A method and system for optimizing fracture network extension

Abstract

This application discloses a method and system for optimizing fracture network propagation, belonging to the field of oil and gas field development technology. The method includes: acquiring oilfield data of the target sandstone and conglomerate reservoir; extracting physical characteristic indicators to characterize the micromechanical mechanism of the sandstone and conglomerate reservoir based on the oilfield data; and selecting the main control feature vector from multiple candidate parameters based on the fracture mechanics energy competition mechanism and the physical characteristic indicators. This application, by constructing four physical characteristic indicators with clear fracture mechanics significance—gravel-matrix modulus heterogeneity, cementation surface weakening potential energy, gravel topological stacking index, and perigravel path curvature entropy—is the first to quantify the microscopic control effect of sandstone and conglomerate heterogeneity on fracture propagation from the perspective of energy competition mechanism (penetration, perigravel, and congestion). This modeling based on physical mechanisms makes the model fundamentally closer to the actual geological process. Combined with Bayesian closed-loop iterative correction, it can dynamically absorb field data and correct model deviations in real time.

Description

Technical Field

[0001] This application belongs to the field of oil and gas field development technology, and specifically relates to a method and system for optimizing the expansion of hydraulic fracture networks. Background Technology

[0002] Conglomerate reservoirs possess abundant oil resources and represent a crucial area for increasing reserves and production in unconventional oil and gas. Horizontal well volumetric fracturing is a key technological approach for developing these reservoirs. Unlike conventional homogeneous sandstone, conglomerate reservoirs exhibit strong heterogeneity, containing numerous gravel particles of varying sizes, disordered distribution, and significant differences in strength.

[0003] Existing fracturing network propagation prediction and optimization technologies often directly input more than ten conventional geological parameters, such as porosity, permeability, and Young's modulus, into static models. These methods are primarily based on empirical and statistical screening, lacking engineering and physical significance. They ignore the mechanical mechanisms of fracture propagation in conglomerate, leading to redundant input parameters. Furthermore, as fracturing fluid is injected, the formation stress field and fluid pressure field dynamically change, lacking real-time correction processes and closed-loop control capabilities for "construction-as-you-go optimization." When faced with the unique "gravel wall" congestion characteristic of sandstone and conglomerate, timely identification is often difficult, easily leading to engineering accidents such as sand blockage. Existing methods cannot fully reflect the complex influence of sandstone and conglomerate heterogeneity on fracture propagation; the fitting accuracy is typically only around 60%, failing to meet the needs of precise engineering guidance.

[0004] Regarding the current state of similar technologies, existing methods mostly rely on static parameter calculations or offline numerical simulations. For example, while the traditional empirical formula method represented by Chinese patent CN103699779A (a method for predicting the half-length of hydraulic fractures in sandstone and conglomerate reservoirs) is easy to implement, it is difficult to fully characterize the complex influence of sandstone and conglomerate heterogeneity on the propagation of three-dimensional fracture networks. While Chinese patent CN109374867A (a discrete element method-based simulation method for hydraulic fracturing in sandstone and conglomerate) can depict the interaction mechanism between hydraulic fractures and gravel, the numerical simulation calculations are time-consuming and lack real-time monitoring and linkage. These traditional or similar methods generally suffer from problems such as a high risk of deviation between predicted and actual field solutions, delayed risk identification, and untimely control. Summary of the Invention

[0005] To address the above problems, this application provides a method for optimizing the propagation of hydraulic fracturing networks, the method comprising: Obtain oilfield data for the target sandstone and conglomerate reservoir; Based on oilfield data, physical characteristic indicators were extracted to characterize the micromechanical mechanism of sandstone and conglomerate reservoirs. Based on the energy competition mechanism of fracture mechanics, the main control feature vector is selected from multiple candidate parameters according to physical characteristic indicators; The master control feature vector is input into the fast forward modeling engine to obtain the initial pressure crack network prediction results; the pressure crack network prediction results include the predicted construction pressure curve and the three-dimensional crack network geometry. Field monitoring data was collected, and based on the field monitoring data and the initial fracture network prediction results, the parameters of the sandstone and conglomerate feature prediction model were iteratively corrected in a closed loop, and the corrected sandstone and conglomerate feature prediction model was output. The field monitoring data included field pump pressure data and microseismic / fiber optic monitoring data. The modified sandstone and conglomerate feature prediction model is used to identify construction risks, and corresponding construction parameter optimization instructions are generated based on the identified risks to achieve real-time control of the fracturing process.

[0006] According to some embodiments of this disclosure, the oilfield data includes core data, well logging data, seismic data, and historical fracturing and production data used for model calibration and closed-loop iterative correction for constructing physical characteristic indicators.

[0007] According to some embodiments of this disclosure, the expression for extracting the gravel-based modulus heterogeneity is as follows: ; Where A represents the gravel-matrix modulus heterogeneity; Indicates the elastic modulus of gravel particles; Indicates the modulus of the matrix cementitious material; Indicates the coefficient of local stress heterogeneity; This represents the correction factor.

[0008] According to some embodiments of this disclosure, the expression for the weakening potential energy of the adhesive surface is extracted as follows: ; Where B represents the weakening potential energy of the cementation surface; Indicates the tensile strength of the interface; Represents the bulk strength of the matrix, a constant; Indicates the volume percentage of the weak interface layer; Indicates the reference strength; This represents the attenuation coefficient.

[0009] According to some embodiments of this disclosure, the expression for extracting the gravel topology stacking index is as follows: ; Where C represents the gravel topology stacking index; Indicates the volumetric content of gravel; N represents the total number of gravel in the body. This represents the contact coordination number of the i-th gravel particle.

[0010] According to some embodiments of this disclosure, the expression for the curvature entropy of the path around the gravel is as follows: ; Where D represents the curvature entropy of the path around the gravel; Indicates the median grain size of the gravel; Indicates the crack width; The sorting coefficient represents the grain size distribution of the gravel. Indicates the sorting coefficient; This represents the weighting coefficient.

[0011] According to some embodiments of this disclosure, based on the fracture mechanics energy competition mechanism, a controlling feature vector is selected from multiple candidate parameters according to physical characteristic indices, including: Based on the energy competition mechanism of fracture mechanics, the competition index of the preset mode of crack propagation is determined according to the physical characteristic index. Compare the competition indices of various preset modes to determine the crack propagation mode of the master controller; Feature parameters corresponding to the crack propagation mode of the master are selected to form the master feature vector.

[0012] According to some embodiments of this disclosure, the preset modes include gravel-penetrating mode, gravel-circling mode, and congestion mode.

[0013] According to some embodiments of this disclosure, the expression for the competition index of the gravel-piercing mode is determined as follows: ; in, Indicates the competition index of the gravel-piercing mode; This indicates the gravel-matrix modulus heterogeneity after normalization. This represents the weakened potential energy of the cemented surface after normalization.

[0014] According to some embodiments of this disclosure, the expression for the competition index of the gravel-surrounding mode is determined as follows: ; in, Indicates the competition index of the gravel-surrounding mode; This indicates the gravel-matrix modulus heterogeneity after normalization. This represents the weakened potential energy of the cemented surface after normalization. This represents the gravel topological stacking index after normalization. This represents the curvature entropy of the path around the gravel after normalization.

[0015] According to some embodiments of this disclosure, the expression for the contention index of the congestion mode is determined as follows: ; in, A competition index indicating the congestion pattern; This represents the gravel topological stacking index after normalization. This represents the curvature entropy of the path around the gravel after normalization.

[0016] According to some embodiments of this disclosure, the parameters of the sandstone and conglomerate feature prediction model are corrected through closed-loop iterative modification, and the corrected sandstone and conglomerate feature prediction model is output, including: An initial sandstone and conglomerate feature prediction model for the target sandstone and conglomerate reservoir is established based on oilfield data, and the initial sandstone and conglomerate feature prediction model is parameterized into a model parameter vector. During the fracturing operation, real-time pump pressure data and microseismic / fiber optic monitoring data are collected on site. The current model parameter vector is input into the fast forward modeling engine to obtain the predicted construction pressure curve and the predicted three-dimensional fracture network geometry. An objective function is constructed with the goal of minimizing the difference between the predicted results and the on-site monitoring data, and the model parameter vector is updated using a regularized gradient iteration method. When the relative change of the objective function is less than or equal to a preset value, or when the normalized gradient magnitude of the objective function is less than or equal to a preset value, the iteration stops and the corrected sandstone and conglomerate feature prediction model is output.

[0017] According to some embodiments of this disclosure, the expression of the objective function is: ; in, This represents the objective function for minimizing the error. and These are the predicted pressure and the on-site pump pressure, respectively. and These are the predicted geometric parameters of the seam mesh and the on-site monitored parameters of the seam mesh, respectively. , These are the weighting coefficients; The regularization coefficient is used. is the initial reservoir parameter vector; x represents the reservoir parameter vector to be corrected in the closed loop.

[0018] According to some embodiments of this disclosure, construction risks are identified using a modified sandstone and conglomerate feature prediction model, including: Based on the physical characteristic indicators and the changing trends of fracture geometry output by the modified sandstone and conglomerate characteristic prediction model, it is determined whether there is a risk of sand blockage or increased friction due to excessively tortuous fracture paths.

