Reservoir fracture parameter quantitative inversion method, system and medium

Abstract

The application discloses a reservoir fracture parameter quantitative inversion method, system and medium, relates to the field of seismic exploration technology, and constructs a nonlinear relationship between a shale reservoir rock physical model and fracture parameters and seismic azimuth difference data; based on the shale reservoir rock physical model, the nonlinear relationship between the fracture parameters and the seismic azimuth difference data is taken as a constraint condition, an inversion data set is synthesized, a physical model driven inversion framework is realized, and a core path for breaking through the precision bottleneck of fracture quantitative inversion is formed; the framework generates labeled samples containing real physical responses through rock physical modeling, constrains a network learning process, and realizes multi-parameter collaborative optimization in combination with seismic responses, so that a solution idea for breaking through the bottleneck of fracture quantitative inversion technology is provided.

Description

Technical Field

[0001] This invention relates to the field of seismic exploration technology, specifically to a quantitative inversion method, system, and medium for reservoir fracture parameters. Background Technology

[0002] Quantitative fracture inversion is a core technology in shale oil and gas exploration and development, and its accuracy directly restricts the effectiveness of reservoir evaluation and engineering development. Traditional fracture prediction methods mainly rely on the characteristics of seismic wave amplitude variation with offset and azimuth (AVAZ), and establish reflection coefficient equations by assuming equivalent medium models (such as VTI, OA, and monoclinic models) to invert indirect indicators such as fracture weakness and anisotropy parameters. For parameters that directly describe fracture characteristics, such as fracture density and dip angle, traditional methods lack direct quantitative means: fracture density needs to be indirectly converted from tangential weakness, and the fracture dip angle has a large cumulative error due to the strong nonlinear relationship between the reflection coefficient and the fracture dip angle. Furthermore, multi-parameter collaborative inversion is significantly limited by model coupling.

[0003] With the development of deep learning technology, data-driven methods have provided a new path for fracture parameter inversion. However, the "black box" nature of the models and the lack of geophysical constraints are prominent issues. Early studies using networks such as CNN and U-Net can predict single parameters such as fracture probability and orientation, but there is limited research on quantitative fracture parameter prediction. At the same time, it is difficult to effectively capture the intrinsic physical relationship between "seismic response and fracture parameters", and the generalization ability is insufficient in complex reservoirs such as thin interbedded layers and heterogeneous reservoirs. Summary of the Invention

[0004] The technical problem this invention aims to solve is that while traditional fracture parameter inversion techniques can predict single parameters such as fracture probability and azimuth, there is limited research on quantitative fracture parameter prediction. Furthermore, it struggles to effectively capture the intrinsic physical correlation between seismic response and fracture parameters, and its generalization ability is insufficient in complex reservoirs such as thin interbedded layers and heterogeneous structures. This invention aims to provide a method, system, and medium for quantitative inversion of reservoir fracture parameters. This scheme improves upon traditional fracture parameter inversion techniques by constructing a shale reservoir rock physics model and establishing a nonlinear relationship between fracture parameters and seismic azimuth difference data. Based on the shale reservoir rock physics model, and using the nonlinear relationship between fracture parameters and seismic azimuth difference data as constraints, an inversion dataset is synthesized, realizing a rock physics model-driven inversion framework. This becomes the core path to overcome the bottleneck of quantitative fracture inversion accuracy. This framework generates labeled samples containing real physical responses through rock physics modeling, constrains the network learning process, and combines seismic response to achieve multi-parameter collaborative optimization, providing a solution for earthquake services and overcoming the bottleneck of quantitative fracture inversion technology.

[0005] This invention is achieved through the following technical solution:

[0006] This scheme provides a quantitative inversion method for reservoir fracture parameters, including:

[0007] Collect basic data on shale reservoirs, consider the bedding structure and fracture development of shale reservoirs, and construct a petrophysical model of shale reservoirs with inclined fractures;

[0008] Based on the rock physics model of shale reservoirs, a nonlinear relationship between fracture parameters and seismic azimuth difference data is constructed;

[0009] Construct a step-by-step inversion network model;

[0010] Based on the rock physics model of shale reservoirs, an inversion dataset is synthesized using the nonlinear relationship between fracture parameters and seismic azimuth difference data as a constraint.

[0011] The step-by-step inversion network model is trained based on the inversion dataset. During the training process, the seismic response constraint and the low-frequency model regularization constraint are used as loss functions, and the effects of the seismic response constraint and the low-frequency model regularization constraint are balanced based on the dynamic weight adjustment strategy.

[0012] Shale reservoir fracture parameter inversion is performed based on a trained step-by-step inversion network model.

[0013] A further optimized approach is that the method for constructing the shale reservoir rock physics model includes:

[0014] Based on the VRH model, a brittle mineral mixture model is constructed, and the equivalent elastic modulus is obtained.

[0015] An organic-mineral mixture model was constructed by adding kerogen to clay based on the SCA model.

