lines and surfaces in space is realized. Combined with massive actual oilfield seismic data, a three-dimensional seismic geological database of Bohai is constructed. The constructed three-dimensional actual seismic data and three-dimensional actual fault labels are used as the target domain dataset for deep learning. Specifically, the following steps are included: S41. Compile the data of three-dimensional actual seismic and fault interpretation results corresponding to different oilfields / structures in the Bohai Oilfield over the past 40 years of exploration and development. S42. The large number of discrete fault interpretation points in the fault interpretation results data... The data is converted into standard three-dimensional continuous cross-section label data through a standardized method of reconstructing three-dimensional cross-sections. ,like Figure 8 As shown; The conversion formula is: In the above formula: This represents the operation of finding a line between two points. This represents bilinear interpolation. Indicates different fault numbers. Indicates the numbering of different fault points on the same fault line; Indicates the number is Numbering on the fault The fault interpretation point of the breakpoint; S43. Standardize the actual three-dimensional seismic data of different oil fields / structures to eliminate the influence of amplitude differences under different seismic tectonic backgrounds; The formula for standardization is: In the above formula: This represents standardized 3D actual seismic data; Represents actual three-dimensional seismic data; This represents the operation of finding the minimum value; This represents the operation to find the maximum value; S44. Standardize the actual 3D seismic data. and 3D continuous cross-sectional label data One-to-one correspondence, such as Figure 9 As shown, after data augmentation through mirroring, scaling, and rotation, it serves as the target domain dataset for deep learning. The mirror formula is: The scaling formula is: The rotation formula is: In the above formula: This represents the scaling factor in the x-direction. This represents the scaling factor in the y-direction; Indicates the scaling factor in the z-direction; express Number of directional sampling points; express Number of directional sampling points; Indicates the number of vertical sampling points; This represents the angle of clockwise rotation around the origin in the xOy plane; This represents the spatial x-coordinate before the data augmentation transformation; This represents the spatial y-coordinate before the data augmentation transformation; This represents the spatial z-coordinate before the data augmentation transformation; S5. Construct a densely connected 3D-Unet transfer learning network based on domain adversarial adaptation to link source domain and target domain data. Achieve large-scale transfer learning by confusing the feature distribution patterns of source and target domain data to improve the accuracy of automatic three-dimensional fault identification of new actual seismic data. Specifically, the following steps are included: S51. Based on the traditional U-net network structure, the source domain samples are used as grayscale image inputs. The input grayscale image is converted into R, G, and B components using the RGB conversion method. The R, G, and B components are used as multi-channel input data. The input of different channels is processed by two convolutional neural networks and one pooling layer to complete one downsampling process, and dense connections are formed in different downsampling processes. Then, the downsampling results obtained from multiple channels are fused together and an upsampling convolution is performed. The convolution result is then fused with the downsampling result of the previous layer and two normalized convolutions are performed to complete one upsampling process, thus constructing a 3D-Unet network structure based on dense connections. S52. Based on the densely connected 3D-Unet network structure constructed in step S51, target domain sample input is added to the input end, and a binary classification domain discrimination network is constructed through source domain samples and target domain samples. The adversarial domain adaptive network is added to the densely connected 3D-Unet network through the gradient reversal layer to construct a densely connected 3D-Unet network structure based on adversarial domain adaptive transfer learning. Adversarial domain adaptive networks are used to enhance the confusion between source domain samples and target domain samples, so as to extract feature information without distinguishing whether the sample comes from the source domain or the target domain. The densely connected 3D-Unet network is used to extract feature information for learning and to achieve the function of predicting three-dimensional faults.

[0021] S53. The source domain and target domain sample datasets obtained in steps S3 and S4 are respectively used as grayscale inputs. Based on the RGB conversion of the grayscale, the input three-dimensional seismic data is converted into R, G, and B components as input data. At the same time, by using the data feature distribution patterns of different domain sources, a large-scale transfer learning is performed on the three-dimensional fault feature information in the source domain samples using a large number of actual three-dimensional fault interpretation results, so as to improve the accuracy of automatic three-dimensional fault identification of actual seismic data.

[0022] This invention innovatively incorporates the geological structural features of the Bohai Sea into the construction process of an artificial intelligence big data platform. It fully integrates the complex fault characteristics unique to the Bohai Sea with deep learning, providing a high-precision intelligent fault identification method that conforms to the complex fault characteristics of the Bohai Sea. Compared with fault results obtained by traditional methods, such as… Figure 10 As shown, the high-precision fault results obtained by this invention are significantly improved, especially in the identification of small-displacement faults and strike-slip faults, with remarkable effectiveness. Figure 11 As shown.