[0019] According to some embodiments of this disclosure, based on the physical characteristic indicators output by the modified sandstone and conglomerate characteristic prediction model and the changing trends of fracture geometry, it is determined whether there is a risk of sand blockage, including: When the modified sandstone and conglomerate feature prediction model shows that, within a preset time window, the increase in the physical characteristic index value representing the heterogeneity of the gravel-matrix modulus relative to the previous time window reaches a first preset threshold, and the predicted crack width reaches a second preset threshold, it is determined that there is a risk of sand blockage.

[0020] According to some embodiments of this disclosure, based on the physical characteristic indicators output by the modified sandstone and conglomerate characteristic prediction model and the changing trends of fracture geometry, it is determined whether there is a risk of increased friction due to excessively tortuous fracture paths, including: When the modified sandstone and conglomerate feature prediction model shows that the physical characteristic index value characterizing the curvature entropy of the path around the gravel reaches the third preset threshold, or its relative increase within the preset time window reaches the fourth preset threshold, it is determined that there is a risk of increased friction due to excessive tortuosity of the crack path.

[0021] According to some embodiments of this disclosure, corresponding construction parameter optimization instructions are generated based on the identified risks, including: When a risk is identified as sand blockage risk, instructions are generated to reduce proppant concentration and / or increase fracturing fluid viscosity; When a risk is identified as increased friction due to excessively tortuous crack paths, an instruction is generated to increase the pump flow rate.

[0022] This application also provides a system for optimizing the propagation of hydraulic fracturing networks, the system comprising: The data acquisition module is used to acquire oilfield data of the target sandstone and conglomerate reservoir; The physical characteristic index extraction module is used to extract physical characteristic indicators based on oilfield data to characterize the micromechanical mechanism of sandstone and conglomerate reservoirs. The screening module is used to select the main control feature vector from multiple candidate parameters based on the fracture mechanics energy competition mechanism and physical characteristic indicators. The initial prediction module is used to input the main control feature vector into the fast forward modeling engine to obtain the initial prediction results of the pressure fracture network; the prediction results of the pressure fracture network include the predicted construction pressure curve and the three-dimensional fracture network geometry. The correction module is used to collect field monitoring data and, based on the field monitoring data and the initial pressure fracture network prediction results, to perform closed-loop iterative correction of the parameters of the sandstone and conglomerate feature prediction model, and output the corrected sandstone and conglomerate feature prediction model; among which, the field monitoring data includes field pump pressure data and microseismic / fiber optic monitoring data; The optimization module is used to identify construction risks using the modified sandstone and conglomerate feature prediction model, and generate corresponding construction parameter optimization instructions based on the identified risks, so as to realize real-time control of the fracturing construction process.

[0023] Compared with the prior art, this application has the following advantages: (1) This application constructs four physical characteristic indicators with clear fracture mechanics significance: gravel-matrix modulus heterogeneity, cementation surface weakening potential energy, gravel topological stacking index, and gravel path curvature entropy. For the first time, it quantifies the microscopic control effect of sandstone and conglomerate heterogeneity on fracture propagation from the perspective of energy competition mechanism (penetration, gravel circumference, and congestion). This modeling based on physical mechanism makes the model more closely resemble the actual geological process. Combined with Bayesian closed-loop iterative correction, it can dynamically absorb field data and correct model deviations in real time.

[0024] (2) This application uses the CMI energy competition algorithm to quickly select the dominant feature vector that controls the current fracture propagation mechanism by calculating and comparing the competition index of three fracture propagation modes: congestion type, gravel-penetrating type, and gravel-circling type, thus achieving intelligent dimensionality reduction of the model input. Subsequently, the dimensionality-reduced features are input into a fast forward simulation engine built on a deep neural network. This engine, after pre-training, can transform complex mechanical calculations into millisecond-level forward inference. This enables the entire prediction system to complete a full simulation within 0.3 seconds, achieving a "second-level" response, thereby enabling the embedding of real-time data streams from fracturing operations and providing a computational basis for "simultaneous construction and optimization".

[0025] (3) This application can predict risks in advance by calculating and monitoring the abrupt changes in physical characteristic indicators in real time. For example, when the model shows a surge in the physical characteristic indicator values ​​characterizing the heterogeneity of gravel-matrix modulus and a sharp contraction in crack width, the system can determine the risk of end sand shedding (sand blockage) 45-60 seconds in advance. Based on this, this application can automatically generate specific control instructions (such as "reduce proppant concentration to 120 kg / m³"). 3 "or increase the fracturing fluid viscosity by 5 mPa" In field applications, this application reduced the number of single-stage pressure abnormality alarms from an average of 7 to 2, and effectively avoided pump shutdown accidents caused by gravel blockage, greatly ensuring construction safety and continuity.

[0026] (4) This application, by accurately predicting fracture orientation and optimizing construction parameters (such as displacement, sand ratio, and viscosity) in real time, can more effectively overcome the "gravel wall" barrier in the reservoir, forming a more complex and complete fracture network system and improving the reservoir volume transformation. Field implementation shows that the average test production of the fracturing section optimized by this method is 30% higher than that of the traditional method in adjacent wells. This not only proves the effectiveness of this application in extremely heterogeneous reservoirs, but also directly translates into considerable economic benefits, providing a reliable technical tool for the efficient development of unconventional reservoirs such as sandstone and conglomerate.

[0027] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 A diagram illustrating a method for optimizing the propagation of a hydraulic fracturing network according to an embodiment of this application is shown. Figure 2 A detailed flowchart of the method for optimizing the expansion of hydraulic fracturing networks according to an embodiment of this application is shown; Figure 3a The diagram illustrates the initial data input and physical characteristic calculation in a practical example of fracturing optimization in a sandstone and conglomerate reservoir in an oilfield depression, according to an embodiment of this application. Figure 3b The diagram illustrates the parameter selection stage based on the energy competition mechanism in a practical example of fracturing optimization of a sandstone and conglomerate reservoir in an oilfield depression, according to an embodiment of this application. Figure 4a A schematic diagram illustrating the identification of initial pressure deviation during the initial stage of fracturing operation according to an embodiment of this application is shown; Figure 4b This diagram illustrates a closed-loop iterative correction process using a Bayesian algorithm in the early stages of fracturing operations, according to an embodiment of this application. Figure 4c A schematic diagram of pressure fitting after correction in the early stage of fracturing operation is shown according to an embodiment of this application; Figure 5a This illustration shows a real-time risk identification based on a modified model during the construction phase, according to an embodiment of this application. Figure 5b A schematic diagram illustrating the generation of construction parameter optimization instructions during the construction phase according to an embodiment of this application is shown; Figure 5c A schematic diagram illustrating the final implementation effect during the construction phase according to an embodiment of this application is shown; Figure 6 A diagram of a pressure fracturing network expansion optimization system according to an embodiment of this application is shown. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] like Figure 1 As shown, this application provides a method for optimizing the propagation of a hydraulic fracturing network, the method comprising: Obtain oilfield data for the target sandstone and conglomerate reservoir; Based on oilfield data, physical characteristic indicators were extracted to characterize the micromechanical mechanism of sandstone and conglomerate reservoirs. Based on the energy competition mechanism of fracture mechanics, the main control feature vector is selected from multiple candidate parameters according to physical characteristic indicators; The master control feature vector is input into the fast forward modeling engine to obtain the initial pressure crack network prediction results; the pressure crack network prediction results include the predicted construction pressure curve and the three-dimensional crack network geometry. Field monitoring data was collected, and based on the field monitoring data and the initial fracture network prediction results, the parameters of the sandstone and conglomerate feature prediction model were iteratively corrected in a closed loop, and the corrected sandstone and conglomerate feature prediction model was output. The field monitoring data included field pump pressure data and microseismic / fiber optic monitoring data. The modified sandstone and conglomerate feature prediction model is used to identify construction risks, and corresponding construction parameter optimization instructions are generated based on the identified risks to achieve real-time control of the fracturing process.

[0032] Preferably, the oilfield data includes core data, well logging data, seismic data for constructing physical characteristic indicators, and historical fracturing and production data for model calibration and closed-loop iterative correction.

[0033] In this embodiment, physical characteristic indicators for characterizing the micromechanical mechanism of sandstone and conglomerate reservoirs are extracted based on the core data, well logging data, and seismic data; and prior constraints and calibrations are applied to the initial parameters of the fast forward modeling engine or the geological model parameters based on the historical fracturing and production data.

[0034] Preferably, the physical characteristic indicators include gravel-matrix modulus heterogeneity, cementation surface weakening potential energy, gravel topological stacking index, and gravel path curvature entropy.

[0035] Preferably, the expression for extracting the gravel-based modulus heterogeneity is: ; Where A represents the gravel-matrix modulus heterogeneity; Indicates the elastic modulus of gravel particles; Indicates the modulus of the matrix cementitious material; Indicates the coefficient of local stress heterogeneity; This represents the correction factor.

[0036] Preferably, the expression for the weakening potential energy of the cemented surface is: ; Where B represents the weakening potential energy of the cementation surface; Indicates the tensile strength of the interface; Represents the bulk strength of the matrix, a constant; Indicates the volume percentage of the weak interface layer; Indicates the reference strength; This represents the attenuation coefficient.

