A gas extraction and grouting hole sealing simulation test method

Abstract

The application provides a gas extraction and grouting hole sealing simulation test method, and relates to the technical field of rock mass mechanics experiment. The method combines specific test steps such as field stress evolution path acquisition, true triaxial test, gas extraction simulation test, fracture rock mass grouting and hole sealing effect verification, fuses multiple system data of each test stage through a Bayesian fusion model, carries out data anomaly detection in combination with isolated forest and DBSCAN algorithm, evaluates the reliability of experimental results by using Monte Carlo simulation, supplements experimental data by using Latin hypercube sampling, and realizes closed-loop optimization of the experimental process based on reinforcement learning. The application can effectively improve the sample state judgment accuracy, improve the reliability of the experimental results, and optimize the experimental design, thereby providing accurate guidance for the data and test scheme optimization of the gas extraction and grouting hole sealing simulation test.

Description

Technical Field

[0001] This application relates to the field of experimental technology in geotechnical engineering and mining engineering, and in particular to a simulation test method for gas extraction and grouting sealing. Background Technology

[0002] In the process of deep mineral resource mining, with increasing burial depth, environmental conditions such as in-situ stress, pore fluid pressure, and temperature become increasingly complex, and the mechanical and seepage environment of the rock mass exhibits significant multi-field coupling characteristics. Among these, fluid-solid coupling is one of the key mechanisms inducing major engineering disasters such as coal and gas outbursts, roadway surrounding rock instability, and water inrush. To reveal the occurrence mechanism of such disasters and optimize prevention and control measures, it is particularly important to conduct multi-physics field coupling simulation experiments that can realistically reflect actual working conditions. Especially in the field of coal mine safety production, gas disasters have always been one of the main threats restricting efficient and safe mining, and efficient gas extraction and reliable grouting sealing technology are the core means to achieve gas control.

[0003] Currently, existing technologies employ true triaxial experimental systems with multiphysics coupling to perform such simulations, such as the multiphysics coupling true triaxial (grouting) experimental devices disclosed in patents CN119375041A and CN104792682B. In multiphysics coupling true triaxial experiments, experimental design, parameter control (loading pressure, gas flow rate, flow velocity, etc.), and data processing are crucial aspects of experimental control.

[0004] Current multiphysics coupled true triaxial experiments suffer from several problems, including experimental design that doesn't align with reality, insufficient multiphysics coupling simulation capabilities, difficulty in fusing multi-source data, inadequate outlier identification in data processing, lack of quantitative analysis of the reliability of permeability calculations, reliance on experience for optimization, high costs, and low efficiency. More significantly, the current optimization process for experimental parameters heavily relies on researchers' experience, lacking systematic and intelligent optimization strategies. This results in long experimental cycles, high costs, and low efficiency, making it difficult to achieve coordinated optimization of multiple objectives and parameters. Especially in simulations involving nonlinear and strongly coupled processes such as gas extraction and grouting sealing, the sensitivity of parameter combinations is high, and traditional trial-and-error optimization methods are insufficient to meet the demands for efficient and accurate experiments. Therefore, there is an urgent need to construct a comprehensive simulation experimental method that considers multiple factors to provide more reliable support for both indoor experiments and field practice. Summary of the Invention

[0005] To address the problems in existing simulation tests, such as large deviations from actual field conditions, difficulty in fusing multi-source data, inaccurate outlier identification, lack of quantitative assessment of result reliability, and low efficiency in parameter optimization, this application aims to provide a simulation test method for gas drainage and grouting sealing. This method aims to solve or alleviate the problems existing in the prior art, improve the accuracy, reliability, and efficiency of simulation tests, and provide a more scientific theoretical basis and technical support for on-site gas drainage and grouting sealing projects.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] A simulation test method for gas extraction and grouting sealing includes the following steps:

[0008] S1. Conduct stress monitoring at the target site to obtain the stress evolution path during the excavation and mining disturbance of the test roadway; the stress evolution path refers to the stress history curve of the entire process from the original rock stress → excavation and mining disturbance → steady state or instability, which is an important basis for applying confining pressure and is used to realistically reproduce the stress environment at the site;

[0009] Representative coal and rock samples were selected from the target site and prepared.

[0010] S2. Conduct a true triaxial test based on the stress evolution path to simulate the mining disturbance process, and monitor the fracture propagation of the coal and rock samples using an acoustic emission device; collect the physical and mechanical parameters of the coal and rock samples, including stress, strain, elastic modulus, Poisson's ratio, peak strain, and fracture initiation stress.

[0011] S3. After the coal and rock samples from the true triaxial test are degassed under vacuum under constant temperature conditions, simulated gas is slowly introduced and pressurized stepwise at a rate of 0.05-0.1 MPa / min to the target pressure matching the burial depth at the site. Then, the first gas extraction simulation test is carried out. Data from the first gas extraction simulation test are collected, including gas pressure and gas flow rate (after the extraction is stabilized).

[0012] S4. Grouting is performed on the coal and rock samples extracted in S3 to simulate the sealing test after on-site gas extraction, and sealing samples are obtained. At the same time, grouting pressure, grouting flow rate and diffusion radius are collected.

[0013] S5. After grouting is completed, gas simulation is introduced into the sealing sample again according to the test conditions of the first gas extraction simulation test in S3 to conduct the second gas extraction simulation test; data from the second gas extraction simulation test are collected, including gas pressure and gas flow rate (after extraction is stabilized).

[0014] S6. Calculate the permeability of the coal and rock sample before grouting. and permeability after grouting , , Following a normal distribution; Monte Carlo simulation was used to evaluate the reliability of permeability in the simulated gas extraction test. If the permeability after grouting... The 95% confidence interval and the permeability before grouting If the 95% confidence intervals do not overlap, the sealing effect is reliable; otherwise, a supplementary test should be conducted to verify the sealing effect again.

[0015] Furthermore, in step S6, a Bayesian fusion model is used to fuse the collected data from various sources and to calculate the permeability before grouting. and permeability after grouting .

[0016] Furthermore, the input data of the Bayesian fusion model includes: stress and strain data from the true triaxial test, gas pressure and gas flow rate data from the first gas extraction simulation test, grouting pressure-flow rate-time curve data during the grouting process, and gas pressure and gas flow rate data from the second gas extraction simulation test.

[0017] Furthermore, the construction process of the Bayesian fusion model includes:

[0018] A multi-stage data association model was established, and a prior probability distribution was constructed through a historical experimental database. Among them, the stress-crack development association parameter adopted a Beta distribution, the gas pressure-flow coupling parameter of the gas simulation adopted a log-normal distribution, and the mapping relationship between grouting parameters and sealing effect adopted a Dirichlet distribution.

[0019] By constructing a Markov chain, based on the current parameter state and candidate parameters conforming to a specific distribution, and combining prior probability and likelihood function to calculate acceptance probability, parameters are iteratively updated. After a sufficient number of samplings, the chain reaches a stationary distribution, thereby obtaining a posterior sample distribution (posterior probability) that reflects the true distribution characteristics of the parameters. Based on the posterior sample distribution, the rock mass state of coal and rock samples is assessed and parameters are inverted. By calculating the statistics of the posterior samples, the permeability before grouting is assessed. Permeability after grouting The key indicators of rock mass damage degree and grouting sealing efficiency of coal and rock samples are quantitatively estimated, and the reliability of the estimate is quantified by the dispersion of the posterior sample distribution.