[0016] By adding bound water pores to the organic mineral mixture model using the KT model, and then adding bound water using the Gassmann equation, the organic matter pores in the shale reservoir can be simulated.

[0017] Based on the Backus average interpretation model of mixed organic minerals and brittle minerals, the VTI shale matrix with layered characteristics was simulated, and the elastic modulus of the VTI shale matrix was obtained.

[0018] The pore modulus without fluid was set to 0, and an anisotropic DEM model was added to the VTI shale matrix to construct a shale model containing pores.

[0019] Add inclined fractures to a porous shale model using a linear sliding model;

[0020] A fluid mixture was constructed by using oil-water and gas-water mixtures. Then, the pores in the porous shale model were replaced with the fluid mixture using the BK model to construct a fluid-saturated shale reservoir rock physics model.

[0021] A further optimization scheme involves constructing a nonlinear relationship between crack parameters and seismic azimuth difference data; including the following methods:

[0022] When the porous shale model contains inclined fractures, the medium is equivalent to a monoclinic medium to obtain the reflection coefficient equation that includes the influence of VTI shale matrix and fracture.

[0023] The reflection coefficient amplitude difference is obtained by eliminating the influence of the background medium based on reflection coefficient amplitude data from two different orientations, and then compared with the wavelet. Convolution yields the target's reflectance coefficient;

[0024] The seismic azimuth amplitude difference is expressed based on the target reflection coefficient, representing the difference between N reflecting interfaces and M incident angles.

[0025] A further optimized scheme is as follows: the seismic azimuth amplitude difference between the N reflecting interfaces and the M incident angles is expressed as:

[0026] ;

[0027] Where M represents the number of incident angles; N represents the number of reflection coefficients; and g represents the square of the ratio of transverse to longitudinal wave velocities. Indicates the increment of crack parameters; Indicates a wavelet; and These represent the forward modeling coefficients of the azimuth difference data corresponding to the tangential weakness at the Mth incident angle and the forward modeling coefficients of the azimuth difference data corresponding to the normal weakness; Indicates the increment of crack weakness; Represents the forward operand operator; Indicates the reflection coefficient; Indicates the angle of incidence as Azimuth difference data at time; Indicates the angle of incidence as Azimuth difference data at time.

[0028] A further optimized solution is to construct a step-by-step inversion network model, including the following methods:

[0029] A step-by-step inversion network model cascaded with UNet and ResNet is constructed based on a parameter-characteristic-oriented differentiated modeling strategy. The step-by-step inversion network model cascaded with UNet and ResNet includes: a UNet encoder-decoder structure and a ResNet residual branch. The ResNet residual branch is used to mine the mapping relationship between crack dip angle and seismic azimuth difference data.

[0030] A further optimization scheme is that the method for constructing the inversion dataset includes:

[0031] Configure key parameters of shale reservoir rock physics model, including fracture parameters, surrounding rock parameters and fluid parameters;

[0032] Sample points are generated in a shale reservoir petrophysical model based on the Latin hypercube sampling method.

[0033] The reflection amplitude of P-waves at different incident angles was calculated based on the nonlinear relationship between fracture parameters and seismic azimuth difference data, ultimately forming an inversion dataset covering reservoir rock physical characteristics, geological characteristics, and seismic response characteristics.

[0034] A further optimization scheme is proposed, wherein the loss function includes:

[0035] ;

[0036] in, This indicates the proportion of regularization constraints in the low-frequency model. Indicates the forward operand operator. This represents the inverted crack parameters. This represents earthquake azimuth difference data. This represents the low-frequency model; ||*||2 represents the L2 norm.

[0037] A further optimization scheme is that the dynamic weight adjustment strategy includes: first increasing the weight of seismic response constraints or decreasing the weight of low-frequency model regularization constraints, and then decreasing the weight of seismic response constraints or increasing the weight of low-frequency model regularization constraints.

[0038] This solution also provides a quantitative inversion system for reservoir fracture parameters, used to implement the aforementioned quantitative inversion method for reservoir fracture parameters; the system includes:

[0039] The first construction module is used to collect basic data on shale reservoirs, consider the bedding structure and fracture development of shale reservoirs, and construct a rock physics model of shale reservoirs with inclined fractures.

[0040] The second construction module is used to construct the nonlinear relationship between fracture parameters and seismic azimuth difference data based on the rock physics model of shale reservoirs;

[0041] The third building module is used to build the step-by-step inversion network model;

[0042] The fourth module is used to synthesize an inversion dataset based on a shale reservoir rock physics model, using the nonlinear relationship between fracture parameters and seismic azimuth difference data as constraints.

[0043] The model training module is used to train the step-by-step inversion network model based on the inversion dataset. During the training process, the seismic response constraint and the low-frequency model regularization constraint are used as loss functions, and the effects of the seismic response constraint and the low-frequency model regularization constraint are balanced based on the dynamic weight adjustment strategy.

[0044] The inversion module is used to invert fracture parameters in shale reservoirs based on a trained step-by-step inversion network model.

