Fracturing agent modeling and optimization method fusing data driving and physical mechanism

Abstract

This invention relates to a fracturing intelligent agent modeling and optimization method integrating data-driven and physical mechanism approaches. It includes: constructing a multi-dimensional fracturing parameter space and simulating fracture initiation and propagation mechanisms to obtain fracture spatial morphology and evolution response data; expanding the dataset of finite mechanism simulation samples based on interpolation and statistical methods; extracting geometric, physical, and topological features from fracture point cloud data, classifying and reducing the dimensionality of fracture morphology, and constructing physically meaningful fracture feature vectors; constructing a data-driven fracture prediction model based on the fracture feature vectors, and introducing fracture mechanics constraints to achieve rapid and stable prediction of fracture initiation and propagation behavior; bidirectionally coupling the data-driven fracture prediction model and the physical mechanism model to form a fracturing intelligent agent with reasoning and correction capabilities, outputting the optimal fracturing scheme that meets engineering constraints. This invention achieves integrated intelligent decision-making for fracture propagation simulation and fracturing parameter optimization.

Description

Technical Field

[0001] This invention relates to the field of petroleum geological resource exploration and exploitation, specifically to a fracturing intelligent agent modeling and optimization method that integrates data-driven and physical mechanism approaches. This method is used for simulation of fracture initiation and propagation and intelligent optimization decision-making of fracturing parameters during the fracturing process of unconventional oil and gas reservoirs. Background Technology

[0002] As global oil and gas development gradually enters the stage of deep, unconventional, and complex reservoirs, unconventional resources such as shale oil, shale gas, and tight oil and gas have become an important part of ensuring energy supply. Hydraulic fracturing, as a core stimulation technology for unconventional oil and gas development, directly determines the reservoir stimulation volume, fracture complexity, and final production level of a single well.

[0003] In practical engineering, the fracturing process involves multiple scales and stages, including perforation, fracture initiation, fracture propagation, and multi-fracture interaction. Its evolution is influenced by a combination of factors, including formation mechanical parameters, geostress state, wellbore structure, perforation parameters, and construction parameters, exhibiting significant nonlinear, high-dimensional, and uncertain characteristics. Therefore, accurately simulating fracture propagation behavior under complex geological conditions and rationally optimizing fracturing parameters based on this simulation is a key engineering problem in current fracturing design and construction.

[0004] Existing fracturing design methods mainly rely on physical mechanism models or empirical rules. On the one hand, physical mechanism models based on fracture mechanics and numerical simulation can describe the basic mechanisms of fracture initiation and propagation, but they have high computational costs and low efficiency in parameter space exploration, making it difficult to quickly evaluate different fracturing schemes under multi-parameter combinations. On the other hand, methods based on empirical formulas or static parameter optimization often fail to fully consider reservoir heterogeneity and the coupling effects between multiple parameters, easily leading to engineering problems such as uneven fracture initiation and limited stimulation volume.

[0005] In recent years, with the development of artificial intelligence technology, data-driven methods have been gradually introduced into the fields of fracturing simulation and parameter prediction, achieving rapid estimation of fracture response through machine learning models. However, existing purely data-driven methods typically rely on large amounts of sample data, and their prediction process lacks physical constraints. When extrapolating operating conditions, they are prone to results that do not conform to the laws of fracture mechanics and energy conservation, making them difficult to directly apply to engineering optimization decisions. Furthermore, existing research often focuses on a single aspect of fracture prediction or parameter optimization. Summary of the Invention

[0006] The purpose of this invention is to provide a fracturing agent modeling and optimization method that integrates data-driven and physical mechanisms. This method addresses the problems of low computational efficiency in traditional fracturing simulation, lack of physical constraints in pure data models, and the disconnect between simulation and optimization processes.

[0007] The technical solution adopted by this invention to solve its technical problem is as follows: This fracturing intelligent agent modeling and optimization method integrating data-driven and physical mechanism is as follows: By constructing a multi-dimensional fracturing parameter space and conducting simulation of fracture initiation and propagation mechanisms, fracture spatial morphology and evolution response data are obtained; the dataset of the finite mechanism simulation sample is expanded based on interpolation and statistical methods; geometric, physical, and topological features are extracted from the fracture point cloud data, and the fracture morphology is classified and dimensionality reduced to construct a physically meaningful fracture feature vector; a data-driven fracture prediction model is constructed based on the fracture feature vector, and fracture mechanics constraints are introduced to achieve rapid and stable prediction of fracture initiation and propagation behavior; the data-driven fracture prediction model and the physical mechanism model are bidirectionally coupled to form a fracturing intelligent agent with reasoning and correction capabilities, and the joint optimization of perforation parameters and fracturing construction parameters is achieved by combining intelligent optimization algorithms to output the optimal fracturing scheme that meets engineering constraints.

