Fracture prediction method based on well-to-seismic joint inversion and heterogeneous mechanical modeling

Abstract

The invention discloses a fracture prediction method based on well-to-seismic joint inversion and heterogeneous mechanical modeling, and the method comprises the following steps: collecting and preprocessing basic geological data of a target block; building a high-resolution reservoir rock physical model; establishing a three-dimensional geologic model of the heterogeneous mechanical body; carrying out fracture sensitivity prediction and spatial distribution modeling; and crack network three-dimensional visualization and development decision support. According to the method, well-to-seismic joint inversion, heterogeneous geomechanical modeling and artificial intelligence prediction technologies are deeply fused, and a systematic, intelligent and refined crack prediction solution is formed. Compared with a traditional method, the method has remarkable advantages in the aspects of crack space identification precision, multi-field coupling analysis capability and intelligent level, and provides scientific basis and technical support for efficient development of oil and gas reservoirs.

Description

Technical Field

[0001] This invention belongs to the field of reservoir fracture prediction technology in oil and gas exploration and development, and in particular relates to a fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling. Background Technology

[0002] Buried hill metamorphic oil and gas reservoirs are formed through long-term tectonic compression, hydrothermal alteration, and multiple phases of geological modification. Their lithology is dominated by metamorphic rocks such as gneiss and monzogne, exhibiting significant heterogeneity, anisotropy, and a multi-scale fracture network. Fractures serve as the primary reservoir space and seepage channels, and their distribution is controlled by rock mechanical properties, the evolution of the geostress field, and differences in lithological interfaces, exhibiting strong spatial heterogeneity and mechanical sensitivity.

[0003] Traditional fracture prediction methods face the following limitations when dealing with such complex geological bodies: 1. Insufficient accuracy of a single data source: Although well logging data can provide high-resolution lithological parameters, it has a weak ability to characterize the spatial extension of fractures; although seismic data can depict macroscopic structures, it is difficult to identify fracture networks below the meter level.

[0004] 2. Lack of mechanical-geological coupling modeling: Existing numerical simulations are mostly based on the homogeneous assumption, ignoring rock mechanical parameters such as the spatial variability of Young's modulus and Poisson's ratio and the fracture nucleation mechanism under lithological constraints, resulting in significant deviations between the predicted results and the actual measurements.

[0005] 3. Immature intelligent fusion technology: Although machine learning, such as convolutional neural networks and random forests, has been used for seismic attribute analysis and well logging interpretation, a framework for intelligent fracture prediction under mechanical-lithological coupling constraints and cross-scale fusion of well and seismic data has not yet been established.

[0006] In recent years, although some progress has been made in well-seismic joint inversion and heterogeneous mechanical modeling, their collaborative application still faces the following key technical bottlenecks: 1. Difficulty in fusing multi-source heterogeneous data: The scale difference between well logging data (vertical resolution up to centimeters) and seismic data (wide lateral coverage but low resolution) leads to information loss in joint inversion, making it difficult to construct a high-fidelity three-dimensional rock physics model.

[0007] 2. High complexity of heterogeneous mechanical characterization: The directional arrangement of mineral components and the spatial variation of the brittle-ductile transition threshold of metamorphic rocks require the mechanical model to integrate multimodal data such as microscopic CT scans and rock acoustic emission experiments. Existing modeling methods have low computational efficiency.

[0008] 3. The lithology-stress coupling mechanism is unclear: the development of fractures is jointly controlled by lithological interfaces such as gneiss surfaces and the current geostress field (σ1 direction, differential stress value). Traditional models mostly use empirical weight superposition and lack a dynamic coupling mechanism based on fracture criteria (such as the Griffith criterion).

[0009] 4. Insufficient interpretability of intelligent algorithms: Although deep learning models can uncover hidden patterns in data, their "black box" nature makes it difficult to correlate their prediction results with geomechanical mechanisms, which restricts the reliability of engineering applications.

[0010] Currently, mainstream international research focuses on the following areas: 1. Multi-scale data fusion: For example, Schlumberger's Petrel platform couples seismic attribute volumes with discrete fracture networks (DFN), but does not achieve dynamic prediction driven by mechanics.

[0011] 2. Physically Constrained Machine Learning: The Stanford University team proposed embedding constitutive equations into neural network loss functions to improve the physical consistency of the model, but the problem of anisotropy modeling of metamorphic rocks has not yet been solved.

