Water-inrush monitoring and early warning method for tunnel under water-rock coupling effect

Abstract

The application discloses a tunnel water inrush and gushing monitoring and early warning method under water-rock coupling, and relates to the technical field of karst tunnels.The method can be used for detecting the positions of karst caves and fissures of a tunnel under water-rock coupling, detecting real-time changes of water-bearing fissure stress, analyzing address disaster influence factors based on machine learning to obtain evaluation indexes, and thus establishing a tunnel water inrush monitoring and early warning system, so that a more accurate and rapid early warning method for tunnel water inrush problems is provided.

Description

Technical Field

[0001] This invention relates to the field of karst tunnel technology, specifically to a method for monitoring and early warning of water inrush in tunnels under water-rock coupling. Background Technology

[0002] The focus of transportation infrastructure construction is shifting to the western mountainous areas and water-rich regions with extremely complex terrain and geological conditions. As a result, a number of tunnels (roads) with significant characteristics such as "deep burial, high ground pressure, high water pressure, and large flow rate" will emerge, posing a huge challenge to the safe construction of karst tunnels. Once a disaster occurs, it will cause significant casualties and property losses, and may even lead to the suspension of the project.

[0003] Because current research on karst tunnels is relatively weak, existing technologies mostly rely on single-type sensors (such as piezometers or displacement sensors), making it difficult to fully capture the multi-factor coupling characteristics of water inrush disasters. Moreover, since water inrush in tunnels occurs extremely quickly, traditional early warning systems rely on manual monitoring, resulting in slow response times and failing to meet the sudden needs of water inrush disasters.

[0004] Therefore, a new solution is needed to address the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method for monitoring and early warning of water inrush in tunnels under water-rock coupling, and to propose a more accurate and rapid early warning method for the problem of water inrush in tunnels.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for monitoring and early warning of water inrush in tunnels under water-rock coupling, comprising at least the following steps:

[0007] S100: Location cross-sectional images of tunnel karst caves and water-bearing fissures obtained based on high-density electrical resistivity tomography;

[0008] S200: Real-time sensing of stress changes in fractured rock mass;

[0009] S300: Rock mass water inrush risk prediction based on machine learning. The system sets warning conditions based on whether the prediction results meet the trigger conditions. When the trigger conditions are met, an early warning is issued.

[0010] Furthermore, S1 includes at least the following steps:

[0011] S101: High-density electrical resistivity tomography (EDT) data acquisition of tunnel rock mass;

[0012] S102: Data removal and correction processing;

[0013] S103: Inversion calculation based on the least squares iterative method;

[0014] S104: Cross-sectional diagram of karst caves and fissures in the tunnel rock mass inferred from S103.

[0015] Furthermore, the method of collecting tunnel rock mass data based on high-density electrical resistivity tomography involves arranging electrodes at the tunnel arch in a dipole-dipole manner and then collecting data. This method is sensitive to water-bearing fissures, can clearly identify changes in resistivity gradients, and can enhance abnormal signals by combining polarizability ratio parameters, distinguishing between high resistivity as karst caves and low resistivity as water-bearing fissures.

[0016] The principle of the dipole-dipole method is as follows:

[0017] The lengths of the power supply dipole AB and the measuring dipole MN are both a, and the distance between them is n×a, where n is the isolation coefficient, which is a positive integer.

[0018] The power supply and measurement dipoles are laid out along the same line. The detection depth is positively correlated with n×a, and the empirical formula is as follows: k is the dielectric resistivity influence factor, which is 0.5-0.7 in the low-resistivity region;

[0019] Apparent resistivity is calculated by measuring the potential difference based on the distribution differences of the artificial current field in the underground medium. The formula is:

[0020]

[0021] Where K=6πa is the device coefficient; ΔV is the potential difference between M and N; and I is the supply current.

[0022] When arranging the electrodes, the electrode spacing n×a is usually 1-5 meters, and the isolation coefficient n ranges from 1 to 8.

[0023] Straight survey lines are laid out along the axis of the tunnel arch, and the electrodes are numbered at equal intervals (E1, E2, ..., EN) to ensure that the grounding resistance is <5kΩ;

[0024] Longitudinal rolling measurement is adopted, that is, the AB dipole is fixed and the data of the measuring dipole MN at different intervals are collected sequentially to form an inverted trapezoidal detection section.

[0025] Furthermore, the data removal and correction processing involves removing bad data, data splicing, and distortion correction preprocessing from the raw tunnel data collected by high-density electrical resistivity tomography. Step S102 includes at least the following steps:

[0026] The first step is outlier removal. This involves calculating the relative error of repeatedly collected data from the same measuring point using the single-point relative error method, and then filtering out outliers. The formula is:

[0027]

[0028] If δ i >δ阈值 The threshold, typically set at 5%-10%, is used to identify and remove pixels as bad; di is the i-th observation value. The average of multiple observations;

[0029] Secondly, data stitching and correction are performed, utilizing nonlinear deviation correction of the survey line. That is, when the survey line undergoes geometric distortion due to terrain undulations, a piecewise linear correction model is adopted:

[0030]

[0031]

[0032] By fitting the coordinate deviation of corresponding points in the overlapping area of ​​adjacent survey lines using polynomials, a correction mapping relationship is established to eliminate geometric position deviations caused by terrain undulations.

[0033] The observed potential difference is then converted into an equivalent value for the horizontal terrain:

[0034]

[0035] Map the sloping terrain data onto a horizontal reference surface;

[0036] Where x is the horizontal coordinate of the pixel; y is the vertical coordinate of the pixel; Δx(y) and Δy(x) are piecewise functions; ρ 真 ρ represents the theoretical resistivity under horizontal terrain; ρ is considered the apparent resistivity under actual terrain.