[0045] This solution also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, can implement the quantitative inversion method for reservoir fracture parameters as described above.

[0046] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0047] 1. The present invention provides a quantitative inversion method, system, and medium for reservoir fracture parameters. This scheme improves upon traditional fracture parameter inversion techniques by constructing a shale reservoir rock physics model and establishing a nonlinear relationship between fracture parameters and seismic azimuth difference data. Based on the shale reservoir rock physics model, and using the nonlinear relationship between fracture parameters and seismic azimuth difference data as constraints, an inversion dataset is synthesized, realizing a full-chain inversion framework driven by the rock physics model. This becomes the core path to overcome the bottleneck of quantitative fracture inversion accuracy. This framework generates labeled samples containing real physical responses through rock physics modeling, constrains the learning process of the network, and combines seismic response to achieve multi-parameter collaborative optimization dynamic modeling, providing a solution to overcome the bottleneck of quantitative fracture inversion technology.

[0048] 2. The present invention provides a quantitative inversion method, system and medium for reservoir fracture parameters; it constructs a step-by-step inversion network model, and through a cascaded network architecture, realizes differentiated modeling of fracture density and dip angle, solving the problem of multi-parameter coupling inversion in traditional methods;

[0049] 3. The quantitative inversion method, system and medium for reservoir fracture parameters provided by this invention combine the constraints of the shale reservoir rock physics model with the nonlinear fitting of deep learning (step-by-step inversion network model) to achieve model fusion, break through the limitations of conventional inversion, and achieve high-precision direct quantitative inversion of fracture density and dip angle.

[0050] 4. The quantitative inversion method, system and medium for reservoir fracture parameters provided by this invention are based on a shale reservoir rock physics model. The nonlinear relationship between fracture parameters and seismic azimuth difference data is used as a constraint to synthesize an inversion dataset. By combining pre-training of the synthesized inversion dataset with transfer learning from actual data, the robustness of the model to heterogeneous reservoirs and low signal-to-noise ratio data is improved, providing reliable technical support for shale oil and gas exploration.

[0051] 5. The quantitative inversion method, system, and medium for reservoir fracture parameters provided by this invention combine rock physics theory, seismic wave propagation laws, and deep learning algorithms to achieve high-precision quantitative inversion of fracture density and dip angle in shale reservoirs. Compared to traditional methods, it leverages the advantages of deep learning in handling nonlinear problems, overcoming the limitations of multi-parameter coupling in traditional equivalent medium inversion; and compared to purely data-driven deep learning methods, by introducing rock physics modeling and seismic response constraints, it significantly improves the stability of the model under complex geological conditions. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0053] Figure 1 A schematic diagram of the process for quantitative inversion of reservoir fracture parameters;

[0054] Figure 2 A schematic diagram illustrating the principle of constructing a rock physical model for shale reservoirs;

[0055] Figure 3 Schematic diagram of the system structure for quantitative inversion of reservoir fracture parameters;

[0056] Figure 4 This is a schematic diagram of the inverted dataset;

[0057] Figure 5 A schematic diagram of the test dataset consisting of seismic azimuth difference data;

[0058] Figure 6 This is a schematic diagram of the inversion results for the test set;

[0059] Figure 7 This is a schematic diagram of the inversion error;

[0060] Figure 8 This is a schematic diagram of the loss curve during pre-training;

[0061] Figure 9 This is a schematic diagram illustrating the actual application effect. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are only for explaining this invention and are not intended to limit this invention.

[0063] While traditional fracture parameter inversion techniques can predict single parameters such as fracture probability and orientation, there is limited research on quantitative fracture parameter prediction. Furthermore, they struggle to effectively capture the intrinsic physical correlation between seismic response and fracture parameters, and their generalization ability is insufficient in complex reservoirs such as thin interbedded layers and heterogeneous formations. Therefore, this solution provides the following embodiments to address the aforementioned technical problems:

[0064] Example 1

[0065] This embodiment provides a quantitative inversion method for reservoir fracture parameters, such as... Figure 1 As shown, it includes:

[0066] Step 1: Collect basic data on shale reservoirs and preprocess the basic data; based on the preprocessed basic data, consider the bedding structure and fracture development of shale reservoirs, and construct a petrophysical model of shale reservoirs with inclined fractures.

[0067] The specific basic data for shale reservoirs include typical shale reservoir stratigraphic data, velocity models, pre-stack wide-azimuth seismic data, well logging data (FMI imaging, mineral composition, porosity, hydrocarbon content, elastic parameters), core data, and prior geological information. The preprocessing process includes denoising the seismic data, spherical diffusion compensation, and surface consistency processing, and extracting amplitude difference data at different incident angles (such as 15°, 25°, and 35°) to prepare for subsequent fracture inversion.