[0012] 3. Digital twin technology: Baker Hughes' FracXpert system corrects geological models through real-time microseismic monitoring, but its predictive capabilities are limited by static geological parameter input.

[0013] To address the aforementioned bottlenecks, there is an urgent need for an intelligent prediction method that can overcome the limitations of traditional methods in adapting to complex metamorphic reservoirs and provide core technical support for the efficient development of unconventional oil and gas reservoirs. Summary of the Invention

[0014] The problem this invention aims to solve is to provide a fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling. This method integrates well-seismic joint inversion, heterogeneous mechanical modeling, and lithology-constrained intelligent algorithms to construct a fracture prediction paradigm driven by a "data-mechanism-intelligence" ternary approach. This breaks through the limitations of traditional methods in adapting to complex metamorphic reservoirs and provides core technical support for the efficient development of unconventional oil and gas reservoirs.

[0015] To solve the above-mentioned technical problems, the technical solution adopted by this invention is: a fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling, comprising the following steps: S1: Basic geological data acquisition and preprocessing for the target block; S2: Construction of high-resolution reservoir rock physics model; S3: Establishment of a three-dimensional geological model of a heterogeneous mechanical body; S4: Crack sensitivity prediction and spatial distribution modeling; S5: 3D visualization and development decision support for fracture networks integrates prediction results, seismic interpretation results, geological modeling data, and stress field simulation results to construct a visualized fracture network model.

[0016] Furthermore, S1 includes the following steps: S11: Comprehensively collect multi-source basic geological data of the target block, including 3D seismic data, well logging data, core analysis data and geostress test data; S12: Integrates information on regional tectonic evolution history, sedimentary background and thermal events, and uses data cleaning, standardization transformation, spatial registration and machine learning to repair missing logging curves to achieve preprocessing of multi-source heterogeneous geological data.

[0017] Furthermore, S2 includes the following steps: S21: By applying well-seismic joint inversion technology and combining the geophysical response relationship between natural gamma, density and acoustic response, we can carry out fine lithology identification and inversion, obtain the spatial distribution of stratigraphic lithology parameters, and determine the lithology classification threshold. S22: Extract elastic parameters from different lithological sections, construct a three-dimensional rock physics model with high resolution and geological constraints, and establish a linear or nonlinear response model based on the relationship between inverted wave impedance and lithology.

[0018] Furthermore, in S3, based on the rock physics model, and integrating geological elements such as tectonic boundaries, fracture systems, stratigraphic unconformities, and undulating landforms, a three-dimensional geomechanical model considering heterogeneity, anisotropy, and multi-scale tectonic responses is established using the finite element method. The parameters of the three-dimensional geomechanical model include: Young's modulus 30-40 GPa, Poisson's ratio 0.25-0.40, model accuracy reaching the m level, and stress boundary conditions based on tectonic interpretation and regional stress field constraints.

[0019] Furthermore, S4 includes the following steps: S41: Introduces multiple artificial intelligence algorithms to perform feature fusion and pattern recognition on seismic attribute data, rock physical parameter data, stress field distribution, fluid pressure and historical crack distribution data, and trains a crack intelligent prediction model. S42: The model outputs the weight contributions of various key sensitive factors to predict the fracture tendency, density, tensile or shear characteristics and spatial orientation under different lithological units; S43: By comparing and verifying the results with numerical crack simulations, and by optimizing the parameters based on the error feedback mechanism, the generalization ability and prediction stability of the model are enhanced.

[0020] Furthermore, the training of the crack intelligent prediction model adopts a CNN-LSTM fusion structure, with parameter settings including: convolutional layer kernel size 3×3, pooling layer stride 2, LSTM unit number 128, learning rate 0.001, training epochs of no less than 300 epochs, and cross-validation to improve the robustness of the model.

[0021] Furthermore, in S5, the visualized fracture network model includes the display of fracture orientation, density, connectivity, and tensile shear properties, supports overlay analysis with horizontal well trajectories and fracturing segment optimization functions, and the model is updated in real time and linked with dynamic development data; for large-scale fractures >30m, significant fracture features are directly extracted based on seismic attributes; for medium-scale fractures 1-30m, stress field simulation combined with structural models and well logging data is used for modeling; for small-scale fractures 0.05-1m, fracture models are established based on the equivalent method of microfracture permeability or porosity.