[0037] Finally, distortion correction preprocessing is performed, using a two-dimensional equalization filter designed with the least squares method to suppress single-point and linear distortions:

[0038]

[0039] To address the electric field distortion caused by poor electrode contact or dielectric inhomogeneity, a current density compensation model is established:

[0040]

[0041] By iteratively adjusting the value of k, the corrected electric field distribution satisfies the continuity condition of the Poisson equation, thereby suppressing local electric field distortion and linear interference.

[0042] In the formula: These are the corrected image values; The range for the weighted average is represented by a rectangular window, the size of which is controlled by n and m, W ij For the filter coefficients, minimize the residuals Sure; Indicates the position in the original image The data or pixel value; k is the distortion coefficient; This represents the theoretical potential gradient; This represents the measured potential gradient.

[0043] Furthermore, S103 involves inverting the optimized data to infer the underground resistivity distribution, i.e., using the current continuity equation and least squares optimization to complete the inversion calculation;

[0044] Using the current continuity equation for forward modeling, assuming the resistivity of the medium is ρ(x,z) and the current source is II, the potential is... Satisfies the Poisson equation:

[0045]

[0046] The Poisson equation, expressed as a system of linear equations using the finite element method (FEM), is as follows:

[0047]

[0048] In the formula: δ is the Dirac function; r s Indicates the location of the point current source; m is a model parameter, where the logarithmic transformation of resistivity is: K is the stiffness matrix, which depends on the model parameters; q is the source term vector.

[0049] The objective function for weighted least squares optimization inversion is:

[0050]

[0051] In the formula, d obs For observed data, i.e., the logarithm of the potential difference or apparent resistivity; F(m) is the forward response; W d Wm is the data weighting matrix, usually a diagonal matrix, with elements being the reciprocal of the data standard deviation; Wm is the model regularization matrix, which constrains the smoothness or structure of the model; λ is the regularization parameter, which balances data fit and model complexity.

[0052] The detailed steps of the inversion inference are as follows:

[0053] The first step is to apply the nonlinear problem to the initial model m. k Perform a Taylor expansion at this point, retaining the first-order terms:

[0054]

[0055] In the formula: It is the model in the current iteration; This is the Jacobian matrix, i.e., the sensitivity matrix; It represents the amount of model updates;

[0056] Substituting into the objective function, we obtain the linearized optimization problem:

[0057]

[0058] in, It is the impact of model update quantity on the prediction of observed data; It is the model update amount;

[0059] Solving this equation yields the update values ​​for the model:

[0060]

[0061] In the formula: These are the weighted matrix terms of the Jacobian matrix, describing the sensitivity of the data to the amount of model updates; It is a model regularization matrix term used to control model complexity and avoid overfitting; The product of the data residuals and the Jacobian matrix represents the contribution of data discrepancies to model updates; It is a model regularization term that limits the complexity of the model;

[0062] It is the residual, which is the difference between the current model mk and the observed data;

[0063] Next is the calculation of the Jacobian matrix, specifically the elements J of the Jacobian matrix J. ij This represents the sensitivity of the i-th data point to the j-th model parameter: By calculating the orthogonal field And is an accompanying field The spatial integrals of the two games are obtained:

[0064]

[0065] In the formula: Ω is the model domain; It is the derivative of resistivity with respect to the model parameters; It is the gradient between the orthogonal field and the adjoint field, representing the spatial variation of the field.

[0066] Next is the regularization design, the regularization matrix W. m Smoothing constraints are typically used.

[0067]

[0068] in and It is the second derivative of the model in the x and y directions (i.e., the Laplace operator), which represents the spatial smoothness of the model.

[0069] Finally, the iteration termination conditions are set, which are divided into residual convergence and model change convergence.

[0070] The residuals converge when the relative root mean square error (RMS) is less than a threshold.

[0071]

[0072] in: For the first One observation data; For the first The forward response of an observation data set, i.e., through the model The predicted observations; For the first The standard deviation of each observation data; The total number of observation data;

[0073] The model changes converge to .

[0074] Furthermore, the real-time sensing of stress changes in fractured rock mass is performed using a device for real-time sensing of stress changes in fractured rock mass, which includes a stress sensor, a stress transmission line, a host, rock mass, water-bearing fractures, and a tunnel.

[0075] The application of the real-time sensing device for stress changes in fractured rock mass includes at least the following steps:

[0076] First, based on the location of the water-bearing fracture cross-section obtained from S100, the stress sensor borehole is inserted into the rock mass;

[0077] The depth to which the material is incorporated into the rock mass should reach the water-bearing fissures to measure the stress in the rock mass. and water pressure P;

[0078] The measured stress data is transmitted to the host computer via a stress transmission line, and then the ultimate stress value of the fractured rock mass is set in the host computer. and ultimate shear strength ;

[0079] Finally, based on the real-time stress changes in the monitored rock mass, the effective stress in the water-bearing fractures was analyzed using the water-rock coupling method. and effective shear stress The degree of crack initiation is determined by combining fracture mechanics.

[0080] Ultimate stress value of fractured rock mass and ultimate shear strength The process involves taking samples from the tunnel and using a uniaxial testing instrument to test the uniaxial compressive strength of the rock mass, thereby obtaining the compressive strength of the fractured rock mass, which is then used as the ultimate stress value of the fractured rock mass. and ultimate shear strength The obtained value is then input into the host via the mechanical keypad.

[0081] Furthermore, the machine learning-based rock mass water inrush risk prediction includes at least the following steps:

[0082] First, geological hazard classification data for water inrush caused by tunnel were collected. Then, stress sensors were used to collect tunnel rock mass stress data for nearly half a month. osmotic pressure The changes were observed by collecting data on the arch settlement at the top of the tunnel over a period of nearly half a month. Based on information collected from the meteorological bureau, data was collected on the historical maximum rainfall over the past 10 years, and the collected data was further transmitted to the data preprocessing module.

