Stratum fracture identification method, system and equipment and storage medium

Abstract

The embodiment of the invention relates to the technical field of fracture recognition, and discloses a stratum fracture recognition method, system and device and a storage medium, and the method comprises the steps: carrying out the convolution of a random dominant frequency Ricker wavelet and a fracture geologic model, obtaining a fracture synthesis seismic model, and setting the fracture synthesis seismic model as sample data; performing deep learning on the sample data to obtain a nonlinear corresponding relation between fracture and seismic data, and establishing a deep learning training model; and calculating actual seismic data through the deep learning training model to obtain an actual seismic data fracture identification result. According to the exemplary embodiment of the invention, through seismic data and fracture modeling, low-resolution and high-resolution fracture modeling and data-driven training of the deep network, the nonlinear relationship between seismic data and fracture is established firstly, then the nonlinear relationship between low-resolution fracture and high-resolution fracture is established, and two-step processing is carried out to form a complete fracture identification process. Fracture feature recognition is clear, and interference of stratum information does not exist.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the technical field of fracture identification, and in particular, to a method, a system, a device, and a storage medium for identifying formation fractures. Background Art

[0002] The methods for identifying formation fractures based on seismic data are mainly reflected in the application of various attributes (such as coherence, curvature, edge detection), which are mainly used to depict large-scale fractures. For micro-scale fractures and cracks, in recent years, there are mainly ant attributes, coherence enhancement attributes, and maximum likelihood attributes. The ant tracking technology has now been widely used in the automatic identification of fractures in seismic data. The research on using ant tracking to identify potential fracture development zones in shale has been used in the Huangjinba block and the Changning block, and the application effect is good. The AFE coherence enhancement technology was initially applied to the Paradigm software, and then the CGG's Insight earth software also adopted this method, but the algorithms are different. The maximum likelihood attribute currently has corresponding modules in the Paradigm software, the Landmark software, and the VVA software, but the calculation results are greatly affected by parameters.

[0003] From the perspective of identification accuracy, the current methods such as coherence, curvature, and edge detection have a relatively high accuracy in identifying large-scale formation fractures and are consistent with actual drilling. However, for fractures with a fault throw of less than 20 meters, such attributes cannot be reflected. The ant, coherence enhancement, and maximum likelihood attributes have a relatively high accuracy and can represent small-scale fracture information that cannot be identified by coherence, curvature, and edge detection. However, they are greatly affected by parameters and the calculation credibility is relatively low, and the coincidence rate with actual drilling is only about 60%. In order to improve the accuracy, aggressive calculation parameters are often used, resulting in overdevelopment of predicted fractures. From the perspective of safety production, preventive measures have been taken for these identified fractures in the early stage of drilling and fracturing design. However, in the later engineering practice, it has been confirmed that a large number of predicted fractures do not exist, resulting in some preventive measures affecting the drilling and reservoir stimulation effects and increasing the engineering cost.

[0004] From the perspective of method principles, traditional formation fracture identification methods are divided into two categories. One is based on the principle that the more curved the formation is and the more developed the fractures are, and the morphological undulations of the seismic data reflection event axis are detected to indirectly predict the fracture position, such as attributes like curvature. The other is based on the principle that the formation is offset, and the correlation or similarity between each seismic trace of the seismic data and the adjacent seismic traces is detected to determine whether there is a fracture, such as coherence, edge detection, AFE, and maximum likelihood attributes. The problems with these two types of methods are that the formation fracture identification methods with high credibility have low accuracy, and the formation fracture identification methods with high accuracy have low credibility. Summary of the Invention

[0005] Embodiments of the present disclosure provide a method, system, device, and storage medium for identifying formation fractures to solve or alleviate one or more of the above technical problems in the prior art.

[0006] According to one aspect of the present disclosure, a method for identifying formation fractures is provided, including:

[0007] Form a distorted structure velocity model;

[0008] Randomly specify the position of the fracture center point, generate fractures with arbitrary attitudes, set the fracture plane slices to be thin and thick, and combine them with the distorted structure velocity model to obtain a fracture geological model;

[0009] Convolve a random dominant frequency Ricker wavelet with the fracture geological model to obtain a fracture synthetic seismic model, which is set as sample data;

[0010] Perform deep learning on the sample data to obtain the non-linear correspondence between fractures and seismic data, and establish a deep learning training model;

[0011] Calculate the actual seismic data fracture identification result through the deep learning training model for the actual seismic data.

[0012] In a possible implementation, the forming of the distorted structure velocity model includes:

[0013] Establish a two-dimensional horizontal velocity model based on the velocity and density curves of well logging data, structural interpretation horizons, and seismic stratification data;

[0014] Perform vertical displacement on the two-dimensional horizontal velocity model to form a distorted structure velocity model.

[0015] In a possible implementation, performing vertical displacement on the two-dimensional horizontal velocity model to form a distorted structure velocity model includes:

[0016] Let the two-dimensional horizontal velocity model be V(X,Z);

[0017] Perform vertical displacement on the two-dimensional horizontal velocity model V(X,Z) to generate a distorted structure velocity model V(X ′ ,Z ′ ), and the vertical displacement formula is:

[0018]

[0019] In the formula, X and Z are the horizontal coordinate and depth coordinate of the two-dimensional horizontal velocity model V(X,Z) respectively, X′ and Z′ are the horizontal coordinate and depth coordinate of the distorted structure velocity model V(X′,Z′) respectively, Z m (X) is the vertical displacement amount, M(Z m) represents the displacement field generated after the vertical displacement.

[0020] In a possible implementation, randomly specifying the position of the fracture center point, generating fractures with arbitrary attitudes, and combining with the distorted tectonic velocity model, the fracture geological model obtained includes:

[0021] In the distorted tectonic velocity model, randomly specifying the position of the fracture center point (x 0 , y 0 , z 0 ), the major axis L x of the ellipse and the minor axis l y to generate a horizontal elliptical fracture surface;

[0022] When the horizontal elliptical fracture surface is generated, assign 1, otherwise assign 0;

[0023] Rotate the horizontal elliptical fracture surface by a random azimuth angle and dip angle to obtain fractures with arbitrary attitudes;

[0024] Displace the distorted tectonic velocity model at the fracture to obtain a fracture geological model containing fractures.

[0025] In a possible implementation, when the horizontal elliptical fracture surface is generated, assign 1, otherwise assign 0 includes:

[0026] Establish a distance formula:

[0027]

[0028] In the formula, l x is the major axis of the ellipse, l y is the minor axis of the ellipse, x 0 is the abscissa of the fracture center, y 0 is the ordinate of the fracture center, x is the abscissa of the point to be calculated, y is the ordinate of the point to be calculated, and dis(x, y) is the distance from the point to be calculated (x, y) to the fracture center (x 0 , y 0 );

[0029] Assign values to the horizontal elliptical fracture surface:

[0030] F(x, y, z_0) = 0 when dis(x, y) > 0;

[0031] F(x, y, z_0) = 1 when dis(x, y) ≤ 0;

[0032] In the formula, F(x, y, z_0) represents the elliptical fracture surface.

