Intelligent prediction method for absolute mining-induced stress based on borehole oil pillow stress meter monitoring

Abstract

The application discloses an absolute mining stress intelligent prediction method based on a borehole oil pillow stress gauge monitoring, and belongs to the technical field of mining stress monitoring. The method comprises the following steps: manufacturing a coal rock surrounding rock test piece with a borehole and installing a borehole oil pillow stress gauge; applying confining pressure to the test piece and initial pressure to the borehole oil pillow stress gauge through a loading test device, and performing multi-working condition loading to collect corresponding stress data; establishing a linear regression model between absolute stress, relative stress and loading parameters; pre-processing and standardizing the data, and dividing a training set and a verification set; based on the training set, using Gaussian process regression combined with Bayesian optimization to construct an absolute stress prediction model; using the verification set to evaluate the precision and reliability of the model; finally, inputting the collected relative stress, initial pressure and borehole diameter data into the trained prediction model, so that high-precision surrounding rock absolute stress prediction values and their confidence intervals can be obtained, and intelligent quantitative prediction of mining stress is realized.

Description

Technical Field

[0001] This invention belongs to the field of mining stress monitoring technology, and specifically relates to an intelligent prediction method for absolute mining stress based on borehole oil pillow stress gauge monitoring. Background Technology

[0002] During underground coal mining, a mining-induced stress field is prevalent around the coal face. With the continuous mining of large areas of coal seam, the surrounding coal-bearing strata are strongly disturbed, generating significant mining-induced stress. This stress disturbance leads to severe deformation and damage to the surrounding rock of roadways, chambers, and the working face, and can even induce various disasters such as roof falls, rock bursts, and caving. This not only seriously threatens the lives of miners but also easily causes equipment damage and production interruptions, thus affecting the implementation of production plans, increasing the burden of safety management, reducing resource recovery rates, and causing huge economic losses and social impacts. Therefore, improving the accurate monitoring and scientific prediction capabilities of the surrounding rock stress in mining areas is a key technological support for ensuring safe and efficient coal mine production.

[0003] Currently, the main methods used for monitoring surrounding rock stress include strain measurement, ultrasonic methods, electromagnetic methods, distributed fiber optic stress monitoring, microseismic monitoring, and borehole stress gauge methods. Strain measurement offers high sensitivity and accuracy, making it suitable for long-term deployment, but it has strict requirements for installation techniques and measurement point coupling conditions, and its stability is relatively poor. Ultrasonic and electromagnetic methods, due to their non-contact and interference-resistant advantages, are suitable for trend analysis and early warning, but their measurement results are relatively indirect and highly dependent on inversion. Distributed fiber optic stress monitoring offers high resolution and remote deployment capabilities, making it suitable for intelligent mining environments, but its cost is high and its construction process is complex. Microseismic monitoring can identify precursory information of surrounding rock instability, making it suitable for dynamic early warning, but it cannot directly obtain quantitative data on stress magnitude. Each of these methods involves varying degrees of trade-offs in terms of monitoring accuracy, stability, applicable scenarios, and cost, and a stress monitoring technology that combines high accuracy, wide applicability, and quantitative analysis capabilities has not yet been formed.

[0004] Borehole oil-pillar stress gauges have been widely used in underground engineering fields such as coal mines due to their simple structure, convenient installation, low cost, and excellent adaptability to downhole environments. This method involves injecting oil into a hydraulic pillow to expand it and conform it to the borehole wall. During subsequent application of surrounding rock pressure, the change in hydraulic pressure reflects the change in surrounding rock stress. However, this hydraulic pressure value reflects a relative value, not the absolute value of the original stress state of the surrounding rock. Furthermore, because the measurement results of borehole oil-pillar stress gauges are affected by various factors such as initial pressure, confining pressure conditions, loading conditions, and the physical and mechanical properties of the surrounding rock, the relationship between the monitored parameters and the actual stress state of the surrounding rock under different conditions is complex. It is difficult to directly and accurately invert the absolute stress of the surrounding rock from the monitored values, thus limiting its quantitative application in guiding coal mine safety production.

[0005] Therefore, it is urgent to study the quantitative relationship between borehole stress and monitoring parameters under different working conditions, and to propose a method that can accurately quantify the stress state of coal body through monitoring parameters, so as to meet the needs of monitoring the actual stress state of coal body in the field. Summary of the Invention

[0006] The purpose of this invention is to propose an intelligent prediction method for absolute mining stress based on borehole oil pillow stress gauge monitoring, so as to realize intelligent analysis of monitoring data and high-precision prediction of absolute stress in surrounding rock.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for intelligent prediction of absolute mining-induced stress based on borehole oil tank stress gauge monitoring, the method comprising the following steps:

[0009] S1. Prepare coal and rock surrounding rock specimens with boreholes and install borehole oil-filled stress gauges in the boreholes.

[0010] S2. Apply confining pressure to the coal and rock surrounding rock specimen using a loading test device, and apply initial pressure to the borehole oil pillow stress gauge. Collect relative stress data through multi-condition loading; where confining pressure is the absolute stress.

[0011] S3. Establish a linear regression model between absolute stress, relative stress, and loading parameters;

[0012] S4. Preprocess and standardize the data obtained from S2 and S3, and divide them into training and validation sets.

[0013] S5. Based on the training set, an absolute stress prediction model is constructed using Gaussian process regression combined with Bayesian optimization.

[0014] S6. Use the validation set to evaluate the accuracy and reliability of the prediction model;

[0015] S7. Collect on-site data and input it into the prediction model to obtain the absolute stress prediction results.

[0016] An electronic device includes a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, it implements the above-described intelligent prediction method for absolute mining stress based on borehole oil conservator stress gauge monitoring.

[0017] A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the above-described intelligent prediction method for absolute mining stress based on borehole oil conservator stress gauge monitoring.

[0018] The present invention has the following advantages:

[0019] This invention creatively constructs a high-precision, intelligent prediction method for converting relative stress to absolute stress by combining in-situ relative stress monitoring data from borehole oil-filled stress gauges with a linear regression model and a Gaussian process regression prediction model established through multi-condition indoor loading tests. This method effectively solves the fundamental technical problem of existing borehole stress gauge methods, which can only provide relative stress and cannot directly obtain the original absolute stress of the rock mass.

[0020] Specifically, this invention, through rigorous coal and rock specimen preparation and rock mechanics loading tests, systematically collected relative and absolute stress data under the influence of multiple parameters such as confining pressure, initial pressure, borehole diameter, and uniaxial compressive strength of rock, and established preliminary linear quantitative relationships based on these data. Utilizing Gaussian process regression, a nonlinear machine learning model, and integrating Bayesian optimization to adaptively optimize the model's hyperparameters, it can accurately characterize the relative stress and the complex nonlinear mapping relationships between each loading parameter and absolute stress. This model not only outputs high-precision point prediction values ​​but also provides reliable prediction intervals, quantitatively characterizing the uncertainty of the prediction results.