[0037] Preferably, the expression for extracting the gravel topological stacking index is: ; Where C represents the gravel topology stacking index; Indicates the volumetric content of gravel; N represents the total number of gravel in the body. This represents the contact coordination number of the i-th gravel particle.

[0038] Preferably, the expression for extracting the curvature entropy of the path around the gravel is: ; Where D represents the curvature entropy of the path around the gravel; Indicates the median grain size of the gravel; Indicates the crack width; The sorting coefficient represents the grain size distribution of the gravel. Indicates the sorting coefficient; This represents the weighting coefficient.

[0039] Preferably, based on the fracture mechanics energy competition mechanism, the controlling feature vector is selected from multiple candidate parameters according to physical characteristic indicators, including: Based on the energy competition mechanism of fracture mechanics, the competition index of the preset mode of crack propagation is determined according to the physical characteristic index. Compare the competition indices of various preset modes to determine the crack propagation mode of the master controller; Feature parameters corresponding to the crack propagation mode of the master are selected to form the master feature vector.

[0040] Preferably, the preset modes include gravel-penetrating mode, gravel-circling mode, and congestion mode.

[0041] Preferably, the expression for determining the competition index of the gravel-piercing mode is: ; in, Indicates the competition index of the gravel-piercing mode; This indicates the gravel-matrix modulus heterogeneity after normalization. This represents the weakened potential energy of the cemented surface after normalization.

[0042] Preferably, the expression for determining the competition index of the gravel-circling mode is: ; in, Indicates the competition index of the gravel-surrounding mode; This indicates the gravel-matrix modulus heterogeneity after normalization. This represents the weakened potential energy of the cemented surface after normalization. This represents the gravel topological stacking index after normalization. This represents the curvature entropy of the path around the gravel after normalization.

[0043] Preferably, the expression for determining the contention index of the congestion mode is: ; in, A competition index indicating the congestion pattern; This represents the gravel topological stacking index after normalization. This represents the curvature entropy of the path around the gravel after normalization.

[0044] Preferably, the parameters of the sandstone and conglomerate feature prediction model are corrected through closed-loop iterative modification, and the corrected sandstone and conglomerate feature prediction model is output, including: An initial sandstone and conglomerate feature prediction model for the target sandstone and conglomerate reservoir is established based on oilfield data, and the initial sandstone and conglomerate feature prediction model is parameterized into a model parameter vector. During the fracturing operation, real-time pump pressure data and microseismic / fiber optic monitoring data are collected on site. The current model parameter vector is input into the fast forward modeling engine to obtain the predicted construction pressure curve and the predicted three-dimensional fracture network geometry. An objective function is constructed with the goal of minimizing the difference between the predicted results and the on-site monitoring data, and the model parameter vector is updated using a regularized gradient iteration method. When the relative change of the objective function is less than or equal to a preset value, or when the normalized gradient magnitude of the objective function is less than or equal to a preset value, the iteration stops and the corrected sandstone and conglomerate feature prediction model is output.

[0045] In this embodiment, an initial sandstone and conglomerate characteristic prediction model for the target sandstone and conglomerate reservoir is established based on core data, well logging data, seismic data, and historical fracturing and production data. This initial sandstone and conglomerate characteristic prediction model is then parameterized into a model parameter vector. Among them, core data is used to constrain the elastic modulus of gravel particles, the modulus of matrix cement, the tensile strength of the interface, and the strength of the matrix itself; well logging data is used to constrain the gravel size distribution, gravel volume content, and sorting factor along the wellbore direction; seismic data is used to constrain the direction of the formation principal stress and the local stress heterogeneity coefficient; and historical fracturing and production data are used to constrain the initial value range of the fracturing pressure, displacement, and fracture response.

[0046] During fracturing operations, real-time data from on-site pump pressure and microseismic / fiber optic monitoring are collected. The current model parameter vector is input into a fast forward modeling engine to obtain the predicted construction pressure curve and the predicted 3D fracture network geometry. Subsequently, an objective function is constructed to minimize the difference between the predicted results and the on-site monitoring data, and a regularized gradient iteration method is used to update the model parameter vector. When the relative change of the objective function is less than or equal to 0.01, or the normalized gradient modulus of the objective function is less than or equal to 0.01, the iteration stops, and the corrected sandstone and conglomerate feature prediction model is output.

[0047] Preferably, the expression for the objective function is: ; in, This represents the objective function for minimizing the error. and These are the predicted pressure and the on-site pump pressure, respectively. and These are the predicted geometric parameters of the seam mesh and the on-site monitored parameters of the seam mesh, respectively. , These are the weighting coefficients; The regularization coefficient is used. is the initial reservoir parameter vector; x represents the reservoir parameter vector to be corrected in the closed loop.

[0048] Preferably, the modified sandstone and conglomerate feature prediction model is used to identify construction risks, including: Based on the physical characteristic indicators and the changing trends of fracture geometry output by the modified sandstone and conglomerate characteristic prediction model, it is determined whether there is a risk of sand blockage or increased friction due to excessively tortuous fracture paths.

[0049] Preferably, based on the changes in physical characteristic indicators and fracture geometry output by the modified conglomerate feature prediction model, the presence of sand blockage risk is determined, including: When the modified sandstone and conglomerate feature prediction model shows that, within a preset time window, the increase in the physical characteristic index value representing the heterogeneity of the gravel-matrix modulus relative to the previous time window reaches a first preset threshold, and the predicted crack width reaches a second preset threshold, it is determined that there is a risk of sand blockage.

[0050] Preferably, based on the physical characteristic indicators and the changing trends of fracture geometry output by the modified sandstone and conglomerate characteristic prediction model, it is determined whether there is a risk of increased friction due to excessively tortuous fracture paths, including: When the modified sandstone and conglomerate feature prediction model shows that the physical characteristic index value characterizing the curvature entropy of the path around the gravel reaches the third preset threshold, or its relative increase within the preset time window reaches the fourth preset threshold, it is determined that there is a risk of increased friction due to excessive tortuosity of the crack path.

[0051] Preferably, the corresponding construction parameter optimization instructions are generated based on the identified risks, including: When a risk is identified as sand blockage risk, instructions are generated to reduce proppant concentration and / or increase fracturing fluid viscosity; When a risk is identified as increased friction due to excessively tortuous crack paths, an instruction is generated to increase the pump flow rate.

[0052] like Figure 2 As shown, the detailed technical content of this method is as follows: S1: Oilfield data acquisition and input; acquiring core data, well logging data, seismic data, and historical fracturing and production data of the target sandstone and conglomerate reservoir; S2: Construction of physical mechanism characteristics; extraction and calculation of four physical characteristic indicators characterizing the micromechanical mechanism of sandstone and conglomerate: gravel-matrix modulus heterogeneity (A), cementation surface weakening potential energy (B), gravel topological stacking index (C), and gravel path curvature entropy (D).

[0053] The specific calculation method is as follows: Calculate the gravel-matrix modulus heterogeneity A to characterize the resistance to gravel penetration when the crack tip encounters high-strength gravel: ; in, The elastic modulus of the gravel particles. The modulus of the matrix cementitious material. This is the coefficient for local stress heterogeneity. This is a correction factor.

[0054] Calculate the weakening potential energy B at the cemented interface: This is used to characterize the tendency of cracks to peel off along the weakly cemented interface. ; in, The tensile strength of the interface; Let be the matrix bulk strength, a constant; This represents the volume percentage of the weak interface layer. Indicates the reference strength; This represents the attenuation coefficient.

[0055] Calculate the gravel topological stacking index C: used to characterize the degree of gravel congestion and the skeleton support effect. ; in, The value represents the gravel volume content, and N represents the total number of gravels in the body. Let be the contact coordination number of the i-th gravel particle.

[0056] Calculate the path curvature entropy D around the gravel: used to characterize the path curvature and energy dissipation of the gravel around the crack. ; in, The median grain size of the gravel; The width of the crack; The sorting coefficient represents the grain size distribution of the gravel. This is the sorting coefficient; These are the weighting coefficients.

[0057] S3: Parameter selection based on energy competition mechanism; establish a parameter selection model based on fracture mechanics energy competition mechanism (CMI algorithm), calculate the competition weights of three modes: gravel penetration, gravel entanglement, and congestion, and select the main control feature vector. The CMI parameter selection model calculates the competition index of the three crack propagation modes and selects the main control features. The energy competition mechanism refers to the competition between the driving energy generated by fracturing fluid at the fracture tip and the energy dissipated required for fracture propagation along different paths. When hydraulic fractures encounter gravel particles and the gravel-matrix interface, three candidate modes exist: fracture propagation through the gravel, propagation along the interface around the gravel, and congestion due to gravel accumulation. For any candidate mode... Its energy margin Represented as:

[0058] In the formula, fracturing fluid acting on the model The effective energy release rate corresponding to the extended path, The equivalent energy resistance required for crack propagation, interface debonding, or localized congestion deformation along this path. Energy margin. The larger the value, the more likely the crack propagation mode is to occur; the mode with the largest energy margin is the current dominant crack propagation mode.

[0059] Since it is difficult to directly obtain the absolute energy release rate and dissipation energy corresponding to the three paths under field conditions, this application adopts the CMI algorithm to construct a dimensionless equivalent energy competition index based on gravel-matrix modulus heterogeneity, cementation surface weakening potential energy, gravel topological stacking index and gravel path curvature entropy.