[0020] Furthermore, before using the Bayesian fusion model to process the data in step S6, the isolated forest algorithm and the DBSCAN algorithm are first used to detect outliers in the experimental data, remove invalid data, and mark key physical mutations.

[0021] Furthermore, for stress data during the true triaxial test, the isolated forest algorithm is used to calculate anomaly scores to identify stress mutation anomalies; for strain data, gas pressure and gas flow data in the first gas extraction simulation test, and gas pressure and gas flow data in the second gas extraction simulation test, the DBSCAN algorithm is used to identify outliers; for grouting pressure-flow-time curve data during the grouting process, a fusion rule is used to comprehensively determine abnormal data.

[0022] Furthermore, the step of comprehensively judging abnormal data using the fusion rules is as follows: the isolated forest algorithm and the DBSCAN algorithm are used respectively to detect the grouting pressure-flow-time curve data to obtain the set of abnormal points. and ;calculate Extract outliers that are only identified by the DBSCAN algorithm, and then... The combined calculation results form the final set of outliers.

[0023] Furthermore, in step S6, a supplementary experimental scheme is designed based on Latin hypercube sampling, including the following steps:

[0024] (1) Determine the sampling parameter space: including stress path parameters, grouting parameters, and gas simulation extraction test parameters;

[0025] (2) Sub-interval division: Divide each parameter dimension into N sub-intervals evenly. N is dynamically determined according to the current error level. When the initial error is >15%, N=30; when the initial error is between 5% and 15%, N=20; when the initial error is <5%, N=10. Randomly select one sample point in each sub-interval to ensure that the sample projection of each parameter dimension is evenly distributed.

[0026] (3) Experimental scheme optimization: The information entropy of each sampling point is predicted by the Kriging surrogate model. The parameter combination with information entropy > 1.2 times the mean is selected for supplementary experiments to achieve efficient coverage of the parameter space. The supplementary experimental data are integrated into the original dataset according to weights until the error is reduced to within 5%. Note that the "5% error" here is essentially different from the "95% confidence interval" mentioned above, but there is an indirect relationship; the "5% error" is the deviation ratio that measures the model fitting accuracy after Latin hypercube sampling or the accuracy of the sealing effect evaluation after reinforcement learning optimization; the "95% confidence interval" is a probability interval calculated by Monte Carlo simulation and used to quantify the uncertainty of gas permeability estimation and determine the statistical significance of sealing effect.

[0027] Furthermore, step S6 also includes constructing a reinforcement learning closed-loop optimization process to optimize the experimental parameters. The experimental parameters include at least the grouting pressure, extraction negative pressure, and loading rate during the true triaxial test. The optimization steps include:

[0028] (1) Define the state space: Define the state space as follows: ,in For the degree of damage to the sample, The permeability of the coal and rock sample before grouting. The permeability of the coal and rock sample after grouting. This refers to the triaxial stress control error during the loading and unloading process in a true triaxial test. To account for the measurement errors in gas flow rate during the gas extraction simulation test and grouting flow rate during the sealing test, Indicates the current experimental phase;

[0029] (2) Define the action space: The action space includes continuous actions and discrete actions. The continuous action is the stress loading rate adjustment amount. Grouting pressure adjustment amount Gas simulation gas injection pressure adjustment amount The discrete actions are sampling frequency switching and anomaly handling strategy selection;

[0030] (3) Define the reward function: ,in, The value represents the decrease in permeability of the coal and rock sample before and after grouting. For the comprehensive error, To save time in the experiment, This is a penalty item for the range of motion. These are dynamically adjusted weighting coefficients;

[0031] (4) A deep deterministic strategy gradient is adopted. (Ψ,a,r,Ψ') samples are stored in the experience replay pool. The strategy is updated using the Actor-Critic network structure. The strategy is evaluated and updated once every 3 rounds of complete experimental process, so as to achieve adaptive optimization of experimental parameters and operation process, and reduce the sealing effect evaluation error to within 5%.

[0032] This invention utilizes a multiphysics flow-structure coupling true triaxial experimental system to conduct true triaxial experiments; the multiphysics flow-structure coupling true triaxial experimental system comprises:

[0033] A true triaxial stress dynamic control system includes loading modules respectively set in the x, y, and z directions for applying stress to coal and rock samples; the loading module includes a hydraulic cylinder and an integrated pressure head set at the piston end of the hydraulic cylinder; the integrated pressure head is equipped with a stress sensor for real-time monitoring of triaxial stress during the true triaxial loading and unloading process;

[0034] The multi-physics coupling control module includes several seepage channels penetrating the integrated pressure head along the thickness direction, a heating device disposed in the seepage channels, and several acoustic emission probes disposed on the integrated pressure head. The acoustic emission probes are used to locate the fracture propagation of the coal and rock sample.

[0035] The gas extraction simulation system uses inert gas as the gas simulation gas, which is injected into the coal and rock sample through a seepage channel.

[0036] The grouting system is used to realize the preparation, injection and monitoring of grouting slurry;

[0037] A data monitoring and processing system is used for data acquisition, data processing, and result presentation.

[0038] This method first obtains the stress evolution path under the disturbance of the test roadway through on-site stress monitoring, and then conducts true triaxial tests to collect the physical and mechanical parameters of coal and rock. Next, a first gas drainage simulation test is conducted to collect gas pressure and permeability data. Subsequently, grouting simulation sealing is performed on fractured coal and rock mass (coal and rock sample), and grouting flow rate and diffusion radius data are collected. Finally, a second gas drainage simulation test is conducted to collect relevant data. During the experiment, a Bayesian fusion model is used to fuse data from multiple systems to accurately infer the rock mass state and gas migration characteristics of the coal and rock sample. Isolated forest and DBSCAN algorithms are used to detect abnormal data to ensure data reliability. Monte Carlo simulation is used to evaluate the reliability of the gas permeability calculation results to quantify the sealing effect. Latin hypercube sampling design is used to supplement the experimental scheme to improve model accuracy and reduce costs. A reinforcement learning closed-loop optimization process is constructed to achieve adaptive optimization of experimental parameters and operation procedures. Ultimately, the accuracy, reliability, and efficiency of the simulation test are improved, providing a scientific theoretical basis and technical support for on-site gas drainage and grouting sealing engineering.

[0039] The technical solution of this application has the following beneficial effects:

[0040] The method of this invention combines on-site stress evolution path acquisition, true triaxial test, gas extraction simulation test, and verification of grouting and sealing effects in fractured rock mass. It can not only effectively simulate gas extraction and grouting sealing at the work site, but also use Monte Carlo simulation to randomly sample a large number of experimental data to obtain an error range with a certain confidence level, determine the reliability range of the experimental results, and automatically trigger a supplementary experimental mechanism of Latin hypercube sampling when the error exceeds the threshold to supplement experimental data and optimize experimental results.

[0041] This invention also improves the accuracy of sample state determination by fusing multi-system data from various experimental stages through a Bayesian fusion model; it can more comprehensively and accurately discover anomalies in the data by fusing isolated forest and DBSCAN algorithms; and it can obtain the optimal experimental strategy based on reinforcement learning, which can be fed back to adjust experimental parameters and operating procedures to achieve closed-loop optimization of the experimental process.