[0008] The above scheme's fracturing agent modeling and optimization method, which integrates data-driven approaches and physical mechanisms, includes the following steps: S1. For the target reservoir fracturing conditions, a multi-dimensional fracturing parameter space is constructed, which includes geomechanical parameters, wellbore parameters, perforation parameters and fracturing construction parameters. Based on fracture mechanics principles and numerical simulation methods, the mechanism of fracture initiation and propagation is simulated in the parameter space to obtain fracture spatial location, fracture aperture and fracture evolution response data, forming an initial mechanism simulation dataset. S2. The simulation results of crack initiation and propagation mechanism are expanded by using an interpolation method based on geostatistics. The crack response results under the non-simulated working conditions are predicted in the multi-dimensional parameter space, and the expanded crack data augmentation dataset is generated. S3. Extract the geometric, physical, and topological features of the cracks from the enhanced crack data dataset. Use clustering and correlation analysis methods to classify the crack morphology and perform dimensionality reduction on the high-dimensional crack point cloud data to construct a crack feature vector with physical meaning. S4. Based on the fracture feature vector and the corresponding fracturing parameters, a data-driven fracture prediction model is constructed. The model uses a neural network structure to learn the mapping relationship between fracturing parameters and fracture response, and to quickly predict the fracture initiation and propagation behavior. Physical constraints are introduced during the training process of the data-driven fracture prediction model. S5. The data-driven crack prediction model and the physical mechanism model are bidirectionally coupled. The prediction results of the data-driven model are corrected and constrained by the physical mechanism model, and the calculation results of the physical mechanism model are accelerated and compensated by the data-driven model, forming a dual-driven crack propagation simulation model that integrates data-driven and physical mechanisms. S6. Based on the dual-drive fracture propagation simulation model, a fracturing intelligent agent optimization decision module is constructed to jointly optimize the perforation parameters and fracturing construction parameters, output the optimal combination of fracturing parameters, and realize intelligent optimization and decision-making of fracturing scheme.

[0009] The perforation parameters in the above scheme S1 include perforation method, perforation depth, hole diameter, hole density, phase angle and cluster spacing; the fracturing construction parameters include displacement, fluid volume, viscosity, and sand ratio / sand addition intensity.

[0010] In the above scheme, the crack initiation and propagation mechanism simulation in S1 adopts the stress intensity factor criterion of linear elastic fracture mechanics. When the stress intensity factor is greater than or equal to the fracture toughness, it is determined that the crack has started or propagated. Select the key inputs that affect crack morphology as the parameter vector x: (1) in, It is a perforation method; Young's modulus; Poisson's ratio; Tensile strength; For fracture toughness; The horizontal stress ratio; The horizontal overburden stress ratio; This is a subset of perforation parameters, including penetration depth, aperture, orifice density, phase angle, and cluster spacing. This is a subset of construction parameters, including discharge rate, liquid volume, viscosity, and sand ratio / sand addition strength. The output is defined as the crack response vector y: (2) in Let the point cloud coordinates of the crack be... For the corresponding opening / seam width, For the length of the seam, To make the seam higher, To improve comprehensive indicators such as volume; The stress intensity factor criterion of linear elastic fracture mechanics is adopted: (3) When satisfied (4) Then the crack will initiate / propagate, where This is the equivalent crack length. As a feature scale, This is a geometric correction function.

[0011] The geostatistical interpolation method described in Scheme S2 above is the Kriging interpolation method, which is used to predict the crack response results under conditions that are not directly simulated in a multidimensional parameter space.

[0012] In the above scheme S3, the geometric features of the crack include crack length, crack height, and orientation; the physical features include average aperture and modification volume; and the topological features include the number of crack branches and connectivity index. From the crack point cloud Construct a set of geometric features, physical features, and topological features: The average opening is defined as: (9) The connectivity index is constructed based on the connected edges and node relationships: (10) in The number of point cloud nodes. Let be the number of connected edges; To characterize the complexity of the crack, a fractal dimension is introduced: (11) in The required side length to cover the crack is The number of boxes; In terms of morphological classification, a clustering algorithm is used to classify crack features, and the silhouette coefficient is used to determine the number of clusters: (12) in For the sample The average distance to samples in the same cluster For the sample The average distance to the nearest other cluster sample; Simultaneously, feature selection and dimensionality reduction are performed based on parameter-response correlation, and the Sobol variance decomposition sensitivity index is used to measure parameter importance. (13) Pearson correlation coefficient is used to measure redundancy between parameters: (14) Based on this, low-sensitivity parameters are eliminated, combined features are constructed, and principal component analysis is performed on the mechanical parameters to form a crack feature vector for learning. : (15) in For point cloud-feature mapping, The dimension is the dimension after dimensionality reduction.

