A tunnel gushing water and collapse combined prediction and treatment method of four field coupling and multi-source data fusion

Abstract

The application discloses a tunnel water gushing-collapsing combined prediction and treatment method based on four-field coupling and multi-source data fusion, relates to the field of tunnel engineering waterproof and drainage and information construction technology, and discloses that the existing dynamic prediction model does not fuse advanced geological prediction information such as TSP and transient electromagnetic method, leading to strong multi-solution inversion; the model structure is fixed, and internal mechanism transition cannot be automatically identified; only water gushing amount can be predicted, and collapsing risk cannot be evaluated simultaneously. The application realizes that damage evolution is directly associated with microseismic monitoring energy release rate by establishing a heat-water-force-damage four-field coupling model; Bayesian inversion is carried out on the fused microseismic, TSP, transient electromagnetic and conventional monitoring data; water gushing amount and collapsing risk index are predicted, water gushing-collapsing chain disaster mode is identified; and a treatment scheme is automatically recommended according to the risk type, and the method has model self-adaptive selection and migration learning capabilities. The method can fuse multi-source data, realize water gushing-collapsing combined prediction and automatically recommend a treatment scheme.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering drainage and information-based construction technology, specifically to a method for joint prediction and treatment of tunnel water inrush and collapse based on four-field coupling and multi-source data fusion. Background Technology

[0002] In the construction of mountain tunnels, deep-buried tunnels, and fault fracture zones, water inrush and collapse are two major hazards, which often induce and couple with each other. Existing water inrush prediction methods are mainly divided into three categories: empirical formula methods, numerical simulation methods, and data-driven methods. Empirical formula methods (such as the Djubuy formula) make stringent assumptions and cannot reflect complex seepage behavior; numerical simulation methods (such as the finite element method and finite volume method) can handle complex boundaries, but they heavily rely on accurate hydrogeological parameters and are mostly pure seepage models, ignoring the dynamic feedback of the surrounding rock stress field, deformation field, and fracture damage evolution on seepage characteristics; data-driven methods lack physical mechanisms and have poor extrapolation capabilities.

[0003] In recent years, some studies have begun to attempt to establish stress-seepage-damage coupled models. Although a non-Darcy dynamic prediction model coupling stress-damage-seepage three-field model has been established, it still has the following shortcomings: It does not incorporate a temperature field, making it impossible to identify seepage paths using groundwater temperature anomalies; damage evolution mainly relies on mechanical parameters and is not directly coupled with microseismic monitoring energy release rates; it only relies on inflow and pore pressure data, without integrating advanced geological prediction information such as TSP and transient electromagnetic data, leading to strong inversion ambiguity; the model structure is fixed, making it unable to automatically identify the mechanistic transitions from Darcy flow to non-Darcy flow and from intact surrounding rock to fractured media; and it only predicts inflow and fails to simultaneously assess collapse risk. Summary of the Invention

[0004] To address the aforementioned problems, this invention aims to provide a method for joint prediction and treatment of tunnel water inrush and collapse based on four-field coupling and multi-source data fusion. This method can fuse multi-source data, consider multi-field coupling, achieve joint prediction of water inrush and collapse, and adaptively recommend treatment solutions.

[0005] The main idea of ​​the technical solution adopted in this invention is as follows: by establishing a four-field coupling model of heat-water-force-damage, damage evolution is directly correlated with the energy release rate of microseismic monitoring; hierarchical Bayesian inversion is performed on the fusion of microseismic, TSP, transient electromagnetic and conventional monitoring data; at the same time, the inflow volume and collapse risk index are predicted, and the inflow-collapse cascading disaster pattern is identified; and treatment plans are automatically recommended according to the risk type, while having the ability of model adaptive selection and transfer learning.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for joint prediction and treatment of tunnel water inrush and collapse based on four-field coupling and multi-source data fusion includes the following steps:

[0008] A four-field coupled numerical model of the tunnel is constructed, and multiple candidate coupled models are built based on the four-field coupled numerical model.