[0042] This invention can effectively improve the accuracy of sample condition determination, enhance the reliability of experimental results, optimize experimental design, and provide precise guidance for optimizing data and experimental schemes for gas extraction and grouting sealing simulation tests. Attached Figure Description

[0043] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. Wherein:

[0044] Figure 1 This is the overall flowchart of the present invention.

[0045] Figure 2 This is a flowchart of the Bayesian fusion model of the present invention.

[0046] Figure 3 This is a flowchart of the Latin hypercube sampling and reinforcement learning closed-loop optimization of the present invention.

[0047] Figure 4 This is a schematic diagram of the integrated pressure head according to an embodiment of the present invention.

[0048] Figure 5 This is a schematic diagram of the seepage channel in an embodiment of the present invention.

[0049] Explanation of reference numerals in the attached figures:

[0050] 1. Integrated pressure head; 2. Acoustic emission probe; 3. Electric heating rod. Detailed Implementation

[0051] This application will now be described in detail with reference to the accompanying drawings. In fact, those skilled in the art will recognize that modifications and variations can be made to this application without departing from the scope or spirit of this application. For example, features shown or described as part of one embodiment may be used in another embodiment to produce yet another embodiment. Therefore, it is desirable that this application encompass such modifications and variations that fall within the scope of the appended claims and their equivalents.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used herein is for the purpose of describing embodiments of this disclosure only and is not intended to limit this disclosure.

[0053] like Figure 1 As shown, a simulation test method for gas extraction and grouting sealing includes the following steps:

[0054] S1. Conduct stress monitoring at the target site to obtain the stress evolution path during the excavation and disturbance of the test roadway; select representative coal and rock at the target site and prepare coal and rock samples.

[0055] S2. Conduct a true triaxial test based on the stress evolution path to simulate the mining disturbance process, and monitor the fracture propagation of the coal and rock samples using an acoustic emission device; collect the physical and mechanical parameters of the coal and rock samples, including stress, strain, elastic modulus, Poisson's ratio, peak strain, and fracture initiation stress.

[0056] S3. After the coal and rock samples from the true triaxial test are degassed under a constant temperature under simulated field temperature, simulated gas is slowly introduced and pressurized stepwise at a rate of 0.05-0.1 MPa / min to the target pressure matching the field burial depth. Then, the first gas drainage simulation test is conducted. Data from the first gas drainage simulation test are collected, including gas pressure and gas flow rate (after drainage stabilization).

[0057] S4. Grouting is performed on the coal and rock samples extracted in S3 to simulate the sealing test after on-site gas extraction, and sealing samples are obtained. At the same time, grouting pressure, grouting flow rate and diffusion radius are collected.

[0058] S5. After grouting is completed, gas simulation is introduced into the sealing sample again according to the test conditions of the first gas extraction simulation test in S3 to conduct the second gas extraction simulation test; data from the second gas extraction simulation test are collected, including gas pressure and gas flow rate (after extraction is stabilized).

[0059] S6. Calculate the permeability of the coal and rock sample before grouting. and permeability after grouting , , Following a normal distribution; Monte Carlo simulation was used to evaluate the reliability of permeability in the simulated gas extraction test. If the permeability after grouting... The 95% confidence interval and the permeability before grouting If the 95% confidence intervals do not overlap, the sealing effect is reliable; otherwise, a supplementary test should be conducted to verify the sealing effect again.

[0060] Furthermore, in step S6, a Bayesian fusion model is used to fuse the collected data from various sources. The input data for the Bayesian fusion model includes: stress and strain data from true triaxial tests (representing real-time triaxial stress and strain data, respectively). , and And the corresponding strain of the coal and rock samples, denoted as ), gas pressure and gas flow rate data from the first gas extraction simulation test, and grouting pressure-flow rate-time curve data during the grouting process (denoted as ), The gas pressure in the second gas extraction simulation test (denoted as ) ) and gas flow data (denoted as The gas pressure in the first gas extraction simulation test includes at least the gas pressure monitoring values ​​at the inlet, middle section, and outlet of the sample, denoted as . The gas flow rate data in the first gas extraction simulation test are dynamic gas flow rate data, denoted as... .

[0061] Furthermore, such as Figure 2 As shown, the construction process of the Bayesian fusion model includes:

[0062] A multi-stage data association model was established, and a prior probability distribution was constructed through a historical experimental database. Among them, the stress-crack development association parameter adopted the Beta distribution, the gas pressure-flow coupling parameter of the gas simulation adopted the log-normal distribution, and the mapping relationship between grouting parameters and sealing effect adopted the Dirichlet distribution.

[0063] By constructing a Markov chain, based on the current parameter state and candidate parameters conforming to a specific distribution, and combining prior probability and likelihood function to calculate acceptance probability, parameters are iteratively updated. After a sufficient number of samplings, the chain reaches a stationary distribution, thereby obtaining a posterior sample distribution (posterior probability) that reflects the true distribution characteristics of the parameters. Based on the posterior sample distribution, the rock mass state of coal and rock samples is assessed and parameters are inverted. By calculating the statistics of the posterior samples, the permeability of the coal and rock samples before grouting is obtained. Permeability after grouting The key indicators of rock mass damage degree and grouting sealing efficiency of coal and rock samples are quantitatively estimated, and the reliability of the estimate is quantified by the dispersion of the posterior sample distribution.

[0064] Furthermore, the specific calculation steps for the prior probability, likelihood function, and posterior probability are as follows:

[0065] (1) Determine the prior probability based on theoretical analysis and the correct stress-pressure-flow relationship in historical experiments. ,in This is the average of historical data. It is the covariance matrix;

[0066] (2) To improve the Bayesian model's ability to fuse multi-source data, a likelihood function considering the correlation during the experimental phase is constructed. The specific function is as follows:

[0067] The full dataset is... ;

[0068] (3) The posterior probability distribution is calculated using the MCMC (Markov Chain Monte Carlo) algorithm. Simultaneously, the sample state is evaluated and parameters are inverted based on the posterior distribution. The posterior probability distribution is as follows: .

[0069] Furthermore, before fusing the data using the Bayesian fusion model in step S6, the isolated forest algorithm and the DBSCAN algorithm (density-based spatial clustering algorithm) are first used to detect outliers in the experimental data, removing invalid data and marking key physical mutations. For stress data during the true triaxial test, the isolated forest algorithm is used to calculate anomaly scores to identify stress mutation anomalies; for strain data, gas pressure and gas flow data in the first gas drainage simulation test, and gas pressure and gas flow data in the second gas drainage simulation test, the DBSCAN algorithm is used to identify outliers; for grouting pressure-flow-time curve data during the grouting process, fusion rules are used to comprehensively determine outlier data.

[0070] Furthermore, the specific process of using the isolated forest algorithm to detect anomalies in stress data S' during a true triaxial test is as follows:

[0071] (1) Data preprocessing: stress data ( =1, 2, 3 represent the stress directions. (where 1, 2, ..., n are sampling points) is converted into a three-dimensional feature vector. Remove zero or constant values ​​that are obviously caused by vibration of the sensor (stress sensor installed in the integrated pressure head 1); (2) Construction of isolated forest: randomly select 20% of the samples as training set and construct 100 isolated trees to form a forest; when each tree is generated, randomly select a feature dimension, randomly select a split point within the range of values ​​of the dimension, and recursively divide the data until the leaf node contains only one sample or reaches the maximum depth. (3) Calculation of outlier scores: For each data point Calculate its average length among 100 tree species. ; Calculate the correction factor According to the formula Calculate the outlier score when The time point is marked as a stress mutation anomaly.

