Coal mine dynamic disaster hydraulic fracturing optimization method based on global model segmentation

Abstract

The invention provides a coal mine dynamic disaster hydraulic fracturing optimization method based on global model segmentation. Comprising the steps of global model and stress result generation, sub-model cutting and stress inheriting, borehole parameterization information modification, fracture randomness simulation, injection pressure and fracturing fluid material parameter loading, high-precision structural grid division, model operation to generate a stress nephogram and data extraction, digital twinborn construction, and AI proxy model training and optimization decision. And an optimized drilling parameter generation and parameter feedback and iterative optimization step. According to the method, accurate control over the hydraulic fracturing process is achieved by simulating the fracture extension process and intelligently optimizing the injection pressure and fracturing fluid parameters, the rock burst prevention and control capacity of hydraulic fracturing is improved through intelligentization and dynamic optimization, and therefore the risk of rock burst is effectively reduced, and the prevention and control effect and the safety of resource exploitation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent prevention and control of coal mine disasters, and in particular to a coal mine dynamic disaster hydraulic fracturing optimization method based on global model segmentation. Background Art

[0002] Rock burst, one of the most sudden and devastating dynamic hazards in coal mining, poses a serious threat to mine safety and the lives of workers. Rock burst is particularly frequent, sudden, and destructive in deep, high-stress, and complex geological structures, where stress builds up in the rock mass and deformation and damage are severe. To effectively reduce localized high-stress areas and promote early energy release in the rock mass, active pressure relief technologies such as hydraulic fracturing, drilling and blasting, and borehole unloading are widely used in current engineering practices.

[0003] Hydraulic fracturing, currently one of the more effective means of preventing and relieving rock bursts, injects high-pressure fracture fluids into coal and rock masses to induce fracture expansion, reduce overall regional stiffness, and disperse concentrated stress, thereby reducing the risk of rock bursts. However, the inventors of this application have discovered that existing hydraulic fracturing technology still faces the following prominent issues when applied in complex coal and rock environments:

[0004] 1. Lack of multi-scale high-precision modeling capabilities: Most existing methods use single-scale, idealized and simplified modeling methods, which are difficult to truly reflect the stress evolution process under the complex geological structure and mining disturbance conditions of the mining area. This leads to deviations in the actual implementation of the fracturing design plan and affects the unloading effect.

[0005] 2. The sub-model and the working condition cannot be coupled for optimization: In traditional models, it is difficult for the global model to accurately transmit the true boundary stress state to the local fracturing model. The stress initialization of the sub-model is often disconnected from the working condition setting, making it impossible to achieve true inheritance and synchronous adjustment of fracturing parameters, which limits the accuracy and practicality of the model.

[0006] 3. Fracturing parameters rely on manual experience: Key technical indicators such as injection pressure, drilling spacing, and fracturing fluid parameters often rely on field experience, analogy, or repeated adjustments through a large number of experiments. The lack of a systematic, data-driven optimization mechanism leads to unstable and unreliable design schemes.

[0007] 4. Lack of real-time feedback and dynamic decision-making capabilities: Traditional hydraulic fracturing operations have not established a real-time monitoring and feedback model coupled with stress evolution. It is difficult to dynamically adjust fracturing parameters based on multi-source data obtained during the operation. The fracturing process control lags and is obviously "passive" and "closed".

[0008] 5. Lack of intelligent integrated system support: Current fracturing design methods have not fully integrated advanced technologies such as 3D geological modeling, image analysis, deep learning, and big data mining, and lack an integrated intelligent auxiliary design system of "modeling-simulation-optimization".

[0009] In summary, there is an urgent need for a new technical method that integrates the global geomechanical model with sub-model cutting inheritance, digital twin construction, AI agent model-driven optimization and multi-source data feedback mechanism to achieve high-precision modeling, dynamic adjustment and intelligent optimization of hydraulic fracturing parameters. Summary of the Invention

[0010] In response to the technical problems in existing hydraulic fracturing methods, such as the lack of accurate modeling of the fracturing process, reliance on experience for fracturing parameter adjustment, and inability to provide real-time feedback and adjustment, the present invention provides a coal mine dynamic disaster hydraulic fracturing optimization method based on global model segmentation. By simulating the crack expansion process and intelligently optimizing the injection pressure and fracturing fluid parameters, precise control of the hydraulic fracturing process is achieved, thereby effectively reducing the risk of rock burst, improving prevention and control effects, and the safety of resource extraction.