[0060] The CMI algorithm is a feature-mode-index algorithm, which includes three steps: physical feature construction, crack propagation mode competition, and competition index discrimination.

[0061] First, the gravel-matrix modulus heterogeneity A, cementation surface weakening potential B, gravel topological stacking index C, and perigravel path curvature entropy D are normalized:

[0062] In the formula, These are the normalized physical characteristic indicators; and These are the minimum and maximum values ​​of the corresponding indicators in the historical samples or offline simulation samples of the target reservoir, respectively. To prevent positive numbers with zero denominators; A represents gravel-matrix modulus heterogeneity; B represents cementation surface weakening potential energy; C represents gravel topological stacking index; D represents path curvature entropy around gravel; Z represents historical samples or offline simulation samples of the target reservoir.

[0063] Based on the normalized physical characteristic indices, the equivalent energy competition indices for the three modes of gravel penetration, gravel entanglement, and congestion are calculated respectively:

[0064]

[0065]

[0066] In the formula, Indicates the competition index of the gravel-piercing mode; This indicates the gravel-matrix modulus heterogeneity after normalization. This represents the weakened potential energy of the cemented surface after normalization. Indicates the competition index of the gravel-surrounding mode; This represents the gravel topological stacking index after normalization. This represents the normalized curvature entropy of the path around the gravel. The competition index represents the congestion pattern.

[0067] Among them, a smaller gravel-matrix modulus heterogeneity and a higher interfacial strength are conducive to the continuous penetration of cracks through gravel, thus increasing the competition index of the gravel penetration mode; a larger gravel-matrix modulus heterogeneity, a weaker gravel-matrix interface, and a lower degree of gravel stacking are conducive to the propagation of cracks around the gravel along the interface, thus increasing the competition index of the gravel-around mode; a higher gravel topological stacking index and a larger ratio of gravel size to crack width will increase the possibility of crack obstruction and gravel congestion, thus increasing the competition index of the congestion mode.

[0068] Furthermore, the three equivalent energy competition indices are converted into competition weights:

[0069] In the formula, For pattern The competitive weight; This is the sensitivity coefficient of the competition index; This represents the competition index for the k-th crack propagation mode; Indicates the competition index of the gravel-piercing mode; Indicates the competition index of the gravel-surrounding mode; The competition index represents the congestion mode; P represents the gravel-crossing mode; U represents the gravel-circling mode; and H represents the congestion mode.

[0070] The crack propagation mode with the highest competitive weight is determined as the controlling crack propagation mode:

[0071] Select the corresponding master control feature vector based on the master control crack propagation mode:

[0072] in, Represents the controlling feature vector; This indicates the gravel-matrix modulus heterogeneity after normalization. This represents the weakened potential energy of the cemented surface after normalization. This represents the gravel topological stacking index after normalization. This represents the normalized curvature entropy of the path around the gravel. The current master crack propagation mode is indicated by P; P indicates gravel penetration mode; U indicates gravel entanglement mode; and H indicates congestion mode.

[0073] S4: Fast forward modeling simulation; Input the master control feature vector X into the fast forward modeling simulation engine (ROM), output the initial pressure crack network state prediction results, and quickly output the predicted construction pressure curve and three-dimensional crack network geometry through nonlinear mapping; The rapid forward modeling engine is a reduced-order surrogate model trained using high-fidelity hydraulic fracturing numerical simulation samples. In the offline phase, multiple sets of fluid-structure interaction hydraulic fracturing numerical simulations are conducted by changing reservoir physical characteristics, geostress conditions, and construction parameters to establish a sample set between input parameters and fracturing response. Specifically, this includes: (1) Basic parameter collection and parameter range determination.

[0074] Explanation of parameter sources: Core experiments: gravel modulus, matrix modulus, interfacial tensile strength, matrix tensile strength; CT or imaging logging: gravel volume fraction, particle size, sorting parameters, particle contact coordination number; Well logging and seismic data: geostress, reservoir thickness, elastic parameters; Historical fracturing data: displacement, fluid viscosity, proppant concentration, and operating pressure.

[0075] The upper and lower limits of each parameter are determined based on the measured values ​​of the target layer, data from adjacent wells, and the experimental error range.

[0076] The original parameter vector can be written as:

[0077] In the formula: For the first The original parameter vector of a high-fidelity numerical simulation; Number the sample; For the first Elastic modulus of gravel particles in the sample group; The elastic modulus of the matrix cementitious material; Tensile strength at the gravel-matrix interface; The tensile strength of the matrix itself; This represents the volume fraction of gravel. To control the average contact coordination number of gravel particles in vivo; The median grain size of the gravel; For gravel particle size sorting parameters; The difference is the horizontal principal stress. This refers to the pump displacement. This refers to the dynamic viscosity of the fracturing fluid. This refers to the proppant concentration; superscript. This represents the transpose of a vector.

[0078] (2) Numerical simulation sample generation Generate samples using Latin hypercube sampling, uniform design, or other explicitly parametric sampling methods. Group parameter combinations:

[0079] In the formula: This represents all combinations of parameters obtained through parameter sampling; The total number of valid numerical simulation samples after quality control is completed; This refers to the range of sample numbers.

[0080] Generate an offline training dataset for each set of parameters, resulting in:

[0081] In the formula: For the first Construction pressure vector obtained from offline training dataset; superscript This represents the results of a high-fidelity numerical simulation. For the first Group simulation at time The construction pressure; For the first A unified discrete time node; m represents the number of time points after discretizing the pressure curve.

[0082] as well as:

[0083] In the formula: For the first The geometries of the seam mesh obtained from the offline training dataset; The length of the crack; The height of the crack; The average or characteristic width of the crack; The azimuth of the main crack; This refers to the volume of the crack or seam network envelope.

[0084] If the offline training dataset also outputs the number of crack branches, you can increase... If there is no stable output, it should not be written.

[0085] Forming a single sample:

[0086] In the formula: For the first A high-fidelity numerical simulation sample; Number the sample; For the first The input vector of each sample; For the first The high-fidelity fracturing response output vector corresponding to each sample; superscript This represents the results of a high-fidelity numerical simulation.

[0087] in:

[0088] In the formula: For the first The complete output vector of a high-fidelity hydraulic fracturing simulation; This is the construction pressure vector; This is the geometric parameter vector of the mesh; outer superscript Used to combine the pressure vector and the geometric parameter vector into a single column vector.

[0089] (3) The input vector is fixed as follows:

[0090] In the formula: For the first Fixed-dimensional principal feature vectors of a group of samples; This represents the normalized gravel-matrix modulus heterogeneity. The normalized relative strength index of the gravel-matrix interface; The normalized gravel topological stacking index; This is the normalized curvature index of the path around the gravel; , , , These are the mask variables corresponding to the four physical characteristics; the mask corresponding to the main control feature is set to 1, and the mask corresponding to the non-main control feature is set to 0. , , These are the one-hot encoded components for gravel penetration, gravel wrapping, and congestion modes, respectively.

[0091] (4) Sample quality control and dataset partitioning The valid samples are divided into training set, validation set and test set according to a preset ratio. The training set is used for parameter update, the validation set is used for sample correctness verification and hyperparameter selection, and the test set is only used for independent accuracy evaluation after training is completed.

[0092] (5) Data standardization and training process Both input and output are standardized using training set statistics:

[0093] In the formula: These are the original values ​​of the input or output parameters to be standardized. For parameters Standardized values; For parameters The mean in the training set; For parameters The standard deviation in the training set; The positive number set to prevent the standard deviation from being zero is much smaller than... This formula is used to standardize the numerical scale of different parameters. The validation set, test set, and field input data are all calculated using the training set. and Standardize it.

[0094] The training process is based on the following formula, the fracturing response vector formula:

[0095] In the formula: For the first The fracturing response vector obtained from the predicted samples; the symbol above the " " indicates that the parameter is a predicted value; Number the sample; For fast forward surrogate model; The parameters to be trained for the surrogate model include the weights and biases of each layer of the model; For the first The complete input vector of each sample. The sample input vector, which has been defined above, is composed of the main control feature vector and the construction and boundary condition vector. It is composed of the predicted construction pressure vector and the predicted joint mesh geometric parameter vector.

[0096] The training parameters are obtained by minimizing the training error:

[0097] in, These are the optimal surrogate model parameters obtained after training. These are the parameters of the surrogate model to be optimized during training. This represents the model parameters that minimize the objective function within the parentheses. Number the training samples; This represents the summation over all samples in the training set; The weighting coefficient for the construction pressure prediction error term; The weighting coefficients for the error term in the prediction of the mesh geometry parameters; For the proxy model to the first The construction pressure vector obtained from the prediction of each sample; For the first Each sample corresponds to a high-fidelity numerical simulation construction pressure vector; For the proxy model to the first The geometries of the seam predicted from each sample; For the first The high-fidelity numerical simulation mesh geometric parameter vector corresponding to each sample; superscript This represents the results of a high-fidelity numerical simulation. For the regularization coefficients of the surrogate model parameters; The second norm of a vector; This represents the sum of squares of the vector components; This is a regularization term for model parameters, used to constrain the magnitude of model parameters and reduce the risk of overfitting.