[0072] Furthermore, the specific process of using the DBSCAN algorithm (density-based spatial clustering algorithm) to perform anomaly detection during the gas extraction and grouting sealing simulation test is as follows: (1) Data formatting: gas pressure data Convert to space-time features ( Number the monitoring points. (for time intervals), gas flow data Convert to flow-rate of change feature (2) Parameter settings: Calculate the standard deviation of gas pressure data Set neighborhood radius , the minimum number of points (3) Clustering and outlier identification: Traverse all data points and calculate the clustering characteristics of unvisited points. The number of points in the neighborhood, if the number of points If the number of points is high, then it is marked as a core point, and it forms a cluster with the points that its density can reach; if the number of points is low... And not affected by any core point If the neighborhood contains the point, it is marked as an outlier. This situation may be due to a sudden drop in pressure caused by a gas leak, etc.

[0073] Furthermore, the step of the fusion rule to comprehensively determine abnormal data is as follows: (1) using the isolated forest algorithm to process slurry flow data. The set of anomalies is obtained. (2) The DBSCAN algorithm was used to process the slurry flow data to obtain outliers. (3) Calculation Extract outliers that are only identified by the DBSCAN algorithm, and then... The combined calculation results form the final set of outliers.

[0074] The formula for calculating penetration rate is: In the formula, For penetration rate, For gas flow rate, The cross-sectional area of ​​the sample is... Let be the side length of the cubic coal and rock sample. For gas viscosity, Due to export pressure, For inlet pressure.

[0075] Specifically, the process of using Monte Carlo simulation to evaluate the reliability of permeability in simulated gas extraction tests is as follows:

[0076] (1) Determine the error distribution of each parameter: apparent gas flow rate The measurement error follows a normal distribution with a mean of 0 and a standard deviation of 2% of the measured value, i.e. , For flow rate measurement; gas viscosity Due to the influence of temperature, it is considered to follow a uniform distribution, with a value range of [value range missing]. , The viscosity value is under standard conditions; the measurement error of the sample length is determined by the accuracy of the measuring tool and follows a normal distribution. , The measured length of the coal and rock sample; the pressure at both ends of the coal and rock sample. and The measured values ​​all follow a normal distribution: , , and These are the pressure measurements at both ends of the coal and rock sample;

[0077] (2) Generate random error samples and calculate the corresponding penetration rate. Generate 10,000 sets of random error samples. Using a pseudo-random number generator, generate 10,000 samples according to the error distribution of the above parameters. and The random values ​​form 10,000 parameter combinations. ,in 1, 2, ..., 10000; Substitute each set of random parameters into the permeability calculation formula to calculate the corresponding 10000 permeability values. ;

[0078] (3) The permeability distribution was obtained statistically. For 10,000 Perform statistical analysis and calculate its mean. and standard deviation The penetration rate is determined to follow the mean. Standard deviation is The normal distribution is obtained, and the 95% confidence interval is calculated accordingly. (4) Comparison of permeability error ranges from two gas extraction simulation tests To evaluate the reliability of the sealing effect. , This represents the lower limit of the 95% confidence interval for the initial sampling penetration rate. This represents the upper limit of the 95% confidence interval for the secondary gas simulation extraction test. If This indicates that the 95% confidence intervals do not overlap. At the 95% confidence interval level, the permeability of the second extraction is significantly lower than that of the first extraction, and the sealing effect is reliable; otherwise, further verification of the sealing effect is required.

[0079] Furthermore, such as Figure 3 As shown, in step S6, a supplementary experimental scheme based on Latin hypercube sampling is designed, including the following steps:

[0080] (1) Determine the sampling parameter space: including stress path parameters, grouting parameters, and gas simulation extraction test parameters; among the stress path parameters The range is 10–30 MPa. The range is 0.3–0.8, the loading rate is 0.05–0.5 MPa / min; the grouting parameters are grouting pressure 1–5 MPa, grout water-cement ratio 0.8–1.5, and grouting time 30–180 s; the gas simulation extraction test parameters are gas simulation gas injection pressure 0.5–2 MPa, adsorption equilibrium time 60–300 min, and extraction negative pressure 0.02–0.1 MPa.

[0081] (2) Sub-interval division: Divide each parameter dimension into N sub-intervals evenly. N is dynamically determined according to the current error level. When the initial error is >15%, N=30; when the initial error is between 5% and 15%, N=20; when the initial error is <5%, N=10. Randomly select one sample point in each sub-interval to ensure that the sample projection of each parameter dimension is evenly distributed.

[0082] (3) Experimental scheme optimization: The information entropy of each sampling point is predicted by the Kriging surrogate model. The parameter combination with information entropy > 1.2 times the mean is selected for supplementary experiments to achieve efficient coverage of the parameter space. The supplementary experimental data is integrated into the original dataset according to the weight until the error is reduced to within 5%.

[0083] Furthermore, step S6 also includes constructing a reinforcement learning closed-loop optimization process to optimize the experimental parameters. The experimental parameters include at least the grouting pressure, extraction negative pressure, and loading rate during the true triaxial test. The optimization steps include:

[0084] (1) Define the state space: Define the state space as follows: ,in For the damage degree of the coal and rock sample, The permeability of the coal and rock sample before grouting. The permeability of the coal and rock sample after grouting. This refers to the triaxial stress control error during the loading and unloading process in a true triaxial test. To encompass the gas flow measurement error in the gas extraction simulation test and the grouting flow measurement error in the sealing test, a comprehensive index is denoted as flow measurement error ( ). ), Indicates the current experimental phase;

[0085] (2) Define the action space: The action space includes continuous actions and discrete actions. The continuous action is the stress loading rate adjustment amount. (Step-by-step pressurization rate in step S3), grouting pressure adjustment amount Gas simulation gas injection pressure adjustment amount The discrete actions are sampling frequency switching and anomaly handling strategy selection;

[0086] (3) Define the reward function: ,in, The value represents the decrease in permeability of the coal and rock sample before and after grouting. For the comprehensive error, To save time in the experiment, This is a penalty item for the range of motion. The weighting coefficients are dynamically adjusted; the overall error is... yes (Triaxial stress control error) and The weighted integration of (flow measurement error) is essentially about making the overall error... Prioritizes reflecting the key error sources in the current stage; the calculation formula is as follows: , , , It needs to be dynamically adjusted based on the experimental phase;

[0087] (4) Deep deterministic policy gradient (DDPG) is adopted. (Ψ,a,r,Ψ') samples are stored in the experience replay pool. The policy is updated using the Actor-Critic network structure. The policy is evaluated and updated once every 3 rounds of complete experimental process, so as to achieve adaptive optimization of experimental parameters and operation process, and reduce the sealing effect evaluation error to within 5%.