[0021] This invention significantly improves the accuracy, reliability, and engineering applicability of mining-induced stress prediction. The resulting intelligent prediction model can quickly and accurately infer the absolute stress state of the surrounding rock based on a few easily measurable parameters such as relative stress and initial pressure collected in real time on-site. This enables quantitative and intelligent perception and prediction of the stress field in the mining area of ​​coal mines. This provides direct and reliable data support for optimizing roadway support design, controlling the advance speed of longwall mining faces, and providing accurate early warning of dynamic disasters such as rockbursts. Therefore, it effectively ensures safe and efficient coal mining and has significant theoretical implications and broad engineering application prospects. Attached Figure Description

[0022] Figure 1 This is a flowchart of an embodiment of the present invention regarding an intelligent prediction method for absolute mining stress based on borehole oil pillow stress gauge monitoring;

[0023] Figure 2 This is a flowchart illustrating the machine model learning process of the intelligent prediction method for absolute mining stress based on borehole oil conservator stress gauge monitoring, as described in an embodiment of the present invention.

[0024] Figure 3 This is a schematic diagram of the hydraulic connection of the borehole oil pillow stress gauge according to an embodiment of the present invention;

[0025] Figure 4 This is a diagram of the borehole stress monitoring and analysis test system according to an embodiment of the present invention;

[0026] Figure 5 This is a schematic diagram of the field application and installation of the borehole oil pillow stress gauge according to an embodiment of the present invention;

[0027] Explanation of reference numerals in the attached figures:

[0028] 11. Hydraulic pump; 12. Rock mechanics loading frame; 13. Loading cylinder; 14. Reaction pad; 2. Drilling oil pillow stress gauge; 21. Oil guide pipe; 22. Three-way valve; 23. Pillar body; 24. Oil injection check valve; 25. Oil bladder; 3. Coal and rock surrounding rock specimen; 4. Electronic pressure gauge; 5. Pressure pump; 6. Computer control system. Detailed Implementation

[0029] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0030] Example 1

[0031] like Figures 1 to 5 As shown, this embodiment proposes an intelligent prediction method for absolute mining-induced stress based on borehole oil tank stress gauge monitoring. The method includes the following steps:

[0032] S1. Prepare coal and rock surrounding rock specimens with boreholes and install borehole oil pillow stress gauges in the boreholes.

[0033] Prepare concrete mixtures whose mechanical properties represent those of the target coal and rock mass surrounding rock;

[0034] A detachable core mold is set in the molding die to reserve cylindrical drill holes. After pouring concrete mixture and curing, a coal and rock surrounding rock specimen with a pre-set diameter drill hole is formed inside.

[0035] S11. To simulate the overall stiffness and strength of the surrounding coal and rock, C42.5 cement was selected in this embodiment. By adjusting parameters such as the mud-sand ratio, the elastic modulus and compressive strength of the concrete were adjusted. Three types of specimens with equivalent uniaxial compressive strengths of approximately 20 MPa, 30 MPa, and 40 MPa were prepared. Cement, sand, and water were added to a mixer in proportions of 1:1.8:0.55, 1:1.5:0.45, and 1:1.2:0.38, respectively, and stirred until the mixture was homogeneous.

[0036] S12. A 300 mm × 300 mm × 300 mm steel mold is selected as the forming mold. Removable cylindrical core molds are arranged inside the mold at designed positions to reserve drilling locations on-site. Preferably, the core mold diameters are set to 40 mm, 45 mm, and 50 mm, corresponding to different drilling diameter conditions. Its axis is arranged at the center of the cubic specimen, but the minimum clear distance between the outer edge of the core mold and the edge of the specimen is not less than 3 to 4 times the hole diameter to reduce the influence of boundary effects.

[0037] S13. Pour the mixed concrete into the mold in layers, with each layer controlled within the range of 100-150 mm in thickness. Use an immersion vibrator or a table vibrating table to compact the concrete, ensuring that it is dense and free of obvious honeycomb and voids, while also ensuring that the reserved core mold position does not shift. After pouring, smooth the surface of the specimen and mark it with specimen number, pouring date, etc.

[0038] S14. After pouring, cure for 14 days in an environment with a temperature of 20±2 ℃ and a relative humidity greater than 95% to allow the concrete strength and elastic modulus to stabilize. After curing, remove the formwork and pull out the core mold with the pre-drilled hole to form a semi-through hole. If necessary, the outer surface of the specimen can be simply ground to ensure that all surfaces are flat and the dimensional deviations meet the installation requirements of the loading test device. Measure the dimensions and inspect the appearance of the formed cubic specimens. After confirming that the geometric dimensions of the specimens are 300 mm × 300 mm × 300 mm, the drilling position and diameter meet the design requirements, and there are no obvious cracks, spalling, or other defects, number them according to the working condition plan for use in subsequent loading tests.

[0039] S2. Apply confining pressure to the coal and rock surrounding rock specimen using a loading test device, and apply initial pressure to the borehole oil pillow stress gauge. Collect relative stress data through multi-condition loading; where confining pressure is the absolute stress.

[0040] The coal and rock surrounding rock specimen is placed in the loading test device (hydraulic pump 11 and rock mechanics loading frame 12). The loading cylinder 13 on the loading frame 12 applies the initial confining pressure of the set value to the coal and rock surrounding rock specimen in two orthogonal directions, so that the coal and rock surrounding rock specimen is in the intended triaxial stress state. The initial pressure is applied to the borehole oil pillow stress gauge 2 to make it fit against the borehole wall. The confining pressure on the coal and rock surrounding rock specimen is increased in stages, and the relative stress output by the confining pressure and the borehole oil pillow stress gauge is collected and recorded simultaneously.

[0041] The above process is repeated under different loading parameters such as confining pressure, initial pressure, and borehole diameter, and the corresponding relative stress is output by the borehole oil tank stress gauge to form a multi-condition stress response dataset.

[0042] S21. Place the coal and rock surrounding rock specimen 3 in the rock mechanics loading frame 12 and apply controlled normal pressure in each direction to simulate the triaxial stress state of the underground surrounding rock. During the clamping process, adjust the position of the reaction pad 14 to ensure that each surface of the coal and rock surrounding rock specimen 3 is fully in contact with the reaction pad 14, so as to ensure uniform force during loading.