[0083] Secondly, the data is organized, filtered, and optimized. In the data preprocessing module, the rock mass stress is... osmotic pressure vault settlement The expression for outlier detection using the Z-score method is as follows:

[0084]

[0085] In the formula: It is the Z-score of the i-th data point; It is the value of the i-th data point; is the mean of the dataset; e is the standard deviation of the dataset;

[0086] The data is normalized and output using the following expression:

[0087]

[0088] In the formula: x is the original data collected; min(x) is the minimum value in the dataset; max(x) is the maximum value in the dataset;

[0089] For rainfall feature extraction, the historical maximum rainfall R is extracted. max Based on the statistical characteristics, the ratio of the current rainfall R(t) to the historical maximum value is first calculated.

[0090]

[0091] Then, a decay factor λ is introduced to weight the impact of historical rainfall:

[0092]

[0093] Where R(tk) is the past rainfall at time step k;

[0094] Multi-factor weighting was used to construct a data matrix based on indicators such as rock mass stress, osmotic pressure, settlement, rainfall ratio, and cumulative rainfall.

[0095]

[0096] in, This represents the value of the j-th indicator at the i-th time point, where n is the number of time steps and m is the number of indicators, indicating how many different indicators there are.

[0097] Calculate the information entropy of the j-th indicator:

[0098]

[0099] in, It is a normalization constant that ensures the information entropy value is between [0,1]. It is the normalized probability, representing the relative weight of the j-th indicator at the i-th time point; It is the standardized value of the j-th indicator at the i-th time point; It is the sum of the j-th index over all time steps; It is information entropy, which measures the uncertainty of the j-th indicator;

[0100] Then calculate the index weight W. j This reflects their contribution:

[0101] ,

[0102] In the formula: m is the number of indicators;

[0103] Next, a risk probability model is constructed, defining the water inrush event label Y∈{0,1}, and the input feature vector is... , This represents the standardized value of rock mass stress. This is the standardized value of osmotic pressure; This is the standardized value for settlement. The ratio of rainfall; This refers to cumulative rainfall.

[0104] The logistic regression model is as follows:

[0105]

[0106] In the formula: The target variable for prediction is denoted as whether a sudden flood disaster will occur. Indicates that it has occurred. This indicates that it did not occur; β0 is the intercept term; βj is the characteristic coefficient; It is a linear combination in the logistic regression model, which is transformed into probability through an exponential function;

[0107] Finally, a risk probability model is constructed to predict the risk of water inrush in rock mass. The maximum likelihood method is used to construct the log-likelihood function and optimize the solution parameters.

[0108]

[0109] in, It is the first The predicted probability of a sample, i.e., the probability of water inrush calculated by the logistic regression model; The actual label of the i-th sample; the parameter estimate that maximizes lnL(β) is obtained by iteratively solving using the Newton-Raphson method. ;

[0110] Once the risk model is built, a dynamic warning threshold is set based on it. The threshold is first calibrated, and then the optimal threshold θ is determined using ROC curves based on historical water inrush event data. The threshold must satisfy the following conditions:

[0111]

[0112] Set dynamic thresholds for early warning and calculate risk probability in real time. Warning trigger conditions:

[0113] .

[0114] Compared with the prior art, the beneficial effects of the present invention are:

[0115] This invention, through the design of a method, can detect the location of karst caves and fissures in tunnels with complex geological conditions such as karst caves and fissures under water-rock coupling, detect the real-time changes in stress in water-bearing fissures, and obtain evaluation indicators by analyzing the influencing factors of geological hazards based on machine learning. Thus, a tunnel water inrush monitoring and early warning system is established, providing a more accurate and rapid early warning method for tunnel water inrush problems. Attached Figure Description

[0116] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0117] Figure 1 This is a flowchart of the present invention;

[0118] Figure 2 This is a flowchart illustrating the location cross-sectional views of tunnel karst caves and water-bearing fissures obtained using high-density electrical resistivity tomography (EDT).

[0119] Figure 3This is a schematic diagram of the device for real-time sensing of stress changes in fractured rock mass according to the present invention.

[0120] Figure 4 This is a stress analysis diagram of the water-bearing fissures in this invention.

[0121] In the diagram: 1. Stress sensor; 2. Stress transmission line; 3. Main unit; 4. Rock mass; 5. Water-bearing fissure; 6. Tunnel. Detailed Implementation

[0122] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0123] Please see Figure 1 A method for monitoring and early warning of water inrush and water surge in tunnels under water-rock coupling effects includes at least the following steps:

[0124] S100: Location cross-sectional images of tunnel karst caves and water-bearing fissures obtained based on high-density electrical resistivity tomography;

[0125] S200: Real-time sensing of stress changes in fractured rock mass;

[0126] S300: Rock mass water inrush risk prediction based on machine learning. The system sets warning conditions based on whether the prediction results meet the trigger conditions. When the trigger conditions are met, an early warning is issued.

[0127] In this embodiment, an additional step can be added to obtain a tunnel water inrush monitoring and early warning system;

[0128] S400: Tunnel Water Inrush Monitoring and Early Warning System Based on Multi-Factor Coupling Effects.

[0129] S1 includes at least the following steps:

[0130] S101: High-density electrical resistivity tomography (EDT) data acquisition of tunnel rock mass;

[0131] S102: Data removal and correction processing;

[0132] S103: Inversion calculation based on the least squares iterative method;

[0133] S104: Cross-sectional diagram of karst caves and fissures in the tunnel rock mass inferred from S103.

[0134] High-density electrical resistivity tomography (EDT) is used to collect tunnel rock mass data. This method involves placing electrodes at the tunnel arch in a dipole-dipole manner and then collecting data. This method is sensitive to water-bearing fissures, can clearly identify changes in resistivity gradients, and can enhance abnormal signals by combining polarizability ratio parameters. It can also distinguish between high resistivity (cave) and low resistivity (water-bearing fissures).