[0098] (6) On-site deployment Fixed after training:

[0099] After the input is processed using the same sorting, masking, and normalization methods, the output is:

[0100] In the formula, The fracturing response vector output by the fast forward surrogate model. To use the optimal model parameters The fast forward proxy model, This is the fixed-dimensional master control feature vector for the current well section. This represents the current construction and boundary condition vector.

[0101] in:

[0102] In the formula, This represents the vector of on-site construction and boundary conditions. The horizontal principal stress difference in the target well section, This refers to the injection and discharge rate of fracturing fluid. This refers to the dynamic viscosity of the fracturing fluid. This refers to the proppant concentration, indicated by the superscript. This represents the transpose of a vector.

[0103]

[0104] middle, The fracturing response vector output by the fast forward surrogate model. To predict the construction pressure vector, , and These are the predicted crack length, crack height, and crack width, respectively. These are the geometric representation parameters of the three-dimensional seam mesh, indicated by superscript. This represents the transpose of a vector.

[0105] During the field phase, the neural network is not retrained; only the reservoir parameter vector is corrected. When field parameters exceed the training sample range, an out-of-bounds flag should be set, and automatic control commands should not be generated directly from model results.

[0106] S5: Closed-loop iterative correction; The sandstone and conglomerate feature prediction model is not a second neural network trained independently, but is composed of the reservoir parameter vector to be corrected by closed-loop, the mapping relationship between reservoir parameters and the master control feature vector, the trained fast forward modeling engine, and field monitoring data constraints.

[0107] The conglomerate feature prediction model is used to dynamically correct reservoir parameters based on field construction data and fracture monitoring data while keeping the parameters of the fast forward modeling engine unchanged, and to update the construction pressure and fracture network geometry prediction results using the corrected reservoir parameters.

[0108] The sandstone and conglomerate feature prediction model is expressed as follows:

[0109] In the formula, A predictive model for the characteristics of sandstone and conglomerate; A fast forward modeling simulation engine for offline training; The optimal model parameters for the fast forward modeling engine remain unchanged during the on-site prediction process; The reservoir parameter vector to be corrected for loop closure; This refers to the main control feature vector calculated based on the current reservoir parameters and filtered using the CMI method; This represents the vector of on-site construction and boundary conditions.

[0110] (1) Establishment of initial reservoir parameters Based on core test data, well logging interpretation data, seismic interpretation data, and historical fracturing and production data from adjacent wells in the target well section, the reservoir parameters to be corrected in the closed loop and their initial values ​​are determined.

[0111] The reservoir parameter vector to be corrected by loop closure is represented as:

[0112] In the formula, The reservoir parameter vector to be corrected for loop closure; This is the local mechanical heterogeneity coefficient, used to characterize the spatial dispersion of the local mechanical properties of the target well section; This is the gravel-matrix modulus heterogeneity correction factor; The strength attenuation coefficient at the gravel-matrix interface; This is the curvature correction factor for the path around the gravel; superscript. This represents the transpose of a vector.

[0113] The initial reservoir parameter vector before the start of the closed-loop iteration is represented as:

[0114] In the formula, This is the initial reservoir parameter vector; , , and These are the initial values ​​of the corresponding reservoir parameters; the subscript 0 indicates the initial state before the closed-loop iteration begins.

[0115] The allowable range of values ​​for each parameter is determined jointly by core experiments, well logging interpretation, seismic interpretation, and data from adjacent wells, and is expressed as follows:

[0116] In the formula, This refers to the preset physical value range for reservoir parameters; This is the lower bound vector for reservoir parameters; Let be the upper limit vector of reservoir parameters; vector inequalities indicate that each parameter component satisfies its corresponding upper and lower limit constraints.

[0117] (2) Establishment of on-site monitoring samples During fracturing operations, data on pump pressure, pump flow rate, fracturing fluid viscosity, proppant concentration, microseismic monitoring data, and / or distributed fiber optic monitoring data are collected synchronously within a preset time window.

[0118] The first The results of construction and boundary conditions, on-site pump pressure, and on-site mesh geometry monitoring within a time window are combined into a single on-site monitoring sample:

[0119] In the formula, For the first One on-site monitoring sample; Number the samples for on-site monitoring; For the first Construction and boundary condition vectors corresponding to each time window; For the first The field-observed pressure vector within a time window; For the first Geometric monitoring vectors of the sewing mesh within a time window; subscript This represents the on-site observation value.

[0120] The construction and boundary condition vector is represented as follows:

[0121] In the formula, For the first The horizontal principal stress difference in the well section corresponding to each time window; For the first Pumping discharge rate within a time window; This refers to the dynamic viscosity of the fracturing fluid. This refers to the proppant concentration.

[0122] The pressure vector observed in the field is represented as:

[0123] In the formula, For the first Within the first time window On-site pump pressure at a unified time point; To unify discrete time nodes; ; This represents the number of discrete pressure nodes within a single time window.

[0124] The on-site mesh geometric monitoring vector is represented as:

[0125] In the formula, For the first Crack lengths monitored or interpreted within a time window; The height of the crack; The azimuth of the main crack; This represents the volume of the sewn mesh envelope.

[0126] For seam mesh geometric parameters that cannot be reliably obtained from on-site monitoring, they will not be included in the geometric error calculation for the corresponding time window.

[0127] All valid field monitoring samples constitute the field monitoring dataset:

[0128] In the formula, The field monitoring dataset used for the sandstone and conglomerate feature prediction model; To determine the effective number of field monitoring samples after time synchronization and quality control; subscript This indicates on-site data.

[0129] (3) Quality control of on-site monitoring data Before establishing the field monitoring dataset, time synchronization and quality control are performed on various types of field data.

[0130] For on-site pump pressure data, missing values ​​caused by communication interruptions and abnormal values ​​exceeding the sensor range are removed, and isolated noise points are filtered; real pressure changes caused by crack initiation, propagation, or sand blockage are retained.

[0131] For microseismic monitoring data, events with positioning errors exceeding preset values ​​are removed, and crack length, crack height, main crack orientation, and crack network envelope volume are calculated based on valid microseismic events.

[0132] For distributed fiber optic monitoring data, the extent of crack influence is determined based on the starting location, propagation range, and duration of the strain or acoustic response.

[0133] On-site pump pressure, construction parameters, and joint mesh monitoring parameters are aligned according to a unified time window. On-site monitoring samples that cannot be synchronized in time, have missing key parameters, or clearly do not conform to physical laws are not included in the on-site monitoring dataset.

[0134] (4) Division of on-site monitoring dataset Unlike fast forward modeling engines that use high-fidelity numerical simulation samples to divide training, validation, and test sets, the sandstone and conglomerate feature prediction model does not retrain the neural network. Its field monitoring data is divided into initial parameter determination dataset, online calibration dataset, and independent validation dataset according to its purpose:

[0135] In the formula, Determine the dataset for initial parameters; For online calibration of datasets; For independent validation datasets; This represents the union of non-overlapping sets.

[0136] The three data subsets satisfy:

[0137] In the formula, Represents the intersection of sets; This indicates that the empty set can only be included in one data subset; this formula means that the same field monitoring sample can only be included in one data subset.

[0138] The three data subsets are divided according to the construction time sequence. Among them, Composed of field monitoring samples from the initial construction phase, these samples, along with core, logging, seismic, and adjacent well data, are used to determine the initial reservoir parameter vector. Physical value range of reservoir parameters ; The field monitoring samples, consisting of samples from the initial parameter determination stage and before the current closed-loop correction time, are used to construct the objective function and update the reservoir parameter vector. It consists of subsequent field monitoring samples that did not participate in parameter updates, and is only used to evaluate the ability of the corrected sandstone and conglomerate feature prediction model to predict subsequent construction pressure and joint mesh geometric parameters.

[0139] Information from independent validation datasets should not be used in advance for reservoir parameter correction to avoid using subsequent construction information to correct previous model parameters.

[0140] (5) Determination of initial reservoir parameters Initial reservoir parameter vector The dataset was determined by core experiments, well logging interpretation, seismic interpretation, and historical data from adjacent wells, and the initial parameters were used to define the dataset. Perform verification.

[0141] Specifically, the candidate initial reservoir parameters are converted into corresponding master control feature vectors and input into the trained fast forward simulation engine to compare the predicted pressure and fracture network geometry parameters with... The differences between the field monitoring results were selected based on whether they met the initial error requirements and were within the physical value range. The parameters within are used as the initial reservoir parameter vector. .

[0142] Before inputting the data into the fast forward modeling engine, the data is processed in the same way as the input parameter arrangement, feature masking method, and standardized parameters determined during the offline training phase of the fast forward modeling engine.

[0143] (6) Model predictions corresponding to current reservoir parameters In the In the next closed-loop iteration, based on the current reservoir parameter vector Recalculate the physical characteristic indices and generate the current master control feature vector using the CMI method:

[0144] In the formula, This represents the number of closed-loop iterations. For the first The reservoir parameter vector for the next iteration; For the first The main control feature vector corresponding to the next iteration; This refers to the process of calculating physical characteristic indicators from reservoir parameters and then screening the main control characteristics using the CMI method.