[0022] Furthermore, the present invention also provides an apparatus for performing the above-described data processing method.

[0023] Furthermore, the present invention also provides an electronic device, including a memory, a processor, and an algorithm stored in the memory and executable on the processor, wherein the processor performs the data processing method described above.

[0024] Furthermore, the present invention also provides a computer-readable storage medium storing a computer algorithm, which, when executed by a processor, implements the above-described data processing method.

[0025] The advantages and positive effects of this invention are: 1. This invention deeply integrates well-seismic joint inversion, heterogeneous geomechanical modeling, and artificial intelligence prediction technology to form a systematic, intelligent, and refined fracture prediction solution. Compared with traditional methods, this method has significant advantages in fracture spatial identification accuracy, multi-field coupling analysis capability, and intelligence level, providing a scientific basis and technical support for the efficient development of oil and gas reservoirs.

[0026] 2. This invention, by combining well-seismic joint inversion technology and heterogeneous mechanical coupling modeling method, can accurately describe and predict the distribution of fracture networks and their effectiveness under complex geological conditions. Compared with traditional single geophysical exploration or geostatistical methods, this technology significantly improves the spatial resolution and accuracy of fracture prediction, especially demonstrating superior predictive ability in multilithic strata.

[0027] 3. This invention combines artificial intelligence algorithms to intelligently analyze and evaluate fracture networks, enabling the identification of key fracture regions and optimization of fracturing design. The lithology-constrained prediction model can provide more scientific decision support for oil and gas development, improve reservoir productivity utilization, and significantly enhance resource development efficiency.

[0028] 4. The accurate fracture effectiveness evaluation and optimized exploitation scheme of this invention can reduce unnecessary fracturing and drilling operations, thereby significantly reducing development costs. Simultaneously, by increasing oil and gas production, extending reservoir life, and optimizing resource utilization efficiency, this invention can bring greater economic benefits to oil and gas field development projects. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall process of an embodiment of the present invention.

[0030] Figure 2 This is a mineral composition diagram of plagioclase gneiss according to an embodiment of the present invention.

[0031] Figure 3 This is a mineral composition diagram of two long gneiss rocks according to an embodiment of the present invention.

[0032] Figure 4 This is a diagram showing the response relationship of logging parameters for different lithologies in a metamorphic reservoir in a certain region, according to an embodiment of the present invention.

[0033] Figure 5 This is a map showing the lithological inversion results of a metamorphic reservoir in a certain region according to an embodiment of the present invention.

[0034] Figure 6 This is a schematic diagram of a heterogeneous mechanical body constrained by lithological inversion of metamorphic reservoir lithology in a certain region, according to an embodiment of the present invention.

[0035] Figure 7 This is a schematic diagram illustrating the determination of the magnitude and direction of the ancient and modern stress fields in a certain area of ​​the Bohai Bay Basin according to an embodiment of the present invention.

[0036] Figure 8 This is a schematic diagram illustrating the current spatial distribution characteristics of the stress field in a certain area of ​​the Bohai Bay Basin, according to an embodiment of the present invention.

[0037] Figure 9 This is a schematic diagram of the spatial distribution characteristics of fracture parameters in a certain area of ​​the Bohai Bay Basin according to an embodiment of the present invention.

[0038] Figure 10 This is a schematic diagram of a three-dimensional crack model visualization in a certain area of ​​the Bohai Bay Basin according to an embodiment of the present invention.

[0039] Figure 11 This is a schematic diagram illustrating the verification of fracture results in a key well section in a certain area of ​​the Bohai Bay Basin according to an embodiment of the present invention. Detailed Implementation

[0040] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] The embodiments of the present invention will be further described below with reference to the accompanying drawings: like Figure 1 As shown, the fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling includes the following steps.

[0042] S1: Basic geological data acquisition and preprocessing for the target block.

[0043] Specifically, a comprehensive collection of multi-source basic geological data for the target block is undertaken, including 3D seismic data such as gathers, velocity models, and reflected wave imaging; well logging data such as sonic transit time, P-wave and S-wave velocities, density, resistivity, and natural gamma rays; core analysis data such as fracture density, pore structure, and mineral composition; and geostress testing data such as imaging logging. Simultaneously, information on regional tectonic evolution history, sedimentary background, and thermal events is integrated. Data cleaning, standardization transformation, spatial registration, and machine learning-driven missing value repair are employed to achieve high-quality preprocessing of multi-source heterogeneous geological data, providing a consistent and accurate data foundation for modeling.