[0135] The principle of the dipole-dipole method is as follows:

[0136] The lengths of the power supply dipole AB and the measuring dipole MN are both a, and the distance between them is n×a, where n is the isolation coefficient, which is a positive integer.

[0137] The power supply and measurement dipoles are laid out along the same line. The detection depth is positively correlated with n×a, and the empirical formula is as follows: k is the dielectric resistivity influence factor, which is 0.5-0.7 in the low-resistivity region;

[0138] Apparent resistivity is calculated by measuring the potential difference based on the distribution differences of the artificial current field in the underground medium. The formula is:

[0139]

[0140] Where K=6πa is the device coefficient; ΔV is the potential difference between M and N; and I is the supply current.

[0141] When arranging the electrodes, the electrode spacing n×a is usually 1-5 meters, and the isolation coefficient n ranges from 1 to 8.

[0142] Straight survey lines are laid out along the axis of the tunnel arch, and the electrodes are numbered at equal intervals (E1, E2, ..., EN) to ensure that the grounding resistance is <5kΩ;

[0143] Longitudinal rolling measurement is adopted, that is, the AB dipole is fixed and the data of the measuring dipole MN at different intervals are collected sequentially to form an inverted trapezoidal detection section.

[0144] Data removal and correction processing involves removing bad data, data splicing, and distortion correction preprocessing of the raw tunnel data collected by high-density electrical resistivity tomography. S102 includes at least the following steps:

[0145] The first step is outlier removal. This involves calculating the relative error of repeatedly collected data from the same measuring point using the single-point relative error method, and then filtering out outliers. The formula is:

[0146]

[0147] If δ i >δ 阈值 The threshold, typically set at 5%-10%, is used to identify and remove pixels as bad; di is the i-th observation value. The average of multiple observations;

[0148] Secondly, data stitching and correction are performed, utilizing nonlinear deviation correction of the survey line. That is, when the survey line undergoes geometric distortion due to terrain undulations, a piecewise linear correction model is adopted:

[0149]

[0150]

[0151] By fitting the coordinate deviation of corresponding points in the overlapping area of ​​adjacent survey lines using polynomials, a correction mapping relationship is established to eliminate geometric position deviations caused by terrain undulations.

[0152] The observed potential difference is then converted into an equivalent value for the horizontal terrain:

[0153]

[0154] Map the sloping terrain data onto a horizontal reference surface;

[0155] Where x is the horizontal coordinate of the pixel; y is the vertical coordinate of the pixel; Δx(y) and Δy(x) are piecewise functions; ρ 真 ρ represents the theoretical resistivity under horizontal terrain; ρ is considered the apparent resistivity under actual terrain.

[0156] Finally, distortion correction preprocessing is performed, using a two-dimensional equalization filter designed with the least squares method to suppress single-point and linear distortions:

[0157]

[0158] To address the electric field distortion caused by poor electrode contact or dielectric inhomogeneity, a current density compensation model is established:

[0159]

[0160] By iteratively adjusting the value of k, the corrected electric field distribution satisfies the continuity condition of the Poisson equation, thereby suppressing local electric field distortion and linear interference.

[0161] In the formula: These are the corrected image values; The range for the weighted average is represented by a rectangular window, the size of which is controlled by n and m, W ij For the filter coefficients, minimize the residuals Sure; Indicates the position in the original image The data or pixel value; k is the distortion coefficient; This represents the theoretical potential gradient; This represents the measured potential gradient.

[0162] S103 involves inverting the optimized data to infer the underground resistivity distribution, which is achieved by using the current continuity equation and least squares optimization to complete the inversion calculation.

[0163] Using the current continuity equation for forward modeling, assuming the resistivity of the medium is ρ(x,z) and the current source is II, the potential is... Satisfies the Poisson equation:

[0164]

[0165] The Poisson equation, expressed as a system of linear equations using the finite element method (FEM), is as follows:

[0166]

[0167] In the formula: δ is the Dirac function; r s Indicates the location of the point current source; m is a model parameter, where the logarithmic transformation of resistivity is: K is the stiffness matrix, which depends on the model parameters; q is the source term vector.

[0168] The objective function for weighted least squares optimization inversion is:

[0169]

[0170] In the formula, d obs For observed data, i.e., the logarithm of the potential difference or apparent resistivity; F(m) is the forward response; W d Wm is the data weighting matrix, usually a diagonal matrix, with elements being the reciprocal of the data standard deviation; Wm is the model regularization matrix, which constrains the smoothness or structure of the model; λ is the regularization parameter, which balances data fit and model complexity.

[0171] The detailed steps of the inversion inference are as follows:

[0172] The first step is to apply the nonlinear problem to the initial model m. k Perform a Taylor expansion at this point, retaining the first-order terms:

[0173]

[0174] In the formula: It is the model in the current iteration; This is the Jacobian matrix, i.e., the sensitivity matrix; It represents the amount of model updates;

[0175] Substituting into the objective function, we obtain the linearized optimization problem:

[0176]

[0177] in, It is the impact of model update quantity on the prediction of observed data; It is the model update amount;

[0178] Solving this equation yields the update values ​​for the model:

[0179]

[0180] In the formula: These are the weighted matrix terms of the Jacobian matrix, describing the sensitivity of the data to the amount of model updates; It is a model regularization matrix term used to control model complexity and avoid overfitting; The product of the data residuals and the Jacobian matrix represents the contribution of data discrepancies to model updates; It is a model regularization term that limits the complexity of the model;

[0181] It is the residual, which is the difference between the current model mk and the observed data;

[0182] Next is the calculation of the Jacobian matrix, specifically the elements J of the Jacobian matrix J. ij This represents the sensitivity of the i-th data point to the j-th model parameter: By calculating the orthogonal field And is an accompanying field The spatial integrals of the two games are obtained:

[0183]

[0184] In the formula: Ω is the model domain; It is the derivative of resistivity with respect to the model parameters; It is the gradient between the orthogonal field and the adjoint field, representing the spatial variation of the field.