[0145] For the first in the online correction dataset For each on-site monitoring sample, the current master feature vector and the construction and boundary conditions corresponding to that time window are input into the trained fast forward simulation engine:

[0146] In the formula, For the first Under the condition of the nth iteration, for the th Fracturing response prediction vector obtained from a field monitoring sample; subscript Indicates the predicted value; For the completed fast forward simulation engine; The parameters of the fast forward simulation engine remain unchanged during the field phase; For the first The construction and boundary condition vectors corresponding to each on-site monitoring sample.

[0147] The fracturing response prediction vector consists of the predicted construction pressure and the predicted fracture network geometry parameters:

[0148] In the formula, For the first During the nth iteration, the ... Construction pressure vector predicted from a single on-site monitoring sample; This is the corresponding predicted seam mesh geometric parameter vector.

[0149] (7) Online calibration objective function Based on online calibration dataset To differentiate between the predicted results and the field monitoring results, a closed-loop correction objective function for reservoir parameters is constructed:

[0150] In the formula, The objective function is established based on the online calibration dataset; This is the set of sample IDs in the online calibration dataset; The weight of the construction pressure error term; Weights for the geometric error term of the seam mesh; For the current reservoir parameters, the first The predicted construction pressure vector for each sample; This corresponds to the field-observed pressure vector; To predict the geometric parameter vector of the seam mesh; This is a vector of geometric parameters for the seam mesh used in on-site monitoring. For reservoir parameters, regularization coefficients are used. The term represents the vector L2 norm; the third term is used to prevent the corrected reservoir parameters from deviating excessively from the initial reservoir parameters.

[0151] (8) Iterative update of reservoir parameters Update reservoir parameters using the projection gradient method:

[0152] In the formula, For the first The reservoir parameter vector for the next iteration; For the first The update step size for each iteration; For the objective function in gradient at; This is a parameter range projection operator used to restrict the updated reservoir parameters to a preset physical value range. Internal; Fast forward simulation engine parameters It remains unchanged during the process.

[0153] When the relative change of the objective function satisfies:

[0154] Or the normalized gradient magnitude of the objective function satisfies:

[0155] Alternatively, stop updating parameters when the number of iterations reaches the preset maximum number of iterations.

[0156] In the formula, This is the convergence threshold for the relative change of the objective function; The convergence threshold for the normalized gradient magnitude; A positive number set to prevent the denominator from being zero.

[0157] (9) Independent verification After the parameters are corrected, an independent validation dataset that was not involved in the parameter update is used. Evaluate the revised sandstone and conglomerate feature prediction model.

[0158] Independent verification error is expressed as:

[0159] In the formula, The prediction error of the corrected sandstone and conglomerate feature prediction model on the independent validation dataset; For the set of sample IDs in the independent validation dataset; This is the final reservoir parameter vector after online calibration; the meanings of the other symbols are the same as before.

[0160] When the independent verification error is less than or equal to the preset verification threshold, the result of this parameter correction is accepted; when the independent verification error is greater than the preset verification threshold, the result of this correction is not used directly to generate automatic construction parameter control instructions, and on-site monitoring data is collected for the next round of parameter correction.

[0161] (10) Output of the sandstone and conglomerate feature prediction model After closed-loop iteration and independent verification are completed, the final corrected reservoir parameter vector is output:

[0162] In the formula, The final reservoir parameter vector that satisfies the convergence condition and passes independent verification.

[0163] The revised sandstone and conglomerate feature prediction model is expressed as follows:

[0164] In the formula, This is a closed-loop corrected prediction model for sandstone and conglomerate characteristics. To finally correct the reservoir parameter vector; This represents the current construction and boundary condition vector.

[0165] The corrected fracturing response prediction results are expressed as follows:

[0166] In the formula, This is the predicted fracturing response vector after closed-loop correction; To ultimately correct the main control feature vector corresponding to the reservoir parameters; superscript " "" indicates the final result after the closed-loop iteration converges and passes independent verification.

[0167] The final output of the sandstone and conglomerate feature prediction model includes: the final corrected reservoir parameter vector. The corrected parameters include: gravel-matrix modulus heterogeneity, interfacial relative strength index, gravel topological stacking index, and gravel-circling path curvature index; corrected competition weights for gravel penetration, gravel-circling, and congestion modes; corrected predicted construction pressure curves; and corrected crack length, crack height, crack width, crack orientation, and crack mesh envelope volume.

[0168] Among them, the parameters of the fast forward simulation engine The reservoir characteristic prediction model remained unchanged throughout the field initialization, online calibration, and independent validation processes; only the reservoir parameter vector was updated during the field stage. .

[0169] S6: Real-time optimization of construction parameters: Based on the corrected model prediction results, potential risks are identified and control instructions are generated. When the modified sandstone and conglomerate feature prediction model shows that, within a preset time window, the increase in the physical characteristic index value characterizing the heterogeneity of the gravel-matrix modulus relative to the previous time window reaches the first preset threshold, and the predicted fracture width reaches the second preset threshold, it is determined that there is a risk of sand blockage. When the modified sandstone and conglomerate feature prediction model shows that the physical characteristic index value characterizing the curvature entropy of the path around the gravel reaches the third preset threshold, or its relative increase within the preset time window reaches the fourth preset threshold, it is determined that there is a risk of increased friction due to excessive tortuosity of the fracture path, and an instruction to increase the pumping discharge rate is generated to increase the net pressure.

[0170] Specifically, the preset time window is a sliding time window for continuous data acquisition during the fracturing operation, preferably 30 seconds. 120s. The first preset threshold is the relative increase threshold of the gravel-matrix modulus heterogeneity index value, preferably 20%. 50%. When the relative increase of the gravel-matrix modulus heterogeneity index value within the preset time window reaches this range, it indicates that the crack tip encounters a high-modulus gravel accumulation zone or a locally highly heterogeneous zone, posing a risk of crack propagation obstruction and proppant blockage.

[0171] The corresponding formula is: ; In the formula, The gravel-matrix modulus heterogeneity within the current time window; This represents the gravel-matrix modulus heterogeneity within the previous time window; Preset time window; The first preset threshold is preferably... .

[0172] The third preset threshold can be determined based on the average curvature entropy of the gravel path in the historical fracturing sections of the same block, preferably 1.2 times the historical average. 1.5 times; the fourth preset threshold is the relative increase in the curvature entropy of the path around the gravel within a preset time window, preferably 15%. 40%. When the curvature entropy of the path around the gravel reaches the above threshold, it indicates that the tortuosity of the crack path is enhanced, the fluid flow path along the crack is lengthened, and the risk of increased local friction is raised.

[0173] The corresponding formula can be written as: ; or: ; In the formula, The curvature entropy of the current path around the gravel; The curvature entropy of the path around the gravel within the previous time window; The mean value of the curvature entropy of the fracturing path around the gravel in the historical fracturing section of the same block; The third preset threshold coefficient is preferred. ; The fourth preset threshold is preferably... .

[0174] This embodiment focuses on optimizing fracturing operations in a horizontal well within a tight sandstone and conglomerate reservoir in a depression of an oilfield. For example... Figure 3aAs shown, the reservoir in this area has a high gravel content (average 30%-50%) and is extremely heterogeneous. Traditional fracturing faces engineering challenges such as inaccurate model predictions and a high risk of sand blockage. This method implements closed-loop optimization through the following steps: S1: Source Data Acquisition and Input Before fracturing operations, the system first accesses multi-source heterogeneous data from the target well area to establish a basic geological dataset: Core data: Full-diameter cores from the target strata were selected for uniaxial / triaxial compression tests and Brazilian splitting tests to obtain the elastic modulus of gravel particles (E=60-80 GPa), the modulus of the matrix cement (E=20-30 GPa), and the interfacial tensile strength (E=60-80 GPa). ) and matrix strength ( ).

[0175] Well logging data: Using Formation MicroScanner Image (FMI) and Nuclear Magnetic Resonance (NMR) logging data, we interpret and obtain gravel size distribution curves and gravel volume content along the wellbore direction. and sorting coefficient .

[0176] Seismic data: Obtaining the principal stress direction and local stress heterogeneity coefficient of the formation using pre-stack seismic inversion data. ; Historical data: Collect pump pressure curves, discharge data, and microseismic monitoring event point data of the fractured sections of adjacent wells.

[0177] S2: Construction of Physical Mechanism Features: In order to accurately characterize the interference mechanism of gravel on crack propagation, this embodiment does not directly use conventional physical property parameters, but calculates four characteristic indicators with clear fracture mechanics significance: Calculate the gravel-matrix modulus heterogeneity A: This characterizes the stress concentration effect that occurs when a crack tip enters hard gravel from a soft matrix. The calculation formula is: ; Where A represents the gravel-matrix modulus heterogeneity; Indicates the elastic modulus of gravel particles; Indicates the modulus of the matrix cementitious material; Indicates the coefficient of local stress heterogeneity; This represents the empirical correction factor (0.15 in this embodiment). The higher the value (usually >3.0), the greater the resistance the cracks face to penetrating the gravel, and the more likely they are to be blocked or deflected.