[0033] In a possible implementation, the method includes: after the random main frequency Ricker wavelet is convolved with the fault geological model to obtain a synthetic fault seismic model and the model is set as sample data, the method includes: simplifying the synthetic fault seismic model, including:

[0034] The seismic profile of 128×128 elements is used as the feature image input, two 3×3 convolution layers and activation layers are added, and the image size before and after convolution is set unchanged in the neural network framework of the fault synthetic seismic model;

[0035] Perform 2×2 pooling, halve the size of the feature image, and continue to add convolutional layers and ReLU activation layers;

[0036] This cycle is repeated four times, with the number of feature images doubled each time;

[0037] When the characteristic image size of the fault synthetic earthquake model is 16×16, it is upsampled to 32×32 and merged with the result of the third cycle, and the convolution layer activation layer is continued and upsampled;

[0038] When the feature image size is restored to 128×128, after passing through the convolution layer and activation layer, 3×3 convolution and activation are performed;

[0039] Use 1×1 convolution to reduce the dimension and reduce the number of feature images to one 128×128 image.

[0040] In a possible implementation, the position of the fracture center point is randomly given to generate a fracture of any occurrence, and the fracture cross-section is set to be thin or thick, and combined with the twisted structure velocity model to obtain a fracture geological model including:

[0041] Generate a horizontal plane, randomly generate inclination and azimuth;

[0042] Rotate the horizontal plane according to the dip and azimuth to form the initial occurrence of the fault surface;

[0043] Apply a random range of displacement on the fracture surface to produce cross-section fluctuations;

[0044] Generate two three-dimensional 0-value discrete data, set two distance thresholds dz1, dz2, and dz1>dz2. In one of the three-dimensional 0-value discrete data, the discrete points whose distance from the fracture surface is less than dz1 are assigned a value of 1 to form a low-resolution fracture attribute body. In the other three-dimensional discrete data, the discrete points whose distance from the fracture surface is less than dz2 are assigned a value of 1 to form a high-resolution fracture attribute body;

[0045] The twisted structural velocity model is displaced along both sides of the fault surface of the low-resolution fault attribute body and the high-resolution fault attribute body to obtain the low-resolution fault geological model and the high-resolution fault geological model.​

[0046] In a possible implementation, it includes:

[0047] Taking the low-resolution fracture attribute volume as the input label and the high-resolution attribute volume as the output label, performing deep network training to form a high-resolution fracture conversion model.

[0048] According to one aspect of the present disclosure, a formation fracture identification system is provided, including:

[0049] A formation unit for forming a distorted structural velocity model;

[0050] A combination unit for randomly specifying the position of the fracture center point, generating fractures with arbitrary attitudes, setting the fracture plane slices to be thin and thick, and combining them with the distorted structural velocity model to obtain a fracture geological model;

[0051] A convolution unit for convolving a randomly generated dominant frequency Ricker wavelet with the fracture geological model to obtain a fracture synthetic seismic model, which is set as sample data;

[0052] A establishment unit for performing deep learning on the sample data to obtain a non-linear correspondence between fractures and seismic data, and establishing a deep learning training model;

[0053] A calculation unit for calculating the actual seismic data fracture identification result through the deep learning training model for the actual seismic data.

[0054] In a possible implementation, the formation unit includes:

[0055] A establishment module for establishing a two-dimensional horizontal velocity model according to the velocity and density curves of well logging data, structural interpretation horizons, and seismic stratification data;

[0056] A formation module for performing vertical displacement on the two-dimensional horizontal velocity model to form a distorted structural velocity model.

[0057] In a possible implementation, the formation module is used for:

[0058] Let the two-dimensional horizontal velocity model be V(X,Z);

[0059] Performing vertical displacement on the two-dimensional horizontal velocity model V(X,Z) to generate a distorted structural velocity model V(X ′ ,Z ′ ), and the vertical displacement formula is:

[0060]

[0061] Wherein, X and Z are respectively the horizontal coordinate and the depth coordinate of the two-dimensional horizontal velocity model V(X, Z), X' and Z' are respectively the horizontal coordinate and the depth coordinate of the distorted structural velocity model V(X', Z'), and Z m (X) is the vertical displacement, and M(Z m ) represents the displacement field generated after the vertical displacement.

[0062] In a possible implementation, the combining unit is configured to:

[0063] In the distorted structural velocity model, randomly assign the position of the fracture center point (x 0 , y 0 , z 0 ), the major axis L of the ellipse x and the minor axis l of the ellipse y to generate a horizontal elliptical fracture surface;

[0064] Assign 1 when the horizontal elliptical fracture surface is generated, otherwise assign 0;

[0065] Rotate the horizontal elliptical fracture surface by a random azimuth angle and dip angle to obtain fractures with arbitrary attitudes;

[0066] Displace the distorted structural velocity model at the fracture to obtain a fractured geological model containing fractures.

[0067] In a possible implementation, assigning 1 when the horizontal elliptical fracture surface is generated and otherwise assigning 0 includes:

[0068] Establish a distance formula:

[0069]

[0070] Wherein, l x is the major axis of the ellipse, l y is the minor axis of the ellipse, x 0 is the abscissa of the fracture center, y 0 is the ordinate of the fracture center, x is the abscissa of the point to be determined, y is the ordinate of the point to be determined, and dis(x, y) is the distance from the point to be determined (x, y) to the fracture center (x 0 , y 0 );

[0071] Assign a value to the horizontal elliptical fracture surface:

[0072] F(x, y, z_0) = 0 when dis(x, y) > 0;

[0073] F(x, y, z_0) = 1 when dis(x, y) ≤ 0;

[0074] Wherein, F(x, y, z_0) represents the elliptical fracture surface.

[0075] In a possible implementation, it includes: a simplification unit for simplifying the fracture synthesis seismic model, including:

[0076] Taking a seismic profile of 128×128 elements as a feature image input, adding two 3×3 convolutional layers and activation layers, and setting the image size before and after convolution to be unchanged in the neural network framework of the fracture synthesis seismic model;

[0077] Performing 2×2 pooling, halving the size of the feature image, and continuously adding convolutional layers and ReLU activation layers;

[0078] Iterating in this way four times, doubling the number of feature images each time;

[0079] When the size of the feature image of the fracture synthesis seismic model is 16×16, performing upsampling to restore it to 32×32, merging it with the result of the third cycle, and continuing with convolutional layers and activation layers, and then performing upsampling;

[0080] When the size of the feature image is restored to 128×128, after passing through convolutional layers and activation layers, performing 3×3 convolution and activation;

[0081] Reducing the dimension by using 1×1 convolution to reduce the number of feature images to one 128×128 image.

[0082] In a possible implementation, the combination unit is used for:

[0083] Generating a horizontal plane and randomly generating dip angles and azimuth angles;

[0084] Rotating the horizontal plane according to the dip angle and azimuth angle to form the preliminary attitude of the fracture surface;

[0085] Applying a displacement amount within a random range to the fracture surface to generate fracture surface undulations;

[0086] Generating two three-dimensional discrete data with a value of 0, setting two distance thresholds dz1 and dz2, and dz1>dz2. In one of the three-dimensional discrete data with a value of 0, assigning a value of 1 to the discrete points whose distance from the fracture surface is less than dz1 to form a low-resolution fracture attribute body, and in the other three-dimensional discrete data, assigning a value of 1 to the discrete points whose distance from the fracture surface is less than dz2 to form a high-resolution fracture attribute body;

[0087] Displacing the distorted tectonic velocity model along both sides of the fracture surface of the low-resolution fracture attribute body and the high-resolution fracture attribute body to obtain a low-resolution fracture geological model and a high-resolution fracture geological model.