[0023] This invention first utilizes 3D modeling technology based on geological stress constraints in the Bohai Sea. For the first time, it considers the characteristics of strike-slip faults during the 3D modeling process, avoiding the low identification accuracy problem caused by the lack of geological background in conventional AI big data due to purely mathematical analysis. Building upon this, it achieves data domain transfer by aliasing the amplitude and phase information of the modeling data and actual seismic data from the Bohai Sea, further enhancing the generalization ability of deep learning networks in the fault identification process of the Bohai oilfield. This ensures the identification accuracy of large-displacement faults while improving the identification accuracy of small-displacement and strike-slip faults. Furthermore, by constructing the first seismic geological big data database for the Bohai oilfield, and fully utilizing the massive local fault interpretation results of the Bohai oilfield, a domain-adversarial adaptive densely connected 3D-Unet transfer learning network is established to link the modeling data and actual seismic data from the Bohai Sea, achieving adaptive large-scale transfer learning. This further improves the accuracy of 3D complex fault interpretation in the actual seismic work area of ​​the Bohai oilfield.

[0024] This invention is not only applicable to the Bohai Bay Basin but can also be extended to other regions. As a typical representative of terrestrial sedimentary deposits in my country, the Bohai Bay Basin has an extremely complex fault system. By incorporating the complex fault characteristics of the Bohai Bay Basin into a large database, this invention can achieve good application results in the Bohai Bay Basin and also realize high-precision fault identification in other regions. The process of extending from the Bohai Bay Basin to other regions is like going from "complex" to "simplified".

[0025] The applicant declares that the above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention fall within the protection and disclosure scope of the present invention.

Claims

1. A three-dimensional fault identification method based on geological stress constraints and adaptive transfer learning, characterized in that: Includes the following steps: S1. Based on the geological movement patterns and tectonic phases of the study area, analyze the geological stress and tectonic movement of the study area to determine the location, direction, and background of geological tectonic stress and tectonic movement in the study area. S2. By combining two-dimensional seismic profiles with the location and direction of geological structural stress, three-dimensional modeling based on geological stress constraints in the study area is achieved, generating three-dimensional seismic data and corresponding three-dimensional fault labels. S3. Perform amplitude and phase decomposition on the 3D seismic data. By aliasing the amplitude and phase information respectively, the amplitude and noise features in the actual seismic data are introduced into the 3D seismic data generated by the 3D modeling in step S2. The 3D seismic data and the corresponding 3D fault labels are used as the source domain dataset for deep learning. S4. The three-dimensional fault interpretation schemes accumulated in the exploration and development process of the oilfield in the study area are made into three dimensions, and the Bohai three-dimensional seismic geological database is constructed in combination with the actual seismic data of the oilfield in the study area. The constructed three-dimensional actual seismic data and three-dimensional actual fault labels are used as the target domain dataset for deep learning. S5. Construct a densely connected 3D-Unet transfer learning network based on domain adversarial adaptation to link source domain and target domain data. Achieve large-scale transfer learning by confusing the feature distribution patterns of source and target domain data, and complete the automatic identification of three-dimensional faults in actual seismic data.

2. The three-dimensional fault identification method based on geological stress constraints and adaptive transfer learning according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11. Conduct a comprehensive review of the geological activity background of the study area, analyze the geological stress and tectonic movement periods of the study area, and clarify the geological tectonic movement patterns of the study area. S12. Based on the geological tectonic movement patterns of the work area to be studied as defined in step S11, determine the stress location and stress direction of different geological movements, and obtain the tensile stress, compressive stress, or shear stress of different phases, locations, and directions.

3. The three-dimensional fault identification method based on geological stress constraints and adaptive transfer learning according to claim 1, characterized in that: Step S2 specifically includes the following steps: S21. The seismic data and fault interpretation data corresponding to the two-dimensional seismic profile are extended to three dimensions through dimensional extension to obtain a three-dimensional model. The two-dimensional seismic profile is a two-dimensional seismic profile extracted during the exploration and development of the oilfield in the study area, which has the characteristics of the study area and typical fault features. S22. By applying the stress location and stress direction of different geological movements determined in step S12 to the three-dimensional model obtained by dimensional extension in step S21, the three-dimensional model is subjected to three-dimensional spatial folding deformation under geological stress constraints, thereby realizing three-dimensional modeling based on the stress constraints of the geological structure of the work area to be studied, and generating three-dimensional seismic data and corresponding three-dimensional fault labels.