[0068] Shale rocks have complex microstructures, and their elastic properties are influenced not only by their mineral composition and porosity but also by factors such as pore connectivity, grain morphology, and fluid adhesion. Based on collected basic data on shale reservoirs and considering the bedding structure and fracture development, a petrophysical model suitable for shale reservoirs with inclined fractures is constructed:

[0069] A. Based on the VRH model, a brittle mineral mixture model is constructed by mixing brittle minerals (such as quartz, feldspar, calcite, and pyrite), and the equivalent elastic modulus is obtained. Since the elastic moduli of brittle minerals are similar, they are classified into one category and the equivalent elastic modulus is obtained.

[0070] B. Based on the SCA model, kerogen is added to clay to construct an organic-mineral mixture model; the organic-mineral mixture model is suitable for cases with high inclusion content.

[0071] C. By adding bound water pores to the organic mineral mixture model using the KT model, and then adding bound water using the Gassmann equation, the organic matter pores in the shale reservoir can be simulated.

[0072] D. Based on the Backus model of mixing organic minerals and brittle minerals in actual proportions, the VTI shale matrix with layered characteristics was simulated, and the elastic modulus of the VTI shale matrix was obtained.

[0073] E, set the pore modulus without fluid to 0, and add an anisotropic DEM model to the VTI shale matrix to construct a shale model with pores;

[0074] F, adding inclined fractures to a porous shale model using a linear sliding model;

[0075] G is a fluid mixture constructed from an oil-water and gas-water mixture (the equivalent elastic modulus of the oil-water and gas-water mixture can be calculated using the Wood formula). Then, the BK model is used to replace the pores in the porous shale model with the fluid mixture to construct a fluid-saturated shale reservoir rock physics model.

[0076] Step 2: Based on the rock physics model of the shale reservoir, construct a nonlinear relationship between fracture parameters and seismic azimuth difference data; this includes the following methods:

[0077] S21, When the porous shale model contains inclined fractures, the medium is equivalent to a monoclinic medium to obtain the reflection coefficient equation that includes the influence of VTI shale matrix and fracture.

[0078] ;

[0079] in, Indicating the influence of VTI shale matrix:

[0080] ;

[0081] Indicating the impact of cracks:

[0082] ;

[0083] in,

[0084] ;

[0085] ;

[0086] in, and These represent the transverse wave modulus and density of the isotropic medium, respectively. Indicates the angle of incidence of seismic waves. Indicates the crack dip angle. Indicates the actual location of the collected seismic data. Indicates the normal orientation of the crack. ; , These represent the normal and tangential crack weakness parameters, respectively; This represents the forward modeling coefficients corresponding to tangential weakness; The forward modeling coefficients corresponding to the phase weakness are represented; This represents the increment of the transverse wave modulus; Indicates the increment of density; This represents the average value of the longitudinal wave modulus of the upper and lower layers; This represents the increment of the longitudinal wave modulus; This represents the average value of the transverse wave modulus of the upper and lower layers. This represents the average density of the upper and lower layers; and These represent the increments of the two anisotropic parameters; g represents the square of the ratio of transverse to longitudinal wave velocities.

[0087] S22, the reflection coefficient amplitude difference is obtained by eliminating the influence of the background medium based on the reflection coefficient amplitude data from two different orientations, and the reflection coefficient amplitude difference is then compared with the wavelet. Convolution yields the target's reflectance coefficient;

[0088] The difference in reflection coefficient amplitude is obtained by eliminating the influence of the background medium through reflection coefficient amplitude data from two different orientations:

[0089]

[0090] in, This represents the difference in reflection coefficient amplitude; This represents the reflection coefficient when the azimuth angle is 0°. This represents the reflection coefficient at an azimuth angle of 90°. This represents the forward modeling coefficients of the azimuth difference data corresponding to the tangential weakness; Indicates the increment of tangential weakness; The forward modeling coefficients of the azimuth difference data corresponding to the normal weakness are represented. Indicates the increment of normal weakness;

[0091] The amplitude difference of the reflection coefficient and the wavelet Convolution has:

[0092] ;

[0093] in, This represents the difference in earthquake amplitude at azimuth 0° and 90°. This represents seismic data with an azimuth of 0°. This represents earthquake data with an azimuth angle of 90°. This represents a sub-wave.

[0094] When the fracture is saturated with fluid, the fracture density e can be obtained through... Indicate:

[0095] ;

[0096] The target reflectance coefficient is:

[0097]

[0098] S23, based on the target reflection coefficient, the seismic azimuth amplitude differences for N reflecting interfaces and M incident angles are:

[0099] ;

[0100] Where M represents the number of incident angles; N represents the number of reflection coefficients; and g represents the square of the ratio of transverse to longitudinal wave velocities. Indicates the increment of crack parameters; Indicates a wavelet; and These represent the forward modeling coefficients of the azimuth difference data corresponding to the tangential weakness at the Mth incident angle and the forward modeling coefficients of the azimuth difference data corresponding to the normal weakness; Indicates the increment of crack weakness; Represents the forward operand operator; Indicates the reflection coefficient; Indicates the angle of incidence as Azimuth difference data at time; Indicates the angle of incidence as Azimuth difference data at time.