[0088] In a possible implementation, it includes:

[0089] A training unit for performing deep network training with the low-resolution fracture attribute volume as the input label and the high-resolution attribute volume as the output label to form a high-resolution fracture conversion model.

[0090] According to one aspect of the present disclosure, a formation fracture identification device is provided, including:

[0091] A processor and a memory;

[0092] The memory is used to store a computer program, and the processor calls the computer program stored in the memory to execute the formation fracture identification method described in any one of the above.

[0093] According to one aspect of the present disclosure, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the processor can execute the formation fracture identification method described in any one of the above.

[0094] The exemplary embodiments of the present disclosure have the following beneficial effects: The exemplary embodiments of the present disclosure, through seismic data and fracture, low-resolution and high-resolution fracture modeling, and data-driven training of a deep network, first establish a non-linear relationship between seismic data and fractures, and then establish a non-linear relationship between low-resolution fractures and high-resolution fractures. Through these two-step processes, a complete fracture identification process is formed, enabling clear fracture feature identification without interference from formation information.

[0095] Details of one or more embodiments of the present application are set forth in the following drawings and description. Other features and advantages of the present application will become apparent from the drawings of the specification. It should be understood that the above general description and the following detailed description are merely exemplary and explanatory and do not limit the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] The drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present disclosure and used together with the specification to explain the principles of the present disclosure. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0097] Figure 1 is a flowchart of a formation fracture identification method according to this exemplary embodiment;

[0098] Figure 2 is a schematic diagram of fractures and seismic data labels according to this exemplary embodiment;

[0099] Figure 3 is a schematic diagram of a deep convolutional neural network based on Unet according to this exemplary embodiment;

[0100] Figure 4 is a schematic diagram of the fracture high-resolution identification process of this exemplary embodiment;

[0101] Figure 5 is a schematic diagram of the high- and low-resolution fracture model labels of this exemplary embodiment;

[0102] Figure 6 is a schematic diagram of the original seismic profile of this exemplary embodiment;

[0103] Figure 7 is a schematic diagram of the fracture attribute surface of the third-generation coherence algorithm of this exemplary embodiment;

[0104] Figure 8 is a schematic diagram of the fracture attribute profile of the deep learning algorithm of this exemplary embodiment;

[0105] Figure 9 is a schematic diagram of the coherence along-layer attribute of this exemplary embodiment;

[0106] Figure 10 is a schematic diagram of the machine learning slice of this exemplary embodiment;

[0107] Figure 11 is a schematic diagram of the specific implementation process of a fracture identification method based on machine learning of this exemplary embodiment;

[0108] Figure 12 is a block diagram of a formation fracture identification system of this exemplary embodiment;

[0109] Figure 13 is a schematic diagram of the structure of a formation fracture identification device of this exemplary embodiment. Detailed Implementation Manner

[0110] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of the present disclosure. However, those skilled in the art will recognize that the technical solutions of the present disclosure can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. can be used. In other cases, well-known technical solutions are not shown or described in detail to avoid obscuring the various aspects of the present disclosure.

[0111] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software form, or in one or more hardware units or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0112] The flowcharts shown in the accompanying drawings are only illustrative and do not necessarily include all steps. For example, some steps can be further decomposed, while some steps can be combined or partially combined. Therefore, the actual execution order may be changed according to the actual situation.

[0113] The terms "first", "second", etc. in the description and claims of this application and the above accompanying drawings are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of the present application described herein can be implemented in an order different from those illustrated or described herein.

[0114] In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or sub-modules does not necessarily have to be limited to those steps or sub-modules clearly listed, but may include other steps or sub-modules not clearly listed or inherent to these processes, methods, products or devices.

[0115] Fracture is an important factor in the preferential selection of favorable areas for shale gas, the formulation of development policies, the optimization of drilling trajectories, and the design of fracturing programs. The Longmaxi Formation shale in the Sichuan Basin has experienced multiple tectonic movements, resulting in complex tectonic deformation of the strata and the development of multi-scale fractures from fractures to microfractures. The effects of fractures of different scales on shale gas are different. Large-scale fractures can break through the overlying caprock and become channels for shale gas loss, causing the gas reservoir to be damaged or the gas content to decrease. When deploying horizontal wells, try to avoid the drilling trajectory crossing such fracture zones; micro-scale fracture zones do not affect the gas-bearing property of the reservoir and provide channels for shale gas production during the later development process, which are important drilling targets. Therefore, under the basic principle that the azimuth of the horizontal section trajectory in well placement should intersect at a large angle with the direction of the maximum horizontal principal stress, it should also intersect at a large angle with medium and small-scale natural fracture zones as much as possible, so as to promote more hydraulic fractures to be generated during fracturing construction and improve the reservoir stimulation effect. In fracturing design, measures such as avoiding perforation or reducing the displacement are usually adopted for large-scale fracture zones to avoid excessive concentration of fracturing stress and fracture slip.

[0116] At present, the achievements of fracture identification run through the entire life cycle of shale gas exploration and development. During the well location deployment stage, it is required that the well location be 1500 meters away from the Grade I fractures and 700 meters away from the Grade II fractures. During the development stage, it is necessary to utilize the distribution areas of the network fractures and single-direction fractures determined by the fracture prediction results, which are important reference bases for the division of development units and the determination of the horizontal well trajectory direction. During the drilling stage, fracture prediction is carried out to anticipate the fracture zones that the trajectory may encounter, and targeted measures for possible well leakage are taken in advance to reduce severe well leakage and downhole complex accidents. During the fracturing stage, accurately characterize natural fractures before fracturing to improve the accuracy of the segmented multi-cluster fracturing scheme design and optimize the fracturing design; during the fracturing implementation, dynamically master the relationship between the propagation of microseismic events and fractures to meet the needs of real-time adjustment of the construction technology and parameters during fracturing.

[0117] Fractures are manifested as discontinuous event axes on seismic profiles. Traditional fracture identification is completed by human-computer interaction and interpreted based on attributes such as coherence, curvature, and edge detection. There are problems such as low efficiency, large uncertainty in results and strong multiple solutions caused by human factors, which increase the cost and risk of oil and gas exploration and development. In order to improve the identification ability of small fractures with weak responses in seismic data, this embodiment proposes a fracture identification method at different scales based on deep learning analysis of seismic data.

[0118] Instead of starting from the perspective of similarity detection, this embodiment starts from the perspective of deep learning analysis of seismic data, uses a convolutional neural network to build a network model for fracture identification in seismic data, applies multi-scale and multi-level feature extraction technology, and finally realizes fracture identification. During the learning process, the fracture sample information does not adopt artificial interpretation of fractures and traditional fracture prediction attributes (such as coherence, curvature, etc.), but uses synthetic seismic data determined according to the fracture geological model to improve the credibility of identification.

