Tunnel water inrush monitoring and early warning method under water-rock coupling effect

Abstract

The invention belongs to the technical field of tunnel water inrush monitoring, and particularly relates to a tunnel water inrush monitoring and early warning method under a water-rock coupling effect, which comprises the following steps: S1, obtaining tunnel data and rainfall data of a historical time sequence, carrying out data processing, and training a preset logistic regression model; s2, acquiring a position cross-section diagram of the water-containing fracture of the tunnel by using a high-density electrical method; s3, selecting the maximum K water-containing fractures based on the position cross-section diagram in the S2, respectively calculating effective method corresponding force sigma'n and shear stress tau'n, and then calculating a stress intensity factor K II; corresponding cracking early warning judgment is carried out; s4, current tunnel data and rainfall data are obtained and processed, and the trained logistic regression model is used for predicting the current water inrush risk probability; and corresponding water inrush early warning judgment is carried out. According to the method, multi-source data can be fused, a water-rock coupling action mechanism is considered, and tunnel water inrush monitoring and early warning are efficiently carried out in real time.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of tunnel water inrush monitoring, and particularly relates to a tunnel water inrush monitoring and early warning method under water-rock coupling. BACKGROUND

[0002] In the western complex mountainous area and water-rich karst area, the rapid development of transportation infrastructure construction has given birth to a large number of deep and long tunnels. Such projects often have the characteristics of "large buried depth, high ground pressure, high water pressure and large flow", and the risk of water inrush and gushing is dramatically increased, which brings unprecedented severe challenges to tunnel safety construction. Tunnel water inrush and gushing refers to the disaster phenomenon that when a tunnel passes through aquifer, karst pipeline, fault fracture zone and other unfavorable geological structures, groundwater will suddenly or short-time gush into the tunnel under high pressure. This phenomenon not only seriously impacts the construction progress and forces the project to be interrupted, but also directly threatens the safety of on-site personnel, often accompanied by significant equipment damage and economic losses, and even leads to project suspension.

[0003] At present, the monitoring and early warning technology research for tunnel water inrush disaster in water-rich strata such as karst is still relatively weak. The main problems and their causes are analyzed as follows:

[0004] Single monitoring means and lack of multi-factor coupling information: existing technologies generally rely on single type sensors (such as only setting up osmometer to monitor water pressure changes or only using displacement sensor to monitor surrounding rock deformation). However, tunnel water inrush is essentially a disastrous result of water-rock strong coupling, and is the product of complex interaction between hydrogeological conditions (groundwater pressure, seepage field changes) and rock mass mechanical state (geostress field, structure surface strength, crack propagation). Single parameter sensor is difficult to comprehensively, real-time and cooperatively capture the dynamic, nonlinear and multi-physical field coupling disaster evolution characteristics, leading to one-sided monitoring information and difficulty in accurately depicting the real process of disaster incubation, which is one of the core reasons for inaccurate early warning.

[0005] Early warning relies on manual operation and response is slow: most existing early warning systems still rely on manual reading of data and manual experience identification. Since tunnel water inrush disaster often occurs, develops and causes damage in a very short time (a few minutes or even shorter), this manual reading and early warning information transmission process has obvious lag, and its response time is much longer than the key time window of disaster development. The delay of early warning and the suddenness and instantaneity of disaster constitute a sharp contradiction, making it difficult for existing methods to meet the timeliness requirements of disaster prevention and reduction.

[0006] The root of these problems lies in the complexity of the water-rock dynamic coupling process and the insufficient comprehensiveness of the monitoring and early warning system. Overcoming these difficulties requires solving key technical bottlenecks such as real-time fusion processing of multi-source heterogeneous information, efficient identification and quantification of key state variables of water-rock interaction, and rapid intelligent early warning algorithms adapted to the characteristics of sudden disasters. Especially under the harsh environment of tunnel construction and high confining pressure and high water pressure conditions, achieving real-time early warning with high reliability and low false alarm rate faces higher technical and engineering implementation difficulties.

[0007] The root of these problems lies in the complexity of the water-rock dynamic coupling process and the insufficient comprehensiveness of the monitoring and early warning system. Overcoming these difficulties requires solving key technical bottlenecks such as real-time fusion processing of multi-source heterogeneous information, efficient identification and quantification of key state variables of water-rock interaction, and rapid intelligent early warning algorithms adapted to the characteristics of sudden disasters. Especially under the harsh environment of tunnel construction and high confining pressure and high water pressure conditions, achieving real-time early warning with high reliability and low false alarm rate faces higher technical and engineering implementation difficulties.

[0008] Therefore, how to fuse multi-source data, consider the water-rock coupling mechanism, and realize real-time and efficient tunnel water inrush monitoring and early warning has become a problem to be solved. SUMMARY

[0009] In view of the above problems of the prior art, the present application provides a tunnel water inrush monitoring and early warning method under water-rock coupling effect, which can fuse multi-source data, consider the water-rock coupling mechanism, and realize real-time and efficient tunnel water inrush monitoring and early warning.

[0010] In order to solve the above technical problems, the present application adopts the following technical scheme:

[0011] A tunnel water inrush monitoring and early warning method under water-rock coupling effect, comprising the following steps:

[0012] S1, obtaining historical time series tunnel data and rainfall data, and performing data processing, grouping each data at the same time after processing into a risk feature vector, taking whether a water inrush disaster occurs at the corresponding time as a label, obtaining training data; training a preset logistic regression model using the training data; the logistic regression model is used to predict a water inrush risk probability according to the risk feature vector;

[0013] S2, using high-density electrical method to obtain a position cross section diagram of water-bearing fissures of the tunnel;

[0014] S3, based on the position cross section diagram of S2, selecting the largest K water-bearing fissures, respectively calculating effective stress σ' n and shear stress τ' n , and then calculating stress intensity factor K Ⅱ; and based on stress intensity factor K Ⅱ a corresponding cracking early warning judgment is made;

[0015] S4, current tunnel data and rainfall data are acquired and processed to obtain a real-time risk feature vector, a trained logistic regression model is used to predict a current water inrush risk probability, and a corresponding water inrush early warning judgment is made based on the current water inrush risk probability;

[0016] The order of S1 and S2 can be mutually changed, and the order of S3 and S4 can be mutually changed.