[0011] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0012] The coal mine power disaster hydraulic fracturing optimization method based on global model segmentation includes the following steps:

[0013] S1. Global model and stress result generation:

[0014] S11. Based on the existing geological structural parameters, a global geomechanical model of the coal mining area is constructed. The input parameters include rock stratum mechanical properties, ground stress field parameters, geometric boundary conditions and structural joint information.

[0015] S12. Use finite element numerical simulation software to perform static or quasi-static analysis on the global geomechanical model, calculate the distribution of the ground stress field, and obtain the global stress result σ global (x,y,z), the stress results of a single node are expressed by the following stress tensor expression:

[0016]

[0017] Among them, σ i is the stress tensor of the i-th element or node; σ xx , σ yy , σ zz is the stress component on the main diagonal, representing the normal stress acting in the x, y, and z directions; τ xy , τ xz , τ yx , τ yz , τ zx, τ zy is the stress component on the non-diagonal line, indicating the shear force;

[0018] σ global (x,y,z)=σ i {σ i |i=1,2,3…,N}

[0019] S2. Submodel cutting and stress inheritance:

[0020] S21. Based on the calculation results of the global geomechanical model and the mining layout, a local sub-model is cut out in the high-risk area of ​​rock burst based on the global geomechanical model in step S1. The sub-model size meets the requirements for covering the drilling, hydraulic fracturing and surrounding impact zones.

[0021] S22, the sub-model inherits the boundary stress field of the global geomechanical model as the boundary condition input, and the sub-model stress field σ sub (x,y,z) is guaranteed to be consistent with σ through interpolation and mapping global (x, y, z) are consistent, and dynamic modification of working conditions is allowed;

[0022] S3. Modification of drilling parameter information:

[0023] Based on existing hydraulic fracturing design or empirical data, the load includes the position coordinates (x i ,y i ), aperture Φ i , depth l i , spacing d i , injection pressure p i , fracturing fluid material property parameters including viscosity μ, and the sub-model of step S2 is parameterized. i , density ρ i , rheological parameter α i Embed the borehole geometry into the sub-model grid using geometric modeling tools; modify the model boundary conditions and material properties to simulate the impact of the borehole on ground stress and structural integrity;

[0024] S4. Simulating crack randomness:

[0025] In the sub-model of step S3, a random fracture surface is constructed to simulate the non-structural distribution of natural joints or fracture-induced fractures in the field. The fractures can be expressed by ellipsoids, fracture clouds, or randomly distributed surfaces. The fracture surface is described in the following way:

[0026]

[0027] Among them, A i represents the influence intensity coefficient of the i-th crack; Indicates that the crack is at σ x Scale control parameters in direction; Indicates that the crack is at σ y Scale control parameters in direction; Indicates that the crack is at σ z Scale control parameters in direction;

[0028] S5. Loading injection pressure and fracturing fluid material parameters:

[0029] In step S4, set the boundary loading condition, injection pressure boundary p inj (t) and the rheological properties of the fracturing fluid material, the shear thinning is described using a power law model:

[0030]

[0031] Where τ is the shear stress, K is the rheological parameter, is the shear rate;

[0032] S6. High-precision structural meshing:

[0033] For the sub-model adjusted in step S5, structured meshing technology is used to achieve high-precision meshing, and the mesh size is adaptively refined according to the drill hole size and stress gradient area to ensure calculation accuracy;

[0034] S7. Run the model to generate stress cloud diagram and extract data:

[0035] S71: Run the sub-model after meshing in step S6 to calculate the stress evolution process under hydraulic fracturing and obtain the three-dimensional stress tensor σ frac (x, y, z, t); where t represents the dynamic evolution of the stress tensor during the fracturing loading process;