[0119] Figure 1 is a flowchart of a formation fracture identification method according to an exemplary embodiment of the present invention. As Figure 1 shown, an exemplary embodiment of the present disclosure provides a formation fracture identification method, including:

[0120] S1 Form a distorted structural velocity model;

[0121] S2 Randomly specify the position of the fracture center point, generate fractures with arbitrary attitudes, set the fracture surface slices to be thin and thick, and combine them with the distorted structural velocity model to obtain a fracture geological model;

[0122] S3 Convolve a random dominant frequency Ricker wave with the fracture geological model to obtain a fracture synthetic seismic model, which is set as sample data;

[0123] S4 Perform deep learning on the sample data to obtain the non-linear correspondence between fractures and seismic data, and establish a deep learning training model;

[0124] S5 calculates the actual seismic data fracture identification result by using the deep learning training model for the actual seismic data.

[0125] Specifically, the formation of the distorted structure velocity model includes:

[0126] Based on the velocity and density curves of well logging data, structural interpretation horizons, and seismic stratigraphic data, a two-dimensional horizontal velocity model is established;

[0127] Vertical displacement is performed on the two-dimensional horizontal velocity model to form a distorted structure velocity model.

[0128] Specifically, performing vertical displacement on the two-dimensional horizontal velocity model to form a distorted structure velocity model includes:

[0129] Let the two-dimensional horizontal velocity model be V(X, Z);

[0130] Vertical displacement is performed on the two-dimensional horizontal velocity model V(X, Z) to generate a distorted structure velocity model V(X′, Z′), and the vertical displacement formula is:

[0131]

[0132] In the formula, X and Z are the horizontal coordinate and depth coordinate of the two-dimensional horizontal velocity model V(X, Z) respectively, X′ and Z′ are the horizontal coordinate and depth coordinate of the distorted structure velocity model V(X′, Z′) respectively, Z m (X) is the vertical displacement amount, and M(Z m ) represents the displacement field generated after vertical displacement.

[0133] Specifically, randomly specifying the position of the fracture center point, generating fractures with arbitrary attitudes, and combining with the distorted structure velocity model to obtain a fracture geological model includes:

[0134] In the distorted structure velocity model, randomly specify the position of the fracture center point (x 0 , y 0 , z 0 ), the major axis l of the ellipse x and the minor axis l of the ellipse y , and generate a horizontal elliptical fracture surface;

[0135] When the horizontal elliptical fracture surface is generated, assign 1, otherwise assign 0;

[0136] Randomly rotate the azimuth and dip angle of the horizontal elliptical fracture surface to obtain fractures with arbitrary attitudes;

[0137] Displace the distorted structure velocity model at the fracture to obtain a fracture geological model containing fractures.

[0138] Specifically, when the horizontal elliptical fracture surface is generated, assign a value of 1, otherwise assign a value of 0, including:

[0139] Establish a distance formula:

[0140]

[0141] In the formula, l x is the major axis of the ellipse, l y is the minor axis of the ellipse, x 0 is the abscissa of the fracture center, y 0 is the ordinate of the fracture center, x is the abscissa of the point to be calculated, y is the ordinate of the point to be calculated, and dis(x, y) is the distance from the point to be calculated (x, y) to the fracture center (x 0 , y 0 );

[0142] Assign a value to the horizontal elliptical fracture surface:

[0143] F(x, y, z_0) = 0 when dis(x, y) > 0;

[0144] F(x, y, z_0) = 1 when dis(x, y) ≤ 0;

[0145] In the formula, F(x, y, z_0) represents the elliptical fracture surface.

[0146] Specifically, it includes: after convolving the random dominant frequency Ricker wave with the fracture geological model to obtain the fracture synthetic seismic model and setting it as the sample data, it includes: performing a simplification process on the fracture synthetic seismic model, including:

[0147] Input a seismic profile with 128×128 elements as the feature image, add two 3×3 convolutional layers and activation layers, and set the image size before and after convolution to be unchanged in the neural network framework of the fracture synthetic seismic model;

[0148] Perform 2×2 pooling, halving the size of the feature image, and continue to add convolutional layers and ReLU activation layers;

[0149] Iterate in this way four times, doubling the number of feature images each time;

[0150] When the size of the feature image of the fracture synthetic seismic model is 16×16, perform upsampling to restore it to 32×32, merge it with the result of the third cycle, continue with convolutional layers and activation layers, and perform upsampling;

[0151] When the size of the feature image is restored to 128×128, after passing through convolutional layers and activation layers, perform 3×3 convolution and activation;

[0152] Reduce the dimension by using 1×1 convolution to reduce the number of feature images to one 128×128 image.

[0153] Specifically, randomly given the position of the fracture center point to generate fractures with arbitrary attitudes, and setting the fracture surface slices to be thin and thick. Combining with the twisted structure velocity model, the fracture geological model obtained includes:

[0154] Generate a horizontal plane and randomly generate the dip angle and azimuth angle;

[0155] Rotate the horizontal plane according to the dip angle and azimuth angle to form the preliminary attitude of the fracture surface;

[0156] Apply a displacement amount within a random range to the fracture surface to generate undulations on the fracture surface;

[0157] Generate two three-dimensional discrete data with a value of 0, set two distance thresholds dz1 and dz2, and dz1>dz2. Assign a value of 1 to the discrete points in one of the three-dimensional discrete data with a distance from the fracture surface less than dz1 to form a low-resolution fracture attribute body, and assign a value of 1 to the discrete points in the other three-dimensional discrete data with a distance from the fracture surface less than dz2 to form a high-resolution fracture attribute body;

[0158] Displace the twisted structure velocity model along both sides of the fracture surface of the low-resolution fracture attribute body and the high-resolution fracture attribute body to obtain a low-resolution fracture geological model and a high-resolution fracture geological model.

[0159] Specifically, it includes:

[0160] Use the low-resolution fracture attribute body as the input label and the high-resolution attribute body as the output label for deep network training to form a high-resolution fracture conversion model.

[0161] In this example, a method for identifying formation fractures using seismic data is provided. This method can be divided into two steps. The first step is to establish a non-linear relationship between fractures and seismic data through a deep learning network, and the second step is to establish a non-linear relationship between the high-resolution fracture attributes and the low-resolution fracture attributes through a deep learning network.

[0162] (1) Establishing a non-linear relationship between fractures and seismic data through deep learning:

[0163] Input the post-stack seismic data into the deep learning network to calculate the fracture data. The essence of the deep learning network training is to enable the network to have the ability to fit the functional relationship between the seismic data and the corresponding fractures. Training requires a large amount of sample data, that is, a sample label library containing seismic data and corresponding fracture data needs to be established. Whether the sample data has the training target characteristics and can sufficiently represent the corresponding relationship between the seismic data and the fractures is crucial for the training results.

[0164] There are two methods to obtain fracture information from actual seismic data: 1. Using manual interpretation, which has high labor costs and subjective errors, and it is difficult to accurately and efficiently establish a label library; 2. Conducting traditional fracture calculations on actual seismic data, such as fracture identification using coherence, curvature, ant tracking, etc., and using the results as labels. Training samples with such labels can only approximately fit traditional fracture characteristics and it is difficult to achieve breakthrough results. In short, it is relatively difficult to directly extract accurate fracture information from actual seismic data. Therefore, a label library is established by synthesizing seismic data. The advantage of the method is that a geological model is first established, and fractures are used as known given information to guide the structural form of the geological model. Then, seismic data is synthesized, and the synthesized seismic data corresponds to dip information, with accurate and known fracture information.