[0009] Collect multi-source data during tunnel construction and preprocess the multi-source data to obtain preprocessed multi-source data;

[0010] Based on the preprocessed multi-source data, hierarchical Bayesian inversion is performed to calibrate the posterior distribution of parameters of each candidate coupled model and to calculate the Bayesian model evidence of each candidate coupled model.

[0011] Based on Bayesian model evidence, the current optimal coupling model is adaptively selected from multiple candidate coupling models.

[0012] Using the optimal coupling model, we can perform joint prediction of water inrush and landslide, and output the predicted water inrush volume, confidence interval, and landslide risk index.

[0013] The risk type is determined based on the predicted water inflow and the landslide risk index;

[0014] Based on the risk type and the selected optimal coupling model, the system automatically matches and outputs treatment plans from the preset treatment rule base.

[0015] After implementing the treatment plan, new multi-source data is continuously collected, and subsequent steps are repeated to achieve a closed-loop dynamic update of prediction and treatment.

[0016] This invention also provides a joint prediction and treatment system for tunnel water inrush and collapse based on four-field coupling and multi-source data fusion, comprising:

[0017] The model building module is used to construct a coupled numerical model of the tunnel's thermal-hydraulic-mechanical-damage fields, and to build multiple candidate coupled models of different complexities based on this model; the data acquisition and preprocessing module is used to collect multi-source data during the tunnel construction process and to preprocess the multi-source data to obtain preprocessed multi-source data.

[0018] The Bayesian inversion module is used to perform hierarchical Bayesian inversion based on preprocessed multi-source data, calibrate the posterior distribution of parameters of each candidate coupled model, and calculate the Bayesian model evidence of each candidate coupled model.

[0019] The model adaptive selection module is used to adaptively select the current optimal coupling model from multiple candidate coupling models based on Bayesian model evidence.

[0020] The joint prediction module uses the optimal coupling model to perform joint prediction of water inrush and landslide, and outputs the predicted water inrush volume and its confidence interval, as well as the landslide risk index.

[0021] The treatment recommendation module is used to determine the risk type based on the predicted water inflow and the collapse risk index; and automatically matches and outputs a treatment plan from the preset treatment rule base according to the risk type and the selected optimal coupling model.

[0022] The closed-loop update and transfer learning module is used to trigger the iterative operation of the Bayesian inversion module, model adaptive selection module, joint prediction module and treatment recommendation module using newly collected monitoring data after the treatment plan is implemented, and to perform transfer learning of model parameters and cases.

[0023] The present invention also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to said at least one processor;

[0024] The memory stores a computer program that can be executed by the at least one processor. When the computer program is executed by the at least one processor, the at least one processor performs the tunnel water inrush-collapse joint prediction and treatment method as described above, which combines four-field coupling and multi-source data fusion.

[0025] The beneficial effects of this invention are:

[0026] (1) Existing methods are mostly pure seepage or three-field coupling (lacking temperature field, damage is not associated with microseismic events). This invention establishes a four-field coupling model of heat-water-mechanical-damage, in which the fracture permeability dynamically evolves with stress and damage, and the damage equation is directly coupled with microseismic energy. It can simultaneously output water inflow and collapse risk index, and identify water inflow-dominated, collapse-dominated, or cascading disasters.

[0027] (2) Existing inversion methods rely on inflow and pore pressure, resulting in multiple solutions. This invention is the first to incorporate microseismic, TSP, transient electromagnetic, and conventional hydrological and temperature monitoring into a three-layer Bayesian framework (TSP / TEM prior generation → MCMC sampling → marginal likelihood model selection), reducing the parameter variation coefficient by 30% to 50% and significantly narrowing the confidence interval.

[0028] (3) Existing methods fix a single model structure. This invention constructs four candidate coupling models C1 to C4, calculates the posterior probability in real time to automatically select the optimal coupling model or weighted average, and defines the model transfer intensity. When the threshold is exceeded, a sudden change warning of seepage-mechanical mechanism is issued to achieve early warning at the physical level.

[0029] (4) Existing treatments are based solely on looking up tables according to the water inflow level. This invention defines a collapse risk index and automatically determines three risk modes (water inflow-dominated / collapse-dominated / chain reaction) based on the combination of water inflow and the index. It outputs targeted measures and parameters (grouting pressure, pipe roof length, etc.) from the rule base, making the decision more accurate and executable.