[0017] Compared with the prior art, the present application has the following beneficial effects:

[0018] 1. Multi-source data fusion and collaborative early warning are realized. The method innovatively fuses historical time series monitoring data (tunnel sensor data, rainfall) and instant geophysical detection data (water-bearing fracture position section map obtained by high-density electrical method). Compared with the prior art mainly relying on a single type of sensor (such as only using water pressure or displacement), the method integrates multi-element information of tunnel state, environmental driving (rainfall) and key geological structure (water-bearing fracture), constructs a “risk feature vector” containing hydrological, geological, mechanical and other multi-factor coupled information, and significantly enhances the capture range and integrity of disaster precursor information. Through the logical regression model, multi-source data are associated, learned and probabilistically predicted, so that the risk assessment is established on the basis of richer and more three-dimensional information.

[0019] 2. Deeply coupled water-rock interaction core mechanism. The method not only relies on surface parameters (such as water pressure, displacement), but also directly cuts into the key link of water-rock coupling effect - the mechanical response of structural surface (water-bearing fracture). The dominant water-bearing fracture (the largest K) is accurately positioned by high-density electrical method, effective normal stress and shear stress are calculated, and then the key index stress intensity factor (K) reflecting the stability of the fracture is obtained. This quantitative analysis based on physical and mechanical mechanism can directly reflect the risk of rock mass fracture under high-pressure water (cracking early warning), so that the early warning judgment has a solid mechanical mechanism support. This overcomes the defects of the prior art that the internal mechanism of disaster gestation (i.e. stress-driven fracture instability and expansion) is fuzzy and quantitatively insufficient.

[0020] 3. Construct a "probability + mechanism" dual early warning system to improve the effectiveness of early warning. Real-time risk probability assessment (S4): Using a trained logistic regression model, the current real-time monitoring data (processed to form a feature vector) can be quickly calculated to output the probability value of water inrush. This probabilistic result provides an objective and quantitative basis for different levels of risk, making up for the shortcomings of traditional empirical early warning. Fracture critical state early warning (S3): Based on the stress intensity factor, the dominant fracture is evaluated for cracking risk. Essentially, it is to detect the approach to the critical point of the potential water inrush channel (fracture). This provides a forward-looking, physically-based early warning dimension. The combination of water inrush risk probability prediction (S4, focusing on macro trends) and fracture cracking risk assessment (S3, focusing on micro key points) in parallel (order can be changed) or in coordination forms a dual early warning system that complements and verifies each other. This synergy greatly improves the reliability (reduces false positives and false negatives) and depth of early warning information.

[0021] 4. Significantly improve the real-time and automation level of early warning. This method realizes the automatic training and learning of historical data (S1) and the automatic processing, modeling calculation, and early warning judgment of real-time data (S3, S4). The entire process is highly automated, greatly reducing the dependence on manual monitoring and data interpretation. In the face of the strong and fast development of water inrush disasters, this automated process can instantly complete the entire process from data collection, feature extraction, model calculation to early warning output, achieving a response time of seconds or even milliseconds, effectively making up for the fatal shortcoming of traditional manual monitoring's slow reaction, and ensuring that early warning information can be issued in time before catastrophic water inrush occurs.

[0022] In summary, this method can integrate multi-source data and consider the water-rock coupling mechanism to perform real-time and efficient tunnel water inrush monitoring and early warning. It combines data-driven risk probability rapid assessment and physically-based fracture critical state judgment to build a more comprehensive, accurate, timely, and reliable early warning system, effectively addressing the high-risk challenges of karst tunnel water inrush disasters in complex geological environments.

[0023] Preferably, in S2, the uniaxial compressive strength of the rock mass at the tunnel vault is also tested to obtain the ultimate shear strength and the stress allowance value [σ u ];

[0024] In S3, the calculated effective normal stress σ' n and shear stress τ' n of the water-bearing fracture are compared with the ultimate shear strength and the stress allowance value [σ u ] to determine whether there is a risk.

[0025] Such a setting, by uniaxial compressive strength test on rock mass, and combined with high-density electrical method to obtain water-filled fracture information, the scheme realizes the full range analysis from macroscopic geological structure to microscopic rock mass mechanical properties. This fine risk management method helps to find and handle possible safety hazards in advance, avoids sudden disasters caused by local rock mass instability, and ensures the safety and stability of tunnel construction. This method not only considers the influence of groundwater dynamic change on rock mass, but also fully considers the mechanical properties of rock mass. By comparing the calculated stress parameters with the limit shear strength and stress allowable value of rock mass, the warning criteria can be adjusted flexibly according to different geological conditions and engineering environment, which enhances the adaptability and flexibility of the warning system.

[0026] Preferably, in S3, the effective normal stress σ' n and shear stress τ' n are calculated as follows:

[0027] σ' n = σ n - αP; P = r w (H-z);

[0028] τ' n = c t + μ t (σ n - αP);

[0029] where σ n is the measured total stress of rock mass; P is the water pressure of water-filled fractured rock mass; r w is the water bulk density, H is the water head, and z is the position elevation; c t and μ t are the single-fracture friction factor and cohesion considering the time effect of water-rock interaction, respectively;

[0030] When σ n ≥ p, α = 0; When σ n < p, α = 1;

[0031] where k ao and u nmax are the initial stiffness and maximum normal closure of the fracture, respectively;

[0032] The calculation formula of stress intensity factor K Ⅱ is as follows:

[0033]

[0034] where a is the crack length.