[0043] S22. Install a borehole oil pillow stress gauge 2 inside the pre-drilled hole in the specimen, and connect it to an electronic pressure gauge 4 through an oil guide pipe 21. A pressure pump 5 is connected to the borehole oil pillow stress gauge 2 through the oil guide pipe 21 and a three-way valve 22. During installation, use a guide rod or positioning device to place the borehole oil pillow stress gauge 2 at a predetermined depth, ensuring that the pillow body 23 of the borehole oil pillow stress gauge 2 remains coaxial with the hole wall, avoiding uneven stress caused by eccentric installation.

[0044] After installation, the range of confining pressure P is set according to the original in-situ stress level of the target mining area. c The confining pressure is applied in stages from 5 MPa to 25 MPa to bring the specimen into a predetermined three-dimensional stress state. Then, oil is gradually injected into the oil bladder 25 of the borehole oil pillow stress gauge 2 through the oil injection check valve 24. The initial stress is applied to the borehole oil pillow stress gauge 2 by the pressure pump 5, and the pressure inside the hole is slowly increased to the set initial pressure P0, so that the pillow body 23 is fully attached to the hole wall and the gap at the contact surface is eliminated. The initial pressure P0 is empirically selected as 4 to 8 MPa.

[0045] S23. After the confining pressure and initial pressure have stabilized, an external loading stress is applied to the specimen to simulate the stress evolution process during mining. During the loading process, the horizontal stress and vertical stress are synchronously controlled according to a constant horizontal stress coefficient, where the horizontal stress coefficient is defined as:

[0046] ;

[0047] That is, the horizontal stress σ experienced by the specimen throughout the entire loading process. h Always vertical stress σ v 2 times.

[0048] Preferably, the vertical stress σ is gradually increased using a graded loading method. v And apply the corresponding horizontal stress σ synchronously according to the said horizontal stress coefficient. h The loading continues until the predetermined maximum stress level is reached or the specimen is close to its ultimate bearing capacity within a safe range. During loading, the electronic pressure gauge 4 outputs the relative stress in real time. The computer control system 6 synchronously records time t and confining pressure P. c Equal amounts, as follows:

[0049] , ;

[0050] in, P represents the confining pressure corresponding to the i-th record; 0,i D represents the initial pressure corresponding to the i-th record; c,i Let the diameter of the borehole be the diameter corresponding to the i-th record; Let be the uniaxial compressive strength corresponding to the i-th record.

[0051] The above loading process was repeated under different confining pressure levels, different initial pressures, different borehole diameters, and different uniaxial compressive strengths to form a relative stress response dataset with multiple working conditions and multiple loading stages, providing rich samples for subsequent linear regression modeling and Gaussian process modeling.

[0052] S3. Establish a linear regression model between absolute stress, relative stress, and loading parameters.

[0053] S31. Organize the multi-condition stress response dataset obtained in S2 according to the loading sequence to form a data record set consisting of relative stress, initial pressure, confining pressure, borehole diameter, and uniaxial compressive strength of rock. Each data record contains... ;

[0054] Where, σ r,i P represents the relative stress corresponding to the i-th record; 0,i P represents the initial pressure corresponding to the i-th record; L，i D represents the uniaxial compressive strength of the rock corresponding to the i-th record; c,i σ is the borehole diameter corresponding to the i-th record; a,i The confining pressure corresponding to the i-th record;

[0055] S32, the confining pressure, i.e., the absolute stress σ a As the dependent variable, σ r P0, D c P L As the independent variable, construct a linear regression model:

[0056] ;

[0057] Where, σ a For confining pressure; σ r P0 is the relative stress; D is the initial pressure; c P is the borehole diameter; L β1 is the uniaxial compressive strength of rock; β2 is the regression coefficient of initial pressure; β3 is the regression coefficient of borehole diameter; β4 is the regression coefficient of uniaxial compressive strength of rock; ε is the error term.

[0058] S33. Convert the N records into matrix form, construct the variable matrix X, the absolute stress matrix Y, and the regression coefficient β, and solve for them. The specific formula is as follows:

[0059] ;

[0060] ;

[0061] ;

[0062] in, The relative stress measured by the borehole oil tank stress gauge in the i-th data set; Let be the initial pressure in the i-th data set; The diameter of the borehole in the i-th data set; The uniaxial compressive strength of the rock in the i-th data set; The absolute stress is in the i-th set of data;

[0063] The coefficient vector β is solved using least squares estimation, as shown in the following formula:

[0064] ;

[0065] in, Indicates matrix transpose; This represents matrix inversion; Y is the target vector, containing the actual absolute stress for each record; The optimal regression coefficients are obtained to reflect the linear contribution of the input variables to the absolute stress.

[0066] S34. Using the optimal regression coefficients Calculate the predicted absolute stress value for each sample, and then calculate the residual between the predicted and actual absolute stress values. The specific formula is as follows:

[0067] ;

[0068] in, X represents the predicted absolute stress value; X is the input matrix, including all independent variables. This represents the optimal regression coefficient vector;

[0069] residual e i Defined as the difference between the actual value and the predicted value for each sample, the formula is as follows:

[0070] ;

[0071] Among them, e i Let be the residual of the i-th sample. Let be the actual absolute stress of the i-th sample. Let be the predicted absolute stress of the i-th sample;

[0072] S35. Calculate the Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE) to evaluate the error of the regression model and provide a basis for the next step of model optimization.

[0073] in,

[0074] MSE is used to reflect the mean square degree of the overall error, and the specific formula is as follows:

[0075] ;

[0076] Where N is the total number of samples; It is the measured or calibrated absolute stress value corresponding to the i-th test sample; It is the predicted absolute stress value calculated by the linear regression model; the smaller the MSE, the smaller the overall fitting error of the model, indicating that the model prediction is closer to the true value.

[0077] RMSE, as the square root of MSE, more directly characterizes the average deviation of the prediction error from its original dimensions. The specific formula is as follows:

[0078] ;

[0079] Where RMSE is the square root of MSE, and remains constant with absolute stress. The same dimensions (MPa); the smaller the RMSE, the more accurate the linear fit of the model to the absolute stress; in engineering applications, RMSE is often used to directly assess whether the prediction deviation is within an acceptable range.

[0080] To avoid the possibility that high-stress samples might dominate the error assessment if only absolute error indicators are used, the Mean Relative Error (MAPE) index is introduced to normalize the error of each sample under different stress levels. Its calculation formula is as follows:

[0081] ;

[0082] in, It is the measured or calibrated absolute stress value corresponding to the i-th test sample; MAPE is the predicted absolute stress value calculated by the linear regression model; the MAPE value reflects the relative proportion of the prediction error to the measured value, and is particularly suitable for multi-condition test conditions with a large range of stress amplitude variations. It is used to characterize the overall prediction reliability of the model under different stress levels. The smaller the MAPE, the better the overall prediction stability and balance of the model.