[0036] S72. Extract the coordinates and principal stress values ​​of key nodes to form the following data set:

[0037] D frac ={(x j ,y j ,z j ,σ j )|j=1,2,…,N}

[0038] Among them, (x j ,y j ,z j ) represents the three-dimensional space coordinates of node j, σ j represents the stress value of node j;

[0039] Use cloud maps and post-processing technology to analyze stress concentration and changes in key areas;

[0040] S8. Digital Twin Construction:

[0041] S81, input the point coordinates and stress value I1 in step S7 = {(x j ,y j ,z j ,σ j )};

[0042] Structuring discrete stress data into a three-dimensional matrix through three-dimensional interpolation methods Represents a structured three-dimensional tensor, which represents the stress value of each voxel in a three-dimensional space grid;

[0043] A three-dimensional convolutional neural network is used to extract features and obtain a high-dimensional feature tensor F 3D =3DCNN(M 3D ), 3DCNN represents three-dimensional convolutional neural network;

[0044] Use 3D pooling layer to reduce dimension, reduce feature dimension, and obtain dimensionality reduction features

[0045] The reduced dimensionality features are further extracted through a two-dimensional convolutional neural network to obtain a two-dimensional feature tensor. 2DCNN stands for two-dimensional convolutional neural network;

[0046] S82, input drilling parameter matrix I2 = {(x i ,y i ,Φ i ,l i ,d i ,p i ,μ i ,α i )};

[0047] Map the drilling parameter matrix into a two-dimensional matrix format and input it into 2DCNN to extract the feature F hole =2DCNN(I2);

[0048] S83. Feature fusion and matrix generation:

[0049] The feature F of step S81 2D and feature F of step S82 hole Perform feature fusion F fuse =F use (F 2D ,F hole ), F use Represents the feature fusion operation, which is to stack the features and then perform convolution fusion;

[0050] Generate a 2D stress-fracture impact matrix f represents the fused output feature tensor;

[0051] Perform a three-dimensional deconvolution operation on the two-dimensional stress-fracture influence matrix to reconstruct the three-dimensional stress-hydraulic fracturing matrix Deconv means deconvolution;

[0052] S9. AI agent model training and optimization decision-making:

[0053] S91, using the real stress results, drilling parameters and fracturing conditions obtained in steps S2 and S7 as supervision signals, construct an AI agent model F with the following expression: AI :

[0054]

[0055] Among them, θ is the model parameter;

[0056] S92, reconstructing the three-dimensional stress-hydraulic fracturing matrix of the digital twin in step S8 As training data for the AI ​​agent model, the AI ​​agent model parameters θ are optimized through gradient descent or genetic algorithm. The training goal is to minimize the error between the predicted stress and the actual stress:

[0057]

[0058] in, represents the pressure relief effect index of the i-th sample predicted by the surrogate model; represents the real pressure relief effect index of the i-th sample;

[0059] S10. Generation of optimized drilling parameters:

[0060] Based on the trained AI agent model, an evolutionary search strategy is used to optimize drilling parameters and fracturing conditions to generate an optimized parameter set.

[0061] S11. Parameter feedback and iterative optimization:

[0062] The optimization parameters generated in step S10 are fed back to step S3 to modify the sub-model working conditions and perform the next round of simulation and stress analysis. This cycle is repeated until the pressure relief effect meets the optimization convergence conditions and the optimal hydraulic fracturing parameter combination is obtained.

[0063] Furthermore, the meshing in step S6 utilizes volume units or hexahedron-dominated meshes to ensure simulation stability.