[0088] Specifically, (Triaxial stress control error) is obtained through "unidirectional stress control error calculation → weighted integration of triaxial errors", in two steps:

[0089] (1) Calculate the uniaxial stress control error:

[0090] For stress in the x-direction (along the tunnel direction) ), y-direction (stress perpendicular to the tunnel direction) z-direction (vertical stress) The stress control error at a given time point t is calculated using the following formula:

[0091] ;

[0092] In the formula: For a certain direction in x / y / z, in time Unidirectional stress control error (%) For a certain direction in x / y / z, in time The actual loaded stress value (MPa) is collected by an external sensor of the true triaxial stress dynamic control system, specifically a stress sensor. The stress sensor is installed in the integrated pressure head 1 of the triaxial system, and its function is to monitor the triaxial stress in real time during the true triaxial loading and unloading process. For a certain direction in x / y / z, in time The preset target stress value (MPa) is obtained from the field stress evolution path (such as the target stress in the original stress stage, the tunneling disturbance stage, etc.) obtained in step S1.

[0093] (2) Calculate the final :

[0094] The errors in the x, y, and z axes are weighted and integrated to obtain the overall triaxial stress control error. The formula is:

[0095] ;

[0096] In the formula, For time Corresponding triaxial stress control error (%) , and The weighting coefficients for the three axial stresses along the x, y, and z axes are respectively. Based on the degree of influence of on-site stress on the development of coal and rock fractures, and with reference to empirical values, the settings are set. , , Specifically, (Flow measurement error) is obtained through "single-parameter raw flow error calculation → weighted integration after outlier correction". The core is to integrate the gas extraction flow error and the grouting flow error, which is done in two steps:

[0097] (1) Single parameter correction of gas flow rate error in gas simulation is divided into correction of gas extraction flow rate error in gas simulation and correction of grouting flow rate error;

[0098] For the gas extraction flow rate error in the gas simulation, the first extraction is first eliminated using the DBSCAN algorithm. ), secondary extraction ( The abnormal traffic flow values ​​are then used to calculate the corrected average error, using the following formula:

[0099] ;

[0100] In the above formula, the original error of the gas flow rate in a single group of gas simulations The formula for calculation is:

[0101] ;

[0102] In the above formula, The corrected average error of the simulated gas extraction flow rate (unit: %). The measured flow rate of the simulated gas extraction system (unit: L / min) is collected by the flow meter of the simulated gas extraction system. The actual gas extraction flow rate (unit: L / min) in the gas simulation is calculated based on Darcy's law:

[0103] ;

[0104] In the above formula, For penetration rate, The cross-sectional area of ​​the sample is... For inlet pressure, Due to export pressure, For gas viscosity, The length is the sample length.

[0105] For grouting flow rate errors, outliers are first removed using the "Isolated Forest + DBSCAN fusion rule," and then the corrected average error is calculated using the following formula:

[0106] ;

[0107] Among them, the original error of grouting flow rate at a single time node The formula for calculation is:

[0108] ;

[0109] In the formula, This represents the total grouting time. For time The grouting flow rate measurement value (unit: L / min) is collected by the flow sensor of the grouting system; For time The actual grouting flow rate (L / min) was obtained based on the fracture grouting theory and the grout diffusion radius obtained from CT scans. In the formula, Where is the diffusion radius, For coal and rock porosity, For slurry density, Density of solid particles;

[0110] (2) The corrected gas extraction flow rate error and grouting flow rate error are integrated according to their weights to obtain the following result. The formula is:

[0111] ;

[0112] In the formula, This is the weighting coefficient for the gas extraction flow rate error in the gas simulation. The weighting coefficient for grouting flow rate error must satisfy the following conditions: Based on extensive experimental calculations, the empirical value is obtained. , .

[0113] In practice, some of the aforementioned data processing methods can be flexibly selected as needed.

[0114] The preferred embodiment of the present invention incorporates all of the above data processing methods into the gas extraction and grouting sealing simulation test:

[0115] (1) The Bayesian fusion model is used to fuse the multi-system data collected in the gas extraction and grouting simulation test. By integrating the multi-stage related data such as stress, gas extraction simulation and grouting simulation and quantifying the uncertainty, the accurate and reliable inference of the state of coal and rock samples and the gas migration characteristics of gas simulation is realized, providing a unified and high-quality parameter basis for the test analysis.

[0116] (2) The isolated forest and DBSCAN algorithms are combined to detect outliers in the experimental data. By combining the advantages of the two algorithms, outliers in the data are identified in a comprehensive and accurate manner, invalid data is removed and key physical mutations are marked, providing a reliable data foundation for subsequent analysis and ensuring the accuracy and scientific nature of the experimental results.

[0117] (3) The reliability of the gas permeability calculation results is evaluated by Monte Carlo simulation. Based on the permeability calculation formula, the error distribution of each parameter is determined, random error samples are generated and the corresponding permeability is calculated. The permeability distribution is statistically obtained. By comparing the permeability error intervals of two gas extraction simulation tests, the reliability of the sealing effect is evaluated, and a quantitative basis is provided for the scientific judgment of the grouting sealing simulation test effect.

[0118] (4) Design a supplementary experimental scheme based on Latin hypercube sampling; determine the sampling parameter space, divide the sub-intervals and randomly select sample points, predict the information entropy of each sampling point through the Kriging surrogate model, and prioritize the parameter combination with higher information entropy for supplementary experiments, so as to achieve efficient coverage of the key parameter space in the experiment under a limited number of experiments, improve the accuracy of the data optimization model, and reduce the experimental cost.

[0119] (5) Construct a reinforcement learning closed-loop optimization process to optimize experimental parameters and operation procedures. Define the state space, action space and reward function, adopt the deep deterministic policy gradient (DDPG), store samples through the experience replay pool, and use the Actor-Critic network structure to update the policy to achieve adaptive optimization of experimental parameters and operation procedures.

[0120] This invention utilizes a multiphysics flow-structure coupling true triaxial experimental system to conduct true triaxial experiments; the multiphysics flow-structure coupling true triaxial experimental system comprises:

[0121] A true triaxial stress dynamic control system includes loading modules respectively located in the x, y, and z directions for applying stress to coal and rock samples. Each loading surface in the triaxial loading module includes nine evenly distributed small servo hydraulic cylinders (referred to as cylinders), with an integrated pressure head 1 fixed to the piston end of each cylinder. The integrated pressure head 1 is equipped with a stress sensor for real-time monitoring of triaxial stress during the true triaxial loading and unloading process. The system monitors the stroke of each cylinder and sets one of the nine cylinders as a reference cylinder (e.g., ...). Figure 4 In this process, the cylinder located in the center is used as the reference cylinder. Its lifting speed is adjusted to meet the required working requirements and kept at a constant speed. The other eight cylinders are used as follow-up cylinders. When the difference between the stroke data of the follow-up cylinder and the reference cylinder exceeds the set range, the operation of the follow-up cylinder is stopped or adjusted by control elements such as electromagnetic proportional directional valves until all cylinders reach the specified stroke.