[0165] Let the two-dimensional horizontal velocity model be V(X,Z), and perform vertical displacement according to formula (1) to generate a distorted structural velocity model V(X ′ ,Z ′ ).

[0166]

[0167] Where X and Z are the horizontal coordinate and depth coordinate of the horizontal velocity model respectively, and X ′ ,Z ′ are the horizontal and depth coordinates of the distorted structural velocity model respectively, and Z m (X) is the vertical displacement amount, which is different at different horizontal coordinates X, generating a displacement field M(Z m ), forming undulations.

[0168] Fractures are generated based on the distorted structural velocity model V(X ′ ,Z ′ ). Randomly specify the position of the fracture center point (x 0 ,y 0 ,z 0 ). First, generate a horizontal elliptical fracture surface. To ensure model diversity, the major and minor axes of the ellipse are randomly generated as l x and l y . When generating the fracture surface, assign a value of 1 inside the ellipse and 0 otherwise. Establish the distance formula (2):

[0169]

[0170] Elliptical fracture surface assignment:

[0171] F(x,y,z_0) = 0 when dis(x,y)>0

[0172] F(x,y,z_0) = 1 when dis(x,y)≤0 (3);

[0173] Figure 2It is a schematic diagram of fracture and seismic data labels in this exemplary embodiment; Figure 2 (a) is the formation velocity model, Figure 2 (b) is the synthetic seismic data, Figure 2 (c) is the fracture label. As Figure 2 shown, after generating a three-dimensional horizontal section, the section is rotated with random azimuth and dip angles to obtain fractures with arbitrary attitudes. Then, a three-dimensional discrete 0-value data is generated. Since the section may not pass through the discrete points, the discrete points closest to the section position are assigned a value of 1, and the rest are assigned a value of 0, thus forming a three-dimensional data volume containing fracture information. However, this results in certain discrete patches. Therefore, a slight smoothing is performed to obtain a three-dimensional fracture attribute volume, Figure 2 (c) shown. The velocity model V(X ′ ,Z ′ ) is displaced near the fracture to obtain a formation model containing fractures. As Figure 2 (a) shown.

[0174] Synthetic seismic records are generated from the formation model. The random main frequency Ricker wavelet is convolved with the formation model information to generate synthetic records. Setting the main frequency to vary randomly is the key, which can avoid the interference of seismic data frequency on fracture identification and enable the trained convolutional neural network to be applicable to seismic data with different main frequencies. Figure 2 (b) shows the generated seismic data profile.

[0175] Since calculating fractures does not require the same high complexity as the original Unet model, we simplified the model to ensure the efficiency of the training process and the prediction process. As Figure 3 shown, a seismic profile of 128×128 elements is used as the input, and two 3×3 convolutional layers and activation layers are added (at the upper left of the figure). In Keras, it is set that the image size remains unchanged before and after convolution. Next, 2×2 pooling is performed, and the size of the feature image is halved. Then, convolutional layers and ReLU activation layers are continuously added. This loop is iterated four times, and the number of feature images is doubled each time. When the size of the feature image of the Unet model is 16×16, upsampling is performed to restore it to 32×32 and merged with the result of the third loop. Then, convolutional layers and activation layers are continued, and upsampling is performed. When the size of the feature image is restored to 128×128, after passing through convolutional layers and activation layers, 3×3 convolution and activation are performed. Finally, dimensionality reduction is performed in the 1×1 convolution manner to reduce the number of feature images to one 128×128 image. The structure of the deep convolutional neural network is as Figure 3 shown.

[0176] The data labels are trained, and after convergence, the non-linear correspondence between fractures and seismic data can be obtained. Save the trained model, and using the actual seismic data as the input, the fracture characteristics can be output.

[0177] (2) Establishing the non - linear relationship between fracture high - resolution attributes and fracture low - resolution attributes through deep learning

[0178] The strategies of up - sampling and image - sharing weights in the deep convolutional neural network enable the convolutional network to achieve efficient training, but also bring the problem of up - sampling blurring. In fracture recognition, the fracture width is relatively rough and the resolution is low. Therefore, simultaneous modeling of high - resolution and low - resolution is carried out. Taking the fracture low - resolution model as the input and the high - resolution model as the output, the fracture high - resolution recognition network is trained.

[0179] The intelligent recognition process of fracture high - resolution is as Figure 4 shown. The process of cross - section generation is similar to the above - mentioned embodiment. The difference is that the given cross - section is no longer just a plane, but a body with thickness. The high - resolution cross - section is thinner, while the low - resolution cross - section is thicker. The initial horizontal cross - section formula (4) is as follows:

[0180] F(x,y,z_0±dz)=0 when dis(x,y)>0

[0181] F(x,y,z_0±dz)=1 when dis(x,y)≤0(4);

[0182] Similarly, after generating the three - dimensional horizontal cross - section body, the cross - section is rotated by a random azimuth angle and dip angle to obtain fractures with any attitude. To make the fractures more realistic, the cross - section is displaced with a certain degree of undulation.

[0183] Figure 5 shows the three - dimensional stereogram of high - and low - resolution fractures. Figure 5 (a) is the synthetic seismic data. Figure 5 (b) is the low - resolution fracture data. Figure 5(c) High-resolution fracture data is obtained as follows: 1. First, generate a horizontal plane, randomly generate an inclination angle ranging from 45 degrees to 135 degrees, and randomly generate an azimuth angle ranging from 0 degrees to 360 degrees. Rotate the horizontal plane according to the inclination and azimuth angles to form the initial attitude of the fracture surface. 2. Apply a random range of displacement amounts to the fracture surface to generate undulations on the fracture surface, making the generated fracture surface characteristics closer to the actual fracture data. 3. Generate two three-dimensional discrete data with a value of 0. Set two distance thresholds dz1 and dz2, where dz1 > dz2. In one of the three-dimensional discrete data, assign a value of 1 to the discrete points whose distance from the fracture surface is less than dz1 to form a low-resolution fracture attribute body. In the other three-dimensional discrete data, assign a value of 1 to the discrete points whose distance from the fracture surface is less than dz2 to form a high-resolution fracture attribute body. The high- and low-resolution labels are thus generated. 4. Displace the velocity model along both sides of the fracture surface and convolve it with a random frequency wavelet to generate the seismic data corresponding to the fracture. Add random Gaussian noise to the seismic data to improve the simulation degree of the synthetic seismic data to a certain extent. Displacing the formation through the fracture surface and then generating seismic data ensures the consistency between the fracture attitude and the seismic data characteristics and guarantees the reliability of the samples.

[0184] For low-resolution fractures, high-resolution fractures can be output after being processed by a deep neural network. The training uses three-dimensional data with a size of 128×128×128 for both the input data and the output data of the samples. In the stage of training the non-linear relationship between the training seismic data and the fractures, the number of label pairs in the training set is 7000, and the number of label pairs in the validation set is 3000. In the stage of training the non-linear relationship between low-resolution and high-resolution fractures, the number of label pairs in the training set is 5000, and the number of label pairs in the validation set is 2000.