[0064] Compared with the existing technology, the coal mine dynamic disaster hydraulic fracturing optimization method based on global model segmentation provided by the present invention accurately controls various parameters of the hydraulic fracturing process by combining digital twin technology with AI intelligent optimization algorithm, and has the following significant advantages: 1) Precise control: By simulating the crack expansion process through digital twins, the fracturing effect can be evaluated more accurately, thereby improving the control accuracy of hydraulic fracturing. 2) Intelligent optimization: The AI ​​algorithm dynamically optimizes parameters such as injection pressure and fracturing fluid materials, and adjusts the fracturing scheme according to real-time feedback, effectively reducing dependence on traditional experience. 3) Real-time adjustment: Based on real-time data of mine geological conditions, the fracturing scheme is quickly responded to and adjusted, effectively reducing the risk of rock burst and improving the prevention and control effect. Therefore, this method improves the ability of hydraulic fracturing to prevent and control rock burst through intelligence and dynamic optimization, and enhances the safety and economy of coal mining. DETAILED DESCRIPTION

[0065] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below.

[0066] The present invention provides a coal mine power disaster hydraulic fracturing optimization method based on global model segmentation, comprising the following steps:

[0067] S1. Global model and stress result generation:

[0068] S11. Based on the existing geological structural parameters, a global geomechanical model of the coal mining area is constructed. The input parameters include rock mechanical properties (elastic modulus E, Poisson's ratio v), ground stress field parameters (initial stress tensor σ0), geometric boundary conditions and structural joint information;

[0069] S12. Use finite element numerical simulation software such as Abaqus to perform static or quasi-static analysis on the global geomechanical model, calculate the distribution of the ground stress field, and obtain the global stress result σ global (x,y,z), the stress results of a single node are expressed by the following stress tensor expression:

[0070]

[0071] Among them, σ i is the stress tensor of the i-th element or node; σ xx , σ yy , σ zz is the stress component on the main diagonal, representing the normal stress acting in the x, y, and z directions; τ xy , τ xz , τ yx , τ yz , τ zx , τ zyis the stress component on the non-diagonal line, indicating the shear force;

[0072] σ global (x,y,z)=σ i {σ i |i=1,2,3…,N}.

[0073] S2. Submodel cutting and stress inheritance:

[0074] S21. Based on the calculation results of the global geomechanical model and the mining layout, a local sub-model is cut out in the high-risk area of ​​rock burst based on the global geomechanical model in step S1. The sub-model size meets the requirements for covering the drilling, hydraulic fracturing and surrounding impact zones.

[0075] S22, the sub-model inherits the boundary stress field of the global geomechanical model as the boundary condition input, and the sub-model stress field σ sub (x,y,z0 are guaranteed to be consistent with σ through interpolation and mapping global (x, y, z0 are consistent, and dynamic modification of working conditions is allowed; unlike existing technologies, sub-model working conditions are dynamically modified and inherit stress, achieving more accurate simulation.

[0076] S3. Modification of drilling parameter information:

[0077] Based on existing hydraulic fracturing design or empirical data, the load includes the position coordinates (x i ,y i ), aperture Φ i , depth l i , spacing d i , injection pressure p i , fracturing fluid material property parameters including viscosity μ, and the sub-model of step S2 is parameterized. i , density ρ i , rheological parameter α i etc.; embed the borehole geometry into the sub-model mesh through geometric modeling tools such as Abaqus; modify the model boundary conditions and material properties to simulate the impact of the borehole on ground stress and structural integrity.

[0078] S4. Simulating crack randomness:

[0079] In the sub-model of step S3, a random fracture surface is constructed to simulate the non-structural distribution of natural joints or fracture-induced fractures in the field. The fractures can be expressed by ellipsoids, fracture clouds, or randomly distributed surfaces. The fracture surface is described in the following way:

[0080]

[0081] Among them, A irepresents the influence intensity coefficient of the i-th crack; Indicates that the crack is at σ x Scale control parameters in direction; Indicates that the crack is at σ y Scale control parameters in direction; Indicates that the crack is at σ z Scale control parameter in direction.

[0082] S5. Loading injection pressure and fracturing fluid material parameters:

[0083] In step S4, set the boundary loading condition, injection pressure boundary p inj (t) and the rheological properties of the fracturing fluid material, the shear thinning is described using a power law model:

[0084]

[0085] Where τ is the shear stress, K is the rheological parameter, is the shear rate.