[0030] (5) The model posterior and parameter posterior of historical cases are stored in a knowledge base, and dual transfer is achieved for new segments through similarity retrieval. Engineering verification shows that it can reduce the initial monitoring data requirement by 30% to 50% and improve the inversion convergence speed by about 40%.

[0031] (6) A complete closed loop of “data collection → inversion → model selection → prediction → treatment → feedback → knowledge base update” is realized. Each successful case is automatically entered into the database, so that the method can continuously evolve in multiple projects and the prediction ability can be continuously improved. Attached Figure Description

[0032] Figure 1 A flowchart of a method for joint prediction and treatment of tunnel water inrush and collapse using four-field coupling and multi-source data fusion;

[0033] Figure 2 This is a schematic diagram illustrating the coupling relationship between the four fields of heat, water, force, and damage.

[0034] Figure 3 A diagram illustrating the hierarchical Bayesian inversion framework for multi-source data;

[0035] Figure 4 This diagram illustrates the evolution of posterior probabilities over time and adaptive switching for different coupling models (C1~C4).

[0036] Figure 5 This is a combined output graph of the water inflow prediction curve and the landslide risk index. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0038] Example 1

[0039] See Figures 1-5 This application discloses a method for joint prediction and treatment of tunnel water inrush and collapse based on four-field coupling and multi-source data fusion, including:

[0040] Step 1: Construct a four-field coupled numerical model of the tunnel, and build multiple candidate coupled models based on this model;

[0041] A coupled numerical model of thermal-hydraulic-mechanical-damage fields is established: The surrounding rock is considered as a dual-medium system consisting of a discrete fracture network and a continuous porous medium. The following four physical fields and their coupling relationships are established:

[0042] Seepage field equations: The flow of water in the fracture adopts the Forchheimer type non-Darcy equations:

[0043] ;

[0044] In the formula The value is the fissure water pressure (Pa). The value is the dynamic viscosity (Pa·s). The velocity is (m / s). Non-Darcy coefficient ( ), The density of water is (kg / m³). This represents the fracture permeability.

[0045] Among them, fracture permeability With effective normal stress and damage variables Dynamic evolution:

[0046] ;

[0047] in: , The coupling coefficients are determined experimentally or through inversion. The initial fracture permeability.

[0048] The matrix pore flow follows Darcy's law, and the matrix permeability... Temperature-related:

[0049] ;

[0050] In the formula: Temperature sensitivity coefficient ( ), Temperature (K) The initial temperature. This represents the initial matrix permeability.

[0051] Stress field equations equilibrium equations:

[0052] ;

[0053] In the formula: Total stress represents the force per unit area within the surrounding rock (including the portion shared by the rock-soil skeleton and pore water pressure), and its unit is Pa. For divergence operators, For the density of water, This is the acceleration due to gravity.

[0054] The constitutive relation adopts an elastoplastic damage model, and the effective stress satisfies Terzaghi's principle:

[0055] ;

[0056] in For effective stress, For Biot coefficient, Pore ​​water pressure, It is a unit tensor (second-order identity matrix) used to convert scalar orifice pressure It is transformed into an isotropic tensor form, so that the total stress and the effective stress are matched in the tensor dimension.

[0057] Damage field equation:

[0058] Define damage variables Its evolution is due to the accumulation of plastic strain and cumulative energy release rate of microseismic monitoring Joint control:

[0059] ;

[0060] In the formula: The plastic strain at failure The characteristic energy is (J). , , For material parameters. Damage leads to a reduction in elastic modulus: , The initial elastic modulus, This represents the elastic modulus after damage, and also affects the permeability according to the aforementioned equation.

[0061] Temperature field equation: Temperature changes in groundwater can indicate seepage paths. Energy conservation equation:

[0062] ;

[0063] In the formula: The volumetric heat capacity of the porous medium is (J / (m³·K)). The density of groundwater, For the specific heat capacity of water, Equivalent thermal conductivity (W / (m·K)) This is a viscous dissipative heat generation term, reflecting the frictional temperature rise in the high flow velocity region.