[0013] The physical constraints in the above scheme S4 ​​include at least fracture mechanics constraints and energy conservation constraints, in order to suppress the generation of non-physical prediction results by the data-driven model; The neural network is: (16) in Output a low-dimensional representation of the crack response or point cloud; Output the crack feature vector; Introducing the combined loss of data error, fracture criterion residual, and energy constraint residual: (17) in (18) Fracture mechanics constraint terms are penalized for violations based on stress intensity factor criterion: (19) Energy constraints are constructed based on energy release rate or work-dissipation consistency: (20) in , It is obtained by back-calculation of the predicted crack response. These are the weighting coefficients.

[0014] The above-mentioned scheme S5 specifically involves bidirectionally coupling the data-driven crack prediction model with the physical mechanism model, correcting and constraining the prediction results of the data-driven model through the physical mechanism model, and accelerating and compensating the calculation results of the physical mechanism model using the data-driven model, thereby forming a dual-driven crack propagation simulation model that integrates data-driven and physical mechanisms to achieve efficient and stable prediction of crack evolution.

[0015] Constructing a prediction-correction closed loop: The data-driven crack prediction model first provides a fast prediction. ; Physical mechanism model As initial / prior values, key physical quantities are corrected to obtain... ; The corrected residuals are fed back to the data model for online / offline updates or to constrain the next round of predictions; A stable fusion of two models is achieved using a gradient matching coupling mechanism, resulting in a dual-drive crack propagation simulation model. (twenty one) This ensures that the data model not only approximates the mechanistic results at the output level, but also maintains consistency at the gradient level, avoiding non-physical extrema caused by gradient drift during optimization search.

[0016] The optimization decision module in the above scheme S6 adopts an intelligent optimization algorithm to jointly optimize the perforation parameters and fracturing operation parameters. Its optimization objectives include at least one of minimizing fracturing pressure, maximizing reservoir stimulation volume, and improving fracture propagation uniformity.

[0017] Specifically, S6 of the above scheme is as follows: Based on the dual-drive fracture propagation simulation model, an optimization decision module of the fracturing agent is constructed. Under the premise of meeting engineering constraints, the perforation parameters and fracturing operation parameters are jointly optimized to minimize fracture pressure, maximize reservoir stimulation volume, or improve fracture propagation uniformity as optimization objectives. The optimal combination of fracturing parameters is output to achieve intelligent optimization and decision support for the fracturing scheme. First, sensitivity analysis is used to determine the key optimization variables and their value ranges. Regarding the optimization objectives, the following multi-objective form is adopted: (twenty two) in To predict rupture pressure; To predict the volume of the modification; The crack uniformity / complexity penalty term is defined based on the number of branches, connectivity, and aperture variance. The feasible domain includes equipment capacity constraints, shaft safety constraints, and construction upper limit constraints. Parameter combination optimization is performed using multi-swarm particle swarm optimization: (23) ; in For the first Substitute particle position; For speed; Inertial weights; For learning factors; is a random number; c is the individual's historical best. To achieve global optimality; for cases with multiple populations, differentiated settings are applied to different subgroups. And interact periodically; Finally, the engineering output format of multiple candidate schemes is provided for each cluster of perforation parameters to obtain the optimized results. Beneficial effects

[0018] 1. By constructing a multidimensional fracturing parameter space and conducting high-fidelity simulations of fracture initiation and propagation mechanisms, fracture spatial morphology and evolution response data are obtained. Secondly, geostatistical interpolation methods are used to expand the data of the limited mechanism simulation samples, improving the coverage of the parameter space. Furthermore, geometric, physical, and topological features of the fractures are extracted, and physically meaningful fracture feature vectors are constructed by combining clustering and correlation analysis. Then, a data-driven fracture prediction model incorporating fracture mechanics constraints is established based on these feature vectors to achieve rapid prediction of fracture propagation behavior. Finally, the data-driven model and the physical mechanism model are bidirectionally coupled, and intelligent optimization algorithms are used to jointly optimize perforation parameters and fracturing operation parameters, outputting the optimal combination of fracturing parameters. This invention achieves integrated intelligent decision-making for fracture propagation simulation and fracturing parameter optimization, significantly improving the efficiency of fracturing scheme evaluation and the reliability of parameter selection while ensuring physical rationality, providing an efficient and stable intelligent technical means for the design and construction of complex reservoir fracturing.

[0019] 2. This invention is an intelligent fracturing method that integrates physical mechanism models and data-driven models. By constructing a fracturing intelligent agent with the ability to simulate fracture evolution and optimize parameters, a unified intelligent system that integrates physical mechanisms and data-driven approaches and has autonomous reasoning and optimization capabilities is formed. This enables efficient modeling and optimization decision-making in the fracturing process and provides reliable intelligent technical means for fracturing design and construction under complex reservoir conditions.