[0044] S2: Construction of high-resolution reservoir rock physics model.

[0045] Specifically, based on the results of the first step, well-seismic joint inversion techniques, such as precondition inversion, statistical inversion, or deep learning inversion methods, are applied. Combining the geophysical response relationship between natural gamma, density, and acoustic response, refined lithology identification and inversion are conducted to obtain the spatial distribution of stratigraphic lithology and other parameters. Furthermore, elastic parameters such as Young's modulus, shear modulus, Poisson's ratio, and brittleness index are extracted from different lithological segments to construct a high-resolution, geologically constrained three-dimensional rock physics model, providing physical property evidence for fracture simulation and stress field analysis.

[0046] S3: Establishment of a three-dimensional geological model of a heterogeneous mechanical body.

[0047] Specifically, based on the rock physics model obtained in the second step, and integrating geological elements such as tectonic boundaries, fracture systems, stratigraphic unconformities, and undulating landforms, a three-dimensional geomechanical model considering heterogeneity, anisotropy, and multi-scale tectonic responses is established using finite element methods, such as three-dimensional mechanical simulation techniques based on geological structure meshes. The model can dynamically express the distribution, evolution, and concentration trends of the regional tectonic stress field, and characterize the influence mechanism of the fracture system on stress transmission paths and fracture mechanical behavior.

[0048] S4: Crack sensitivity prediction and spatial distribution modeling.

[0049] Specifically, various artificial intelligence algorithms are introduced, especially a fusion architecture of convolutional neural networks (CNN) and long short-term memory networks (LSTM), to perform feature fusion and pattern recognition on seismic attribute data, rock physical parameter data, stress field distribution, fluid pressure, and historical fracture distribution data, training an intelligent fracture prediction model. The model can output the weight contributions of various key sensitive factors and predict fracture dip, density, tensile / shear characteristics, and spatial orientation under different lithological units. Simultaneously, the model is validated by comparison with numerical fracture simulation results, and parameter tuning is performed based on an error feedback mechanism to enhance its generalization ability and predictive stability.

[0050] S5: 3D visualization and development decision support for crack networks.

[0051] By integrating prediction results, seismic interpretation findings, geological modeling data, and stress field simulation results, a visualized fracture network model is constructed. This model not only supports dynamic display of fracture distribution in three-dimensional space but also supports overlay analysis with development engineering parameters such as horizontal well trajectories, fracturing section locations, and injection-production well relationships. The system can interface with a dynamic development database to load production data and fracturing response curves in real time, enabling inverse evaluation of development effects, assisting in optimizing reservoir development plans, and improving the intelligence level of production capacity prediction and injection-production adjustment decisions.

[0052] The invention will now be specifically illustrated using a metamorphic rock reservoir in a certain area of ​​the Bohai Bay Basin as an example: S1: Data collection and lithological analysis.

[0053] Before performing fracture prediction, an accurate lithological model must be established. Therefore, the first step is to collect basic data of the study area and conduct a detailed analysis of the reservoir lithology. Specifically, S1 includes the following steps: S11: Data collection is fundamental to research and mainly includes the following categories: Seismic data volume: Acquire complete 3D seismic data, covering seismic reflection information of the entire study area; eliminate noise and improve signal quality through high-resolution processing; and use time-depth conversion method to convert seismic time data into actual depth to improve data matching accuracy.

[0054] Key well logging data: Log logging data from key wells in the study area were collected, including: sonic transit time (DT): reflecting the formation velocity characteristics of rocks and can be used for lithology identification; density (RHOB): used to calculate porosity and indirectly characterize lithology and fluid type; natural gamma (GR): can be used to distinguish rocks containing potassium and clay minerals; resistivity (RT): reflecting the water content and pore structure of rocks; and other key parameters such as porosity.

[0055] Core analysis data: Core description, thin section identification, scanning electron microscopy (SEM) analysis, and X-ray diffraction (XRD) testing were used to finely characterize the mineral composition, structure, and porosity of the rocks. SEM analysis was used to identify microscopic fracture features, and XRD was used for quantitative mineral analysis.

[0056] In-situ stress measurement data: Wellbore imaging technology is used to analyze fracture orientation and morphology; combined with microfracture experiments, pump injection tests and other methods, the stress state of the formation is determined to provide constraints for fracture prediction.