[0083] By using MSE and RMSE to control the absolute magnitude of prediction error and MAPE to evaluate the relative consistency of prediction error among different working conditions, we can comprehensively judge the fitting ability of the linear regression model to the monitoring data and its reliability as a priori basis for subsequent Gaussian process modeling, providing a quantitative basis for data weighted fusion and model training.

[0084] S4. Preprocess and standardize the data obtained from S2 and S3, and divide them into training and validation sets.

[0085] S41. Classify the data obtained in steps S2 and S3 according to the measurement points and loading stages, calculate the statistical characteristics of each variable, and use the 3σ principle to detect and remove outliers. The outlier judgment formula is as follows:

[0086] >3s;

[0087] Where, x i Let be the measured value of a certain variable in the i-th sample; Let be the mean of the i samples; s be the sample standard deviation; the formula for calculating s is as follows:

[0088] ;

[0089] Where N is the total number of samples;

[0090] If the anomaly detection formula is satisfied, then x is considered... i Abnormalities should be removed.

[0091] The principle is based on the statistical property of the normal distribution, that is, when the population of data approximately follows a normal distribution, about 68.27% of the data falls within the range of... Within the range, 95.45% fell within... Within the range, 99.73% of the data fell within... Within the range. Therefore, data points exceeding ±3σ can be considered extremely unlikely outliers. The normal range for this variable is... This interval covers an estimated 99.73% of normal data points and can be considered a reasonable fluctuation range for this variable under current operating conditions. For each data point... If its deviation from the mean exceeds 3 times the standard deviation, it satisfies the condition. Then the data point is determined to be an outlier.

[0092] S42. To reduce fluctuation noise in the stress acquisition signal, a moving average method is used to smooth the data and suppress noise. The specific formula is as follows:

[0093] ;

[0094] in, x represents the i-th data point after smoothing. j The original j-th data point; k is the half width of the sliding window;

[0095] S43. Using the residuals obtained in S3 as weighting factors, the data are fused. Let the root mean square error (RMSE) of the j-th working condition be denoted as RMSE. j Then its weight is defined as:

[0096] ;

[0097] Where, ω j δ represents the weight of the j-th working condition; δ is a small constant to prevent the denominator from being zero; RMSE j The root mean square error for the j-th working condition is calculated using the following formula:

[0098] ;

[0099] Where, n j Let j be the number of samples for the j-th working condition. Let j be the sample index set for the j-th working condition;

[0100] The fused weighted stress data is calculated using the following formula:

[0101] ;

[0102] in, The absolute stress after fusion; m is the total number of working conditions; This represents the average absolute stress value corresponding to the j-th working condition;

[0103] S44. To eliminate dimensional differences and magnitude biases, the zero-mean, unit-variance Z-score standardization method is used to transform each variable into dimensionless data with zero mean and unit standard deviation. The standardization formula is as follows:

[0104] ;

[0105] in, The standardized variable value; x i The original variable value; is the statistical mean of the variable calculated in the corresponding sample set; s is the sample standard deviation of the variable calculated in the corresponding sample set;

[0106] S45. Divide the standardized sample data proportionally to form the model training set and validation set:

[0107] ;

[0108] Where, N train N is the number of samples in the training set;val The number of validation set samples; N is the total number of standardized samples.

[0109] S5. Based on the training set, an absolute stress prediction model is constructed using Gaussian process regression combined with Bayesian optimization.

[0110] S51. Construct the structure of the Gaussian process regression model, and represent the standardized input variable vector X as follows:

[0111] ;

[0112] in, , , , The relative stress, initial pressure, borehole diameter, and uniaxial compressive strength of rock are S4 standardized values.

[0113] A Gaussian process regression method is used to establish a nonlinear mapping relationship between absolute stress and input variables. The objective function y(X) is set as a Gaussian process model with a mean of 0 and a covariance determined by a kernel function, expressed as:

[0114] ;

[0115] Where y(X) is the standardized absolute stress, This is the covariance kernel function, used to measure the correlation between different sample inputs;

[0116] S52. Using the radial basis kernel function as the covariance function, its expression is:

[0117] ;

[0118] in, The signal variance characterizes the amplitude of absolute stress variation. The length scale controls the sensitivity of the function to changes in the input. This represents the noise variance, used to characterize experimental measurement noise. For the Kronecker delta function, when The value is 1 if it is true, and 0 otherwise. The Euclidean distance between the input variables;

[0119] S53. Construct the training covariance matrix and the training target vector;

[0120] Suppose the training set contains N train The training set input matrix consists of 10 samples and their corresponding standardized input vectors. :

[0121] ;

[0122] The training covariance matrix is:

[0123] ;

[0124] in, This is the set of hyperparameters for the radial basis function kernel. The signal variance is used to characterize the magnitude of change in the objective function. This is the length scale parameter, used to control how sensitive the model is to changes in input variables; N represents the noise variance, used to characterize the uncertainty of measurement noise on the model output; train The number of training samples. As the input matrix for the training set, X i K is the standardized input vector for the i-th training sample, containing relative stress, initial pressure, borehole diameter, and uniaxial compressive strength of the rock as characteristic variables; train To train the set covariance matrix, This represents the element in the i-th row and j-th column of the covariance matrix. Its value is calculated by the radial basis kernel function and reflects the similarity between the i-th and j-th training sample input vectors. Given a set of hyperparameters θ, the radial basis function (RBF) is used to compute the input vector X. i With X j The covariance between them;

[0125] The target vector of the training set is composed of the standardized absolute stress values ​​corresponding to each training sample, and its expression is:

[0126] ;

[0127] in, Let be the standardized absolute stress target value corresponding to the i-th training sample;

[0128] S54. Radial basis function hyperparameter optimization and model training: During the training process, the hyperparameters of the radial basis function are optimized by maximizing the log-marginal likelihood function as follows:

[0129] ;

[0130] The expression for the logarithmic marginal likelihood function is:

[0131] ;

[0132] Among them, y train The training set consists of standardized absolute stress target vectors; The matrix transpose of the standardized absolute stress target vector in the training set; For the target vector y in the training settrain The logarithmic marginal likelihood function; X train Input matrix for the training set; K is the set of hyperparameters for the radial basis function kernel; train For kernel function The calculated training set covariance matrix; Let be the determinant of the covariance matrix; The inverse of the covariance matrix; N train The number of training samples is represented by the constant log(2π), which is used to ensure the normalization of the probability distribution form.