[0185] Traditional fracture attribute calculation methods are based on the geometric characteristics of the data and establish an artificial simplified mathematical calculation model, resulting in relatively large fracture errors. With the development of artificial intelligence technology, data-driven deep learning methods have natural advantages in automatic modeling and can be effectively applied to the identification of fractures in seismic data. However, upsampling in convolutional neural networks in deep learning will reduce the image resolution. To address this problem, this paper proposes a method for high-resolution fracture identification. By modeling the seismic data, fractures, low-resolution and high-resolution fractures, and training a deep network driven by data, first establish the non-linear relationship between the seismic data and the fractures, and then establish the non-linear relationship between low-resolution and high-resolution fractures. Through these two steps of processing, a complete fracture identification process is formed. In addition, in subsequent applications, adding manually marked labels of actual data to the synthetic samples and using transfer learning with actual data will help further increase the applicability of the algorithm.

[0186] Use this method and the third-generation coherence algorithm in commercial software to respectively conduct method comparison and verification on the three-dimensional actual seismic data of a certain research work area in the Sichuan Basin. Figure 6 is the original seismic profileFigure 7 It is the fracture identification result of the third-generation coherent algorithm, Figure 8 It is the fracture identification result of the machine learning algorithm, Figure 9 It is the schematic diagram of the coherent along-layer attribute, Figure 10 It is the schematic diagram of the machine learning slice. It can be seen from the comparison that the traditional algorithm is seriously interfered by the formation, especially in weak signal areas, and the identification effect is significantly worse; the machine learning only responds to real fractures, can effectively exclude many interference factors, not only reflects the location of the fractures, but also has a higher level of detail, the fractures converge, and is more accurate.

[0187] Generally speaking, for small fractures with minor misalignments of the event axis, the fracture characteristics of the coherent algorithm are not obvious, the fracture characteristics of this method are clear, and there is no interference from formation information.

[0188] One embodiment of the present application proposes a fracture identification method based on machine learning. By using the synthetic seismic record models of various fractures as samples, deep learning is completed to establish a non-linear correspondence between fractures and seismic data, and then it is applied to actual seismic data to identify fractures. The specific operation can be divided into 5 steps, as Figure 11 shown:

[0189] Step 1: According to the velocity and density curves of well logging data, structural interpretation horizons, and seismic stratigraphic data, establish a two-dimensional horizontal velocity model, and on this basis, perform appropriate vertical displacement to obtain a distorted structural geological model.

[0190] Step 2: Randomly specify the position of the fracture center point, generate elliptical fracture surfaces with arbitrary attitudes, and set the thickness of the fault slices to two types: thick and thin. Combine with the geological model in Step 1 to obtain a fracture geological model.

[0191] Step 3: Given seismic wavelets with different dominant frequencies, convolve them with the fracture geological model in Step 2 to obtain fracture synthetic seismic models with different azimuth angles, different dips, different fault offsets, different thicknesses, and different dominant frequencies, which are set as sample data.

[0192] Step 4: Perform deep learning on the sample data to obtain the non-linear correspondence between faults and seismic data, and establish a deep learning training model.

[0193] Step 5: For actual seismic data, use the deep learning training model in Step 4 to perform calculations to obtain the fracture identification result of the actual seismic data.

[0194] Figure 12 It is the block diagram of a formation fracture identification system according to an exemplary embodiment of the present application, as Figure 12 shown. The exemplary embodiment of the present disclosure provides a formation fracture identification system, including:

[0195] A formation unit 10 for forming a distorted structural velocity model;

[0196] A combination unit 20 for randomly specifying the positions of fracture center points, generating fractures with arbitrary attitudes, setting the fracture surface slices to be thin and thick, and combining them with the distorted structural velocity model to obtain a fracture geological model;

[0197] A convolution unit 30 for convolving a randomly generated dominant frequency Ricker wavelet with the fracture geological model to obtain a fracture synthetic seismic model, which is set as sample data;

[0198] A building unit 40 for performing deep learning on the sample data to obtain a non-linear correspondence between fractures and seismic data, and establishing a deep learning training model;

[0199] A calculation unit 50 for calculating actual seismic data through the deep learning training model to obtain a fracture identification result of the actual seismic data.

[0200] Specifically, the formation unit includes:

[0201] A building module for establishing a two-dimensional horizontal velocity model according to the velocity and density curves of well logging data, structural interpretation horizons, and seismic stratification data;

[0202] A formation module for performing vertical displacement on the two-dimensional horizontal velocity model to form a distorted structural velocity model.

[0203] Specifically, the formation module is used for:

[0204] Let the two-dimensional horizontal velocity model be V(X, Z);

[0205] Perform vertical displacement on the two-dimensional horizontal velocity model V(X, Z) to generate a distorted structural velocity model V(X ′ , Z ′ ), and the vertical displacement formula is:

[0206]

[0207] In the formula, X and Z are respectively the horizontal coordinate and depth coordinate of the two-dimensional horizontal velocity model V(X, Z), X ′ , Z ′ are respectively the horizontal coordinate and depth coordinate of the distorted structural velocity model V(X ′ , Z ′ ), Z m (X) is the vertical displacement amount, and M(Z m ) represents the displacement field generated after vertical displacement.

[0208] Specifically, the combination unit is used for:

[0209] In the twisted structure velocity model, the position of the fracture center point (x 0 , y 0 , z 0 ) is randomly given, the major axis length l x of the ellipse and the minor axis length l y of the ellipse are used to generate a horizontal elliptical fracture surface;

[0210] When the horizontal elliptical fracture surface is generated, it is assigned 1, otherwise it is assigned 0;

[0211] The horizontal elliptical fracture surface is rotated by a random azimuth angle and dip angle to obtain fractures with arbitrary attitudes;

[0212] At the fracture, the twisted structure velocity model is displaced to obtain a fractured geological model containing fractures.

[0213] Specifically, when the horizontal elliptical fracture surface is generated, it is assigned 1, otherwise it is assigned 0 includes:

[0214] Establish a distance formula:

[0215]

[0216] In the formula, l x is the major axis of the ellipse, l y is the minor axis of the ellipse, x 0 is the abscissa of the fracture center, y 0 is the ordinate of the fracture center, x is the abscissa of the point to be calculated, y is the ordinate of the point to be calculated, and dis(x, y) is the distance from the point to be calculated (x 0 , y 0 ) to the fracture center;

[0217] Assign a value to the horizontal elliptical fracture surface:

[0218] F(x, y, z_0) = 0 when dis(x, y)>0;

[0219] F(x, y, z_0) = 1 when dis(x, y)≤0;

[0220] In the formula, F(x, y, z_0) represents the elliptical fracture surface.

[0221] Specifically, it includes: a simplification unit for simplifying the fracture synthetic seismic model, including for:

[0222] Taking a seismic profile of 128×128 elements as the feature image input, adding two 3×3 convolutional layers and activation layers, and setting the image size before and after convolution to be unchanged in the neural network framework of the fracture synthetic seismic model;

[0223] Perform 2×2 pooling, halving the size of the feature image, and continue to add a convolutional layer and a ReLU activation layer;

[0224] Iterate this process four times, doubling the number of feature images each time;

[0225] When the size of the feature image of the fracture synthesis seismic model is 16×16, perform upsampling to restore it to 32×32, merge it with the result of the third cycle, continue with the convolutional layer and activation layer, and then perform upsampling;

[0226] When the size of the feature image is restored to 128×128, after passing through the convolutional layer and activation layer, perform 3×3 convolution and activation;

[0227] Reduce the dimension using 1×1 convolution, reducing the number of feature images to one 128×128 image.