[0185] Next is the regularization design, the regularization matrix W. m Smoothing constraints are typically used.

[0186]

[0187] in and It is the second derivative of the model in the x and y directions (i.e., the Laplace operator), which represents the spatial smoothness of the model.

[0188] Finally, the iteration termination conditions are set, which are divided into residual convergence and model change convergence.

[0189] The residuals converge when the relative root mean square error (RMS) is less than a threshold.

[0190]

[0191] in: For the first One observation data; For the first The forward response of an observation data set, i.e., through the model The predicted observations; For the first The standard deviation of each observation data; The total number of observation data;

[0192] The model changes converge to .

[0193] Real-time sensing of stress changes in fractured rock mass is achieved through a device for real-time sensing of stress changes in fractured rock mass, which includes a stress sensor 1, a stress transmission line 2, a main unit 3, a rock mass 4, a water-bearing fracture 5, and a tunnel 6.

[0194] The application of a device for real-time sensing of stress changes in fractured rock masses includes at least the following steps:

[0195] First, based on the location of the water-bearing fracture cross-section obtained from S100, the stress sensor 1 is drilled and inserted into the rock mass (4);

[0196] The depth of the incorporation into rock mass 4 should reach the water-bearing fracture 5 to measure the stress in the rock mass. and water pressure P;

[0197] The measured stress data is transmitted to the host computer 3 via stress transmission line 2. Then, the ultimate stress value of the fractured rock mass is set in the host computer (3). and ultimate shear strength ;

[0198] Finally, based on the real-time stress changes in the monitored rock mass, the effective stress in the water-bearing fractures was analyzed using the water-rock coupling method. and effective shear stress The degree of crack initiation is determined by combining fracture mechanics.

[0199] Ultimate stress value of fractured rock mass and ultimate shear strength On-site sampling was conducted at Tunnel 6, and the uniaxial compressive strength of the rock mass was tested using a uniaxial testing instrument to obtain the compressive strength of the fractured rock mass, which was then used as the ultimate stress value of the fractured rock mass. and ultimate shear strength Then, the obtained value is input into host 3 via the mechanical keypad panel;

[0200] Analysis of effective stress in water-bearing fractures using water-rock coupling method and effective shear stress ,like Figure 4 As shown, according to the effective stress law of saturated porous elastic media, the effective stress of rock mass fractures is: ;

[0201] In the formula: is the total stress of the rock mass, measured by a stress sensor; P is the water pressure in the water-bearing fractured rock mass, also measured by a stress sensor, and its calculation formula is: , Let H be the specific weight of water, H be the head of the water, and z be its elevation. This represents the effective stress coefficient of the crack.

[0202] For fully saturated rough rock fissures, when hour, ,when hour, ;

[0203] In the formula: These are the initial stiffness of the crack and the maximum normal closure, respectively.

[0204] The influence of water-rock coupling on the stability of water-bearing fractures in rock masses is investigated. Based on the aforementioned expression for effective stress, the classical Coulomb criterion is modified as follows: ;

[0205] In the formula: These are the friction factor and cohesion of a single fracture, respectively, taking into account the time effect of water-rock interaction.

[0206] The specific steps for determining whether a crack has initiated based on fracture mechanics are as follows:

[0207] The crack propagation is a type II compression crack, which is a pure type II crack problem. Therefore, considering the water-rock coupling effect, the formula for calculating the stress intensity factor at the tip of the water-bearing fracture in the tunnel surrounding rock is as follows: In the formula, a is the crack length;

[0208] When the rock mass stress intensity factor reaches the point of crack initiation, then This is used to determine whether a crack has opened.

[0209] Machine learning-based rock mass water inrush risk prediction includes at least the following steps:

[0210] First, geological hazard classification data for water inrush caused by tunnel were collected. Tunnel rock mass stress was collected over nearly half a month using stress sensor 1. osmotic pressure The changes were observed by collecting data on the arch settlement at the top of the tunnel over a period of nearly half a month. Based on information collected from the meteorological bureau, data was collected on the historical maximum rainfall over the past 10 years, and the collected data was further transmitted to the data preprocessing module.

[0211] Secondly, the data is organized, filtered, and optimized. In the data preprocessing module, the rock mass stress is... osmotic pressure vault settlement The expression for outlier detection using the Z-score method is as follows:

[0212]

[0213] In the formula: It is the Z-score of the i-th data point; It is the value of the i-th data point; is the mean of the dataset; e is the standard deviation of the dataset;

[0214] The data is normalized and output using the following expression:

[0215]

[0216] In the formula: x is the original data collected; min(x) is the minimum value in the dataset; max(x) is the maximum value in the dataset;

[0217] For rainfall feature extraction, the historical maximum rainfall R is extracted. max Based on the statistical characteristics, the ratio of the current rainfall R(t) to the historical maximum value is first calculated.

[0218]

[0219] Then, a decay factor λ is introduced to weight the impact of historical rainfall:

[0220]

[0221] Where R(tk) is the past rainfall at time step k;

[0222] Multi-factor weighting was used to construct a data matrix based on indicators such as rock mass stress, osmotic pressure, settlement, rainfall ratio, and cumulative rainfall.

[0223]

[0224] in, This represents the value of the j-th indicator at the i-th time point, where n is the number of time steps and m is the number of indicators, indicating how many different indicators there are.