[0101] Step 3: Constructing a step-by-step inversion network model; To overcome the technical bottleneck of quantitative inversion of fracture parameters in shale oil and gas exploration, this solution provides a step-by-step inversion network architecture cascaded with UNet and ResNet, employing a parameter-characteristic-oriented differentiated modeling strategy to achieve high-precision prediction. Methods include:

[0102] A step-by-step inversion network model cascaded with UNet and ResNet is constructed based on a parameter-characteristic-oriented differentiated modeling strategy. The step-by-step inversion network model cascaded with UNet and ResNet includes: a UNet encoder-decoder structure and a ResNet residual branch. The ResNet residual branch is used to mine the mapping relationship between crack dip angle and seismic azimuth difference data.

[0103] For fracture density and normal weakness parameters, which are significantly constrained by geological background and exhibit continuous spatial distribution, UNet's classic encoder-decoder structure is used for modeling. In the encoder, a four-level downsampling module progressively expands the receptive field, effectively extracting regional geological features while suppressing data noise. The decoder, through deconvolution operations and cross-layer connections, fuses multi-scale features, accurately reconstructs the spatial details of the parameters, and ultimately outputs stable and reliable predicted results for fracture density and normal weakness.

[0104] To address the issues of strong nonlinearity and poor stability in fracture dip inversion, a ResNet residual branch is constructed to achieve refined estimation. The ResNet residual branch integrates the fracture density and normal weakness parameters output from the UNet encoder-decoder structure with seismic azimuth data into a five-channel feature as input. Leveraging the unique cross-layer connection mechanism of the ResNet residual branch, the complex mapping relationship between fracture dip and seismic azimuth data is deeply explored. This design, which integrates relatively stable parameter information with raw seismic observation data, provides a new information dimension for solving the challenge of fracture dip inversion.

[0105] Step 4: Based on the shale reservoir rock physics model, and using the nonlinear relationship between fracture parameters and seismic azimuth difference data as constraints, synthesize the inversion dataset; specific methods include:

[0106] S41, Configure the key parameters of the shale reservoir rock physics model, the key parameters of the shale reservoir include: fracture parameters, surrounding rock parameters and fluid parameters;

[0107] S42, generating sample points in a shale reservoir petrophysical model based on the Latin hypercube sampling method;

[0108] S43 calculates the reflection amplitude of P-waves at different incident angles based on the nonlinear relationship between fracture parameters and seismic azimuth difference data, ultimately forming an inversion dataset covering reservoir rock physical characteristics, geological characteristics, and seismic response characteristics.

[0109] This scheme, based on a shale reservoir petrophysical model, can generate training samples with physical constraints, providing high-quality data for subsequent step-by-step inversion of the network model. First, key parameters of the shale reservoir are configured based on the collected data. The Latin hypercube sampling (LHS) method is used to generate sample points in a multidimensional parameter space. A stratified randomization strategy ensures that the samples cover different geological scenarios; for example, specific dip angle combinations are preferentially sampled for areas with high fracture density to simulate fracture development patterns in real reservoirs.

[0110] The reflection amplitude of P-waves at different incident angles was calculated based on the nonlinear relationship between fracture parameters and seismic azimuth data. A multi-label dataset encompassing reservoir rock physical characteristics, geological characteristics, and seismic response parameters was ultimately formed, divided into training and test sets in an 8:2 ratio. These samples retain the theoretical constraints of the rock physics model while simulating the complex characteristics of actual seismic data, providing a physically interpretable training foundation for the step-by-step inversion network model and facilitating accurate fracture parameter inversion.

[0111] Step 5: Train the step-by-step inversion network model based on the inversion dataset. During pre-training, the L1 norm is used. Fine-tuning uses seismic response constraints and low-frequency model regularization constraints as loss functions, and a dynamic weight adjustment strategy is used to balance the effects of these constraints. Specific loss functions include: ;

[0112] in, This indicates the proportion of regularization constraints in the low-frequency model. Indicates the forward operand operator. This represents the inverted crack parameters. This represents earthquake azimuth difference data. This represents the low-frequency model; ||*||2 represents the L2 norm.

[0113] After training the step-by-step inversion network model based on theoretical data, parameter fine-tuning using actual seismic data is necessary to ensure the model accurately adapts to real geological scenarios. However, actual data often suffers from limited sample size and low signal-to-noise ratio, making it difficult to directly use for model training. To address these challenges, this solution uses seismic response constraints and low-frequency model regularization constraints as loss functions, effectively improving the model's inversion performance in real-world data environments.

[0114] Dynamic weight adjustment strategies include: first increasing the weight of seismic response constraints or decreasing the weight of low-frequency model regularization constraints, and then decreasing the weight of seismic response constraints or increasing the weight of low-frequency model regularization constraints.