[0122] Multiphysics coupling control module, such as Figure 4 , Figure 5 As shown, it includes several seepage channels penetrating the integrated pressure head 1 along the thickness direction, a heating device set in the seepage channels, and several acoustic emission probes 2 set on the integrated pressure head 1; the heating device adopts an electric heating rod 3; the simulated gas enters through the seepage channels, and the temperature environment of the gas is adjusted by the electric heating rod 3 to realize the simulation of the real environment of the target site; the acoustic emission probes 2 are located at the center of every 4 cylinders, and 4 probes are arranged on the two loading surfaces in the same axis, which can realize the eight-channel positioning of the coal and rock fracture expansion morphology;

[0123] The gas extraction simulation system uses inert gas as the simulated gas to ensure experimental safety. The system includes a high-pressure gas cylinder and a gas collection cylinder. The flow process of the simulated gas is as follows: the simulated gas in the high-pressure cylinder is first reduced to a suitable pressure by a pressure reducing valve, and then stabilized by a pressure regulating valve to ensure pressure balance in subsequent processes. Next, the simulated gas enters a pressure pump for further pressurization to simulate the pressure conditions of an actual extraction environment. The pressurized simulated gas is then filtered to remove impurities, preventing contaminants from affecting experimental accuracy. Afterward, the simulated gas first passes through an inlet pressure sensor to monitor the pressure value entering the extraction system in real time, and then flows into the coal and rock sample through a seepage channel to ensure the consistency of the simulation conditions. Before the simulated gas reaches the gas outlet, an outlet pressure sensor records the outlet pressure, while a differential pressure sensor simultaneously measures the pressure difference between the gas inlet and outlet. The sample permeability is calculated based on Darcy's law. Finally, the simulated gas extracted during the extraction simulation flows into the gas collection cylinder, which can be used for gas composition analysis. Preferably, a check valve and a safety valve are installed in the gas line between the gas outlet and the gas collecting cylinder. The check valve prevents gas backflow from affecting the safety of the test and the accuracy of the data, while the safety valve automatically opens to release pressure when the pressure exceeds the set threshold, ensuring the safe operation of the entire gas extraction simulation system.

[0124] The grouting system is used to realize the preparation, injection, and monitoring of grouting slurry. The grouting system mainly includes a storage tank, a multi-stage booster pump, a throttle valve, and a three-stage filtration device. The storage tank is a dynamic stirring storage tank, where the slurry is temporarily stored after being manually prepared. The slurry is drawn by the multi-stage booster pump and impurities are removed by the throttle valve and the three-stage filtration device before being injected into the simulated cavity of the coal and rock sample extracted in step S3. A spring-loaded safety valve is installed on the liquid line. In case of overpressure, the safety valve quickly opens to release pressure and ensure safety. Specifically, during the test, when the coal and rock sample is prepared in step S1, a hole is drilled on one side and a gas inlet is installed. The slurry is injected into the coal and rock sample through the gas inlet.

[0125] A data monitoring and processing system is used for data acquisition, data processing, and result presentation.

[0126] The details of each step in this application will be described in detail below with reference to the embodiments. The target site is Zhaogu No. 2 Coal Mine, which mainly mines the No. 21 coal seam; it is located at the junction of Xinxiang and Jiaozuo cities in Henan Province; the No. 21 coal seam is the main coal seam, with an average thickness of 6.16m, an average dip angle of 5°, and a simple and stable coal seam structure; the average burial depth of each seam is 730m.

[0127] In S1, the stress monitoring scheme at the target site is as follows:

[0128] Six three-dimensional stress sensors are installed in the transport roadway of the working face, one every 50m along the roadway. The monitoring period is from 30 days before the working face advances to the sensor position to 30 days after the advance. The sampling frequency is 10Hz. The range of the three-dimensional stress sensors is 0~50MPa, and the accuracy is ±0.1MPa.

[0129] Record stress values ​​at different stages of mining ( For stress along the tunnel direction, Stress perpendicular to the direction of the tunnel. (For vertical stress), the following stress evolution path is obtained:

[0130] (1) Initial stress stage (more than 100m away from the working surface): =12MPa, =8MPa, =20MPa;

[0131] (2) Tunneling disturbance stage (50-100m from the working face): Increased to 15MPa Reduced to 6 MPa Fluctuated to 22 MPa;

[0132] (3) Mining-induced impact stage (0-50m from the working face): Peak pressure 25 MPa Peak value 12MPa Peak pressure 28 MPa;

[0133] (4) Stabilization phase (after the working face has been pushed 50m): =18MPa =10MPa, =25MPa;

[0134] The preparation of coal and rock samples involves selecting representative coal and rock samples from the target site, processing the coal samples into 100mm×100mm×100mm cubic specimens, and treating them with vacuum saturation for 24 hours to simulate the water content of the coal and rock at the site.

[0135] In step S2, a multiphysics fluid-structure coupled true triaxial experimental system is used, which has a stress and strain acquisition system with a sampling frequency of 20 Hz. The loading process of the true triaxial test is carried out in stages according to the acquired stress evolution path, with the loading rate as follows: =0.2MPa / min, =0.15MPa / min, =0.3MPa / min, data were collected after each stage of loading stabilized for 30 minutes, until the peak stress of the mining-induced impact stage was reached. Data output yielded the following coal and rock physical and mechanical parameters: elastic modulus 3.2GPa, Poisson's ratio 0.28, peak strain 2.1×10 -3 Crack initiation stress =18MPa =9MPa, providing mechanical boundary conditions for subsequent experiments.

[0136] In step S3, a gas extraction simulation system is used. This system includes a gas cylinder (simulating gas purity of 99.9%), a pressure controller (accuracy ±0.01MPa), and a flow meter (range 0~10L / min, accuracy ±0.05L / min). Monitoring point layout: at the coal and rock sample inlet end ( ), 1 / 3 of the way ( ), 2 / 3 of the way ( ), export end ( Each outlet is equipped with one pressure sensor, and a flow sensor is installed at the outlet. The experimental parameters for the first gas drainage simulation test are as follows:

[0137] Gas injection pressure: 1.5 MPa (simulating on-site gas pressure), adsorption equilibrium time: 240 min;

[0138] Sampling negative pressure: 0.08MPa, sampling time: 120min;

[0139] Data recording: After the sampling stabilizes, =1.45MPa, =0.8MPa, =0.3MPa, =0.05MPa, =3.2L / min, the permeability before grouting was calculated. =2.1×10 -15 m 2 The permeability is calculated using the permeability formula given above: ,in The simulated gas viscosity is 0.0108 MPa•s. The specimen length is 0.1m. The cross-sectional area is 0.01m. 2 .

[0140] In step S4, the grout used was a cement-water glass grout (cement grade P.O42.5, water glass concentration 35 Baume, cement:water glass volume ratio 1:1), with a water-cement ratio of 1.2 and a grout viscosity of 50 MPa•s (25℃). In the grouting system, the grouting pump had a range of 0–10 MPa and a flow rate of 0–5 L / min; the grouting pipe had an inner diameter of 6 mm, and the diffusion radius was observed in real-time using CT scanning. The grouting parameters were: grouting pressure 4 MPa, grouting rate 2 L / min, grouting time 90 s, and curing for 24 h after grouting (temperature 20℃, humidity 90%). Data acquisition: The grouting pressure-flow rate-time curve was recorded. The initial flow rate was 2 L / min, decreasing to 1.2 L / min after 30 s, decreasing to 0.5 L / min after 60 s, and stabilizing at 0.3 L / min after 90 s. CT scanning showed a grout diffusion radius of 12 mm and a crack filling rate of 85%.