[0057] S12: Perform lithological analysis.

[0058] Through core observation, thin section identification, and well logging data analysis, the main lithologies in the study area were determined to be plagioclase gneiss and granodiorite gneiss, and the influence of their lithological characteristics on fracture development was studied.

[0059] Plagioclase gneiss: plagioclase accounts for more than 60%, potassium feldspar is less than 10%; brittle mineral content is close to 80%, and cracks are easily affected by stress; It has strong toughness and deformation capacity, and may undergo plastic deformation under high stress.

[0060] Gramine gneiss: plagioclase accounts for about 40%, and potassium feldspar exceeds 35%; the content of brittle minerals is as high as 90%, and cracks are easily formed; due to its high brittleness, it is more easily controlled by geostress, forming a well-developed crack network, such as... Figure 2 , Figure 3 As shown.

[0061] Based on the above analysis, the differences in stress response among different lithologies are clarified, providing important constraints for subsequent crack prediction.

[0062] S2: Lithology Inversion and 3D Spatial Modeling To construct a three-dimensional lithological distribution model for the study area, a combined well-seismic inversion technique was employed, integrating seismic and well logging data to achieve spatial prediction of lithology. Specifically, this includes the following steps: S21: Preprocessing seismic data.

[0063] Noise reduction: High-resolution filtering is used to eliminate noise and improve the signal-to-noise ratio of seismic signals; Time-depth conversion: Depth calibration is performed using well logging data to match the seismic data with the actual well depth; Amplitude compensation: Amplitude attenuation caused by absorption effect is eliminated to improve inversion accuracy.

[0064] S22: Well-seismic joint inversion.

[0065] Simulated annealing (SA) algorithm is used for lithology inversion to improve accuracy; lithology classification standards are established based on well logging curves such as GR (natural gamma), RHOB (density), and DT (sonic); seismic attributes are extracted, and volumetric lithology inversion is performed to obtain a lithology spatial distribution model.

[0066] S23: Analysis of lithological distribution patterns.

[0067] The northeastern part of the study area is mainly composed of plagioclase gneiss, while the southwestern part is dominated by granodiorite gneiss. Analysis of the relationship between lithological distribution and the fault system reveals the lithological distribution characteristics under tectonic control, such as... Figure 4 and Figure 5 As shown.

[0068] S3: Modeling of non-homogeneous mechanical parameters.

[0069] Based on the lithological spatial model, and combined with laboratory rock mechanics tests and numerical simulations, a three-dimensional mechanical parameter field is constructed for the study area to quantify the mechanical properties of different lithologies and further constrain the accuracy of fracture prediction. Specifically, S3 includes the following steps: S31: Determination of rock mechanical parameters.

[0070] To accurately characterize the mechanical properties of different lithologies in the study area, a series of rock mechanics tests need to be conducted in the laboratory, including the following: Young's modulus (E): reflects the elastic deformation capacity of rock and is used to assess the degree of deformation of rock strata under stress.

[0071] Poisson's ratio (ν): describes the ratio of lateral deformation to longitudinal deformation in rocks, and has an important influence on the direction of crack development.

[0072] Uniaxial compressive strength (UCS): measures the load-bearing capacity of rock without lateral restraint and is an important indicator of rock strength.

[0073] Tensile strength (T0): reflects the rock's ability to resist tensile stress and is closely related to the initial formation conditions of cracks.

[0074] The experimental methods include: Uniaxial Compression Test (UCS Test): Determines the compressive strength and stress-strain curve of the rock, and obtains Young's modulus and Poisson's ratio.

[0075] Triaxial Compression Test: Tests the mechanical response of rocks under different confining pressures and calculates their yield criterion.

[0076] The Brazilian test is used to measure the tensile strength of rocks, providing constraints for predicting crack opening.

[0077] S32: Optimize mechanical parameters and interpret the model through finite element numerical simulation.

[0078] To compensate for errors caused by the size limitations of laboratory test samples, the finite element method (FEM) was used to simulate and calculate the stress-strain state within the formation. The simulation process is as follows: Establish a rock mechanics parameter input model: Define material properties of different lithologies based on experimental data.

[0079] Loading regional geological stress field: Determine regional stress boundary conditions using current geostress data (wellbore fracture, seismic reflection characteristics, etc.).