[0115] In the process of fine-tuning the parameters of the step-by-step inversion network model, a dynamic weight adjustment strategy is adopted to balance the effects of the two constraint mechanisms. In the initial stage of fine-tuning, the weight of the seismic response constraint is increased, i.e., the weight of the seismic response constraint is decreased. The value of the value prompts the step-by-step inversion network model to converge quickly to a solution space that conforms to physical laws. As training progresses, the proportion of regularization constraints on low-frequency models is gradually increased, further optimizing the spatial smoothness and geological rationality of parameter distribution. Through this phased optimization strategy, the model can effectively adapt to complex reservoir conditions (such as heterogeneous shale and low signal-to-noise ratio areas), significantly improving the accuracy and reliability of fracture parameter inversion, and providing strong data support for shale reservoir evaluation and development.

[0116] To evaluate the ability of the step-by-step inversion network model to invert crack parameters, this embodiment also includes a theoretical data validation experiment: azimuth superposition data at different incident angles (15°, 25°, 35°) in the training set are cut into 256×256 segments and used as input samples into the step-by-step inversion network model; the L1 norm loss function is selected to reduce the impact of outliers on training. The training process is iterated 200 times, and the Adam optimizer is used to dynamically adjust the learning rate to ensure stable convergence of the model. Next, the trained step-by-step inversion network model is fine-tuned using a validation set to achieve prediction of the data from the step-by-step inversion network model. The validation results show that the step-by-step inversion network model's prediction of crack parameters matches the theoretical values ​​well, providing a reliable basis for subsequent application of practical data.

[0117] Step 6: Perform shale reservoir fracture parameter inversion based on the trained step-by-step inversion network model.

[0118] Example 2

[0119] This embodiment provides a quantitative inversion system for reservoir fracture parameters, characterized in that it is used to implement the quantitative inversion method for reservoir fracture parameters described in Embodiment 1; such as Figure 3 As shown, the system includes:

[0120] The first construction module is used to collect basic data of shale reservoirs and preprocess the basic data; based on the preprocessed basic data, considering the bedding structure and fracture development of shale reservoirs, a rock physics model of shale reservoirs with inclined fractures is constructed.

[0121] The second construction module is used to construct the nonlinear relationship between fracture parameters and seismic azimuth difference data based on the rock physics model of shale reservoirs;

[0122] The third building module is used to build the step-by-step inversion network model;

[0123] The fourth module is used to synthesize an inversion dataset based on a shale reservoir rock physics model, using the nonlinear relationship between fracture parameters and seismic azimuth difference data as constraints.

[0124] The model training module is used to train the step-by-step inversion network model based on the inversion dataset. During the training process, the seismic response constraint and the low-frequency model regularization constraint are used as loss functions, and the effects of the seismic response constraint and the low-frequency model regularization constraint are balanced based on the dynamic weight adjustment strategy.

[0125] The inversion module is used to invert fracture parameters in shale reservoirs based on a trained step-by-step inversion network model.

[0126] Example 3

[0127] This embodiment provides a computer-readable medium having a computer program stored thereon. The computer program, when executed by a processor, can implement the quantitative inversion method for reservoir fracture parameters as described in Embodiment 1. Figure 1 As shown, the specific steps are as follows:

[0128] Step 1: Collect basic data on shale reservoirs and preprocess the basic data;

[0129] Based on the preprocessed basic data, considering the bedding structure and fracture development of shale reservoirs, a petrophysical model of shale reservoirs with inclined fractures is constructed.

[0130] Step 2: Based on the rock physics model of shale reservoirs, construct the nonlinear relationship between fracture parameters and seismic azimuth difference data;

[0131] Step 3: Construct a step-by-step inversion network model;

[0132] Step 4: Based on the rock physics model of shale reservoirs, and using the nonlinear relationship between fracture parameters and seismic azimuth difference data as constraints, synthesize the inversion dataset;

[0133] Step 5: Train the step-by-step inversion network model based on the inversion dataset. During the process, L1 norm is used for pre-training. The fine-tuning process uses seismic response constraints and low-frequency model regularization constraints as loss functions, and balances the effects of seismic response constraints and low-frequency model regularization constraints based on a dynamic weight adjustment strategy.

[0134] Step 6: Perform shale reservoir fracture parameter inversion based on the trained step-by-step inversion network model.

[0135] Example 4

[0136] This embodiment provides a specific example, which constructs... Figure 2The shale reservoir petrophysical model shown fully considers the microstructure of shale oil and gas reservoirs, dividing reservoir pores into intergranular pores and fracture pores, and introducing fracture dip angle when considering fractures, which is more consistent with the actual reservoir conditions; when considering mineral composition, the model introduces the interlocking of kerogen and clay mineral particles to simulate shale bedding; in addition, bound water is introduced through non-connected pores to simulate the adsorption state of fluids in shale reservoirs.

[0137] The test set real model is generated using a shale reservoir petrophysical model, and the inverted dataset is as follows: Figure 4 The figures show the fracture density, fracture weakness, and fracture dip angle, respectively. The fracture density is less than 0.01, the fracture weakness is less than 0.1, and the fracture dip angle is mainly concentrated in the medium to high angles, which is in good agreement with the statistical results of the actual core samples.