[0178] Calculate the weakening potential energy B at the cemented surface: This characterizes the probability of cracks peeling off along the gravel-matrix interface and forming a fracture network. The calculation formula is: ; Where B represents the weakening potential energy of the cementation surface; Indicates the tensile strength of the interface; Represents the bulk strength of the matrix, a constant; This indicates the volume percentage of the weakly bonded interface layer; Indicates the reference strength; This represents the attenuation coefficient. When B < 0.6, it indicates that the interface is very easy to open, which is conducive to the formation of complex meshes.

[0179] Calculate the gravel topological stacking index C: This index quantifies the density of the "skeleton support" structure formed by gravel particles, indicating the risk of crack tips encountering "gravel walls." The calculation formula is: ; Where C represents the gravel topology stacking index; Indicates the volumetric content of gravel; N represents the total number of gravel in the body. This represents the contact coordination number of the i-th gravel particle (i.e., how many surrounding particles this particle is in direct contact with). When C is between 0.5 and 1, it is determined that there is an extremely high risk of congestion.

[0180] Calculate the curvature entropy D of the path around the gravel: This index describes the degree of tortuosity of the path and fluid friction when a crack is forced to bypass gravel. The calculation formula is: ; Where D represents the curvature entropy of the path around the gravel; Indicates the median particle size. Indicates the crack width; The sorting coefficient represents the grain size distribution of the gravel. Indicates the sorting coefficient; This represents the weighting coefficient. The larger the D value, the more difficult it is for the fracture to extend and the higher the near-wellbore friction.

[0181] S3: Parameter selection based on energy competition mechanism utilizes the CMI algorithm to select master control parameters. This algorithm, based on the principle of minimum energy, calculates the competition index for three expansion modes: Congestion-type competition index: ; Gravel-type competition index: ; Competition index around gravel: ; in, Indicates the competition index of the gravel-piercing mode; This indicates the gravel-matrix modulus heterogeneity after normalization. This represents the weakened potential energy of the cemented surface after normalization. Indicates the competition index of the gravel-surrounding mode; This represents the gravel topological stacking index after normalization. This represents the normalized curvature entropy of the path around the gravel. The competition index represents the congestion pattern.

[0182] The system automatically compares the three indices mentioned above. For example... Figure 3b As shown, in the third fracturing stage of this embodiment, the calculation is as follows: The system determined that the main control mechanism for this segment was "gravel congestion" because it was significantly higher than the other two. Therefore, it automatically selected C and B as the core input features and temporarily masked them. Secondary parameters are used to generate the dimensionality-reduced feature vector X.

[0183] Step S4: Fast forward simulation The feature vector X is input into a pre-set fast forward simulation engine (ROM). This engine is based on a deep neural network (DNN) architecture and has been pre-trained on thousands of high-fidelity finite element simulations. In this embodiment, the ROM engine outputs the following results under the current operating conditions within 0.3 seconds: Predicted construction pressure curve: indicates that a pressure rise may occur 15 minutes after pumping; Predicted fracture geometry: It shows that the development of fracture flanks is hindered and they mainly extend along the principal stress direction.

[0184] S5: After the closed-loop iterative correction fracturing begins, the system collects the actual pump pressure transmitted from the on-site pump truck in real time. And downhole microseismic monitoring data. The system performs a Bayesian update every 30 seconds, and the following objective function is constructed: ; in, This represents the objective function for minimizing the error. and These are the predicted pressure and the on-site pump pressure, respectively. and These are the predicted geometric parameters of the seam mesh and the on-site monitored parameters of the seam mesh, respectively. , These are the weighting coefficients; The regularization coefficient is used. is the initial reservoir parameter vector; x represents the reservoir parameter vector to be corrected in the closed loop.

[0185] like Figure 4aAs shown in the figure, the red solid line represents the actual trajectory of the construction pressure over time as monitored at the fracturing site, and the blue dashed line represents the theoretical pressure curve calculated based on the initial sandstone and conglomerate characteristic prediction model without closed-loop iterative correction. The red shaded area is enclosed by the red solid line and the blue dashed line, which intuitively shows the deviation range between the initial model prediction value and the actual measured value at the site. In the early stage of construction, it was found that the measured pressure was 2 MPa higher than the predicted pressure (also the initial simulated pressure).

[0186] like Figure 4b As shown, the measured pressure (P) 实测 ) and model-generated simulated pressure (P) 模拟 Using ) as the initial input, calculate the current model prediction value (P) 模拟 ) and actual on-site monitoring value (P) 实测 The difference between the error and the target function J(x) is used to generate the objective function J(x). Based on the calculated error J(x), the process enters the "Bayesian update" stage. Using Bayes' theorem, the current error information is transformed into a posterior probability estimate of the model parameters. Subsequently, the system executes the "update parameters (...)" step. 拥塞 +15%). This step transforms the abstract probability distribution obtained from the Bayesian update into specific physical parameter adjustment instructions. (The text in parentheses in the diagram is missing.) 拥塞 "+15%" is a specific engineering implementation example, indicating that the system determines a certain key parameter (such as crack tortuosity or stress shadowing effect coefficient) at present. The pressure (P) was too high, so a clear correction instruction was issued to reduce it by 15%. The updated parameters were immediately fed back to the "ROM simulation," and the ROM model used this new set of parameters, which better reflected the current operating conditions, to quickly recalculate a new pressure curve (P). 模拟 This new curve, once again acting as an input arrow pointing upwards, is compared with the measured data in a new round, thus initiating the next round of the "calculation-comparison-correction" cycle until the error converges.

[0187] like Figure 4c As shown, the system inversely calculates the local P of the strata by minimizing the objective function J(x). 实测 The actual value was 15% higher than the initial model. The system then corrected the parameters of the sandstone and conglomerate feature prediction model, and the updated prediction curve had an overlap of more than 96% with the measured curve.

[0188] S6: Real-time optimization of construction parameters like Figure 5a As shown, based on the revised model, the system issues a yellow alert 20 minutes into the construction process: Risk identification: The predictive model shows that the crack tip is about to enter a high A (gravel congestion) area, and the crack width shrinks sharply from 4 mm to 2.5 mm, posing a serious risk of end sand shedding.

[0189] like Figure 5b As shown, the control command is generated: The system automatically generates control commands: 1. Prop ratio: It is recommended to reduce the proppant concentration in the next stage from 180 kg / m³. 3 Temporarily reduced to 120 kg / m 3 .

[0190] 2. Increase viscosity: It is recommended to turn on the online thickener injection pump to increase the fracturing fluid viscosity by 5 mPa. s.

[0191] like Figure 5c As shown, after the on-site engineer confirmed and executed the instructions, the subsequent pressure curve transitioned smoothly without sand blockage. Ultimately, the fracturing operation in this section was successfully completed, and the sand addition reached 105% of the design value.

[0192] Implementation effect verification: The method of this embodiment was applied in the field, and the results showed that: 1. Significantly improved prediction accuracy: The error between simulated net pressure and on-site G-function analysis results is less than 7%, and the spatial matching error of microseismic event points is less than 5m.

[0193] 2. Significantly reduced engineering risks: By identifying signs of congestion 45-60 seconds in advance, the number of abnormal pressure alarms in this well area was reduced from an average of 7 per segment to 2, and no pump stoppage accidents caused by gravel congestion occurred.

[0194] 3. Increased production capacity: The average test production of the optimized fracturing section increased by 30% compared with adjacent wells, confirming the practicality and advanced nature of this method in the stimulation of highly heterogeneous sandstone and conglomerate reservoirs.

[0195] like Figure 6 As shown, this application also provides a system for optimizing the propagation of hydraulic fracturing networks, the system comprising: The data acquisition module is used to acquire oilfield data of the target sandstone and conglomerate reservoir; The physical characteristic index extraction module is used to extract physical characteristic indicators based on oilfield data to characterize the micromechanical mechanism of sandstone and conglomerate reservoirs. The screening module is used to select the main control feature vector from multiple candidate parameters based on the fracture mechanics energy competition mechanism and physical characteristic indicators. The initial prediction module is used to input the main control feature vector into the fast forward modeling engine to obtain the initial prediction results of the pressure fracture network; the prediction results of the pressure fracture network include the predicted construction pressure curve and the three-dimensional fracture network geometry. The correction module is used to collect field monitoring data and, based on the field monitoring data and the initial pressure fracture network prediction results, to perform closed-loop iterative correction of the parameters of the sandstone and conglomerate feature prediction model, and output the corrected sandstone and conglomerate feature prediction model; among which, the field monitoring data includes field pump pressure data and microseismic / fiber optic monitoring data; The optimization module is used to identify construction risks using the modified sandstone and conglomerate feature prediction model, and generate corresponding construction parameter optimization instructions based on the identified risks, so as to realize real-time control of the fracturing construction process.

[0196] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for optimizing the propagation of a hydraulic fracturing network, characterized in that, The method includes: Obtain oilfield data for the target sandstone and conglomerate reservoir; Based on oilfield data, physical characteristic indicators were extracted to characterize the micromechanical mechanism of sandstone and conglomerate reservoirs. These physical characteristic indicators include gravel-matrix modulus heterogeneity, cementation surface weakening potential energy, gravel topological stacking index, and gravel path curvature entropy. Based on the energy competition mechanism of fracture mechanics, the main control feature vector is selected from multiple candidate parameters according to physical characteristic indicators; The master control feature vector is input into the fast forward modeling engine to obtain the initial pressure crack network prediction results; the pressure crack network prediction results include the predicted construction pressure curve and the three-dimensional crack network geometry. Field monitoring data was collected, and based on the field monitoring data and the initial fracture network prediction results, the parameters of the sandstone and conglomerate feature prediction model were iteratively corrected in a closed loop, and the corrected sandstone and conglomerate feature prediction model was output. The field monitoring data included field pump pressure data and microseismic / fiber optic monitoring data. The modified sandstone and conglomerate feature prediction model is used to identify construction risks, and corresponding construction parameter optimization instructions are generated based on the identified risks to achieve real-time control of the fracturing process.