[0080] Simulating rock strata deformation and fracture response under different stress conditions: Investigating the fracture modes of rocks under different stress states (tensile fracture, shear fracture). Calculating the stress concentration effect near fracture structures for different lithologies.

[0081] Optimize the mechanical parameter model: By comparing the numerical simulation results with the measured data, adjust the mechanical parameters to improve the model accuracy.

[0082] S33: Three-dimensional mechanical parameter modeling.

[0083] After obtaining accurate rock mechanics parameters, a high-precision three-dimensional mechanical parameter field is constructed based on seismic inversion results and well logging data, including: Density Model: A density distribution model is established using well logging density data and lithology inversion results.

[0084] Poisson's Ratio Model: Calculates Poisson's ratio using well logging data (sonic transit time DT, density RHOB) and performs spatial interpolation.

[0085] Young's Modulus Model: Based on lithological classification, well logging data, and experimental data, a spatial distribution model of Young's modulus is constructed, such as... Figure 6 As shown.

[0086] Furthermore, by integrating structural features (faults, folds) and seismic attribute data (curvature, ant-like structures), the accuracy of stress field modeling is further optimized to ensure the reliability of crack prediction.

[0087] S34: Analysis of the recovery of the geostress field and the development stages of cracks.

[0088] By analyzing the tectonic evolution history of the study area to study stress changes during different geological periods, the formation mechanisms of fractures at different stages are assessed. Combining wellbore fracture analysis and production dynamic data, the magnitude and direction of the current stress field are determined, providing key constraints for fracture prediction. Through structural geological analysis, the formation periods of fractures are delineated, and combined with stress field distribution, dominant fracture development areas are identified, such as... Figure 7 As shown.

[0089] S4: Deep learning-driven intelligent crack prediction.

[0090] Based on a heterogeneous mechanical parameter volume model, numerical simulations of the ancient and modern stress fields were conducted to clarify the distribution characteristics of the three-dimensional stress field in the target layer. Taking a region in the Bohai Bay Basin as an example, under near-east-west compressive stress, the minimum and maximum horizontal principal stresses are significantly controlled by folds and faults, exhibiting differential distributions along both sides of the faults. With increasing depth, the range of high-value areas at fault junctions and fold hinges gradually increases, such as... Figure 8 As shown.

[0091] After completing lithological and mechanical parameter modeling, deep learning technology is introduced. Using multi-source data such as seismic data, well logging data, stress fields, and rock physical parameters as input, an intelligent fracture prediction model is constructed to improve the accuracy of fracture distribution prediction. Specifically, this includes the following steps: S41: Data input and feature extraction.

[0092] Input data: rock physical parameters (Young's modulus, Poisson's ratio, compressive strength); three-dimensional stress field distribution (maximum principal stress, minimum principal stress, horizontal stress difference); seismic attribute data (ant body, curvature, seismic discontinuity); well logging data (GR, DT, RHOB, RT); fluid pressure data (production data, fracturing test).

[0093] Feature extraction methods: Convolutional Neural Networks (CNNs): used to extract spatial features from 2D / 3D seismic data and identify complex geological structures such as faults and fracture zones. Long Short-Term Memory Networks (LSTMs): used to analyze the temporal dynamics of stress fields and optimize time series prediction capabilities.

[0094] S42: Crack prediction model training.

[0095] Deep learning framework: TensorFlow / PyTorch is used for model building. Adaptive regularization, dropout, and iterative optimization are combined to prevent overfitting and improve the model's generalization ability. Backpropagation is used to adjust model weights and improve crack prediction accuracy.

[0096] Training process: Data augmentation is performed to improve the representativeness of the dataset. K-fold cross-validation is used to evaluate the model's generalization ability. Gradient descent optimization algorithm (Adam / SGD) is used for parameter optimization.

[0097] S43: Simulation and Verification of Crack Distribution.

[0098] The prediction model was validated based on numerical simulation results: the stress concentration in the crack development region was calculated using the finite element method (FEM) and compared with the prediction results of the deep learning model.

[0099] The model was validated using reservoir production data: Fluid response of fracture channels was analyzed using historical production data to optimize prediction results. Fracture connectivity was assessed using fracturing stimulation data to improve model adaptability. Case studies revealed that in multi-lithological formations, the deep learning model can effectively distinguish the influence of different rock types on fracture formation, quantify the sensitivity of each parameter, and clarify the three-dimensional spatial distribution characteristics of fracture parameters in the study area, such as… Figure 9 As shown.