[0035] Such a setting, 1. The traditional method often ignores the time effect of water pressure on rock mass strength parameters (such as long-term water-rock interaction leading to c t 、μ t Weakening), or using static empirical coefficient correction. This method reflects the nonlinear change of effective stress of fracture under the coupling action of high ground pressure and high water pressure (such as the reduction of permeability caused by fracture closure under high ground pressure) by dynamically calculating α. In addition, the design of piecewise function (α n When p < σ t , μ t ), quantifying the attenuation of rock mass strength caused by long-term seepage-stress coupling, can make the model closer to the actual disaster process.

[0036] 2. Existing early warning relies on macroscopic parameter threshold (such as sudden displacement increase, water pressure sudden rise), which cannot directly associate with the mechanical nature of fracture expansion. This method directly represents the stress concentration degree of fracture tip by stress intensity factor K Ⅱ , anchors the early warning standard in the critical mechanical state of fracture penetration, and replaces the empirical criterion with fracture mechanics theory , so that the early warning result has strict physical interpretability and can reduce misjudgment.

[0037] Preferably, in S3, when making corresponding cracking early warning judgment based on stress intensity factor K Ⅱ , compare K Ⅱ with the preset fracture threshold K ⅡC .

[0038] If 0.8K ⅡC ≤ K Ⅱ ≤ 0.9K ⅡC , subcritical crack propagation warning is carried out;

[0039] If 0.9K ⅡC ≤ K Ⅱ ≤ K ⅡC , emergency reinforcement warning is carried out;

[0040] If K ⅡC ≤ K Ⅱ , immediate evacuation warning is carried out.

[0041] Such a setting, the traditional early warning is mostly used single threshold, can not distinguish the crack from stable expansion (low risk) to critical instability (high risk) of continuous evolution state, resulting in rough early warning information (only "safe / dangerous" binary judgment). The three-level early warning mechanism of the method accurately maps the nonlinear evolution law of crack propagation (subcritical → accelerated → critical breakthrough); can provide progressive response time window for the construction party (such as subcritical early warning stage can optimize the construction scheme in advance), greatly improve the forward-looking of risk control.

[0042] Preferably, in S1, the tunnel data includes the rock mass stress σ(t), the osmotic pressure P(t) and the vault settlement D(t) of the last X days; the rainfall data includes the rainfall R(t) of the last X days and the historical maximum rainfall R max .

[0043] Preferably, in S1, the data processing includes:

[0044] After the anomaly value check and normalization processing of the tunnel data, the rock mass stress σ norm , osmotic pressure P norm and settlement D norm are obtained as indicators;

[0045] For rainfall data, first calculate the ratio of rainfall R(t) at each time to the historical maximum value Then introduce the decay factor λ to weight the historical rainfall influence, calculate the cumulative rainfall influence value Where, T is the number of rainfall R(t) data obtained;

[0046] Integrate the above data to obtain the wind direction feature vector X = [σ norm , P norm , D norm , R ratio , R cum ].

[0047] Such a setting, 1. Break through the limitation of single point data, build disaster correlation characteristics with clear physical meaning. Traditional methods directly use original monitoring data (such as untreated σ(t), P(t)), ignoring the time-varying coupling effect of data noise, dimension difference and environmental factors (rainfall). This method improves the anti-interference and comparability of rock mass state parameters through anomaly value cleaning + normalization, avoiding misjudgment; R ratio (t) converts instantaneous rainfall into relative risk measure relative to historical extreme value (such as ratio > 1, which breaks through the historical extreme value), enhances the response sensitivity of extreme weather; R cum(t) Quantify the time delay hazard of rainwater continuous infiltration by decay factor λ (such as 0.8-0.95) (e.g. the contribution of rainstorm three days ago to the current surrounding rock water saturation), solve the defects of traditional methods in modeling the correlation of "rain-inrush".

[0048] 2、Precise quantification of "water-rock-rain" dynamic coupling mechanism. Existing warning models often separate the dynamic correlation of rainfall and rock mass response (such as using only the daily rainfall), or simply superimpose static parameters. In this method, the feature vector X explicitly integrates three key disaster driving factors: real-time state of rock mass (σ norm ,P norm ,D norm → high ground pressure, high water pressure, deformation instability); rainfall instantaneous impact (R ratio → extreme precipitation induced instantaneous seepage); hydrological cumulative effect (R cum → long-term infiltration reduces rock mass strength). The logic regression model can learn the coupled disaster evolution path of multiple time scales and multiple physical fields, significantly improving the prediction interpretability.

[0049] 3、Provide high-value input for machine learning model and reduce the risk of overfitting. Existing technologies directly input raw data, which can easily lead the model to fall into noise learning or feature collinearity trap (such as the large dimensional difference and strong correlation between original stress and osmotic pressure). This method normalizes and constructs features (R ratio 、R cum ) to compress the feature value range and accelerate model convergence; and the constructed features have clear physical meaning (such as R cum representing rock mass saturation), enhancing the model generalization ability and avoiding overfitting in data scarce tunnel scenarios.

[0050] Preferably, in S1, a risk data matrix X is also constructed wherein x ij represents the jth index in the ith wind direction feature vector, n is the number of risk feature vectors, and m is the number of indexes in the risk feature vector;

[0051] and the weight of each index in the wind direction feature vector is calculated:

[0052]

[0053] In the formula, E j is the information entropy of the jth index;

[0054]

[0055] Such a setting, 1. solve the index contribution ambiguity problem, realize objective weighting. Traditional methods rely on expert experience to set weights subjectively (such as analytic hierarchy process AHP), which is prone to human bias and cannot dynamically respond to changes in data distribution. In this method, information entropy E j Quantitatively characterize the amount of index information, automatically identify key indicators with high discrimination (such as osmotic pressure mutation data); weight formula W j Dynamic adjustment, self-adaptive weight distribution update with historical data accumulation, ensure that the model always focuses on the most informative disaster precursor signals, and avoids subjective bias.