[0133] A Bayesian optimization algorithm is used to iteratively optimize the hyperparameter θ. In each iteration, the marginal likelihood is used to calculate the current parameter performance, and the parameter is adjusted accordingly to finally obtain the optimal hyperparameter. Optimal hyperparameters Specifically , and ;

[0134] In the specific implementation process, this step uses the Bayesian optimization algorithm to search for hyperparameters.

[0135] First, set a reasonable search range for each hyperparameter:

[0136] , , .

[0137] Then, several initial sample points are selected within the parameter space, and the corresponding log-marginal likelihood values ​​are calculated. A surrogate model is constructed based on the existing evaluation samples to approximately express the response relationship between hyperparameters and the log-likelihood function. By setting the expected improvement function EI, the next set of candidate hyperparameter points is selected in the parameter space for evaluation. The surrogate model is iteratively updated continuously, causing the search region to gradually converge towards the global optimum.

[0138] The iterative process may terminate when any of the following conditions are met:

[0139] a. The logarithmic marginal likelihood increment is lower than a set threshold for several consecutive iterations;

[0140] b. Reach the preset maximum number of iterations;

[0141] c. The variation range of hyperparameters tends to stabilize.

[0142] After training, the final Gaussian process regression model is obtained as follows:

[0143] ;

[0144] in, Indicates optimal hyperparameters A defined Gaussian process regression model; This indicates that the mean function of the Gaussian process regression model is zero, and the covariance function is the radial basis function after substituting the optimal hyperparameters. Given; X and Let each represent any two input vectors, whose elements are standardized relative stress, initial pressure, borehole diameter, and characteristic variables of uniaxial compressive strength of rock.

[0145] S55. Predict the absolute stress and construct the prediction range;

[0146] To obtain the optimal hyperparameters of S54 , and Then, the input vector of the validation set is... Perform absolute stress prediction;

[0147] Calculate the covariance vector between the test input and the training set input based on the radial basis kernel function:

[0148] ;

[0149] in, This is the covariance vector between the test samples and the training samples; X represents the radial basis function value after substituting the optimal hyperparameters; i Let N be the input vector for the i-th training sample; train This represents the number of training samples;

[0150] The standardized prediction of absolute stress is given by the following formula:

[0151] ;

[0152] in, This is a standardized prediction of the absolute stress. To train the inverse of the covariance matrix; The absolute stress vector after standardization of the training set;

[0153] The prediction variance is:

[0154] ;

[0155] in, The kernel function value of the test point itself; The prediction variance of the validation set samples is used to quantify the uncertainty of the model's prediction results for that sample. This is the covariance vector between the test samples and the training samples; Input vector for the validation set; The inverse of the covariance matrix; This is the matrix transpose of the covariance matrix;

[0156] The predicted values ​​were denormalized to obtain the final absolute stress prediction results as follows:

[0157] ;

[0158] in, To validate the predicted absolute stress values ​​corresponding to the sample set; and These are the mean and standard deviation of the absolute stress, respectively.

[0159] To reflect the prediction confidence interval, a 95% prediction interval for the absolute stress is constructed based on the prediction variance:

[0160] , ;

[0161] in, and These are the lower and upper limits of the prediction, respectively.

[0162] S6. Use the validation set to evaluate the accuracy and reliability of the prediction model.

[0163] S61. Input feature vectors of the validation set Input the Gaussian process regression model trained with S5 and calculate the standardized predicted values. and prediction variance And by inverse standardization, we obtain:

[0164] ;

[0165] in, These are the standardized input variables: relative stress, initial pressure, borehole diameter, and uniaxial compressive strength of rock. The standardized absolute stress prediction value output by the Gaussian process regression model; To train the standard deviation of concentrated absolute stress; To train the mean of the concentrated absolute stress; The absolute stress prediction value after destandardization;

[0166] S62. Collect the measured absolute stress value of each sample point in the verification set:

[0167] , ;

[0168] Following the prediction process in S61, the trained Gaussian process regression model is used to predict the input vectors of each sample in the validation set, obtaining the predicted absolute stress value corresponding to each measured absolute stress value:

[0169] , ;

[0170] Calculate the average measured absolute stress of the validation set:

[0171] ;

[0172] Calculate the coefficient of determination R 2 The following formula is used to evaluate the model's fit to the overall trend of the validation set:

[0173] ;

[0174] in, Let be the measured absolute stress of the i-th verification sample; The absolute stress prediction value is obtained from S61; To verify the mean of the absolute stress of the set; The number of samples in the validation set; the coefficient of determination R. 2 The value range is [0, 1], R 2 The closer the value is to 1, the higher the model fitting accuracy.

[0175] when This indicates that the predicted results are highly consistent with the measured values; This indicates a good fit and can be used to assist in analysis and judgment; when When the model is considered to have insufficient fitting accuracy, it is necessary to readjust the hyperparameters or supplement the training samples.

[0176] S63. Calculate the root mean square error, mean absolute error, and mean relative error, respectively, to comprehensively quantify the prediction accuracy of the Gaussian process regression model under different confining pressure conditions and stress levels. The model is considered to have engineering-usable accuracy when both of the following conditions are met:

[0177] Condition a: Both the root mean square error and the mean absolute error are within the allowable absolute error control range;

[0178] Condition b: The average relative error remains below the preset threshold, indicating that the model has stable predictive performance in both high-stress and low-stress regions.

[0179] S64. For each validation sample point, the predicted standard deviation is calculated using the Gaussian process model:

[0180] ;

[0181] in, Let be the predicted standard deviation of the i-th validation sample; To verify the standardized input feature vector corresponding to the i-th sample in the set; This is the covariance vector between the test point and all training samples; To train the set covariance matrix;

[0182] Assuming the prediction error approximately follows a normal distribution, construct the 95% prediction interval for the i-th sample:

[0183] ;

[0184] ;

[0185] in, To predict absolute stress; and These represent the lower and upper limits of the prediction interval, respectively; the coefficient 1.96 is the quantile coefficient corresponding to the standard normal distribution at the 95% confidence level;

[0186] The measured absolute stress value Compare with the constructed prediction interval, when the condition is met When the predicted results of a sample are statistically consistent with the measured results, it is determined that the predicted results are statistically consistent.

[0187] The interval coverage C of all validation samples is calculated:

[0188] ;

[0189] Where, N in N represents the number of samples whose measured values ​​fall within the corresponding prediction interval. val To verify the total number of samples;

[0190] When coverage At that time, it was believed that the predictive model's ability to characterize uncertainty met the needs of safety assessment and on-site early warning applications.

[0191] S7. Collect on-site data and input it into the prediction model to obtain the absolute stress prediction results.