[0020] 3. By integrating data-driven methods and physical mechanism models, this invention overcomes the problems of low computational efficiency in traditional fracturing simulation, lack of physical constraints in pure data models, and the disconnect between simulation and optimization processes, and achieves integrated intelligent decision-making for fracture propagation simulation and fracturing parameter optimization.

[0021] 4. This invention organically integrates physical mechanism models with artificial intelligence methods to achieve integrated intelligent decision-making for fracture propagation simulation and fracturing parameter optimization. While ensuring physical rationality, it significantly improves the efficiency of fracturing scheme evaluation and the reliability of parameter selection, and is applicable to fracturing design and construction under complex reservoir conditions. Attached Figure Description

[0022] Figure 1 This is a flowchart of the overall process for modeling and optimizing fracturing intelligent agents that integrates data-driven and physical mechanisms, as provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the crack propagation simulation results of the mechanism model shown in the embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the expansion of Kriging interpolation data according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating crack point cloud feature extraction and dimensionality reduction in an embodiment of the present invention; Figure 5 This is a schematic diagram of the data-driven model structure for crack morphology prediction shown in an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the fusion of the data model and the mechanism model as shown in an embodiment of the present invention; Figure 7 This is a schematic diagram of the particle swarm optimization process shown in an embodiment of the present invention. Detailed Implementation The present invention will be further described below with reference to the accompanying drawings: like Figure 1 As shown, this fracturing agent modeling and optimization method integrating data-driven and physical mechanism approaches constructs a multi-dimensional fracturing parameter space and conducts high-fidelity simulations of fracture initiation and propagation mechanisms to obtain fracture spatial morphology and evolution response data. Secondly, it expands the dataset of limited mechanism simulation samples using interpolation and statistical methods, effectively improving the parameter space coverage and solving the problems of high computational cost and sparse samples in high-fidelity numerical simulations. Furthermore, it extracts geometric, physical, and topological features from fracture point cloud data, and uses clustering and correlation analysis to classify fracture morphology and reduce the dimensionality of high-dimensional data, constructing physically meaningful fracture feature vectors. Then, based on these feature vectors, it constructs a data-driven fracture prediction model and introduces fracture mechanics constraints to achieve rapid and stable prediction of fracture initiation and propagation behavior. Finally, it bidirectionally couples the data-driven model with the physical mechanism model to form a fracturing agent with reasoning and correction capabilities. Based on this, it combines intelligent optimization algorithms to jointly optimize perforation parameters and fracturing construction parameters, outputting the optimal fracturing scheme that meets engineering constraints. Specifically, it includes the following steps: S1. Constructing a multi-dimensional parameter space and conducting initial high-fidelity mechanism simulations. The multi-dimensional parameter space includes geomechanical, wellbore, perforation, and fracturing parameters. Based on fracture mechanics and numerical simulation, fracture responses are obtained to form an initial mechanism simulation dataset. S2, based on geostatistical interpolation, expands simulation results to generate a more comprehensive and evenly distributed data augmentation dataset, addressing the problem of limited initial samples.

[0023] S3 involves crack feature extraction, classification, and dimensionality reduction. Geometric, physical, and topological features are extracted, and clustering and correlation analysis are used to reduce the dimensionality of high-dimensional point clouds and construct representative feature vectors.

[0024] S4 constructs a data-driven prediction model that incorporates physical constraints, uses neural networks to learn mapping relationships, and introduces physical constraints such as fracture mechanics and energy conservation to achieve rapid prediction.

[0025] S5 is a data-driven and physical mechanism bidirectional coupling model construction. The physical model corrects the data model prediction, and the data model accelerates / compensates the physical model calculation, forming a fusion dual-drive crack propagation simulation model.

[0026] S6, the fracturing intelligent agent optimization decision module, is based on a dual-drive model and uses intelligent optimization algorithms to jointly optimize perforation and fracturing parameters. It takes engineering constraints as input, minimizes fracturing pressure, maximizes reservoir stimulation volume or improves fracture propagation uniformity as optimization objectives, and outputs the optimal combination of fracturing parameters to achieve intelligent optimization and decision support for fracturing schemes.

[0027] Example: This method for modeling and optimizing fracturing agents, which integrates data-driven approaches with physical mechanisms, is as follows: Step S1: For the target reservoir fracturing conditions, construct a multi-dimensional parameter space containing geomechanical parameters, wellbore parameters, perforation parameters, and fracturing operation parameters; based on fracture mechanics principles and numerical simulation methods, conduct high-fidelity mechanism simulations of fracture initiation and propagation within the parameter space to obtain basic response data such as fracture spatial location, fracture aperture, and fracture evolution process, forming an initial mechanism simulation dataset.