[0056] 2, suppress noise index interference, improve model robustness. In the existing unweighted feature vector, high volatility but meaningless noise indicators (such as temporary stress fluctuations caused by construction disturbance) may be strongly correlated with real disaster signals, leading to misjudgment. In this method, high-entropy indicators (such as random noise) automatically attenuate (Ej→1 when Wj→0); low-entropy indicators (such as continuously increasing cumulative rainfall) significantly enhance the weight → make the logistic regression model more focused on stable, interpretable physical driving factors, reducing the risk of overfitting.

[0057] Preferably, in S1, the logistic regression model is:

[0058] Where β0 is the intercept term; β j is the characteristic coefficient, β j = W j .

[0059] Such a setting, by constructing and training a logistic regression model, combining historical data and real-time data, effectively predicts and warns of tunnel water inrush risk. This method not only has strong theoretical basis and practical feasibility, but also provides strong support for the safety management of tunnel engineering.

[0060] Preferably, in S4, after calculating the real-time risk probability P new = P(Y = 1 | X new ), the water inrush risk level is judged according to the following method, and the corresponding level of warning is carried out; where x new represents the real-time risk feature vector;

[0061]

[0062] Where θ is a pre-set water inrush risk reference value.

[0063] Such a setting, by real-time calculation and grading warning, effectively manages and controls the tunnel water inrush risk. This method not only improves the accuracy and timeliness of risk assessment, but also provides reliable protection for the safe operation of tunnel engineering.

[0064] Preferably, in S1, in the process of training the logistic regression model, the maximum likelihood method is used to construct the log-likelihood function and optimize the parameter solution: The parameter estimate value that maximizes lnL(β) is solved by iteration through the Newton-Raphson method

[0065] In S4, the water inrush risk reference value θ is determined based on historical water inrush event data through the ROC curve, and satisfies:

[0066] θ = argmax(Youden index) = arg max(sensitivity + specificity - 1).

[0067] Such a setting realizes accurate estimation of the parameters of the logistic regression model through the maximum likelihood method and the Newton-Raphson method, and determines a reasonable water inrush risk reference value through the ROC curve and the Youden index. This method not only improves the prediction ability and stability of the model, but also provides a scientific basis and effective means for tunnel water inrush warning. BRIEF DESCRIPTION OF DRAWINGS

[0068] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below in combination with the drawings, in which:

[0069] Figure 1 is a flowchart of the method;

[0070] Figure 2 is an example diagram of the detection principle of the stress rod monitoring water-bearing fractured rock mass stress device in the embodiment;

[0071] Figure 3 is a stress analysis diagram of the water-bearing fracture in the embodiment. DETAILED DESCRIPTION

[0072] The application will be further described in detail below through specific embodiments:

[0073] Embodiment:

[0074] As shown in the drawings, the embodiment discloses a tunnel water inrush monitoring and warning method under water-rock coupling effect, which comprises the following steps: Figure 1 S1, obtain historical time series tunnel data and rainfall data, and perform data processing, and each data at the same time after processing is combined into a risk feature vector, and whether water inrush disaster occurs at the corresponding time is taken as a label to obtain training data; the training data is used to train a preset logistic regression model; the logistic regression model is used to predict water inrush risk probability according to the risk feature vector.

[0075]

[0076] ​Wherein, the tunnel data includes tunnel rock mass stress σ (t), osmotic pressure P (t), vault settlement D (t) in the last X days; the rainfall data includes rainfall R (t) in the last X days and the historical maximum rainfall R max .

[0077] The data processing includes:

[0078] After the tunnel data is subjected to outlier checking and normalization processing, the rock mass stress σ norm , osmotic pressure P norm , and settlement D norm are obtained as indexes;

[0079] For the rainfall data, first, the ratio of each time rainfall R (t) to the historical maximum value is calculated Then, the decay factor λ is introduced to weight the historical rainfall influence, and the cumulative rainfall influence value is calculated Wherein, T is the number of obtained rainfall R (t) data;

[0080] Integrating the above data, the wind direction feature vector X = {σ norm , P norm , D norm , R ratio , R cum} is obtained.

[0081] In this way, through outlier cleaning and normalization, the anti-interference and comparability of the rock mass state parameters are improved, and misjudgment is avoided; R ratio (t) converts the instantaneous rainfall into a relative risk measure relative to the historical extreme value (such as the ratio > 1, which breaks through the historical extreme value), and enhances the extreme weather response sensitivity; R cum (t) quantifies the time lag hazard of continuous rainfall infiltration (such as the contribution of the rainstorm three days ago to the current surrounding rock water saturation) through the decay factor λ (such as 0.8-0.95), and solves the defects of the traditional method in modeling the correlation between “rainfall and water inrush”. In addition, the feature vector X explicitly integrates three types of key disaster driving factors: rock mass real-time state (σ norm , P norm , D norm → high ground pressure, high water pressure, deformation instability); rainfall instantaneous impact (R ratio → extreme precipitation induces instantaneous seepage); hydrological cumulative effect (R cum → long-term infiltration reduces rock mass strength). The logic regression model can learn the coupled disaster evolution path of multiple time scales and multiple physical fields, significantly improving the prediction interpretability. Moreover, normalization and feature construction (R ratio , R cum ) compress the feature value range, accelerate model convergence; and the constructed features have clear physical meaning (such as R cumrepresenting rock mass saturation), enhancing the generalization ability of the model, and avoiding overfitting in the tunnel scene with scarce data.