[0228] Specifically, the combining unit is used for:

[0229] Generate a horizontal plane and randomly generate dip angle and azimuth angle;

[0230] Rotate the horizontal plane according to the dip angle and azimuth angle to form the initial attitude of the fracture surface;

[0231] Apply a displacement amount within a random range to the fracture surface to generate fracture surface undulations;

[0232] Generate two three-dimensional discrete data with all 0 values, set two distance thresholds dz1 and dz2, and dz1>dz2. In one of the three-dimensional discrete data with all 0 values, assign 1 to the discrete points whose distance from the fracture surface is less than dz1 to form a low-resolution fracture attribute body. In the other three-dimensional discrete data, assign 1 to the discrete points whose distance from the fracture surface is less than dz2 to form a high-resolution fracture attribute body;

[0233] Displace the distorted tectonic velocity model along both sides of the fracture surface of the low-resolution fracture attribute body and the high-resolution fracture attribute body to obtain a low-resolution fracture geological model and a high-resolution fracture geological model.

[0234] Specifically, it includes:

[0235] A training unit for using the low-resolution fracture attribute body as an input label and the high-resolution attribute body as an output label to perform deep network training to form a high-resolution fracture conversion model.

[0236] Figure 13 It is a schematic structural diagram of a formation fracture identification device according to an exemplary embodiment. As Figure 13As shown, corresponding to the above-provided formation fracture identification method, the present invention also provides a formation fracture identification device. Since the embodiments of this device are similar to the above method embodiments, the description is relatively simple. For the relevant parts, please refer to the description in the above method embodiment section. The device described below is only illustrative. The device may include: a processor 1, a memory 2, a communication bus (i.e., the above device bus), and a search engine. Among them, the processor 1 and the memory 2 complete mutual communication through the communication bus and communicate with the outside through a communication interface. The processor 1 can call the logical instructions in the memory 2 to execute the formation fracture identification method.

[0237] In addition, when the logical instructions in the above-mentioned memory 2 are implemented in the form of a software functional unit and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in the various embodiments of the present invention. The aforementioned storage medium includes: storage chips, USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs, etc., which can store program codes.

[0238] On the other hand, the embodiments of the present invention also provide a processor-readable storage medium. A computer program 3 is stored on the processor-readable storage medium. When the computer program 3 is executed by the processor 1, it is used to execute the formation fracture identification methods provided in the above-mentioned various embodiments.

[0239] The processor-readable storage medium can be any available medium or data storage device accessible by the processor 1, including but not limited to magnetic memories (such as floppy disks, hard disks, magnetic tapes, magneto-optical discs (MO), etc.), optical memories (such as CDs, DVDs, BDs, HVDs, etc.), and semiconductor memories (such as ROM, EPROM, EEPROM, non-volatile memories (NANDFLASH), solid-state drives (SSD)), etc.

[0240] The above is only the preferred implementation manner of the present disclosure. The protection scope of the present disclosure is not limited to the above embodiments. All technical solutions falling within the idea of the present disclosure belong to the protection scope of the present disclosure. It should be noted that for those of ordinary skill in the art in this technical field, several improvements and refinements made without departing from the principle of the present disclosure should be regarded as within the protection scope of the present disclosure.

Claims

1. A method for identifying formation fractures, characterized in that: include: Formation of a twisted structural velocity model; The position of the fracture center point is randomly given to generate fractures of arbitrary occurrence, and the fracture cross-section slices are set to be thin or thick, and combined with the twisted structure velocity model to obtain a fracture geological model; Convolving the random main frequency Ricker wavelet with the fault geological model to obtain a fault synthetic earthquake model, which is set as sample data; Performing deep learning on the sample data to obtain a nonlinear correspondence between fault and seismic data, and establishing a deep learning training model; The actual earthquake data is calculated through the deep learning training model to obtain the actual earthquake data fracture identification result.

2. The formation fracture identification method according to claim 1, characterized in that: The forming of the twisted structure velocity model comprises: A two-dimensional horizontal velocity model is established based on the velocity and density curves of well logging data, structural interpretation horizons and seismic layering data; A vertical displacement is performed on the two-dimensional horizontal velocity model to form a twisted structural velocity model.

3. The formation fracture identification method according to claim 2, characterized in that: Performing vertical displacement on the two-dimensional horizontal velocity model to form a twisted structure velocity model includes: Assume the two-dimensional horizontal velocity model is V(X,Z); A vertical displacement is performed on the two-dimensional horizontal velocity model V(X, Z) to generate a twisted structure velocity model V(X ′ ,Z ′ ), the vertical displacement formula is: Where X and Z are the horizontal coordinate and depth coordinate of the two-dimensional horizontal velocity model V(X,Z), respectively. ′ ,Z ′ They are the twisted structure velocity model V(X ′ ,Z ′ ) horizontal coordinate and depth coordinate, Z m (X) is the vertical displacement, M(Z m ) represents the displacement field generated after vertical displacement.

4. The formation fracture identification method according to claim 3, characterized in that: The randomly given fracture center point position is used to generate fractures of arbitrary occurrence, and combined with the twisted structure velocity model, the fracture geological model is obtained, including: In the twisted structure velocity model, the fracture center position (x0, y0, z0), the ellipse major axis l x and the minor axis l of the ellipse y , generate a horizontal elliptical fracture surface; When the horizontal elliptical fracture surface is generated, a value of 1 is assigned, otherwise a value of 0 is assigned; The horizontal elliptical fracture surface is randomly rotated in azimuth and inclination to obtain a fracture with any occurrence; The twisted structural velocity model is displaced at the fault to obtain a fault geological model containing the fault.

5. The formation fracture identification method according to claim 4, characterized in that: When the horizontal elliptical fracture surface is generated, a value of 1 is assigned, otherwise a value of 0 is assigned, including: Establish the distance formula: In the formula, l x is the major axis of the ellipse, l y is the minor axis of the ellipse, x0 is the horizontal coordinate of the fracture center, y0 is the vertical coordinate of the fracture center, x is the horizontal coordinate of the point to be found, y is the vertical coordinate of the point to be found, and dis(x,y) is the distance between the point to be found (x,y) and the fracture center (x0,y0); Assign values ​​to the horizontal elliptical fracture surface: F(x,y,z_0)=0 dis(x,y)>0; F(x,y,z_0)=1 dis(x,y)≤0; Where F(x,y,z_0) represents the elliptical fracture surface.

6. The formation fracture identification method according to claim 1, characterized in that: include: After the random main frequency Ricker wavelet is convolved with the fault geological model to obtain a fault synthetic seismic model and set as sample data, the method includes: simplifying the fault synthetic seismic model, including: The seismic profile of 128×128 elements is used as the feature image input, two 3×3 convolutional layers and activation layers are added, and the image size before and after convolution is set unchanged in the neural network framework of the fault synthetic seismic model; Perform 2×2 pooling, halve the size of the feature image, and continue to add convolutional layers and ReLU activation layers; This cycle is repeated four times, and the number of feature images is doubled each time; When the feature image size of the fault synthetic earthquake model is 16×16, it is upsampled to 32×32 and merged with the result of the third cycle, and the convolution layer activation layer is continued and upsampled; When the feature image size is restored to 128×128, after passing through the convolution layer and activation layer, 3×3 convolution and activation are performed; Use 1×1 convolution to reduce the dimension and reduce the number of feature images to one 128×128 image.