[0028] Specifically, this embodiment adopts a multi-scale perforation fracture initiation and propagation simulation approach, selecting key inputs affecting fracture morphology as parameter vector x: (1) in, It is a perforation method; Young's modulus; Poisson's ratio; For tensile strength For fracture toughness; The horizontal stress ratio; The horizontal overburden stress ratio; This is a subset of perforation parameters (including penetration depth, aperture, orifice density, phase angle, and cluster spacing). This is a subset of construction parameters (including discharge rate, liquid volume, viscosity, sand ratio / sand addition strength).

[0029] The output of the mechanistic model is defined as the crack response vector y: (2) in Let the point cloud coordinates of the crack be... For the corresponding opening / seam width, For the length of the seam, To make the seam higher, To improve comprehensive indicators such as volume.

[0030] To ensure physical consistency between the crack initiation and propagation criteria, the stress intensity factor criterion of linear elastic fracture mechanics is adopted: (3) When satisfied (4) This indicates that the crack has initiated / propagated. This is the equivalent crack length. As a feature scale, The geometric correction function is shown in the simulation results. Figure 2 As shown.

[0031] Step S2: To address the issues of limited sample size and insufficient parameter space coverage in the initial mechanism simulation data, an interpolation method based on geostatistics is used to expand the mechanism simulation results. This method predicts the crack response results under conditions not directly simulated within a multidimensional parameter space, generating a data augmentation dataset with wider coverage and more uniform distribution, thus providing support for subsequent data-driven model training.

[0032] Specifically, such as Figure 3 As shown, this embodiment uses Kriging interpolation to augment sparse samples. Let's assume several simulated sample points are selected within the parameter space. ,in This represents a scalar quantity representing the response of a crack (average aperture). , Modify volume or seam length For unsimulated points The estimated value is obtained using the optimal linear unbiased estimate: (5) The unbiasedness constraint is: (6) Spatial correlation is characterized by the variogram: (7) Empirical semivariance: (8) The weights are obtained by fitting a theoretical variogram model and solving the Kriging equations. This allows for interpolation and augmentation of the responses to uncovered operating conditions.

[0033] Step S3: For the enhanced crack dataset, extract the crack geometric features, physical features, and topological features. The geometric features include crack length, crack height, and orientation; the physical features include average aperture and modification volume; and the topological features include the number of crack branches and connectivity index. Further, cluster analysis and correlation analysis are used to classify the crack morphology, and the high-dimensional crack point cloud data is dimensionality reduced to construct a representative crack feature vector representation.

[0034] Specifically, such as Figure 4 As shown, this embodiment analyzes crack point clouds, Construct three sets of features: Geometric features: (Seam length, seam height, orientation angle / orientation tensor, etc.); Physical characteristics: (Average opening, modified volume, volume fraction, etc.); Topological features: (Number of branches, connectivity index, etc.)

[0035] The average opening degree is defined as: (9) The connectivity index is constructed based on the connected edges and node relationships: (10) in The number of point cloud nodes. Let be the number of connected edges.

[0036] To characterize the complexity of the crack, a fractal dimension is introduced: (11) in The required side length to cover the crack is The number of boxes.

[0037] In terms of morphological classification, clustering algorithms are used to categorize crack features. The silhouette coefficient is used to determine the number of clusters. (12) in For the sample The average distance to samples in the same cluster For the sample The average distance to the nearest other cluster sample.

[0038] Simultaneously, feature selection and dimensionality reduction are performed based on parameter-response correlation, and the Sobol variance decomposition sensitivity index is used to measure parameter importance. (13) Pearson correlation coefficient is used to measure redundancy between parameters: (14) Based on this, low-sensitivity parameters are eliminated, combined features are constructed, and principal component analysis is performed on the mechanical parameters to form a crack feature vector for learning. : (15) in For point cloud-feature mapping, The dimension is the dimension after dimensionality reduction.

[0039] Step S4: Based on the fracture feature vector and corresponding fracturing parameters obtained in step S3, a data-driven fracture prediction model is constructed. The model uses a neural network structure to learn the mapping relationship between fracturing parameters and fracture response, so as to realize the rapid prediction of fracture initiation and propagation behavior. At the same time, physical constraints are introduced during the model training process to ensure that the prediction results conform to physical mechanisms such as fracture mechanics and energy conservation.

[0040] Specifically, the model structure can adopt Figure 5 The neural network framework shown is used to define a neural network as follows: (16) in Output the crack response (or a low-dimensional representation of the point cloud). Output the crack feature vector.