[0082] In specific implementation, a risk data matrix is also constructed wherein x ij represents the jth index in the ith wind direction feature vector, n is the number of risk feature vectors, and m is the number of indexes in the risk feature vector;

[0083] and the weight of each index in the wind direction feature vector is calculated:

[0084]

[0085] In the formula, E j is the information entropy of the jth index.

[0086]

[0087] In this way, in the present method, the information entropy E j quantitatively represents the amount of information of the index and automatically identifies key indexes with high discrimination (such as sudden change data of osmotic pressure); the weight formula W j is dynamically adjusted, and the weight distribution is automatically updated with the accumulation of historical data, so as to ensure that the model always focuses on the most informative disaster precursor signals and avoids subjective bias. In addition, high-entropy indexes (such as random noise) automatically attenuate (when Ej→1, Wj→0); low-entropy indexes (such as continuously increasing cumulative rainfall) significantly enhance the weight → make the logistic regression model focus more on stable and interpretable physical driving factors, and reduce the risk of overfitting.

[0088] The logistic regression model is:

[0089] wherein β0 is an intercept term; β j is a feature coefficient, and β j = W j .

[0090] In this way, by constructing and training the logistic regression model, the effective prediction and early warning of the tunnel water inrush risk are realized by combining historical data and real-time data. The present method not only has strong theoretical basis and practical feasibility, but also can provide strong support for the safety management of tunnel engineering.

[0091] In the process of training the logistic regression model, the maximum likelihood method is used to construct a log-likelihood function and optimize the parameter solution: The parameter estimate value that maximizes lnL(β) is iteratively solved by the Newton-Raphson method

[0092] S2, using high-density electrical method, obtaining the position section of water-bearing fissure of the tunnel; and performing uniaxial compressive strength test on the rock mass of the tunnel vault to obtain the ultimate shear strength of the rock mass and stress allowable value [σ u ].

[0093] In this way, by performing uniaxial compressive strength test on the rock mass and combining with the water-bearing fissure information obtained by the high-density electrical method, the scheme realizes all-around analysis from macro geological structure to micro mechanical properties of the rock mass. This refined risk management mode helps to discover and handle possible safety hazards in advance, avoids sudden disasters caused by local rock mass instability, and ensures the safety and stability of the tunnel construction. The method not only considers the influence of underground water dynamic change on the rock mass, but also fully considers the mechanical properties of the rock mass. By comparing the calculated stress parameters with the ultimate shear strength and stress allowable value of the rock mass, the warning standard can be flexibly adjusted according to different geological conditions and engineering environment, and the adaptability and flexibility of the warning system are enhanced.

[0094] In specific implementation, the electrodes are arranged in the position of the tunnel vault in a dipole-dipole manner. This method is sensitive to water-bearing fissure, can clearly identify the resistivity gradient change, and can enhance the abnormal signal by combining with the polarization ratio parameter to distinguish between karst cave (high resistance) and water-bearing fissure (low resistance).

[0095] The principle of dipole-dipole method: the length of power supply dipole AB and measurement dipole MN is a, and the interval distance between them is n×a; n is the isolation coefficient, which is a positive integer;

[0096] The power supply and measurement dipoles are arranged in the same line, the detection depth is positively related to n×a, and the empirical formula is D=n×a×k, k is the medium resistivity influence factor, and the low resistance area takes 0.5-0.7;

[0097] According to the distribution difference of artificial current field in underground medium, the potential difference is measured to calculate the apparent resistivity ρ a , the formula is:

[0098] In the formula, K=6πa is the device coefficient; ΔV is the potential difference between MN; I is the power supply current.

[0099] The specific steps of collecting tunnel data by dipole-dipole method are as follows: the electrode spacing a is usually 1-5 meters, and the isolation coefficient n takes the value range 1-8. When arranging, straight lines are arranged along the tunnel axis, and the electrodes are numbered (E1, E2,..., EN) at equal intervals according to the spacing a, and the grounding resistance is ensured to be less than 5kΩ; longitudinal rolling measurement is adopted, that is, the AB dipole is fixed, and the MN data of different n values are collected in turn to form an inverted trapezoidal detection section.

[0100] The raw tunnel data collected by the high-density electrical method is preprocessed by removing bad data, splicing data, and correcting distortion. The specific steps and methods are as follows:

[0101] The first step is to eliminate bad data. The single-point relative error method is used to calculate the relative error of the data collected repeatedly at the same measuring point to filter out abnormal values. The formula is: If δ i >δ 阈值 The threshold is usually 5%-10%, which is considered a bad pixel and removed;

[0102] Where di is the i-th observation value; is the mean of multiple observations.

[0103] The second step is data splicing and correction. Nonlinear deviation correction of the survey line is used. That is, when the survey line is geometrically distorted due to terrain undulation, a piecewise linear correction model is used: x'=x+Δx(y), y'=y+Δy(x). The coordinate deviations of the same-name points in the overlapping area of ​​adjacent survey lines are fitted by polynomials to establish a correction mapping relationship to eliminate the geometric position deviation caused by terrain undulation. The observed potential difference is then converted into an equivalent value of the horizontal terrain: Mapping sloped terrain data to a horizontal reference surface;

[0104] Where Δx(y) and Δy(x) are piecewise functions; ρ 真 is the theoretical resistivity under horizontal terrain; ρ 视 is the apparent resistivity under actual terrain.

[0105] Finally, distortion correction is performed. The least squares method is used to design a two-dimensional equalization filter to suppress single-point and linear distortion: A current density compensation model is established to address the electric field distortion caused by poor electrode contact or uneven dielectric properties: By iteratively adjusting the k value, the corrected electric field distribution satisfies the continuity condition of the Poisson equation, thereby suppressing local electric field distortion and linear interference.