[0225] Calculate the information entropy of the j-th indicator:

[0226]

[0227] in, It is a normalization constant that ensures the information entropy value is between [0,1]. It is the normalized probability, representing the relative weight of the j-th indicator at the i-th time point; It is the standardized value of the j-th indicator at the i-th time point; It is the sum of the j-th index over all time steps; It is information entropy, which measures the uncertainty of the j-th indicator;

[0228] Then calculate the index weight W. j This reflects their contribution:

[0229] ,

[0230] In the formula: m is the number of indicators;

[0231] Next, a risk probability model is constructed, defining the water inrush event label Y∈{0,1}, and the input feature vector is... , This represents the standardized value of rock mass stress. This is the standardized value of osmotic pressure; This is the standardized value for settlement. The ratio of rainfall; This refers to cumulative rainfall.

[0232] The logistic regression model is as follows:

[0233]

[0234] In the formula: The target variable for prediction is denoted as whether a sudden flood disaster will occur. Indicates that it has occurred. This indicates that it did not occur; β0 is the intercept term; βj is the characteristic coefficient; It is a linear combination in the logistic regression model, which is transformed into probability through an exponential function;

[0235] Finally, a risk probability model is constructed to predict the risk of water inrush in rock mass. The maximum likelihood method is used to construct the log-likelihood function and optimize the solution parameters.

[0236]

[0237] in, It is the first The predicted probability of a sample, i.e., the probability of water inrush calculated by the logistic regression model; The actual label of the i-th sample; the parameter estimate that maximizes lnL(β) is obtained by iteratively solving using the Newton-Raphson method. ;

[0238] Once the risk model is built, a dynamic warning threshold is set based on it. The threshold is first calibrated, and then the optimal threshold θ is determined using ROC curves based on historical water inrush event data. The threshold must satisfy the following conditions:

[0239]

[0240] Set dynamic thresholds for early warning and calculate risk probability in real time. Warning trigger conditions:

[0241] .

[0242] Multi-factor coupling effect tunnel water inrush monitoring and early warning system:

[0243] A tunnel water inrush early warning system will be formed by combining early warning of water-bearing fissures under water-rock interaction with early warning of rock mass water inrush risk prediction model based on machine learning.

[0244] Establishment of an early warning system for water inrush caused by the expansion of water-bearing fractures under water-rock interaction;

[0245] First, based on the rock mass stress value transmitted from stress sensor 1 to host 3, then the effective normal stress of water-bearing fractures under water-rock interaction is analyzed. and shear stress The system determines whether a crack is expanding based on fracture mechanics, and then forms a water inrush warning based on the degree of cracking. Secondly, it determines whether a single crack is expanding based on fracture mechanics, and finally forms a water inrush warning based on the degree of cracking.

[0246] Flood Warning Threshold Setting: Level 1 Warning (Yellow): K Ⅱ ≥0.8K ⅡC This indicates that the crack has entered the subcritical propagation stage; Level II warning (orange): K Ⅱ ≥0.9K ⅡC Emergency reinforcement measures need to be initiated; Level III warning (red): K Ⅱ ≥0.9K ⅡC Evacuate personnel immediately and seal off the work area;

[0247] This system closely integrates early warning of water inrush from water-bearing fractures under water-rock interaction with early warning from a machine learning-based rock mass water inrush risk prediction model to form a tunnel water inrush monitoring and early warning system under multiple coupled interactions.

[0248] The stress changes in the rock mass detected by stress sensor 1 are transmitted to host computer 3. The host computer analyzes the effective stress changes in the fractured surrounding rock under water-rock coupling, and obtains the ultimate shear strength of the rock mass through uniaxial tests. and allowable stress value The information is input into the host 3 via a mechanical keyboard. The degree of crack opening is determined by fracture mechanics. The set warning condition threshold is also input into the host 3 via a mechanical keyboard. At the same time, the warning threshold for tunnel water inrush monitoring under the multi-factor coupling effect is input into the host 3 via a mechanical keyboard. Then, alarm sound and light lights are evenly arranged on both sides of the tunnel arch and connected to the host 3.

[0249] When a Level 1 warning, i.e. a potential risk, is issued, it is pushed to the monitoring terminal via API, generating a yellow alert log. When a Level 2 warning, i.e. a high-probability fault, is issued, it triggers an audible and visual alarm device and simultaneously sends an SMS to the responsible person. When a Level 3 warning, i.e. an emergency fault, is issued, it automatically initiates an emergency shutdown protocol, activates the alarm audible and visual response, and uploads fault snapshot data to the cloud.

[0250] It should be noted that the system for detecting the location of karst water and water-bearing fissures in rock masses using high-density detection, real-time sensing of stress changes in the rock mass within fissures, rock mass water inrush risk prediction based on machine learning, and tunnel water inrush monitoring and early warning systems based on the coupling effects of various factors are all connected to the same computer terminal. First, the data obtained from the high-density electrical resistivity tomography (EDT) needs to be transmitted to the computer terminal for processing, resulting in a cross-sectional image of the tunnel fissure location. Second, real-time sensing of stress changes in the rock mass within fissures is performed. Based on the location of water-bearing fissures obtained from the cross-sectional image, stress sensors are inserted at the fissure locations, and the obtained rock mass stress data is input into host computer 3. Host computer 3 is then transmitted to the computer terminal via data transmission. Finally, the system for predicting rock mass water inrush risk and monitoring and early warning systems based on the coupling effects of various factors, after constructing various risk models on the computer terminal, sets dynamic warning thresholds, including those for stress changes in rock mass fissures. The thresholds are first calibrated, and then corresponding warnings are set based on various factors. The computer terminal then controls the alarm lights and sound system to issue warnings according to the tunnel's hazard conditions.