[0041] To ensure physical consistency, a joint loss is introduced, consisting of data error, fracture criterion residual, and energy constraint residual. (17) in (18) Fracture mechanics constraint terms are penalized for violations based on stress intensity factor criterion: (19) Energy constraints are constructed based on energy release rate or work-dissipation consistency: (20) in , It is obtained by back-calculation of the predicted crack response. These are the weighting coefficients.

[0042] Step S5: The data-driven crack prediction model and the physical mechanism model are bidirectionally coupled. The prediction results of the data-driven model are corrected and constrained by the physical mechanism model, and the calculation results of the physical mechanism model are accelerated and compensated by the data-driven model, forming a dual-driven crack propagation simulation model that integrates data-driven and physical mechanism models, so as to achieve efficient and stable prediction of crack evolution process.

[0043] Specifically, such as Figure 6 As shown, this embodiment constructs a prediction-correction closed loop: The data model first provides a fast prediction. ; Mechanism model As initial / prior values, key physical quantities are corrected to obtain... ; The corrected residuals are fed back to the data model for online / offline updates or to constrain the next round of predictions.

[0044] Achieving stable fusion of two models using gradient matching coupling mechanism: (twenty one) This ensures that the data model not only approximates the mechanistic results at the output level, but also maintains consistency at the gradient level, avoiding non-physical extrema caused by gradient drift during optimization search.

[0045] Step S6: Based on the dual-drive fracture propagation simulation model, construct the optimization decision module of the fracturing agent. Under the premise of meeting the engineering constraints, use the intelligent optimization algorithm to jointly optimize the perforation parameters and fracturing construction parameters. The optimization objectives are to minimize the fracture pressure, maximize the reservoir stimulation volume, or improve the uniformity of fracture propagation. Output the optimal combination of fracturing parameters to realize intelligent optimization and decision support of the fracturing scheme.

[0046] Specifically, the key optimization variables and their value ranges were first determined through sensitivity analysis. The optimization parameter settings are shown in Table 1.

[0047] Table 1 Optimization Parameter Settings Parameter name Is it preferred? benchmark value Minimum value Maximum value accuracy Perforation mode 1.0 0.0 0.0 2.0 1.0 Perforation depth 0.0 0.55 aperture 0.0 0.009 Perforation density 1.0 12.0 3.0 20.0 2.0 Phase angle 1.0 75.0 30.0 180.0 15.0 Cluster spacing 1.0 7.0 5.0 8.0 0.2 Regarding the optimization objectives, the following multi-objective form is adopted: (twenty two) in To predict rupture pressure, To predict the volume of the modification, This is a penalty term for crack uniformity / complexity (defined based on factors such as the number of branches, connectivity, and aperture variance). The feasible domain of the project (constraints such as equipment capacity, well safety, and construction limits).

[0048] Parameter combination optimization is performed using multi-swarm particle swarm optimization: (twenty three) ; in For the first Substitute particle position (i.e., parameter combination). For speed, For inertial weights, As a learning factor, It is a random number. For the individual's historical best, This is globally optimal. For cases with multiple populations, differentiation can be applied to different subgroups. And periodic interactions are performed to improve global search capabilities and reduce the risk of premature convergence. The optimization process is as follows: Figure 7 As shown.

[0049] Finally, the engineering output format provides multiple candidate schemes for each cluster of perforation parameters, and the optimization results can be directly correlated with the parameter optimization results given in Table 2.

[0050] Table 2 Parameter optimization results Number of clusters Perforation mode Penetration depth (m) Aperture (m) Pore ​​density (pores / m) Phase angle (deg) Cluster spacing (m) Scheme Ranking 1 directional perforation 0.5 0.02 4 30 7.2 1 2 directional perforation 0.5 0.02 4 30 5 1 3 directional perforation 0.5 0.02 4 30 5 1 1 directional perforation 0.5 0.02 4 30 7 2 2 directional perforation 0.5 0.02 4 30 5 2 3 directional perforation 0.5 0.02 4 30 5 2 1 directional perforation 0.5 0.02 4 30 6.2 3 2 directional perforation 0.5 0.02 4 30 5 3 3 directional perforation 0.5 0.02 4 30 5 3 1 directional perforation 0.5 0.02 4 30 7.4 4 2 directional perforation 0.5 0.02 4 30 5 4 3 directional perforation 0.5 0.02 4 30 5 4 1 directional perforation 0.5 0.02 4 30 6.8 5 2 directional perforation 0.5 0.02 4 30 5 5 3 directional perforation 0.5 0.02 4 30 5 5 This invention is used for simulation of fracture initiation and propagation and intelligent optimization decision-making of fracturing parameters during the fracturing process of unconventional oil and gas reservoirs.