[0192] S71. In the surrounding rock of the target working face roadway, boreholes are arranged according to the designed spacing and depth. Borehole oil-filled stress gauges are installed in the boreholes. After the borehole oil-filled stress gauges are installed, an initial pressure P0 is applied to the borehole oil-filled stress gauges by oil injection to make the stress gauges fit against the borehole wall. The relative stress σ inside the borehole is recorded in real time. r and the uniaxial compressive strength P of the rock L and borehole diameter D cTogether they form the original input vector ;

[0193] S72, change the original input vector Perform zero-mean, unit-variance Z-score standardization in step S4:

[0194] ;

[0195] in, The original input vector The j-th component; This is the mean of the j-th input variable obtained from the statistical analysis of the training set samples in step S4; This is the sample standard deviation of the j-th input variable calculated based on the training set samples in step S4. This represents the standardized value of the j-th input variable.

[0196] After standardizing the four input variables respectively, the standardized input vector is obtained:

[0197] ;

[0198] S73, will Input the pre-trained Gaussian process regression model (S5) and calculate the standardized predicted values. and prediction variance And by inverse standardization, we obtain:

[0199] ;

[0200] in, This is the predicted absolute stress value at the corresponding location on site; The normalized predicted value of absolute stress output by the Gaussian process regression model; To train the standard deviation of concentrated absolute stress; To train the mean of the concentrated absolute stress;

[0201] The predicted absolute stress value of the surrounding rock at the borehole location is obtained, forming the spatial distribution of absolute stress in the surrounding rock of the working face;

[0202] S74. Using Prediction Variance For each monitoring point on site, a 95% prediction interval was constructed:

[0203] , ;

[0204] in, and These are the lower and upper limits of the prediction, respectively.

[0205] Compare the predicted values ​​with the design-allowed surrounding rock stress threshold: when or If the pre-set safety threshold is exceeded, the absolute stress level of the surrounding rock at the borehole location is determined to be too high. It is recommended to take measures such as strengthening support, slowing down the drilling speed, and optimizing the mining sequence to achieve early identification and proactive control of high-stress hazard areas.

[0206] Example 2

[0207] This embodiment 2 describes an electronic device including a memory and one or more processors. Executable code is stored in the memory. When the processor executes the executable code, it implements the steps of the intelligent prediction method for absolute mining stress based on borehole oil conservator stress gauge monitoring as described in embodiment 1 above.

[0208] Example 3

[0209] This embodiment 3 describes a computer-readable storage medium storing a program that, when executed by a processor, is used to implement the steps of the intelligent prediction method for absolute mining stress based on borehole oil pillow stress gauge monitoring as described in embodiment 1 above.

[0210] The computer-readable storage medium can be an internal storage unit of any device or apparatus with data processing capabilities, such as a hard disk or memory, or an external storage device of any device with data processing capabilities, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc.

[0211] The present invention has now been described in detail with reference to the accompanying drawings. Based on the above description, those skilled in the art should have a clear understanding of the present invention. Of course, the above description is merely a preferred embodiment of the present invention, and the present invention is not limited to the embodiments listed above. It should be noted that any equivalent substitutions or obvious modifications made by those skilled in the art under the guidance of this specification fall within the substantial scope of this specification and should be protected by the present invention.

Claims

1. A method for intelligent prediction of absolute mining-induced stress based on borehole oil tank stress gauge monitoring, characterized in that, The method includes the following steps: S1. Prepare coal and rock surrounding rock specimens with boreholes and install borehole oil-filled stress gauges in the boreholes. S2. Apply confining pressure to the coal and rock surrounding rock specimen using a loading test device, and apply initial pressure to the borehole oil pillow stress gauge. Collect relative stress data through multi-condition loading; where confining pressure is the absolute stress. S3. Establish a linear regression model between absolute stress, relative stress, and loading parameters; S4. Preprocess and standardize the data obtained from S2 and S3, and divide them into training and validation sets. S5. Based on the training set, an absolute stress prediction model is constructed using Gaussian process regression combined with Bayesian optimization. S5 includes: S51. Construct the structure of the Gaussian process regression model, and represent the standardized input variable vector X as follows: ； in, , , , The relative stress, initial pressure, borehole diameter, and uniaxial compressive strength of rock are S4 standardized values. A Gaussian process regression method is used to establish a nonlinear mapping relationship between absolute stress and input variables. The objective function y(X) is set as a Gaussian process model with a mean of 0 and a covariance determined by a kernel function, expressed as: ； Where y(X) is the standardized absolute stress, This is the covariance kernel function, used to measure the correlation between different sample inputs; S52. Using the radial basis kernel function as the covariance function, its expression is: ； in, The signal variance characterizes the amplitude of absolute stress variation. The length scale controls the sensitivity of the function to changes in the input. This represents the noise variance, used to characterize experimental measurement noise. For the Kronecker delta function, when The value is 1 if it is true, and 0 otherwise. The Euclidean distance between the input variables; S53. Construct the training covariance matrix and the training target vector; Suppose the training set contains N train The training set input matrix consists of 10 samples and their corresponding standardized input vectors. : ； The training covariance matrix is: ； in, This is the set of hyperparameters for the radial basis function kernel. The signal variance is used to characterize the magnitude of change in the objective function. This is the length scale parameter, used to control how sensitive the model is to changes in input variables; N represents the noise variance, used to characterize the uncertainty of measurement noise on the model output; train The number of training samples. As the input matrix for the training set, X i K is the standardized input vector for the i-th training sample, containing relative stress, initial pressure, borehole diameter, and uniaxial compressive strength of the rock as characteristic variables; train To train the set covariance matrix, This represents the element in the i-th row and j-th column of the covariance matrix. Its value is calculated by the radial basis kernel function and reflects the similarity between the i-th and j-th training sample input vectors. Given a set of hyperparameters θ, the radial basis function (RBF) is used to compute the input vector X. i With X j The covariance between them; The target vector for the training set is composed of the standardized absolute stress values ​​corresponding to each training sample, and its expression is: ； in, Let be the standardized absolute stress target value corresponding to the i-th training sample; S54. Radial basis function hyperparameter optimization and model training: During the training process, the hyperparameters of the radial basis function are optimized by maximizing the log-marginal likelihood function as follows: ； The expression for the logarithmic marginal likelihood function is: ； Among them, y train The training set consists of standardized absolute stress target vectors; The matrix transpose of the standardized absolute stress target vector in the training set; For the target vector y in the training set train The logarithmic marginal likelihood function; X train Input matrix for the training set; K is the set of hyperparameters for the radial basis function kernel; train For kernel function The calculated training set covariance matrix; Let be the determinant of the covariance matrix; The inverse of the covariance matrix; N train The number of training samples is represented by the constant log(2π), which is used to ensure the normalization of the probability distribution form. A Bayesian optimization algorithm is used to iteratively optimize the hyperparameter θ. In each iteration, the marginal likelihood is used to calculate the current parameter performance, and the parameter is adjusted accordingly to finally obtain the optimal hyperparameter. Optimal hyperparameters Specifically , and ; After training, the final Gaussian process regression model is obtained as follows: ； in, Indicates optimal hyperparameters A defined Gaussian process regression model; This indicates that the mean function of the Gaussian process regression model is zero, and the covariance function is the radial basis function after substituting the optimal hyperparameters. Given; X and Let each represent any two input vectors, whose elements are standardized relative stress, initial pressure, borehole diameter, and characteristic variables of uniaxial compressive strength of rock. S55. Predict the absolute stress and construct the prediction range; To obtain the optimal hyperparameters of S54 , and Then, the input vector of the validation set is... Perform absolute stress prediction; Calculate the covariance vector between the test input and the training set input based on the radial basis kernel function: ； in, This is the covariance vector between the test samples and the training samples; X represents the radial basis function value after substituting the optimal hyperparameters; i Let N be the input vector for the i-th training sample; train This represents the number of training samples; The standardized prediction of absolute stress is given by the following formula: ； in, This is a standardized prediction of the absolute stress. To train the inverse of the covariance matrix; The absolute stress vector after standardization of the training set; The prediction variance is: ； in, The kernel function value of the test point itself; The prediction variance of the validation set samples is used to quantify the uncertainty of the model's prediction results for that sample. This is the covariance vector between the test samples and the training samples; Input vector for the validation set; The inverse of the covariance matrix; This is the matrix transpose of the covariance matrix; The predicted values ​​were denormalized to obtain the final absolute stress prediction results as follows: ； in, To validate the predicted absolute stress values ​​corresponding to the sample set; and These are the mean and standard deviation of the absolute stress, respectively. To reflect the prediction confidence interval, a 95% prediction interval for the absolute stress is constructed based on the prediction variance: ， ； in, and These are the lower and upper limits of the prediction, respectively. S6. Use the validation set to evaluate the accuracy and reliability of the prediction model; S7. Collect on-site data and input it into the prediction model to obtain the absolute stress prediction results.