[0100] S5: Multi-scale crack discrete modeling and 3D visualization.

[0101] Based on crack prediction, a multi-scale discrete crack modeling method is adopted to realize the three-dimensional visualization of the crack network in the study area and optimize the crack distribution prediction scheme. Specifically, it includes the following steps.

[0102] S51: Discrete modeling of cracks.

[0103] Based on the predicted fracture distribution, multi-scale fracture models were constructed using different methods: Large-scale fractures (>30m): Significant fracture features were directly extracted based on seismic attributes (ant-like structures, curvature, etc.). Medium-scale fractures (1–30m): Modeling was performed using stress field simulation combined with structural models and well logging data. Small-scale fractures (0.05–1m): Fracture models were established based on the microfracture permeability / porosity equivalence method.

[0104] S52: 3D visualization modeling.

[0105] An interactive 3D fracture visualization model is constructed by combining seismic data, well logging data, stress field distribution, and fracture prediction results. It supports real-time adjustment of fracture parameters, improving the model's applicability and interpretability.

[0106] S53: Crack density distribution analysis.

[0107] By combining wellbore fracture density, fault displacement models, and geostress data, a fracture density prediction model was established. The fracture dip angle and strike were quantified, clarifying the three-dimensional distribution characteristics of fractures in the study area, providing guidance for oil and gas exploration and fracturing development.

[0108] This study integrates seismic data, well logging data, stress field distribution, and fracture prediction results to construct a three-dimensional visualization model of geological structure, stress field, and fracture network. This model supports interactive analysis, providing real-time optimization suggestions for oil and gas reservoir development scheme design, and further validates the model's predictive effectiveness by combining production dynamic data (such as production decline curves and fracturing effect evaluation). Under the guidance of geomechanical theory, the distribution law of fracture parameters is quantitatively predicted to construct a fracture development probability model. Combined with traditional constraint factors such as ant bodies, fractures, and curvature, a multi-scale fracture discrete model is constructed. Large-scale features are relatively obvious and can be directly extracted using seismic attributes such as ant bodies; for small and medium-scale fractures, seismic data is difficult to distinguish, so a phased model is constructed based on tectonic activity under geomechanical theory. A large-scale fracture model is constructed using large-scale fractures identified by seismic attributes as constraints. Small and medium-scale fracture models are constructed using stress field simulation results, tectonic models, well logging data, core samples, and CT data as constraints. Multi-scale fractures are integrated to establish multi-scale fracture discrete models under different constraints, such as… Figure 10 As shown, a fracture density distribution model was established based on geostress model data, fault displacement model, and surface fracture density, and fracture models at different scales were also established. High-angle fractures predominated; mesoscale fractures ranged from 1-30m, with 5m lengths being the most common; and small-scale fractures (0.05m-1m) were distributed relatively evenly. In the construction of the multi-scale fracture discretization model under different constraints, permeability and porosity of microfractures were represented in the grid. The model's predictive performance was further validated by combining production dynamic data from key well sections, such as... Figure 11 As shown.

[0109] The advantages and positive effects of this invention are: 1. This invention deeply integrates well-seismic joint inversion, heterogeneous geomechanical modeling, and artificial intelligence prediction technology to form a systematic, intelligent, and refined fracture prediction solution. Compared with traditional methods, this method has significant advantages in fracture spatial identification accuracy, multi-field coupling analysis capability, and intelligence level, providing a scientific basis and technical support for the efficient development of oil and gas reservoirs.

[0110] 2. This invention, by combining well-seismic joint inversion technology and heterogeneous mechanical coupling modeling method, can accurately describe and predict the distribution of fracture networks and their effectiveness under complex geological conditions. Compared with traditional single geophysical exploration or geostatistical methods, this technology significantly improves the spatial resolution and accuracy of fracture prediction, especially demonstrating superior predictive ability in multilithic strata.

[0111] 3. This invention combines artificial intelligence algorithms to intelligently analyze and evaluate fracture networks, enabling the identification of key fracture regions and optimization of fracturing design. The lithology-constrained prediction model can provide more scientific decision support for oil and gas development, improve reservoir productivity utilization, and significantly enhance resource development efficiency.