Claims

1. A method for modeling and optimizing fracturing intelligent agents that integrates data-driven approaches and physical mechanisms, characterized in that: By constructing a multidimensional fracturing parameter space and simulating the mechanism of fracture initiation and propagation, fracture spatial morphology and evolution response data are obtained. The dataset is expanded based on interpolation and statistical methods to fit the finite mechanism simulation samples. Geometric, physical, and topological features are extracted from the fracture point cloud data, and fracture morphology is classified and its dimensionality reduced to construct physically meaningful fracture feature vectors. A data-driven fracture prediction model is constructed based on these feature vectors, and fracture mechanics constraints are introduced to achieve rapid and stable prediction of fracture initiation and propagation behavior. The data-driven fracture prediction model and the physical mechanism model are bidirectionally coupled to form a fracturing agent with reasoning and correction capabilities. Combined with intelligent optimization algorithms, the perforation parameters and fracturing operation parameters are jointly optimized to output the optimal fracturing scheme that meets engineering constraints.

2. The fracturing intelligent agent modeling and optimization method integrating data-driven and physical mechanism approaches as described in claim 1, characterized in that... Includes the following steps: S1. For the target reservoir fracturing conditions, a multi-dimensional fracturing parameter space is constructed, which includes geomechanical parameters, wellbore parameters, perforation parameters and fracturing construction parameters. Based on fracture mechanics principles and numerical simulation methods, the mechanism of fracture initiation and propagation is simulated in the parameter space to obtain fracture spatial location, fracture aperture and fracture evolution response data, forming an initial mechanism simulation dataset. S2. The simulation results of crack initiation and propagation mechanism are expanded by using an interpolation method based on geostatistics. The crack response results under the non-simulated working conditions are predicted in the multi-dimensional parameter space, and the expanded crack data augmentation dataset is generated. S3. Extract the geometric, physical, and topological features of the cracks from the enhanced crack data dataset. Use clustering and correlation analysis methods to classify the crack morphology and perform dimensionality reduction on the high-dimensional crack point cloud data to construct a crack feature vector with physical meaning. S4. Based on the fracture feature vector and the corresponding fracturing parameters, a data-driven fracture prediction model is constructed. The model uses a neural network structure to learn the mapping relationship between fracturing parameters and fracture response, and to quickly predict the fracture initiation and propagation behavior. Physical constraints are introduced during the training process of the data-driven fracture prediction model. S5. The data-driven crack prediction model and the physical mechanism model are bidirectionally coupled. The prediction results of the data-driven model are corrected and constrained by the physical mechanism model, and the calculation results of the physical mechanism model are accelerated and compensated by the data-driven model, forming a dual-driven crack propagation simulation model that integrates data-driven and physical mechanisms. S6. Based on the dual-drive fracture propagation simulation model, a fracturing intelligent agent optimization decision module is constructed to jointly optimize the perforation parameters and fracturing construction parameters, output the optimal combination of fracturing parameters, and realize intelligent optimization and decision-making of fracturing scheme.

3. The fracturing intelligent agent modeling and optimization method integrating data-driven and physical mechanism as described in claim 2, characterized in that: The perforation parameters in S1 include perforation method, perforation depth, hole diameter, hole density, phase angle, and cluster spacing; the fracturing operation parameters include displacement, fluid volume, viscosity, and sand ratio / sand addition intensity.

4. The fracturing agent modeling and optimization method integrating data-driven and physical mechanism approaches as described in claim 3, characterized in that: The simulation of crack initiation and propagation mechanism in S1 adopts the stress intensity factor criterion of linear elastic fracture mechanics. When the stress intensity factor is greater than or equal to the fracture toughness, it is determined that crack initiation or propagation has occurred. Select the key inputs that affect crack morphology as the parameter vector x: (1) in, It is a perforation method; Young's modulus; Poisson's ratio; Tensile strength; For fracture toughness; The horizontal stress ratio; The horizontal overburden stress ratio; This is a subset of perforation parameters, including penetration depth, aperture, orifice density, phase angle, and cluster spacing. This is a subset of construction parameters, including discharge rate, liquid volume, viscosity, and sand ratio / sand addition strength. The output is defined as the crack response vector y: (2) in Let the point cloud coordinates of the crack be... To correspond to the opening / seam width, For the length of the seam, To make the seam higher, To improve comprehensive indicators such as volume; The stress intensity factor criterion of linear elastic fracture mechanics is adopted: (3) When satisfied (4) Then the crack will initiate / propagate, where This is the equivalent crack length. As a feature scale, This is a geometric correction function.

5. The fracturing agent modeling and optimization method integrating data-driven and physical mechanism approaches according to claim 4, characterized in that: The geostatistical interpolation method mentioned in S2 is the Kriging interpolation method, which is used to predict the crack response results under conditions that are not directly simulated in a multidimensional parameter space.