7. The formation fracture identification method according to claim 1, characterized in that: The randomly given fracture center point position is used to generate fractures of arbitrary occurrence, and the fracture cross-section slices are set to be thin and thick. The fracture geological model is obtained by combining the fracture cross-section slices with the twisted structure velocity model, including: Generate a horizontal plane, and randomly generate inclination and azimuth; Rotate the horizontal plane according to the dip and azimuth to form the preliminary occurrence of the fault surface; Apply a random range of displacements to the fracture surface to produce cross-sectional fluctuations; Generate two three-dimensional 0-value discrete data, set two distance thresholds dz1 and dz2, and dz1>dz2, in which the discrete points whose distance from the fracture surface is less than dz1 in one three-dimensional 0-value discrete data are assigned a value of 1 to form a low-resolution fracture attribute body, and in the other three-dimensional discrete data, the discrete points whose distance from the fracture surface is less than dz2 are assigned a value of 1 to form a high-resolution fracture attribute body; The twisted structural velocity model is displaced along both sides of the fault surface of the low-resolution fault attribute body and the high-resolution fault attribute body to obtain a low-resolution fault geological model and a high-resolution fault geological model.

8. The formation fracture identification method according to claim 7, characterized in that: include: The low-resolution fracture attribute body is used as an input label, and the high-resolution attribute body is used as an output label to perform deep network training to form a high-resolution fracture conversion model.

9. A formation fracture identification system, characterized in that: include: A forming unit for forming a twisted structure velocity model; A combining unit is used to randomly give the position of the fracture center point, generate a fracture of any occurrence, set the fracture section slices to be thin or thick, and combine it with the twisted structure velocity model to obtain a fracture geological model; A convolution unit, used for convolving the random main frequency Ricker wavelet with the fault geological model to obtain a synthetic seismic model of the fault, which is set as sample data; An establishment unit is used to perform deep learning on the sample data, obtain a nonlinear correspondence between fault and seismic data, and establish a deep learning training model; A computing unit is used to calculate actual seismic data through the deep learning training model to obtain actual seismic data fracture identification results.

10. The formation fracture identification system according to claim 9, characterized in that: The forming unit comprises: Establish a module for building a two-dimensional horizontal velocity model based on the velocity and density curves of well logging data, structural interpretation horizons and seismic layering data; A forming module is used to perform vertical displacement on the two-dimensional horizontal velocity model to form a twisted structural velocity model.

11. The formation fracture identification system according to claim 10, characterized in that: The formation module is used to: Assume the two-dimensional horizontal velocity model is V(X,Z); A vertical displacement is performed on the two-dimensional horizontal velocity model V(X, Z) to generate a twisted structure velocity model V(X ′ ,Z ′ ), the vertical displacement formula is: Where X and Z are the horizontal coordinate and depth coordinate of the two-dimensional horizontal velocity model V(X,Z), respectively. ′ ,Z ′ They are the twisted structure velocity model V(X ′ ,Z ′ ) horizontal coordinate and depth coordinate, Z m (X) is the vertical displacement, M(Z m ) represents the displacement field generated after vertical displacement.

12. The formation fracture identification system according to claim 11, characterized in that: The combining unit is used for: In the twisted structure velocity model, the fracture center position (x0, y0, z0), the ellipse major axis L x and the minor axis of the ellipse l y , generate a horizontal elliptical fracture surface; When the horizontal elliptical fracture surface is generated, a value of 1 is assigned, otherwise a value of 0 is assigned; The horizontal elliptical fracture surface is randomly rotated in azimuth and inclination to obtain a fracture with any occurrence; The twisted structural velocity model is displaced at the fault to obtain a fault geological model containing the fault.

13. The formation fracture identification system according to claim 12, characterized in that: When the horizontal elliptical fracture surface is generated, a value of 1 is assigned, otherwise a value of 0 is assigned, including: Establish the distance formula: In the formula, l x is the major axis of the ellipse, l y is the minor axis of the ellipse, x0 is the horizontal coordinate of the fracture center, y0 is the vertical coordinate of the fracture center, x is the horizontal coordinate of the point to be found, y is the vertical coordinate of the point to be found, and dis(x,y) is the distance between the point to be found (x,y) and the fracture center (x0,y0); Assign values ​​to the horizontal elliptical fracture surface: F(x,y,z_0)=0 dis(x,y)>0; F(x,y,z_0)=1 dis(x,y)≤0; Where F(x,y,z_0) represents the elliptical fracture surface.

14. The formation fracture identification system according to claim 9, characterized in that: include: A simplification unit is used to simplify the synthetic seismic model of the fault, including: The seismic profile of 128×128 elements is used as the feature image input, two 3×3 convolutional layers and activation layers are added, and the image size before and after convolution is set unchanged in the neural network framework of the fault synthetic seismic model; Perform 2×2 pooling, halve the size of the feature image, and continue to add convolutional layers and ReLU activation layers; This cycle is repeated four times, and the number of feature images is doubled each time; When the feature image size of the fault synthetic earthquake model is 16×16, it is upsampled to 32×32 and merged with the result of the third cycle, and the convolution layer activation layer is continued and upsampled; When the feature image size is restored to 128×128, after passing through the convolution layer and activation layer, 3×3 convolution and activation are performed; Use 1×1 convolution to reduce the dimension and reduce the number of feature images to one 128×128 image.

15. The formation fracture identification system according to claim 9, characterized in that: The combining unit is used for: Generate a horizontal plane, and randomly generate inclination and azimuth; Rotate the horizontal plane according to the dip and azimuth to form the preliminary occurrence of the fault surface; Apply a random range of displacements to the fracture surface to produce cross-sectional fluctuations; Generate two three-dimensional 0-value discrete data, set two distance thresholds dz1 and dz2, and dz1>dz2, in which the discrete points whose distance from the fracture surface is less than dz1 in one three-dimensional 0-value discrete data are assigned a value of 1 to form a low-resolution fracture attribute body, and in the other three-dimensional discrete data, the discrete points whose distance from the fracture surface is less than dz2 are assigned a value of 1 to form a high-resolution fracture attribute body; The twisted structural velocity model is displaced along both sides of the fault surface of the low-resolution fault attribute body and the high-resolution fault attribute body to obtain a low-resolution fault geological model and a high-resolution fault geological model.

16. The formation fracture identification system according to claim 15, characterized in that: include: The training unit is used to use the low-resolution fracture attribute body as an input label and the high-resolution attribute body as an output label to perform deep network training to form a high-resolution fracture conversion model.

17. A formation fracture identification device, characterized in that: include: Processor and memory; The memory is used to store a computer program, and the processor calls the computer program stored in the memory to execute the formation fracture identification method according to any one of claims 1 to 8.

18. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor is enabled to execute the formation fracture identification method according to any one of claims 1 to 8.

Images