[0106] Where W ij is the filter coefficient, by minimizing the residual ε=||d-Wd|| 2 Determine; k is the distortion coefficient; φ0 is the theoretical potential gradient; is the measured potential gradient.

[0107] The optimized data is inverted and inferred to infer the underground resistivity distribution. The basic principle is to use the current continuity equation and least squares optimization. The detailed inversion steps are as follows:

[0108] Using the current continuity equation for forward calculation, assuming that the dielectric resistivity is ρ(x,z) and the current source is II, the potential φ satisfies the Poisson equation: The equation can be expressed as a linear system of equations by finite element method (FEM): K(m)φ=q;

[0109] where δ is Dirac function; φ(r) represents the position of point current source r s ; m is model parameter, where the log transform of resistivity: m i =lnρ i ; K is stiffness matrix, dependent on model parameter; q is source term vector.

[0110] The objective function is optimized by weighted least squares: min

[0111] where d obs is observed data, i.e. log of potential difference or apparent resistivity; F(m) is forward response; W d is data weighting matrix, usually diagonal, elements are inverse of data standard deviation; Wm is model regularization matrix, constraining model smoothness or structure; λ is regularization parameter, balancing data fitting and model complexity.

[0112] The detailed steps of inversion inference are as follows: firstly, Taylor expansion of nonlinear problem at initial model m k , keeping first order term: F(m k+1 )≈F(m k )+J k Δm k ;

[0113] where J is Jacobian matrix, i.e. sensitivity matrix.

[0114] Substitute into objective function, get linearized optimization problem:

[0115] Δm k =argmin Δm ||W d (r k -J k Δm)|| 2 |λ|W m (m k +Δm)|| 2 ,

[0116] Solve the equation to get the model update:

[0117] where residual r k =d obs -F(m k ).

[0118] Secondly, Jacobian matrix calculation, element of Jacobian matrix J is J ijIndicates the sensitivity of the i-th data to the j-th model parameter: By calculating the forward field φ and the adjoint field ψ, we can get the spatial integral of the two fields:

[0119] Where Ω is the model domain.

[0120] Then comes the regularization design, the regularization matrix W m Smooth constraints are usually used:

[0121] Finally, set the iteration termination conditions, which are divided into residual convergence and model change convergence. Residual convergence is: the relative root mean square error (RMS) is less than the threshold: The model change converges to: ||Δm k ||<∈.

[0122] In summary, the final cross-section diagram of the tunnel cave and cracks is obtained.

[0123] S3. Based on the position cross-section of S2, select the largest K water-bearing fractures and calculate their effective normal stress σ' n and shear stress τ' n , and then calculate the stress intensity factor K Ⅱ ; and based on the stress intensity factor K Ⅱ The corresponding cracking early warning judgment is carried out. The effective normal stress σ' of the water-bearing crack is also calculated. n and shear stress τ' n , and the ultimate shear strength and the allowable stress value [σ u ] to compare and determine whether there is any risk.

[0124] When implementing it specifically, Figure 2 As shown in the figure, 1-stress sensor, 2-stress transmission line, 3-host, 4-rock mass, 5-water-bearing fracture, 6-tunnel.

[0125] First, based on the position of the fracture cross-section, the stress sensor 1 is drilled into the rock mass 4 to a depth that reaches the water-bearing fracture 5 to measure the stress σ of the rock mass. n and water pressure P, the obtained stress data is transmitted to the host 3 through the profit transmission line 2, and then the ultimate stress value of the fractured rock mass [σ u ] and ultimate shear strength Finally, according to the real-time changes of rock mass stress, the effective stress σ' of water-bearing fractures is analyzed by water-rock coupling method. n and effective shear stress τ' n , combined with fracture mechanics to determine the degree of cracking.

[0126] First, according to the rock mass stress value transmitted by the stress sensor 1 to the host computer 3, the effective stress σ' n and shear stress τ' n of the water-bearing fissure under the water-rock interaction are analyzed, and whether the single fissure expands is judged according to the fracture mechanics, and the water inrush early warning is formed according to the cracking degree.

[0127] The rock mass stress change detected by the stress sensor 1 is transmitted to the host computer 3, the effective stress change of the fissure surrounding rock under the water-rock coupling is analyzed in the host computer, the rock mass ultimate shear strength and stress allowable value [σ u ] obtained through uniaxial test are input into the host computer 3 through the mechanical keyboard, the cracking degree of the fissure is judged by using the fracture mechanics, the set early warning condition threshold is also input into the host computer 3 through the mechanical keyboard, and the multi-factor coupling tunnel water inrush monitoring and early warning threshold is also input into the host computer 3 through the mechanical keyboard, then the alarm sound and light lamps are arranged uniformly on both sides of the tunnel haunch, and the alarm sound and light lamps are connected to the host computer 3.

[0128] As shown in Figure 3 , the effective stress σ' n of the rock mass fissure is obtained according to the effective stress law of the saturated porous elastic medium;

[0129] σ' n = σ n - αP; P = r w (H - z);

[0130] In the formula, σ n is the measured total stress of the rock mass; P is the water pressure of the water-bearing fissure rock mass; r w is the water bulk density, H is the water head, and z is the position elevation;

[0131] When σ n ≥ p, when σ n < p, α = 1;

[0132] In the formula, k ao , u nmax are the initial stiffness and maximum normal closure of the fissure respectively;

[0133] The influence of water-rock coupling on the stability of the water-bearing fissure of the rock mass, according to the above expression of the effective stress, the classical Coulomb criterion is modified as follows:

[0134] τ' n = c t + μ t (σ n - αP);

[0135] where c t and μ t are the time-dependent frictional coefficient and cohesion of the single fracture, respectively.

[0136] The stress intensity factor K Ⅱ is calculated as:

[0137]

[0138] where a is the crack length.