[0138] Substituting the generated shale reservoir petrophysical model into the target reflection coefficient formula, and taking the azimuth difference reflection coefficients at incident angles of 15°, 25°, and 35°, and convolving them with a 30 Hz Reich wavelet, we obtain the following: Figure 5 The seismic azimuth difference data shown; through the synthesized seismic data, it can be seen that as the incident angle increases, the amplitude of the seismic azimuth difference gradually increases, proving that the anisotropy is stronger.

[0139] The results obtained by inputting the test set into the step-by-step inversion network model are as follows: Figure 6 As shown, the generated azimuth difference seismic data is fed into the network for training, and the pre-trained network is fine-tuned through seismic response constraints. It can be seen that the prediction results of the test set match the model data well.

[0140] The difference between the predicted value and the actual model is calculated as follows: Figure 7 As shown, the error magnitudes of crack density, crack weakness, and crack inclination angle are 1e-4, 1e-3, and , respectively, with errors less than 10%, which is better than the prediction results of traditional methods.

[0141] Loss curve as shown Figure 8 As shown in the curve, in the initial stage of training (when the number of rounds is small), the loss value decreases rapidly with the increase of rounds. Subsequently, as the number of rounds continues to increase, the rate of decrease of the loss value gradually slows down and tends to stabilize, indicating that the model gradually converges and approaches the optimal state.

[0142] The actual application effect of data is as follows Figure 9As shown in the figure, this diagram presents the fracture parameter inversion profile of well A. The inversion results for fracture weakness and fracture density show good agreement with the measured curves of well A, and the fracture dip angle is consistent with the logging results, verifying the reliability of the method. The inversion results show that both the normal and tangential fracture weakness of the target interval are high, indicating a high degree of fracture development in this interval, dominated by medium-to-high angle fractures.

[0143] This approach enables high-precision processing of seismic data from actual shale oil and gas production areas, achieving quantitative descriptions of key parameters such as fracture density and dip angle. Specifically, the method first utilizes independently developed rock physics modeling technology to generate training samples containing real physical mechanisms for the initial training of a deep learning network. Based on this, a quantitative mapping relationship between fracture parameters and seismic response characteristics is constructed and incorporated as a constraint into the network fine-tuning process, thereby enhancing the model's adaptability to different geological conditions and production area environments.

[0144] In the method validation phase, both theoretical data and actual seismic data were used for testing: theoretical data was used for model training and forward validation to ensure the correctness of the algorithm logic; actual field data was used to further optimize parameters, and the fracture parameters obtained by inversion showed a high degree of agreement with the well logging curves. Figure 9 Taking the fracture parameter inversion profile of well A as an example, the predicted results of fracture weakness and fracture density match well with the well data, and the fracture dip angle is consistent with the logging results, which intuitively demonstrates the reliability and effectiveness of the method under complex geological conditions. This invention provides an efficient and accurate technical solution for the quantitative prediction of fracture parameters in shale reservoirs, and can provide key data support for subsequent oil and gas exploration and development decisions.

[0145] Practical applications show that this method has high accuracy and reliability in fracture density and dip angle inversion. The inversion results are in high agreement with well logging data and geological characteristics, providing more robust technical support for shale reservoir evaluation.

[0146] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A quantitative inversion method for reservoir fracture parameters, characterized in that, include: Collect basic data on shale reservoirs, consider the bedding structure and fracture development of shale reservoirs, and construct a petrophysical model of shale reservoirs with inclined fractures; Based on the rock physics model of shale reservoirs, a nonlinear relationship between fracture parameters and seismic azimuth difference data is constructed; Construct a step-by-step inversion network model; Based on the rock physics model of shale reservoirs, an inversion dataset is synthesized using the nonlinear relationship between fracture parameters and seismic azimuth difference data as a constraint. The step-by-step inversion network model is trained based on the inversion dataset. During the training process, the seismic response constraint and the low-frequency model regularization constraint are used as loss functions, and the effects of the seismic response constraint and the low-frequency model regularization constraint are balanced based on the dynamic weight adjustment strategy. Shale reservoir fracture parameter inversion is performed based on a trained step-by-step inversion network model.

2. The quantitative inversion method for reservoir fracture parameters according to claim 1, characterized in that, The method for constructing the rock physics model of the shale reservoir includes: Based on the VRH model, a brittle mineral mixture model is constructed, and the equivalent elastic modulus is obtained. An organic-mineral mixture model was constructed by adding kerogen to clay based on the SCA model. By adding bound water pores to the organic mineral mixture model using the KT model, and then adding bound water using the Gassmann equation, the organic matter pores in the shale reservoir can be simulated. Based on the Backus average well logging interpretation results, a mixed organic mineral model and a mixed brittle mineral model were used to simulate the VTI shale matrix with layered characteristics, and the elastic modulus of the VTI shale matrix was obtained. The pore modulus without fluid was set to 0, and an anisotropic DEM model was added to the VTI shale matrix to construct a shale model containing pores. Add inclined fractures to a porous shale model using a linear sliding model; A fluid mixture was constructed by using oil-water and gas-water mixtures. Then, the pores in the porous shale model were replaced with the fluid mixture using the BK model to construct a fluid-saturated shale reservoir rock physics model.