[0141] In step S5, the test conditions for the second gas extraction simulation test were the same as the first extraction (injection pressure 1.5 MPa, extraction negative pressure 0.08 MPa, same monitoring points), and the extraction time was 120 min. Data recording: After stabilization... =1.42MPa, =1.0MPa, =0.6MPa, =0.06MPa, =1.1L / min, the permeability before grouting was calculated. =0.7×10 -15 m 2

[0142] In step S6, the data processing method of the preferred embodiment is adopted, and the data processing and optimization process is as follows:

[0143] 1. The steps of the Bayesian fusion model for fusing collected data from multiple systems are as follows:

[0144] (1) The input data includes:

[0145] True triaxial test data: , , and corresponding strain values ​​( );

[0146] Data from the first gas extraction simulation test: (Pressure values ​​at 4 monitoring points represent gas pressure) (The value from the flow sensor at the outlet represents the gas flow rate.)

[0147] Data from the second gas extraction simulation test: ';

[0148] Grouting data: Grouting pressure-flow rate-time curve.

[0149] (2) Model building

[0150] Prior probability: The stress-fracture development parameter follows a Beta distribution ( =3, =7), the gas pressure-flow parameters of the gas simulation follow a log-normal distribution (mean 0.4, standard deviation 0.12), and the grouting parameters follow a Dirichlet distribution ( =[2,3,4]);

[0151] Likelihood function: Calculated following the steps outlined above, the full dataset is... ;

[0152] Posterior probability: After 10,000 iterations of the MCMC algorithm, the posterior sample was obtained and the rock mass damage degree was calculated to be 0.58 (confidence interval [0.55, 0.61]) and the grouting sealing efficiency was 67% (confidence interval [64%, 70%]).

[0153] 2. The steps for outlier detection in the experimental data using the Isolation Forest and DBSCAN algorithms are as follows:

[0154] (1) Anomaly detection of stress data using the isolated forest algorithm:

[0155] Preprocessing: Remove three zero-value data points caused by sensor vibration;

[0156] Model training: 20% of the samples (1200 data points) were randomly selected to construct 100 isolated trees, anomaly scores were calculated, and two stress mutation points were marked (anomaly scores of 0.82 and 0.7), which were manifested as a sudden increase in stress during the corresponding peak mining phase.

[0157] (2) Using the DBSCAN algorithm to detect anomalies in gas extraction data:

[0158] Feature formatting: Convert gas pressure data into spatial-temporal (monitoring point number-time) features, and flow rate data into flow rate-rate of change features;

[0159] Parameter settings: Calculated based on a pressure standard deviation of 0.05, neighborhood radius. =0.15, minimum number of points MinPts=6;

[0160] Clustering and outlier identification results: One outlier was identified, indicating gas pressure. A sudden drop to 0.5 L / min indicates a leak at the pipe connection, and the line is rejected.

[0161] (3) The steps for determining abnormal grouting flow data using fusion rules are as follows:

[0162] The isolated forest algorithm detected two outliers (characterized by sudden changes in traffic), and the DBSCAN algorithm detected one outlier (characterized by persistently low traffic).

[0163] According to the fusion rules Three anomalies were identified, all caused by equipment malfunctions.

[0164] 3. The process of using Monte Carlo simulation to evaluate the permeability reliability in a gas simulation extraction test is as follows:

[0165] (1) Parameter error distribution:

[0166] Gas flow measurement error follows a normal distribution: The first sampling , secondary extraction The subscript "meas" represents the measured value of this parameter.

[0167] Gas pressure error follows a normal distribution: , ;

[0168] Gas viscosity Follows a uniform distribution: , ;

[0169] Specimen length normal distribution: , ;

[0170] (2) Simulation results:

[0171] Generate 10,000 sets of random parameters and calculate... The 95% confidence interval is ;

[0172] 95% confidence interval ;

[0173] because Lower limit (1.7×10) -15 )> Upper limit (0.9×10) -15 This indicates that the sealing effect is reliable.

[0174] 4. A supplementary experimental design based on Latin hypercube sampling is proposed, including the following steps:

[0175] (1) Determine the sampling parameter space:

[0176] Stress path: =10~30MPa, =0.3~0.8, loading rate =0.05~0.5MPa / min;

[0177] Grouting parameters: pressure = 1~5MPa, grout water-cement ratio = 0.8~1.5, grouting time = 30~180s;

[0178] Extraction parameters: gas simulation injection pressure = 0.5~2MPa, adsorption time = 60~300min, negative pressure = 0.02~0.1MPa.

[0179] (2) Sub-interval division: The initial error is 18%, and the sub-intervals are N=30. After dividing the sub-intervals, random sampling is performed.

[0180] (3) Optimization of the experimental design:

[0181] The Kriging surrogate model predicts information entropy, selecting 20 parameter combinations with information entropy greater than 1.2 times the mean (e.g., ...). Supplementary tests were conducted at pressures of 22 MPa, grouting pressure of 3.5 MPa, and negative pressure of 0.07 MPa. After supplementing the data, the model error was reduced to 4.5%, and the parameter coverage was improved to 92%.

[0182] 5. The steps for optimizing experimental parameters by constructing a closed-loop optimization process for reinforcement learning are as follows:

[0183] (1) Define the state space:

[0184] Define the state space as Damage degree of coal and rock samples =0.58, =0.02 (triaxial stress control error). =0.03 (flow measurement error) =5 (Currently in the second sampling stage).

[0185] (2) Define the action space:

[0186] Continuous action: Stress loading rate adjustment =0.05MPa / min, grouting pressure adjustment amount =0.3MPa, injection pressure adjustment amount =0.1MPa;

[0187] Discrete action: The sampling frequency is switched to 10Hz, and the anomaly handling strategy is "recalibrate the sensor".

[0188] (3) Define the reward function:

[0189] Reward function: calculated based on experimental data , , Substituting into the general formula of the reward function, we get The calculation shows that the optimized reward value increases by 25%.

[0190] (3) Optimization results:

[0191] The optimal parameters were determined by iterating the DDPG algorithm 500 times: grouting pressure 3.8MPa, extraction negative pressure 0.07MPa, loading rate 0.25MPa / min, test time was shortened by 20%, and sealing efficiency was increased to 72%.

[0192] This embodiment strictly followed the aforementioned method, using a true triaxial test guided by on-site stress path analysis. Combined with multi-source data fusion, anomaly detection, reliability assessment, and intelligent optimization, it verified that the gas permeability decreased by 67% after grouting and sealing, with stable performance within the 95% confidence interval. The optimized test scheme showed a 90% agreement with the on-site engineering, indicating that this method can effectively improve the accuracy and efficiency of gas extraction and grouting simulation tests, providing a reliable basis for on-site engineering design.