[0139] Traditional methods often ignore the time-dependent effect of water pressure on rock mass strength parameters (e.g., weakening of c t and μ t due to long-term water-rock interaction) or use static empirical coefficients for correction. This method reflects the nonlinear changes in effective stress of the fracture under the coupling effect of high ground pressure and high water pressure (e.g., reduced permeability due to fracture closure under high ground pressure) by dynamically calculating α. In addition, the piecewise function design (α n < p) forces the coverage of the most unfavorable hydraulic fracturing conditions, avoiding the underestimation of traditional methods in such high-risk scenarios. The introduction of time-dependent parameters (c t and μ t quantifies the attenuation of rock mass strength caused by long-term seepage-stress coupling, making the model more realistic in actual disaster processes. In addition, this method directly characterizes the stress concentration at the fracture tip through the stress intensity factor K Ⅱ , anchoring the warning criteria at the critical mechanical state of fracture penetration. The use of fracture mechanics theory instead of empirical criteria makes the warning results strictly physically interpretable, reducing false positives.

[0140] When making the corresponding cracking warning judgment based on the stress intensity factor K Ⅱ , compare K Ⅱ with the preset fracture threshold K ⅡC .

[0141] If 0.8K ⅡC ≤ K Ⅱ ≤ 0.9K ⅡC , a subcritical crack propagation warning is given.

[0142] If 0.9K ⅡC ≤ K Ⅱ ≤ K ⅡC , an emergency reinforcement warning is given.

[0143] If K ⅡC ≤ K Ⅱ , an immediate evacuation warning is given.

[0144] Thus, traditional early warning often uses a single threshold, which cannot distinguish between the continuous evolution state of the fracture from stable expansion (low risk) to critical instability (high risk), resulting in rough early warning information (only "safe / dangerous" binary judgment). The three-level early warning mechanism of the method accurately maps the nonlinear evolution law of fracture expansion (subcritical → accelerated → critical breakthrough); it can provide a progressive response time window for the construction party (such as the subcritical early warning stage can optimize the construction scheme in advance), greatly improving the forward-looking of risk management and control.

[0145] S4, acquiring current tunnel data and rainfall data and processing to obtain a real-time risk feature vector, then using the trained logistic regression model to predict the current water bursting risk probability; and based on the current water bursting risk probability, making a corresponding water bursting early warning judgment.

[0146] In specific implementation, the real-time risk probability P new = P)Y = 1 | X new After that, the water bursting risk level is judged in the following way, and the corresponding level of early warning is carried out; wherein, x new represents the real-time risk feature vector;

[0147]

[0148] Wherein, θ is a preset water bursting risk reference value.

[0149] The water bursting risk reference value θ is determined based on historical water bursting event data through the ROC curve, and satisfies: θ = argmax (Youden index) = arg max (sensitivity + specificity - 1).

[0150] In this way, through real-time calculation and grading early warning, effective management and control of the water bursting risk of the tunnel are realized. This method not only improves the accuracy and timeliness of risk assessment, but also provides reliable protection for the safe operation of the tunnel project.

[0151] Wherein, the order of S1 and S2 can be interchanged, and the order of S3 and S4 can be interchanged.

[0152] The method innovatively combines historical time series monitoring data (tunnel sensor data, rainfall) and real-time geophysical exploration data (water-bearing fracture position section map obtained by high-density electrical method). Compared with the existing technology which mainly relies on a single type of sensor (such as only using water pressure or displacement), the method integrates multiple information of tunnel state, environmental driving (rainfall) and key geological structure (water-bearing fracture), builds a "risk feature vector" containing hydrological, geological, mechanical and other multi-factor coupled information, significantly enhances the capture range and completeness of disaster precursor information. Through the logical regression model, the multi-source data is associated with learning and probability prediction, so that the risk assessment is based on more abundant and more three-dimensional information. In addition, the method not only relies on surface parameters (such as water pressure, displacement), but also directly cuts into the key link of water-rock coupling effect - the mechanical response of structural plane (water-bearing fracture). By accurately positioning the dominant water-bearing fracture (the largest K), the effective normal stress and shear stress are calculated, and then the key indicator stress intensity factor (K) reflecting the stability of the fracture is obtained. This quantitative analysis based on physical and mechanical mechanism can directly reflect the risk of rock mass fracture under high pressure water (cracking warning), so that the warning judgment has a solid mechanical mechanism support. This overcomes the defects of the existing technology in fuzzy understanding and insufficient quantification of the internal mechanism of disaster incubation (i.e. stress-driven fracture instability and expansion).

[0153] In addition, a "probability + mechanism" dual warning system is constructed to improve the warning effectiveness. Real-time risk probability assessment (S4): using the trained logistic regression model, the current real-time monitoring data (processed to form a feature vector) can be quickly calculated, and the probability value of water inrush and water gushing is output. This probabilistic result provides an objective and quantitative basis for different levels of risk, making up for the shortcomings of traditional experience-based warning. Fracture critical state warning (S3): based on the stress intensity factor, the dominant fracture is evaluated for cracking risk. This is essentially a detection of the approach to the critical point of potential water inrush channel (fracture). This provides a forward-looking, physically-based warning dimension. The water inrush risk probability prediction (S4, focusing on macro trends) and the fracture cracking risk assessment (S3, focusing on micro key points) are applied in parallel (the order can be changed) or cooperatively, forming a dual warning system that complements and verifies each other. This cooperation greatly improves the reliability (reduces false positives and false negatives) and depth of warning information.

[0154] The method can integrate multi-source data, consider the water-rock coupling mechanism, and perform real-time and efficient tunnel water inrush monitoring and warning.