3. The quantitative inversion method for reservoir fracture parameters according to claim 2, characterized in that, The nonlinear relationship between the constructed crack parameters and the seismic azimuth difference data; Including methods: When the porous shale model contains inclined fractures, the medium is equivalent to a monoclinic medium to obtain the reflection coefficient equation that includes the influence of VTI shale matrix and fracture. The reflection coefficient amplitude difference is obtained by eliminating the influence of the background medium based on reflection coefficient amplitude data from two different orientations, and then compared with the wavelet. Convolution yields the target's reflectance coefficient; The seismic azimuth amplitude difference is expressed based on the target reflection coefficient, representing the difference between N reflecting interfaces and M incident angles.

4. The quantitative inversion method for reservoir fracture parameters according to claim 3, characterized in that, The seismic azimuth amplitude difference between the N reflecting interfaces and the M incident angles is expressed as: ； Where M represents the number of incident angles; N represents the number of reflection coefficients; and g represents the square of the ratio of transverse to longitudinal wave velocities. Indicates the increment of crack parameters; Indicates a wavelet; and These represent the forward modeling coefficients of the azimuth difference data corresponding to the tangential weakness at the Mth incident angle and the forward modeling coefficients of the azimuth difference data corresponding to the normal weakness; Indicates the increment of crack weakness; Represents the forward operand operator; Indicates the reflection coefficient; Indicates the angle of incidence as Azimuth difference data at time; Indicates the angle of incidence as Azimuth difference data at time.

5. The quantitative inversion method for reservoir fracture parameters according to claim 1, characterized in that, The method for constructing the step-by-step inversion network model includes: A step-by-step inversion network model cascaded with UNet and ResNet is constructed based on a parameter-characteristic-oriented differentiated modeling strategy. The step-by-step inversion network model cascaded with UNet and ResNet includes: a UNet encoder-decoder structure and a ResNet residual branch. The ResNet residual branch is used to mine the mapping relationship between crack dip angle and seismic azimuth difference data.

6. The quantitative inversion method for reservoir fracture parameters according to claim 1, characterized in that, The method for constructing the inversion dataset includes: Configure key parameters of shale reservoir rock physics model, including fracture parameters, surrounding rock parameters and fluid parameters; Sample points are generated in a shale reservoir petrophysical model based on the Latin hypercube sampling method. The reflection amplitude of P-waves at different incident angles was calculated based on the nonlinear relationship between fracture parameters and seismic azimuth difference data, ultimately forming an inversion dataset covering reservoir rock physical characteristics, geological characteristics, and seismic response characteristics.

7. The quantitative inversion method for reservoir fracture parameters according to claim 5, characterized in that, The loss function includes: ； in, This indicates the proportion of regularization constraints in the low-frequency model. Indicates the forward operand operator. This represents the inverted crack parameters. This represents earthquake azimuth difference data. This represents the low-frequency model; ||*||2 represents the L2 norm.

8. The quantitative inversion method for reservoir fracture parameters according to claim 7, characterized in that, The dynamic weight adjustment strategy includes: first increasing the weight of seismic response constraints or decreasing the weight of low-frequency model regularization constraints, and then decreasing the weight of seismic response constraints or increasing the weight of low-frequency model regularization constraints.

9. A quantitative inversion system for reservoir fracture parameters, characterized in that, The system is used to implement the quantitative inversion method for reservoir fracture parameters according to any one of claims 1-8; the system comprises: The first construction module is used to collect basic data on shale reservoirs, consider the bedding structure and fracture development of shale reservoirs, and construct a rock physics model of shale reservoirs with inclined fractures. The second construction module is used to construct the nonlinear relationship between fracture parameters and seismic azimuth difference data based on the rock physics model of shale reservoirs. The third building module is used to build the step-by-step inversion network model; The fourth module is used to synthesize an inversion dataset based on a shale reservoir rock physics model, using the nonlinear relationship between fracture parameters and seismic azimuth difference data as constraints. The model training module is used to train the step-by-step inversion network model based on the inversion dataset. During the training process, the seismic response constraint and the low-frequency model regularization constraint are used as loss functions, and the effects of the seismic response constraint and the low-frequency model regularization constraint are balanced based on the dynamic weight adjustment strategy. The inversion module is used to invert fracture parameters in shale reservoirs based on a trained step-by-step inversion network model.

10. A computer-readable medium having a computer program stored thereon, characterized in that, The computer program, when executed by a processor, can implement the quantitative inversion method for reservoir fracture parameters as described in any one of claims 1-8.

Images