2. The intelligent prediction method for absolute mining-induced stress based on borehole oil conservator stress gauge monitoring according to claim 1, characterized in that, S1 includes: Prepare concrete mixtures whose mechanical properties represent those of the target coal and rock mass surrounding rock; A detachable core mold is set in the molding die to reserve drilling holes. After pouring concrete mixture and curing, a coal and rock surrounding rock specimen with pre-drilled holes inside is formed.

3. The intelligent prediction method for absolute mining-induced stress based on borehole oil conservator stress gauge monitoring according to claim 1 or 2, characterized in that, S2 includes: The coal and rock surrounding rock specimens were placed in the loading test device; The initial confining pressure of the coal and rock surrounding rock specimen is applied to the specimen in two orthogonal directions by the loading test device, so that the specimen is in the desired stress state. The initial pressure is applied to the borehole oil pillow stress gauge to make it fit against the borehole wall. The confining pressure was applied to the coal and rock surrounding rock specimens in stages, and the confining pressure and the relative stress output by the borehole oil pillow stress gauge were collected and recorded simultaneously. The above process is repeated under different confining pressures, initial pressures, and borehole diameter loading parameters, and the corresponding relative stress is output by the borehole oil tank stress gauge to form a multi-condition stress response dataset.

4. The intelligent prediction method for absolute mining-induced stress based on borehole oil conservator stress gauge monitoring according to claim 3, characterized in that, S3 includes: S31. Organize the multi-condition stress response dataset obtained in S2 according to the loading sequence to form a data record set consisting of relative stress, initial pressure, confining pressure, borehole diameter, and uniaxial compressive strength of rock. Each data record contains... ; Where, σ r,i P represents the relative stress corresponding to the i-th record; 0,i P represents the initial pressure corresponding to the i-th record; L，i D represents the uniaxial compressive strength of the rock corresponding to the i-th record; c,i σ is the borehole diameter corresponding to the i-th record; a,i The confining pressure corresponding to the i-th record; S32, the confining pressure σ a As the dependent variable, σ r P0, D c P L As the independent variable, construct a linear regression model: ； Where, σ a For confining pressure; σ r P0 is the relative stress; D is the initial pressure; c P is the borehole diameter; L β1 is the uniaxial compressive strength of rock; β2 is the regression coefficient of initial pressure; β3 is the regression coefficient of borehole diameter; β4 is the regression coefficient of uniaxial compressive strength of rock; ε is the error term. S33. Convert the N records into matrix form, construct the variable matrix X, the absolute stress matrix Y, and the regression coefficient β, and solve for them. The specific formula is as follows: ； ； ； in, The relative stress measured by the borehole oil tank stress gauge in the i-th data set; Let be the initial pressure in the i-th data set; The diameter of the borehole in the i-th data set; The uniaxial compressive strength of the rock in the i-th data set; The absolute stress is in the i-th set of data; The coefficient vector β is solved using least squares estimation, as shown in the following formula: ； in, Indicates matrix transpose; This represents matrix inversion; Y is the target vector, containing the actual absolute stress for each record; The optimal regression coefficients are obtained to reflect the linear contribution of the input variables to the absolute stress. S34. Using the optimal regression coefficients Calculate the predicted absolute stress value for each sample, and then calculate the residual between the predicted and actual absolute stress values. The specific formula is as follows: ； in, X represents the predicted absolute stress value; X is the input matrix, including all independent variables. This represents the optimal regression coefficient vector; residual e i Defined as the difference between the actual value and the predicted value for each sample, the formula is as follows: ； Among them, e i Let be the residual of the i-th sample. Let be the actual absolute stress of the i-th sample. Let be the predicted absolute stress of the i-th sample; S35. Calculate the mean square error, root mean square error, and average relative error to evaluate the error of the regression model and provide a basis for the next step of model optimization.