6. The fracturing agent modeling and optimization method integrating data-driven and physical mechanism approaches according to claim 5, characterized in that: The geometric features of the cracks in S3 include crack length, crack height, and orientation; the physical features include average aperture and modified volume; and the topological features include the number of crack branches and connectivity index. From the crack point cloud Construct a set of geometric features, physical features, and topological features: The average opening degree is defined as: (9) The connectivity index is constructed based on the connected edges and node relationships: (10) in The number of point cloud nodes. Let be the number of connected edges; To characterize the complexity of the crack, a fractal dimension is introduced: (11) in The required side length to cover the crack is The number of boxes; In terms of morphological classification, a clustering algorithm is used to classify crack features, and the silhouette coefficient is used to determine the number of clusters: (12) in For the sample The average distance to samples in the same cluster For the sample The average distance to the nearest other cluster sample; Simultaneously, feature selection and dimensionality reduction are performed based on parameter-response correlation, and the Sobol variance decomposition sensitivity index is used to measure parameter importance. (13) Pearson correlation coefficient is used to measure redundancy between parameters: (14) Based on this, low-sensitivity parameters are eliminated, combined features are constructed, and principal component analysis is performed on the mechanical parameters to form a crack feature vector for learning. : (15) in For point cloud-feature mapping, The dimension is the dimension after dimensionality reduction.

7. The fracturing agent modeling and optimization method integrating data-driven and physical mechanism approaches according to claim 6, characterized in that: The physical constraints in S4 include at least fracture mechanics constraints and energy conservation constraints to suppress non-physical prediction results generated by the data-driven model. The neural network is: (16) in Output a low-dimensional representation of the crack response or point cloud; Output the crack feature vector; Introducing the combined loss of data error, fracture criterion residual, and energy constraint residual: (17) in (18) Fracture mechanics constraint terms are penalized for violations based on stress intensity factor criterion: (19) Energy constraints are constructed based on energy release rate or work-dissipation consistency: (20) in , It is obtained by back-calculation of the predicted crack response. These are the weighting coefficients.

8. The fracturing agent modeling and optimization method integrating data-driven and physical mechanism approaches according to claim 7, characterized in that: S5 specifically involves: bidirectionally coupling the data-driven crack prediction model with the physical mechanism model; using the physical mechanism model to correct and constrain the prediction results of the data-driven model; and using the data-driven model to accelerate and compensate the calculation results of the physical mechanism model, thus forming a dual-driven crack propagation simulation model that integrates data-driven and physical mechanism approaches, achieving efficient and stable prediction of crack evolution. Constructing a prediction-correction closed loop: The data-driven crack prediction model first provides a fast prediction. ; Physical mechanism model As initial / prior values, key physical quantities are corrected to obtain... ; The corrected residuals are fed back to the data model for online / offline updates or to constrain the next round of predictions; A stable fusion of two models is achieved using a gradient matching coupling mechanism, resulting in a dual-drive crack propagation simulation model. (21) This ensures that the data model not only approximates the mechanistic results at the output level, but also maintains consistency at the gradient level, avoiding non-physical extrema caused by gradient drift during optimization search.

9. The fracturing agent modeling and optimization method integrating data-driven and physical mechanism approaches according to claim 8, characterized in that: The optimization decision module in S6 uses an intelligent optimization algorithm to jointly optimize the perforation parameters and fracturing operation parameters. Its optimization objectives include at least one of minimizing fracturing pressure, maximizing reservoir stimulation volume, and improving fracture propagation uniformity.

10. The fracturing agent modeling and optimization method integrating data-driven and physical mechanism approaches according to claim 9, characterized in that: Specifically, S6 involves: based on the dual-drive fracture propagation simulation model, constructing an optimization decision module for the fracturing agent; employing intelligent optimization algorithms to jointly optimize perforation parameters and fracturing operation parameters while meeting engineering constraints; aiming to minimize fracture pressure, maximize reservoir stimulation volume, or improve fracture propagation uniformity; and outputting the optimal combination of fracturing parameters to achieve intelligent optimization and decision support for the fracturing scheme. First, sensitivity analysis is used to determine the key optimization variables and their value ranges. Regarding the optimization objectives, the following multi-objective form is adopted: (22) in To predict rupture pressure; To predict the volume of the modification; The crack uniformity / complexity penalty term is defined based on the number of branches, connectivity, and aperture variance. The feasible domain includes equipment capacity constraints, shaft safety constraints, and construction upper limit constraints. Parameter combination optimization is performed using multi-swarm particle swarm optimization: (23) ； in For the first Substitute particle position; For speed; Inertial weights; For learning factors; is a random number; c is the individual's historical best. It is the global optimum; For multi-population scenarios, differentiate the settings for different subgroups. And interact periodically; Finally, the engineering output format of multiple candidate schemes is provided for each cluster of perforation parameters to obtain the optimized results.

Images