[0155] Finally, it needs to be explained that the above examples are only used to illustrate the technical solutions of the present application but not to limit the technical solutions, and those of ordinary skill in the art should understand that the technical solutions of the present application are modified or equivalently replaced without departing from the purpose and scope of the technical solutions, which should be covered in the scope of claims of the present application.

Claims

1. A method for monitoring and early warning of water inrush in tunnels under water-rock coupling, characterized in that: The following steps are involved: S1. Obtain historical time series tunnel data and rainfall data, perform data processing, and combine the processed data at the same time into a risk feature vector. Use whether a flood disaster occurred at the corresponding time as a label to obtain training data. Use the training data to train the preset logistic regression model; The logistic regression model is used to predict the probability of water inrush risk based on the risk characteristic vector; S2. Using high-density electrical method, obtain the location cross-section of water-bearing fissures in the tunnel; S3. Based on the position cross-section of S2, select the largest K water-bearing fractures and calculate their effective normal stress σ' n and shear stress τ' n , and then calculate the stress intensity factor K Ⅱ ; and based on the stress intensity factor K Ⅱ Make corresponding cracking early warning judgment; S4. Obtain and process the current tunnel data and rainfall data to obtain a real-time risk feature vector, and then use the trained logistic regression model to predict the current water inrush risk probability; And make corresponding water inrush warning judgments based on the current water inrush risk probability; The order of S1 and S2 can be interchanged, and the order of S3 and S4 can be interchanged.

2. The method for monitoring and early warning of water inrush in a tunnel under water-rock coupling according to claim 1, characterized in that: In S2, the uniaxial compressive strength test of the rock mass at the tunnel vault was also carried out to obtain the ultimate shear strength of the rock mass. and the allowable stress value [σ u ]; In S3, the calculated effective normal stress σ' of the water-bearing fracture is also n and shear stress τ' n , and the ultimate shear strength and the allowable stress value [σ u ] to compare and determine whether there is any risk.

3. The method for monitoring and early warning of water inrush in a tunnel under water-rock coupling as claimed in claim 2, characterized in that: In S3, the effective normal force σ' n and shear stress τ' n The calculation formulas are: in n =s n -αP;P=r w (Hz); the n =c t +m t (s n -αP); Where, σ n is the measured total stress of the rock mass; P is the water pressure of the water-containing fractured rock mass; r w is the specific gravity of water, H is the head of water, z is its position elevation; c t 、μ t are the friction factor and cohesion of a single fracture considering the time effect of water-rock interaction; When σ n When ≥p, When σ n When <p, α=1; Where k ao ,u nmax are the initial stiffness and maximum normal closure of the crack, respectively; Stress intensity factor K Ⅱ The calculation formula is: Where a is the crack length.

4. The method for monitoring and early warning of water inrush in a tunnel under water-rock coupling as claimed in claim 3, characterized in that: In S3, based on the stress intensity factor K Ⅱ When making corresponding cracking warning judgment, K Ⅱ and the preset crack threshold K ⅡC Make comparisons; If 0.8K ⅡC ≤K Ⅱ ≤0.9K ⅡC , then a subcritical crack expansion warning is carried out; If 0.9K ⅡC ≤K Ⅱ ≤K ⅡC , then carry out emergency reinforcement warning; If K ⅡC ≤K Ⅱ , an immediate evacuation warning will be issued.

5. The method for monitoring and early warning of water inrush in a tunnel under water-rock coupling according to claim 1, characterized in that: In S1, the tunnel data includes the tunnel rock stress σ(t), seepage pressure P(t), and vault settlement D(t) in the last X days; the rainfall data includes the rainfall R(t) in the last X days and the historical maximum rainfall R(t) in the last Y years. max .

6. The method for monitoring and early warning of water inrush in a tunnel under water-rock coupling according to claim 5, characterized in that: In S1, the data processing includes: After checking the outliers and normalizing the tunnel data, the rock stress σ is obtained as an indicator. norm , osmotic pressure P norm , settlement D norm ; For rainfall data, first calculate the ratio of rainfall R(t) at each time to the historical maximum value Then, the attenuation factor λ is introduced to weight the historical rainfall impact and the cumulative rainfall impact value is calculated. Where T is the number of rainfall R(t) data obtained; Integrating the above data, we can get the wind direction characteristic vector X = [σ norm ,P norm ,D norm ,R ratio ,R cum ].

7. The method for monitoring and early warning of water inrush in a tunnel under water-rock coupling according to claim 6, characterized in that: In S1, a risk data matrix is ​​also constructed Among them, x ij represents the jth indicator in the i-th wind direction feature vector, n is the number of risk feature vectors, and m is the number of indicators in the risk feature vector; And calculate the weight of each indicator in the wind direction eigenvector: Where, E j is the information entropy of the j-th indicator; 8. The method for monitoring and early warning of water inrush in a tunnel under water-rock coupling according to claim 7, characterized in that: In S1, the logistic regression model is: Among them, β0 is the intercept term; β j is the characteristic coefficient, β j =W j .

9. The method for monitoring and early warning of water inrush in a tunnel under water-rock coupling according to claim 8, characterized in that: In S4, the real-time risk probability P is calculated new =P(Y=1|X new ), the water inrush risk level is judged as follows, and the corresponding level of warning is issued; where x new Represents the real-time risk feature vector; Among them, θ is the preset reference value of water inrush risk.

10. The method for monitoring and early warning of water inrush in a tunnel under water-rock coupling according to claim 8, characterized in that: In S1, during the training of the logistic regression model, the maximum likelihood method is used to construct the log-likelihood function and optimize the solution parameters: The parameter estimate that maximizes lnL(β) is solved iteratively using the Newton-Raphson method. In S4, the water inrush risk reference value θ is determined based on historical water inrush event data through the ROC curve and satisfies: θ = argmax (Youden index) = argmax (sensitivity + specificity - 1).

Images