[0251] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for monitoring and early warning of water inrush and seepage in tunnels under water-rock coupling, characterized in that: At least the following steps are included: S100: Location cross-sectional images of tunnel karst caves and water-bearing fissures obtained based on high-density electrical resistivity tomography; S200: Real-time sensing of stress changes in fractured rock mass; Establishment of an early warning system for water inrush caused by the expansion of water-bearing fractures under water-rock interaction; First, based on the location of the water-bearing fracture cross-section obtained from S100, the stress sensor 1 is drilled and inserted into the rock mass (4); the insertion depth into the rock mass (4) should reach the water-bearing fracture (5) to measure the stress of the rock mass. and water pressure P; pass Obtain shear stress ; In the formula: P represents the total stress of the rock mass, measured using a stress sensor; P represents the water pressure in the water-bearing fractured rock mass, also measured using a stress sensor. The effective stress coefficient of the crack, These are the friction coefficient and cohesion of a single fracture considering the time effect of water-rock interaction; The formula for calculating the stress intensity factor is: In the formula, a is the crack length; When the rock mass stress intensity factor reaches the point of crack initiation, then This is used to determine whether a crack has opened, and finally, a water inrush warning is generated based on the degree of cracking. S300: Machine learning-based prediction of rock mass water inrush risk. The machine learning-based rock mass water inrush risk prediction includes at least the following steps: First, geological disaster classification data of water inrush caused by tunnel were collected. Tunnel rock mass stress was collected over the past half month using stress sensors (1). osmotic pressure The changes were observed by collecting data on the arch settlement at the top of the tunnel over a period of nearly half a month. Based on information collected from the meteorological bureau, data was collected on the historical maximum rainfall over the past 10 years, and the collected data was further transmitted to the data preprocessing module. Secondly, the data is organized, filtered, and optimized. Construct a risk probability model, defining the water inrush event label Y∈{0,1}, and the input feature vector is... , This represents the standardized value of rock mass stress. This is the standardized value of osmotic pressure; This is the standardized value for settlement. The ratio of rainfall; This refers to cumulative rainfall. The logistic regression model is as follows: In the formula: The target variable for prediction is denoted as whether a sudden flood disaster will occur. Indicates that it has occurred. This indicates that it did not occur; β0 is the intercept term; βj is the characteristic coefficient; It is a linear combination in the logistic regression model, which is transformed into probability through an exponential function; Finally, a risk probability model is constructed to predict the risk of water inrush in rock mass. The maximum likelihood method is used to construct the log-likelihood function and optimize the solution parameters. in, It is the first The predicted probability of a sample, i.e., the probability of water inrush calculated by the logistic regression model; The actual label of the i-th sample; the parameter estimate that maximizes lnL(β) is obtained by iteratively solving using the Newton-Raphson method. ; Once the risk model is built, a dynamic warning threshold is set based on it. The threshold is first calibrated, and then the optimal threshold θ is determined using ROC curves based on historical water inrush event data. The threshold must satisfy the following conditions: Set dynamic thresholds for early warning and calculate risk probability in real time. Warning trigger conditions: The settings are configured based on whether the prediction results meet the triggering conditions for an early warning; an early warning is issued when the triggering conditions are met. S400: The water inrush warning of water-bearing fracture expansion under water-rock interaction in step S200 is closely combined with the early warning of rock mass water inrush risk prediction model based on machine learning in step S300 to form a tunnel water inrush monitoring and early warning system under multiple coupling effects.

2. The method for monitoring and early warning of tunnel water inrush under water-rock coupling as described in claim 1, characterized in that: S1 includes at least the following steps: S101: High-density electrical resistivity tomography (EDT) data acquisition of tunnel rock mass; S102: Data removal and correction processing; S103: Inversion calculation based on the least squares iterative method; S104: Cross-sectional diagram of karst caves and fissures in the tunnel rock mass inferred from S103.

3. The method for monitoring and early warning of tunnel water inrush under water-rock coupling as described in claim 2, characterized in that: The method of acquiring tunnel rock mass data based on high-density electrical resistivity tomography involves placing electrodes at the tunnel arch in a dipole-dipole manner and then collecting data. This method is sensitive to water-bearing fissures, can clearly identify changes in resistivity gradient, and enhances abnormal signals by combining the polarizability ratio parameter, distinguishing between high resistivity as karst caves and low resistivity as water-bearing fissures. The principle of the dipole-dipole method is as follows: The lengths of the power supply dipole AB and the measuring dipole MN are both a, and the distance between them is n×a, where n is the isolation coefficient, which is a positive integer. The power supply and measurement dipoles are laid out along the same line. The detection depth is positively correlated with n×a, and the empirical formula is as follows: k is the dielectric resistivity influence factor, which is 0.5-0.7 in the low-resistivity region; Apparent resistivity is calculated by measuring the potential difference based on the distribution differences of the artificial current field in the underground medium. The formula is: Where K=6πa is the device coefficient; ΔV is the potential difference between M and N; and I is the supply current. Straight survey lines are laid out along the axis of the tunnel arch, and electrodes are numbered at equal intervals to ensure that the grounding resistance is <5kΩ; Longitudinal rolling measurement is adopted, that is, the AB dipole is fixed and the data of the measuring dipole MN at different intervals are collected sequentially to form an inverted trapezoidal detection section.