[0193] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A gas extraction and grouting hole sealing simulation test method, characterized in that, Includes the following steps: S1. Conduct stress monitoring at the target site to obtain the stress evolution path during the excavation and mining disturbance of the test roadway; Representative coal and rock samples were selected from the target site and prepared. S2. Conduct a true triaxial test based on the stress evolution path to simulate the mining disturbance process, and monitor the fracture propagation of the coal and rock samples using an acoustic emission device; collect the physical and mechanical parameters of the coal and rock samples, including stress, strain, elastic modulus, Poisson's ratio, peak strain, and fracture initiation stress. S3. After the coal and rock samples from the true triaxial test are degassed under vacuum under constant temperature conditions, simulated gas is slowly introduced and pressurized stepwise at a rate of 0.05-0.1 MPa / min to the target pressure matching the burial depth at the site. Then, the first gas extraction simulation test is carried out. Data from the first gas extraction simulation test are collected, including gas pressure and gas flow rate. S4. Grouting is performed on the coal and rock samples extracted in S3 to simulate the sealing test after on-site gas extraction, and sealing samples are obtained. At the same time, grouting pressure, grouting flow rate and diffusion radius are collected. S5. After grouting is completed, gas simulation is introduced into the sealing sample again according to the test conditions of the first gas extraction simulation test in S3 to conduct the second gas extraction simulation test; data from the second gas extraction simulation test are collected, including gas pressure and gas flow rate. S6. A Bayesian fusion model is used to fuse the collected data from various sources, and the permeability before grouting is calculated. and permeability after grouting , , Following a normal distribution; Monte Carlo simulation was used to evaluate the reliability of permeability in the simulated gas extraction test. If the permeability after grouting... The 95% confidence interval and the permeability before grouting If the 95% confidence intervals do not overlap, then the sealing effect is reliable; If not, conduct supplementary tests and verify the sealing effect again; The input data for the Bayesian fusion model includes: stress and strain data from the true triaxial test, gas pressure and gas flow rate data from the first gas extraction simulation test, grouting pressure-flow rate-time curve data during the grouting process, and gas pressure and gas flow rate data from the second gas extraction simulation test. The construction process of the Bayesian fusion model includes: A multi-stage data association model was established, and a prior probability distribution was constructed through a historical experimental database. Among them, the stress-crack development association parameter adopted a Beta distribution, the gas pressure-flow coupling parameter of the gas simulation gas adopted a log-normal distribution, and the mapping relationship between grouting parameters and sealing effect adopted a Dirichlet distribution. By constructing a Markov chain, based on the current parameter state and candidate parameters conforming to a specific distribution, and combining prior probability and likelihood function to calculate acceptance probability, parameters are iteratively updated. After a sufficient number of samplings, the chain reaches a stationary distribution, thereby obtaining a posterior sample distribution that reflects the true distribution characteristics of the parameters. Based on the posterior sample distribution, the rock mass state of coal and rock samples is assessed and parameters are inverted. By calculating the statistics of the posterior samples, the permeability before grouting is assessed. Permeability after grouting The key indicators of rock mass damage degree and grouting sealing efficiency of coal and rock samples are quantitatively estimated, and the reliability of the estimate is quantified by the dispersion of the posterior sample distribution.

2. The gas extraction and grouting hole sealing simulation test method according to claim 1, characterized in that: Before using the Bayesian fusion model to process the data in step S6, the isolated forest algorithm and the DBSCAN algorithm are first used to detect outliers in the experimental data, remove invalid data, and mark key physical mutations.

3. The gas extraction and grouting hole sealing simulation test method according to claim 2, characterized in that: For stress data during the true triaxial test, the isolated forest algorithm is used to calculate anomaly scores to identify stress mutation anomalies; for strain data, gas pressure and gas flow data in the first gas extraction simulation test, and gas pressure and gas flow data in the second gas extraction simulation test, the DBSCAN algorithm is used to identify outliers; for grouting pressure-flow-time curve data during the grouting process, a fusion rule is used to comprehensively determine abnormal data.

4. The gas extraction and grouting hole sealing simulation test method according to claim 3, characterized in that: The steps for comprehensively judging abnormal data using the fusion rules are as follows: The isolated forest algorithm and the DBSCAN algorithm are used respectively to detect the grouting pressure-flow-time curve data to obtain a set of abnormal points. and ;calculate Extract outliers that are only identified by the DBSCAN algorithm, and then... The combined calculation results form the final set of outliers.

5. The gas extraction and grouting hole sealing simulation test method according to claim 1, characterized in that: In step S6, a supplementary experimental design based on Latin hypercube sampling is implemented, including the following steps: Determine the sampling parameter space: including stress path parameters, grouting parameters, and gas simulation extraction test parameters; Sub-interval partitioning: Divide each parameter dimension evenly into N sub-intervals. N is dynamically determined based on the current error level. When the initial error is >15%, N=30; when the initial error is between 5% and 15%, N=20; when the initial error is <5%, N=10. Randomly select one sample point in each sub-interval to ensure that the sample projections of each parameter dimension are evenly distributed. Experimental design optimization: The information entropy of each sampling point is predicted by the Kriging surrogate model. Supplementary experiments are conducted by selecting parameter combinations with information entropy greater than 1.2 times the mean to achieve efficient coverage of the parameter space. The supplementary experimental data are then integrated into the original dataset according to weights until the error is reduced to within 5%.

6. The gas extraction and grouting hole sealing simulation test method according to claim 5, characterized in that: Step S6 also includes constructing a reinforcement learning closed-loop optimization process to optimize the experimental parameters. The experimental parameters include at least the grouting pressure, extraction negative pressure, and loading rate during the true triaxial test. The optimization steps include: Define the state space: Define the state space as Ψ = ,in For the degree of damage to the sample, The permeability of the coal and rock sample before grouting. The permeability of the coal and rock sample after grouting. This refers to the triaxial stress control error during the loading and unloading process in a true triaxial test. To account for the measurement errors in gas flow rate during the gas extraction simulation test and grouting flow rate during the sealing test, Indicates the current experimental phase; Define the action space: The action space includes continuous actions and discrete actions. Continuous actions are the stress loading rate adjustment amount. Grouting pressure adjustment amount Gas simulation gas injection pressure adjustment amount The discrete actions are sampling frequency switching and anomaly handling strategy selection; Define the reward function: ,in, This represents the decrease in permeability of the coal and rock sample before and after grouting. For the comprehensive error, To save time in the experiment, This is a penalty item for the range of motion. These are dynamically adjusted weighting coefficients; A deep deterministic strategy gradient is adopted, and (Ψ,a,r,Ψ') samples are stored in an experience replay pool. The strategy is updated using an Actor-Critic network structure. The strategy is evaluated and updated once every 3 rounds of complete experimental process, so as to achieve adaptive optimization of experimental parameters and operation process, and reduce the sealing effect evaluation error to within 5%.

7. The gas extraction and grouting sealing simulation test method according to claim 1, characterized in that: True triaxial experiments were conducted using a multiphysics flow-structure coupling true triaxial experimental system; the multiphysics flow-structure coupling true triaxial experimental system includes: A true triaxial stress dynamic control system includes loading modules respectively set in the x, y, and z directions for applying stress to coal and rock samples; the loading module includes a hydraulic cylinder and an integrated pressure head set at the piston end of the hydraulic cylinder; the integrated pressure head is equipped with a stress sensor for real-time monitoring of triaxial stress during the true triaxial loading and unloading process; The multi-physics coupling control module includes several seepage channels penetrating the integrated pressure head along the thickness direction, a heating device disposed in the seepage channels, and several acoustic emission probes disposed on the integrated pressure head. The acoustic emission probes are used to locate the fracture propagation of the coal and rock sample. The gas extraction simulation system uses inert gas as the gas simulation gas, which is injected into the coal and rock sample through a seepage channel. The grouting system is used to realize the preparation, injection and monitoring of grouting slurry; A data monitoring and processing system is used for data acquisition, data processing, and result presentation.

Images