2. The method for optimizing the propagation of hydraulic fracturing networks according to claim 1, characterized in that, The oilfield data includes core data, well logging data, seismic data, and historical fracturing and production data used for model calibration and closed-loop iterative correction, which are used to construct physical characteristic indicators. The core data includes the elastic modulus of gravel particles, the modulus of matrix cement, the interfacial tensile strength, and the matrix bulk strength. The logging data includes gravel size distribution curves along the wellbore direction, gravel volume content, and sorting coefficient.

3. The method for optimizing the propagation of hydraulic fracturing networks according to claim 1, characterized in that, The expression for extracting the gravel-matrix modulus heterogeneity is as follows: ； Where A represents the gravel-matrix modulus heterogeneity; Indicates the elastic modulus of gravel particles; Indicates the modulus of the matrix cementitious material; Indicates the coefficient of local stress heterogeneity; This represents the correction factor.

4. The method for optimizing the propagation of hydraulic fracturing networks according to claim 1, characterized in that, The expression for the weakening potential energy of the cemented surface is as follows: ； Where B represents the weakening potential energy of the cementation surface; Indicates the tensile strength of the interface; Represents the bulk strength of the matrix, a constant; Indicates the volume percentage of the weak interface layer; Indicates the reference strength; This represents the attenuation coefficient.

5. The method for optimizing the propagation of hydraulic fracturing networks according to claim 1, characterized in that, The expression for extracting the gravel topology stacking index is: ； Where C represents the gravel topology stacking index; Indicates the volumetric content of gravel; N represents the total number of gravel in the body. This represents the contact coordination number of the i-th gravel particle.

6. The method for optimizing the propagation of hydraulic fracturing networks according to claim 1, characterized in that, The expression for extracting the curvature entropy of the path around the gravel is: ； Where D represents the curvature entropy of the path around the gravel; Indicates the median grain size of the gravel; Indicates the crack width; The sorting coefficient represents the grain size distribution of the gravel. Indicates the sorting coefficient; This represents the weighting coefficient.

7. The method for optimizing the propagation of hydraulic fracturing networks according to claim 1, characterized in that, Based on the energy competition mechanism of fracture mechanics, the controlling feature vector is selected from multiple candidate parameters according to physical characteristic indicators, including: Based on the energy competition mechanism of fracture mechanics, the competition index of the preset mode of crack propagation is determined according to the physical characteristic index. Compare the competition indices of various preset modes to determine the crack propagation mode of the master controller; Feature parameters corresponding to the crack propagation mode of the master are selected to form the master feature vector.

8. The method for optimizing the propagation of hydraulic fracturing networks according to claim 7, characterized in that, The preset modes include gravel-penetrating mode, gravel-circling mode, and congestion mode.

9. The method for optimizing the propagation of hydraulic fracturing networks according to claim 8, characterized in that, The expression for the competition index of the gravel-penetrating mode is as follows: ； in, Indicates the competition index of the gravel-piercing mode; This indicates the gravel-matrix modulus heterogeneity after normalization. This represents the weakened potential energy of the cemented surface after normalization.

10. The method for optimizing the propagation of hydraulic fracturing networks according to claim 8, characterized in that, The expression for determining the competition index of the gravel-surrounding mode is: ； in, Indicates the competition index of the gravel-surrounding mode; This indicates the gravel-matrix modulus heterogeneity after normalization. This represents the weakened potential energy of the cemented surface after normalization. This represents the gravel topological stacking index after normalization. This represents the curvature entropy of the path around the gravel after normalization.

11. The method for optimizing the propagation of hydraulic fracturing networks according to claim 8, characterized in that, The expression for determining the contention index of the congestion mode is: ； in, A competition index indicating the congestion pattern; This represents the gravel topological stacking index after normalization. This represents the curvature entropy of the path around the gravel after normalization.

12. The method for optimizing the propagation of hydraulic fracturing networks according to claim 1, characterized in that, The parameters of the sandstone and conglomerate feature prediction model are corrected through closed-loop iterative modification, and the corrected sandstone and conglomerate feature prediction model is output, including: An initial sandstone and conglomerate feature prediction model for the target sandstone and conglomerate reservoir is established based on oilfield data, and the initial sandstone and conglomerate feature prediction model is parameterized into a model parameter vector. During the fracturing operation, real-time pump pressure data and microseismic / fiber optic monitoring data are collected on site. The current model parameter vector is input into the fast forward modeling engine to obtain the predicted construction pressure curve and the predicted three-dimensional fracture network geometry. An objective function is constructed with the goal of minimizing the difference between the predicted results and the on-site monitoring data, and the model parameter vector is updated using a regularized gradient iteration method. When the relative change of the objective function is less than or equal to a preset value, or when the normalized gradient magnitude of the objective function is less than or equal to a preset value, the iteration stops and the corrected sandstone and conglomerate feature prediction model is output.

13. The method for optimizing the propagation of hydraulic fracturing networks according to claim 12, characterized in that, The expression for the objective function is: ； in, This represents the objective function for minimizing the error. and These are the predicted pressure and the on-site pump pressure, respectively. and These are the predicted geometric parameters of the seam mesh and the on-site monitored parameters of the seam mesh, respectively. , These are the weighting coefficients; The regularization coefficient is used. is the initial reservoir parameter vector; x represents the reservoir parameter vector to be corrected in the closed loop.

14. The method for optimizing the propagation of hydraulic fracturing networks according to claim 1, characterized in that, Construction risks were identified using a modified sandstone and conglomerate feature prediction model, including: Based on the physical characteristic indicators and the changing trends of fracture geometry output by the modified sandstone and conglomerate characteristic prediction model, it is determined whether there is a risk of sand blockage or increased friction due to excessively tortuous fracture paths.

15. The method for optimizing the propagation of hydraulic fracturing networks according to claim 14, characterized in that, Based on the changes in physical characteristic indicators and fracture geometry output by the modified sandstone and conglomerate feature prediction model, the risk of sand blockage is determined, including: When the modified sandstone and conglomerate feature prediction model shows that, within a preset time window, the increase in the physical characteristic index value representing the heterogeneity of the gravel-matrix modulus relative to the previous time window reaches a first preset threshold, and the predicted crack width reaches a second preset threshold, it is determined that there is a risk of sand blockage.

16. The method for optimizing the propagation of hydraulic fracturing networks according to claim 14, characterized in that, Based on the physical characteristic indicators and fracture geometry trends output by the modified sandstone and conglomerate feature prediction model, it is determined whether there is a risk of increased friction due to excessively tortuous fracture paths, including: When the modified sandstone and conglomerate feature prediction model shows that the physical characteristic index value characterizing the curvature entropy of the path around the gravel reaches the third preset threshold, or its relative increase within the preset time window reaches the fourth preset threshold, it is determined that there is a risk of increased friction due to excessive tortuosity of the crack path.

17. The method for optimizing the propagation of hydraulic fracturing networks according to claim 1, characterized in that, Based on the identified risks, corresponding construction parameter optimization instructions are generated, including: When a risk is identified as sand blockage risk, instructions are generated to reduce proppant concentration and / or increase fracturing fluid viscosity; When a risk is identified as increased friction due to excessively tortuous crack paths, an instruction is generated to increase the pump flow rate.

18. A system for optimizing the propagation of hydraulic fracturing networks, characterized in that, The system includes: The data acquisition module is used to acquire oilfield data of the target sandstone and conglomerate reservoir; The physical characteristic index extraction module is used to extract physical characteristic indexes based on oilfield data to characterize the micromechanical mechanism of sandstone and conglomerate reservoirs. The physical characteristic indexes include gravel-matrix modulus heterogeneity, cementation surface weakening potential energy, gravel topological stacking index, and perigravel path curvature entropy. The screening module is used to select the main control feature vector from multiple candidate parameters based on the fracture mechanics energy competition mechanism and physical characteristic indicators. The initial prediction module is used to input the main control feature vector into the fast forward modeling engine to obtain the initial prediction results of the pressure fracture network; the prediction results of the pressure fracture network include the predicted construction pressure curve and the three-dimensional fracture network geometry. The correction module is used to collect field monitoring data and, based on the field monitoring data and the initial pressure fracture network prediction results, to perform closed-loop iterative correction of the parameters of the sandstone and conglomerate feature prediction model, and output the corrected sandstone and conglomerate feature prediction model; among which, the field monitoring data includes field pump pressure data and microseismic / fiber optic monitoring data; The optimization module is used to identify construction risks using the modified sandstone and conglomerate feature prediction model, and generate corresponding construction parameter optimization instructions based on the identified risks, so as to realize real-time control of the fracturing construction process.

Images