5. The intelligent prediction method for absolute mining-induced stress based on borehole oil conservator stress gauge monitoring according to claim 4, characterized in that, S4 includes: S41. Classify the data obtained in steps S2 and S3 according to the measurement points and loading stages, calculate the statistical characteristics of each variable, and use the 3σ principle to detect and remove outliers. The outlier judgment formula is as follows: ＞3s； Where, x i Let be the measured value of a certain variable in the i-th sample; Let be the mean of the i samples; s be the sample standard deviation; the formula for calculating s is as follows: ； Where N is the total number of samples; If the anomaly detection formula is satisfied, then x is considered... i Abnormalities should be removed. S42. To reduce fluctuation noise in the stress acquisition signal, a moving average method is used to smooth the data and suppress noise. The specific formula is as follows: ； in, x represents the i-th data point after smoothing. j The original j-th data point; k is the half width of the sliding window; S43. Using the residuals obtained in S3 as weighting factors, the data are fused. Let the root mean square error (RMSE) of the j-th working condition be denoted as RMSE. j Then its weight is defined as: ； Where, ω j δ represents the weight of the j-th working condition; δ is a small constant to prevent the denominator from being zero; RMSE j The root mean square error for the j-th working condition is calculated using the following formula: ； Where, n j Let j be the number of samples for the j-th working condition. Let j be the sample index set for the j-th working condition; The fused weighted stress data is calculated using the following formula: ； in, The absolute stress after fusion; m is the total number of working conditions; This represents the average absolute stress value corresponding to the j-th working condition; S44. To eliminate dimensional differences and magnitude biases, the zero-mean, unit-variance Z-score standardization method is used to transform each variable into dimensionless data with zero mean and unit standard deviation. The standardization formula is as follows: ； in, The standardized variable value; x i The original variable value; is the statistical mean of the variable calculated in the corresponding sample set; s is the sample standard deviation of the variable calculated in the corresponding sample set; S45. Divide the standardized sample data proportionally to form the model training set and validation set: ； Where, N train N is the number of samples in the training set; val The number of validation set samples; N is the total number of standardized samples.

6. The intelligent prediction method for absolute mining-induced stress based on borehole oil conservator stress gauge monitoring according to claim 1, characterized in that, S6 includes: S61. Input feature vectors of the validation set Input the Gaussian process regression model trained with S5 and calculate the standardized predicted values. and prediction variance And by inverse standardization, we obtain: ； in, These are the standardized input variables: relative stress, initial pressure, borehole diameter, and uniaxial compressive strength of rock. The standardized absolute stress prediction value output by the Gaussian process regression model; To train the standard deviation of concentrated absolute stress; To train the mean of the concentrated absolute stress; The absolute stress prediction value after destandardization; S62. Collect the measured absolute stress value of each sample point in the verification set: ， ； Following the prediction process in S61, the trained Gaussian process regression model is used to predict the input vectors of each sample in the validation set, obtaining the predicted absolute stress value corresponding to each measured absolute stress value: ， ； Calculate the average measured absolute stress of the validation set: ； Calculate the coefficient of determination R 2 The following formula is used to evaluate the model's fit to the overall trend of the validation set: ； in, Let be the measured absolute stress of the i-th verification sample; The absolute stress prediction value is obtained from S61; To verify the mean of the absolute stress of the set; The number of samples in the validation set; the coefficient of determination R. 2 The value range is [0, 1], R 2 The closer the value is to 1, the higher the model fitting accuracy. when This indicates that the predicted results are highly consistent with the measured values; This indicates a good fit and can be used to assist in analysis and judgment; when When the model is considered to have insufficient fitting accuracy, it is necessary to readjust the hyperparameters or supplement the training samples. S63. Calculate the root mean square error, mean absolute error, and mean relative error, respectively, to comprehensively quantify the prediction accuracy of the Gaussian process regression model under different confining pressure conditions and stress levels. The model is considered to have engineering-usable accuracy when both of the following conditions are met: Condition a: Both the root mean square error and the mean absolute error are within the allowable absolute error control range; Condition b: The average relative error remains below the preset threshold, indicating that the model has stable predictive performance in both high-stress and low-stress regions. S64. For each validation sample point, the predicted standard deviation is calculated using the Gaussian process model: ； in, Let be the predicted standard deviation of the i-th validation sample; To verify the standardized input feature vector corresponding to the i-th sample in the set; This is the covariance vector between the test point and all training samples; To train the set covariance matrix; Assuming the prediction error approximately follows a normal distribution, construct the 95% prediction interval for the i-th sample: ； ； in, To predict absolute stress; and These represent the lower and upper limits of the prediction interval, respectively; the coefficient 1.96 is the quantile coefficient corresponding to the standard normal distribution at the 95% confidence level; The measured absolute stress value Compare with the constructed prediction interval, when the condition is met When the predicted results of a sample are statistically consistent with the measured results, it is determined that the predicted results are statistically consistent. The interval coverage C of all validation samples is calculated: ； Where, N in N represents the number of samples whose measured values ​​fall within the corresponding prediction interval. val To verify the total number of samples; When coverage At that time, it was believed that the predictive model's ability to characterize uncertainty met the needs of safety assessment and on-site early warning applications.

7. The intelligent prediction method for absolute mining-induced stress based on borehole oil conservator stress gauge monitoring according to claim 6, characterized in that, S7 includes: S71. In the surrounding rock of the target working face roadway, boreholes are arranged according to the designed spacing and depth. Borehole oil-filled stress gauges are installed in the boreholes. After the borehole oil-filled stress gauges are installed, an initial pressure P0 is applied to the borehole oil-filled stress gauges by oil injection to make the stress gauges fit against the borehole wall. The relative stress σ inside the borehole is recorded in real time. r and the uniaxial compressive strength P of the rock L and borehole diameter D c Together they form the original input vector ; S72, change the original input vector Perform zero-mean, unit-variance Z-score standardization in step S4: ； in, The original input vector The j-th component; This is the mean of the j-th input variable obtained from the statistical analysis of the training set samples in step S4; This is the sample standard deviation of the j-th input variable calculated based on the training set samples in step S4. This represents the standardized value of the j-th input variable. After standardizing the four input variables respectively, the standardized input vector is obtained: ； S73, will Input the pre-trained Gaussian process regression model (S5) and calculate the standardized predicted values. and prediction variance And by inverse standardization, we obtain: ； in, This is the predicted absolute stress value at the corresponding location on site; The normalized predicted value of absolute stress output by the Gaussian process regression model; To train the standard deviation of concentrated absolute stress; To train the mean of the concentrated absolute stress; The predicted absolute stress value of the surrounding rock at the borehole location is obtained, forming the spatial distribution of absolute stress in the surrounding rock of the working face; S74. Using Prediction Variance For each monitoring point on site, a 95% prediction interval was constructed: ， ； in, and These are the lower and upper limits of the prediction, respectively. Compare the predicted values ​​with the design-allowed surrounding rock stress threshold: when or When the absolute stress level of the surrounding rock at the borehole location exceeds the preset safety threshold, it is determined that the absolute stress level is too high.

8. An electronic device comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that, When the processor executes the executable code, it implements the intelligent prediction method for absolute mining stress based on borehole oil pillow stress gauge monitoring as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements the intelligent prediction method for absolute mining stress based on borehole oil tank stress gauge monitoring as described in any one of claims 1 to 7.

Images