[0086] S6. High-precision structural meshing:

[0087] For the sub-model adjusted in step S5, a structured meshing technique is used to achieve high-precision meshing. The mesh size is adaptively refined based on the drill hole size and stress gradient area to ensure calculation accuracy. Preferably, the meshing is performed using volume elements or hexahedron-dominated meshes to ensure simulation stability.

[0088] S7. Run the model to generate stress cloud diagram and extract data:

[0089] S71: Run the sub-model after meshing in step S6 to calculate the stress evolution process under hydraulic fracturing and obtain the three-dimensional stress tensor σ frac (x, y, z, t); where t represents the dynamic evolution of the stress tensor during the fracturing loading process;

[0090] S72. Extract the coordinates and principal stress values ​​of key nodes to form the following data set:

[0091] D frac ={(x j ,y j ,z j ,σ j )|j=1,2,…,N}

[0092] Among them, (x j ,y j ,z j ) represents the three-dimensional space coordinates of node j, σ j represents the stress value of node j;

[0093] Use cloud maps and post-processing technology to analyze stress concentration and changes in key areas.

[0094] S8. Digital Twin Construction:

[0095] S81, input the point coordinates and stress value I1 in step S7 = {(x j ,y j ,z j ,σ j )};

[0096] Structuring discrete stress data into a three-dimensional matrix through three-dimensional interpolation methods Represents a structured three-dimensional tensor, which represents the stress value of each voxel in a three-dimensional space grid;

[0097] A three-dimensional convolutional neural network is used to extract features and obtain a high-dimensional feature tensor F 3D =3DCNN(M 3D ), 3DCNN represents three-dimensional convolutional neural network;

[0098] Use 3D pooling layer to reduce dimension, reduce feature dimension, and obtain dimensionality reduction features

[0099] The reduced dimensionality features are further extracted through a two-dimensional convolutional neural network to obtain a two-dimensional feature tensor. 2DCNN stands for two-dimensional convolutional neural network;

[0100] Through the above process, step S81 can achieve structural processing of the stress results of the sub-model;

[0101] S82, input drilling parameter matrix I2 = {(x i ,y i ,Φ i ,l i ,d i ,p i ,μ i ,α i )};

[0102] Map the drilling parameter matrix into a two-dimensional matrix format and input it into 2DCNN to extract the feature F hole =2DCNN(I2);

[0103] Through the above process, step S82 can realize the extraction of drilling parameter features;

[0104] S83. Feature fusion and matrix generation:

[0105] The feature F of step S81 2Dand feature F of step S82 hole Perform feature fusion F fuse =F use (F 2D ,F hole ), F use Represents the feature fusion operation, which is to stack the features and then perform convolution fusion;

[0106] Generate a 2D stress-fracture impact matrix f represents the fused output feature tensor;

[0107] Perform a three-dimensional deconvolution operation on the two-dimensional stress-fracture influence matrix to reconstruct the three-dimensional stress-hydraulic fracturing matrix Deconv means deconvolution.

[0108] S9. AI agent model training and optimization decision-making:

[0109] S91, using the real stress results, drilling parameters and fracturing conditions obtained in steps S2 and S7 as supervision signals, construct an AI agent model F with the following expression: AI :

[0110]

[0111] Among them, θ is the model parameter;

[0112] S92, reconstructing the three-dimensional stress-hydraulic fracturing matrix of the digital twin in step S8 As training data for the AI ​​agent model, the AI ​​agent model parameters θ are optimized through gradient descent or genetic algorithm. The training goal is to minimize the error between the predicted stress and the actual stress:

[0113]

[0114] in, represents the pressure relief effect index of the i-th sample predicted by the surrogate model; Represents the true pressure relief effect index of the i-th sample.

[0115] S10. Generation of optimized drilling parameters:

[0116] Based on the trained AI agent model, an evolutionary search strategy is used to optimize drilling parameters and fracturing conditions to generate an optimized parameter set.