[0112] 4. The accurate fracture effectiveness evaluation and optimized exploitation scheme of this invention can reduce unnecessary fracturing and drilling operations, thereby significantly reducing development costs. Simultaneously, by increasing oil and gas production, extending reservoir life, and optimizing resource utilization efficiency, this invention can bring greater economic benefits to oil and gas field development projects.

[0113] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling, characterized in that: Includes the following steps, S1: Basic geological data acquisition and preprocessing for the target block; S2: Construction of high-resolution reservoir rock physics model; S3: Establishment of a three-dimensional geological model of a heterogeneous mechanical body; S4: Crack sensitivity prediction and spatial distribution modeling; S5: 3D visualization and development decision support for fracture networks integrates prediction results, seismic interpretation results, geological modeling data, and stress field simulation results to construct a visualized fracture network model.

2. The fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling according to claim 1, characterized in that: S1 includes the following steps: S11: Comprehensively collect multi-source basic geological data of the target block, including 3D seismic data, well logging data, core analysis data and geostress test data; S12: Integrates information on regional tectonic evolution history, sedimentary background and thermal events, and uses data cleaning, standardization transformation, spatial registration and machine learning to repair missing logging curves to achieve preprocessing of multi-source heterogeneous geological data.

3. The fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling according to claim 1 or 2, characterized in that: S2 includes the following steps: S21: By applying well-seismic joint inversion technology and combining the geophysical response relationship between natural gamma, density and acoustic response, we can carry out fine lithology identification and inversion, obtain the spatial distribution of stratigraphic lithology parameters, and determine the lithology classification threshold. S22: Extract elastic parameters from different lithological sections, construct a three-dimensional rock physics model with high resolution and geological constraints, and establish a linear or nonlinear response model based on the relationship between inverted wave impedance and lithology.

4. The fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling according to claim 3, characterized in that: In S3, based on the rock physics model, and integrating geological elements such as tectonic boundaries, fracture systems, stratigraphic unconformities, and undulating landforms, a three-dimensional geomechanical model considering heterogeneity, anisotropy, and multi-scale tectonic responses is established using the finite element method. The parameters of the three-dimensional geomechanical model include: Young's modulus 30-40 GPa, Poisson's ratio 0.25-0.40, model accuracy reaching the m level, and stress boundary conditions based on tectonic interpretation and regional stress field constraints.

5. The fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling according to claim 1 or 2, characterized in that: S4 includes the following steps: S41: Introduces multiple artificial intelligence algorithms to perform feature fusion and pattern recognition on seismic attribute data, rock physical parameter data, stress field distribution, fluid pressure and historical crack distribution data, and trains a crack intelligent prediction model. S42: The model outputs the weight contributions of various key sensitive factors to predict the fracture tendency, density, tensile or shear characteristics and spatial orientation under different lithological units; S43: By comparing and verifying the results with numerical crack simulations, and by optimizing the parameters based on the error feedback mechanism, the generalization ability and prediction stability of the model are enhanced.

6. The fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling according to claim 5, characterized in that: The crack intelligent prediction model is trained using a CNN-LSTM fusion structure, with the following parameter settings: convolutional layer kernel size 3×3, pooling layer stride 2, LSTM unit number 128, learning rate 0.001, training epochs no less than 300, and cross-validation is used to improve the robustness of the model.

7. The fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling according to claim 1 or 2, characterized in that: In S5, the visualized fracture network model includes the display of fracture orientation, density, connectivity, and tensile shear properties, supports overlay analysis with horizontal well trajectories and fracturing segment optimization, and the model is updated in real time and linked with dynamic development data; for large-scale fractures >30m, significant fracture features are directly extracted based on seismic attributes. For mesoscale fractures ranging from 1 to 30 meters, stress field simulation combined with structural models and well logging data was used for modeling. For small-scale cracks ranging from 0.05 to 1 m, crack models are established based on the equivalent method of microcrack permeability or porosity.

8. A fracture prediction method based on well-seismic joint inversion and heterogeneous mechanical modeling, characterized in that: The data processing method described in any one of claims 1 to 7 is executed.

9. An electronic device comprising a memory, a processor, and an algorithm stored in the memory and executable on the processor, characterized in that: The processor performs the data processing method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer algorithm, characterized in that, When the computer algorithm is executed by the processor, it implements the data processing method as described in any one of claims 1 to 7.

Images