4. The three-dimensional fault identification method based on geological stress constraints and adaptive transfer learning according to claim 1, characterized in that: Step S3 specifically includes the following steps: S31. Perform a Fourier transform on the three-dimensional seismic data constructed in step S2 using the Fourier transform formula to obtain the corresponding amplitude information. and phase information ; Amplitude information The calculation formula is: In the above formula: This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Indicates amplitude information; Phase information The calculation formula is: In the formula: This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Represents the arctangent function; For phase information; Amplitude information and phase information In the calculation formula: In the above formula: This represents the Fourier transform operation; The three-dimensional seismic data constructed in step S2; S32. Using the Fourier transform formula, perform a Fourier transform on the actual 3D seismic data used in the oilfield exploration and development process of the study area to obtain the corresponding amplitude information. and phase information ; Amplitude information The calculation formula is: In the above formula: This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Indicates amplitude information; Phase information The calculation formula is: In the above formula: Represents the arctangent function; This indicates the operation of taking the real part of a complex number; This indicates the operation of taking the imaginary part of a complex number; Represents phase information; Amplitude information and phase information In the calculation formula: In the above formula: This represents the Fourier transform operation; The three-dimensional seismic data constructed in step S2; S33. By aliasing the amplitude information A1 and A2, and the phase information F1 and F2 obtained in steps S31 and S32, data domain transfer based on amplitude and phase aliasing is achieved, and the obtained three-dimensional seismic data and the corresponding three-dimensional fault label data are used as the source domain dataset for deep learning.

5. The three-dimensional fault identification method based on geological stress constraints and adaptive transfer learning according to claim 1, characterized in that: Step S4 specifically includes the following steps: S41. Compile the actual three-dimensional seismic and fault interpretation data corresponding to different oilfields or structures during the exploration and development of the oilfield in the research area. S42. Convert a large number of discrete fault interpretation points in the fault interpretation results data into standard three-dimensional continuous cross-section label data by reconstructing three-dimensional cross-sections. The conversion formula is: In the above formula: This represents the operation of finding a line between two points. This represents bilinear interpolation. Indicates different fault numbers. Indicates the numbering of different fault points on the same fault line; Indicates the number is Numbering on the fault The fault interpretation point of the breakpoint; S43. Standardize the actual three-dimensional seismic data of different oil fields or structures; The formula for standardization is: In the above formula: This represents standardized 3D actual seismic data; Represents actual three-dimensional seismic data; This represents the operation of finding the minimum value; This represents the operation to find the maximum value; S44. The standardized 3D actual seismic data and 3D continuous cross-section label data are mapped one by one, and then the data is augmented by mirroring, scaling and rotation, and then used as the target domain dataset for deep learning. The mirror formula is: The scaling formula is: The rotation formula is: In the above formula: This represents the scaling factor in the x-direction. This represents the scaling factor in the y-direction; Indicates the scaling factor in the z-direction; express Number of directional sampling points; express Number of directional sampling points; Indicates the number of vertical sampling points; This represents the angle of clockwise rotation around the origin in the xOy plane; This represents the spatial x-coordinate before the data augmentation transformation; This represents the spatial y-coordinate before the data augmentation transformation; This represents the spatial z-coordinate before the data augmentation transformation.

6. The three-dimensional fault identification method based on geological stress constraints and adaptive transfer learning according to claim 1, characterized in that: Step S5 specifically includes the following steps: S51. Based on the traditional U-net network structure, the source domain samples are used as grayscale image inputs. The input grayscale image is converted into R, G, and B components using the RGB conversion method. The R, G, and B components are used as multi-channel input data. The input of different channels is processed by two convolutional neural networks and one pooling layer to complete one downsampling process, and dense connections are formed in different downsampling processes. Then, the downsampling results obtained from multiple channels are fused together and an upsampling convolution is performed. The convolution result is then fused with the downsampling result of the previous layer and two normalized convolutions are performed to complete one upsampling process, thus constructing a 3D-Unet network structure based on dense connections. S52. Based on the densely connected 3D-Unet network structure constructed in step S51, target domain sample input is added to the input end, and a binary classification domain discrimination network is constructed through source domain samples and target domain samples. The adversarial domain adaptive network is added to the densely connected 3D-Unet network through the gradient reversal layer to construct a densely connected 3D-Unet network structure based on adversarial domain adaptive transfer learning. S53. The source domain and target domain sample datasets obtained in steps S3 and S4 are respectively used as grayscale inputs. Based on the RGB conversion of the grayscale, the input three-dimensional seismic data is converted into R, G, and B components as input data. By using the data feature distribution patterns of different domain sources, a large-scale transfer learning is performed on the three-dimensional fault feature information in the source domain samples using a large number of actual three-dimensional fault interpretation results, so as to improve the accuracy of automatic three-dimensional fault identification of actual seismic data.

Images