[0117] S11. Parameter feedback and iterative optimization:

[0118] The optimization parameters generated in step S10 are fed back to step S3 to modify the sub-model working conditions and perform the next round of simulation and stress analysis. This cycle is repeated (steps S3 to S10 are repeated) until the pressure relief effect meets the optimization convergence conditions and the optimal hydraulic fracturing parameter combination is obtained.

[0119] Compared with the existing technology, the coal mine dynamic disaster hydraulic fracturing optimization method based on global model segmentation provided by the present invention accurately controls various parameters of the hydraulic fracturing process by combining digital twin technology with AI intelligent optimization algorithm, and has the following significant advantages: 1) Precise control: By simulating the crack expansion process through digital twins, the fracturing effect can be evaluated more accurately, thereby improving the control accuracy of hydraulic fracturing. 2) Intelligent optimization: The AI ​​algorithm dynamically optimizes parameters such as injection pressure and fracturing fluid materials, and adjusts the fracturing scheme according to real-time feedback, effectively reducing dependence on traditional experience. 3) Real-time adjustment: Based on real-time data of mine geological conditions, the fracturing scheme is quickly responded to and adjusted, effectively reducing the risk of rock burst and improving the prevention and control effect. Therefore, this method improves the ability of hydraulic fracturing to prevent and control rock burst through intelligence and dynamic optimization, and enhances the safety and economy of coal mining.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A coal mine dynamic disaster hydraulic fracturing optimization method based on global model segmentation, characterized by: The following steps are involved: S1. Global model and stress result generation: S11. Based on the existing geological structural parameters, a global geomechanical model of the coal mining area is constructed. The input parameters include rock stratum mechanical properties, ground stress field parameters, geometric boundary conditions and structural joint information. S12. Use finite element numerical simulation software to perform static or quasi-static analysis on the global geomechanical model, calculate the distribution of the ground stress field, and obtain the global stress result σ global (x,y,z), the stress results of a single node are expressed by the following stress tensor expression: Among them, σ i is the stress tensor of the i-th element or node; σ xx , σ yy , σ zz is the stress component on the main diagonal, representing the normal stress acting in the x, y, and z directions; τ xy , τ xz , τ yx , τ yz , τ zx , τ zy is the stress component on the non-diagonal line, indicating the shear force; s global (x,y,z)=σ i {s} i |i=1,2,3…,N} S2. Submodel cutting and stress inheritance: S21. Based on the calculation results of the global geomechanical model and the mining layout, a local sub-model is cut out in the high-risk area of ​​rock burst based on the global geomechanical model in step S1. The sub-model size meets the requirements for covering the drilling, hydraulic fracturing and surrounding impact zones. S22, the sub-model inherits the boundary stress field of the global geomechanical model as the boundary condition input, and the sub-model stress field σ sub (x,y,z) is guaranteed to be consistent with σ through interpolation and mapping global (x, y, z) are consistent, and dynamic modification of working conditions is allowed; S3. Modification of drilling parameter information: Based on existing hydraulic fracturing design or empirical data, the load includes the position coordinates (x i ,y i ), aperture Φ i , depth l i , spacing d i , injection pressure p i , fracturing fluid material property parameters including viscosity μ, and the sub-model of step S2 is parameterized. i , density ρ i , rheological parameter α i Embed the borehole geometry into the sub-model grid using geometric modeling tools; modify the model boundary conditions and material properties to simulate the impact of the borehole on ground stress and structural integrity; S4. Simulating crack randomness: In the sub-model of step S3, a random fracture surface is constructed to simulate the non-structural distribution of natural joints or fracture-induced fractures in the field. The fractures can be expressed by ellipsoids, fracture clouds, or randomly distributed surfaces. The fracture surface is described in the following way: Among them, A i represents the influence intensity coefficient of the i-th crack; Indicates that the crack is at σ x Scale control parameters in direction; Indicates that the crack is at σ y Scale control parameters