4. The method for monitoring and early warning of tunnel water inrush under water-rock coupling as described in claim 3, characterized in that: The data removal and correction process involves removing bad data, data splicing, and distortion correction preprocessing of the raw tunnel data collected by high-density electrical resistivity tomography. Step S102 includes at least the following steps: The first step is bad data removal. The relative error is calculated using the single-point relative error method on repeatedly collected data from the same measuring point to filter out outliers. The formula is: If δ i >δ 阈值 The threshold, set at 5%-10%, is used to identify and remove pixels as bad; di is the i-th observation value. The average of multiple observations; Secondly, data stitching and correction are performed, utilizing nonlinear deviation correction of the survey line. That is, when the survey line undergoes geometric distortion due to terrain undulations, a piecewise linear correction model is adopted: By fitting the coordinate deviation of corresponding points in the overlapping area of ​​adjacent survey lines using polynomials, a correction mapping relationship is established to eliminate geometric position deviations caused by terrain undulations. The observed potential difference is then converted into an equivalent value for the horizontal terrain: Map the sloping terrain data to a horizontal reference surface; Where x is the horizontal coordinate of the pixel; y is the vertical coordinate of the pixel; Δx(y) and Δy(x) are piecewise functions; ρ 真 ρ represents the theoretical resistivity under horizontal terrain; ρ is considered the apparent resistivity under actual terrain. Finally, distortion correction preprocessing is performed, using a two-dimensional equalization filter designed with the least squares method to suppress single-point and linear distortions: To address the electric field distortion caused by poor electrode contact or dielectric inhomogeneity, a current density compensation model is established: By iteratively adjusting the value of k, the corrected electric field distribution satisfies the continuity condition of the Poisson equation, thereby suppressing local electric field distortion and linear interference. In the formula: These are the corrected image values; The range for the weighted average is represented by a rectangular window, the size of which is controlled by n and m, W ij For the filter coefficients, minimize the residuals Sure; Indicates the position in the original image The data or pixel value; k is the distortion coefficient; This represents the theoretical potential gradient; This represents the measured potential gradient.

5. The method for monitoring and early warning of tunnel water inrush under water-rock coupling as described in claim 4, characterized in that: S103 involves inverting the optimized data to infer the underground resistivity distribution, i.e., using the current continuity equation and least squares optimization to complete the inversion calculation; Using the current continuity equation for forward modeling, assuming the resistivity of the medium is ρ(x,z) and the current source is II, the potential is... Satisfies the Poisson equation: The Poisson equation, expressed as a system of linear equations using the finite element method, is as follows: In the formula: δ is the Dirac function; r s Indicates the location of the point current source; m is a model parameter, where the logarithmic transformation of resistivity is: K is the stiffness matrix, which depends on the model parameters; q is the source term vector. The objective function for weighted least squares optimization inversion is: In the formula, d obs For observed data, i.e., the logarithm of the potential difference or apparent resistivity; F(m) is the forward response; W d is the data weighting matrix, a diagonal matrix, with elements being the reciprocal of the data standard deviation; Wm is the model regularization matrix, constraining the smoothness or structure of the model; λ is the regularization parameter, balancing data fitting and model complexity. The detailed steps of the inversion inference are as follows: The first step is to apply the nonlinear problem to the initial model m. k Perform a Taylor expansion at this point, retaining the first-order terms: In the formula: It is the model in the current iteration; This is the Jacobian matrix, i.e., the sensitivity matrix; It represents the amount of model updates; Substituting into the objective function, we obtain the linearized optimization problem: in, It is the impact of model update quantity on the prediction of observed data; It is the model update amount; Solving this equation yields the update values ​​for the model: In the formula: These are the weighted matrix terms of the Jacobian matrix, describing the sensitivity of the data to the amount of model updates; It is a model regularization matrix term used to control model complexity and avoid overfitting; The product of the data residuals and the Jacobian matrix represents the contribution of data discrepancies to model updates; It is a model regularization term that limits the complexity of the model; It is the residual, which is the difference between the current model mk and the observed data; Next is the calculation of the Jacobian matrix, specifically the elements J of the Jacobian matrix J. ij This represents the sensitivity of the i-th data point to the j-th model parameter: By calculating the orthogonal field And is an accompanying field The spatial integrals of the two games are obtained: In the formula: Ω is the model domain; It is the derivative of resistivity with respect to the model parameters; It is the gradient between the orthogonal field and the adjoint field, representing the spatial variation of the field. Next is the regularization design, the regularization matrix W. m Apply smoothing constraints: in and It is the second derivative of the model in the x and y directions (i.e., the Laplace operator), which represents the spatial smoothness of the model. Finally, the iteration termination conditions are set, which are divided into residual convergence and model change convergence. The residuals converge when the relative root mean square error is less than a threshold. in: For the first One observation data; For the first The forward response of an observation data set, i.e., through the model The predicted observations; For the first The standard deviation of each observation data; The total number of observation data; The model changes converge to .

6. The method for monitoring and early warning of tunnel water inrush under water-rock coupling as described in claim 1, characterized in that: The real-time sensing of stress changes in fractured rock mass is achieved through a device for real-time sensing of stress changes in fractured rock mass, which includes a stress sensor (1), a stress transmission line (2), a host (3), a rock mass (4), a water-bearing fracture (5), and a tunnel (6). The application of the device for real-time sensing of stress changes in fractured rock mass includes at least the following steps: First, based on the location of the water-bearing fracture cross-section obtained from S100, the stress sensor (1) is drilled and inserted into the rock mass (4); The depth of the incorporation into the rock mass (4) should reach the water-bearing fissures (5) to measure the stress in the rock mass. and water pressure P; The measured stress data is transmitted to the host (3) via the stress transmission line (2), and then the ultimate stress value of the fractured rock mass is set in the host (3). and ultimate shear strength ; Finally, based on the real-time stress changes in the monitored rock mass, the effective stress in the water-bearing fractures was analyzed using the water-rock coupling method. and effective shear stress The degree of crack initiation is determined by combining fracture mechanics. Ultimate stress value of fractured rock mass and ultimate shear strength The process involves taking samples from the tunnel (6) on-site and using a uniaxial testing instrument to test the uniaxial compressive strength of the rock mass, thereby obtaining the compressive strength of the fractured rock mass, which is then used as the ultimate stress value of the fractured rock mass. and ultimate shear strength Then the obtained value is input into the host (3) through the mechanical keypad.

Images