in direction; Indicates that the crack is at σ z Scale control parameters in direction; S5. Loading injection pressure and fracturing fluid material parameters: In step S4, set the boundary loading condition, injection pressure boundary p inj (t) and the rheological properties of the fracturing fluid material, the shear thinning is described using a power law model: Where τ is the shear stress, K is the rheological parameter, is the shear rate; S6. High-precision structural meshing: For the sub-model adjusted in step S5, structured meshing technology is used to achieve high-precision meshing, and the mesh size is adaptively refined according to the drill hole size and stress gradient area to ensure calculation accuracy; S7. Run the model to generate stress cloud diagram and extract data: S71: Run the sub-model after meshing in step S6 to calculate the stress evolution process under hydraulic fracturing and obtain the three-dimensional stress tensor σ frac (x, y, z, t); where t represents the dynamic evolution of the stress tensor during the fracturing loading process; S72. Extract the coordinates and principal stress values ​​of key nodes to form the following data set: D frac {(x j ,y j ,z j ,σ j )|j=1,2,…,N] Among them, (x j ,y j ,z j ) represents the three-dimensional space coordinates of node j, σ j represents the stress value of node j; Use cloud maps and post-processing technology to analyze stress concentration and changes in key areas; S8. Digital Twin Construction: S81, input the point coordinates and stress value I1 in step S7 = {(x j ,y j ,z j ,σ j )}; Structuring discrete stress data into a three-dimensional matrix through three-dimensional interpolation methods Represents a structured three-dimensional tensor, which represents the stress value of each voxel in a three-dimensional space grid; A three-dimensional convolutional neural network is used to extract features and obtain a high-dimensional feature tensor F 3D =3DCNN(M 3D ), 3DCNN represents three-dimensional convolutional neural network; Use 3D pooling layer to reduce dimension, reduce feature dimension, and obtain dimensionality reduction features The reduced dimensionality features are further extracted through a two-dimensional convolutional neural network to obtain a two-dimensional feature tensor. 2DCNN stands for two-dimensional convolutional neural network; S82, input drilling parameter matrix I2 = {(x i ,y i ,Φ i ,l i ,d i ,p i ,μ i ,α i )}; Map the drilling parameter matrix into a two-dimensional matrix format and input it into 2DCNN to extract the feature F hole =2DCNN(I2); S83. Feature fusion and matrix generation: The feature F of step S81 2D and feature F of step S82 hole Perform feature fusion F fuse =F use (F 2D ,F hole ), F use Represents the feature fusion operation, which is to stack the features and then perform convolution fusion; Generate a 2D stress-fracture impact matrix f represents the fused output feature tensor; Perform a three-dimensional deconvolution operation on the two-dimensional stress-fracture influence matrix to reconstruct the three-dimensional stress-hydraulic fracturing matrix Deconv means deconvolution; S9. AI agent model training and optimization decision-making: S91, using the real stress results, drilling parameters and fracturing conditions obtained in steps S2 and S7 as supervision signals, construct an AI agent model F with the following expression: AI : Among them, θ is the model parameter; S92, reconstructing the three-dimensional stress-hydraulic fracturing matrix of the digital twin in step S8 As training data for the AI ​​agent model, the AI ​​agent model parameters θ are optimized through gradient descent or genetic algorithm. The training goal is to minimize the error between the predicted stress and the actual stress: in, represents the pressure relief effect index of the i-th sample predicted by the surrogate model; represents the real pressure relief effect index of the i-th sample; S10. Generation of optimized drilling parameters: Based on the trained AI agent model, an evolutionary search strategy is used to optimize drilling parameters and fracturing conditions to generate an optimized parameter set. S11. Parameter feedback and iterative optimization: The optimization parameters generated in step S10 are fed back to step S3 to modify the sub-model working conditions and perform the next round of simulation and stress analysis. This cycle is repeated until the pressure relief effect meets the optimization convergence conditions and the optimal hydraulic fracturing parameter combination is obtained.

2. The method for optimizing hydraulic fracturing of coal mine power disasters based on global model segmentation according to claim 1, characterized in that: In step S6, the meshing is performed using volume elements or hexahedron-dominated meshes to ensure simulation stability.

Images