[0064] The above four equations are coupled in the following ways: the stress field affects porosity and fracture aperture, which in turn affects permeability; the seepage field changes pore pressure, which affects effective stress; the damage field affects both mechanical stiffness and permeability; and the temperature field affects the stress field through thermal expansion and the seepage field through viscosity and density.

[0065] To avoid the rigidity of a single coupled model structure, this invention constructs multiple candidate coupled models with different complexities, as shown in Table 1.

[0066] Table 1 Candidate Coupling Models

[0067]

[0068] Step 2: Collect multi-source data during tunnel construction and preprocess the multi-source data to obtain preprocessed multi-source data;

[0069] The collected data includes conventional hydrological monitoring data (water inflow, pore water pressure, water temperature), microseismic monitoring data (number of events, cumulative energy release rate, source location), TSP advanced geological prediction data (wave velocity anomaly, reflection horizon), and transient electromagnetic data (apparent resistivity profile, low resistivity anomaly). After preprocessing such as time synchronization alignment, outlier removal, filtering and denoising (e.g., wavelet transform), and feature extraction (e.g., moving average of microseismic energy release rate), the preprocessed multi-source data are obtained and used for inversion basis verification, damage variable calibration, prior geological zoning construction, and water-bearing structure identification.

[0070] Step 3: Based on the preprocessed multi-source data, perform hierarchical Bayesian inversion, calibrate the posterior distribution of parameters of each candidate coupled model, and calculate the Bayesian model evidence of each candidate coupled model.

[0071] To overcome the difficulty of sampling in high-dimensional parameter spaces, a three-layer Bayesian framework is established:

[0072] Layer 1 (Prior Generation Layer): Using TSP and TEM data, a strongly constrained prior distribution of the parameters to be inverted is generated through empirical relationships or simplified rock physics models. The TSP wave velocity reduction zone is superimposed with the TEM low resistivity anomaly zone to identify the spatial location of the water-bearing fractured zone. Within this region, the prior mean of permeability is increased by 1-2 orders of magnitude, and the prior variance is appropriately amplified to retain uncertainty. This spatial variation information is encoded as a Gaussian process prior, replacing the globally uniform prior.

[0073] Layer 2 (Data Assimilation Layer): Utilizing the collected water flow rate pore pressure Microseismic energy release rate ,temperature The posterior distribution of the parameters is sampled using MCMC (Markov Chain Monte Carlo) sampling. The likelihood function includes error terms for various data types.

[0074] ;

[0075] in The energy release rate is given by the damage evolution equation. Calculated from the temperature field equation. Let t be the measured value of water inflow. Let t be the predicted water inflow value at time t. The measured value of pore water pressure at time t. Let t be the predicted value of pore water pressure. The measured value of the cumulative energy release rate from microseismic monitoring at time t. The measured temperature value at time t. This is the predicted temperature value at time t. The standard deviation of the observed water flow rate. The standard deviation of the pore pressure observations. The standard deviation of the observed microseismic energy release rate This represents the standard deviation of the temperature observations.

[0076] Layer 3 (Model Evidence Layer): Used to compare coupled models of different complexities, calculate the marginal likelihood of each model, and thus obtain the posterior probability of the model.

[0077] Step 4: Based on Bayesian model evidence, adaptively select the current optimal coupling model from multiple candidate coupling models;

[0078] Real-time calculation of the marginal likelihood of each candidate coupled model This updates the model's posterior probability.

[0079] ;

[0080] In the formula: For the k-th candidate coupling model, For observation data, For the model The posterior probability, To normalize the denominator, For the model The prior probability.

[0081] The model with the highest posterior probability is selected as the current optimal coupled model for joint prediction. If the maximum probability is less than 0.7, the weighted result of the Bayesian model's average output is used. When the posterior probability of the model changes drastically (e.g., C1→C3), a "seepage-mechanical mechanism abrupt change warning" is issued.

[0082] Step 5: Using the optimal coupling model, perform joint prediction of water inrush and landslide, and output the predicted water inrush volume and its confidence interval, as well as the landslide risk index;

[0083] Based on the optimal parameters (or posterior parameter distribution) obtained through inversion, a four-field coupled model is run, outputting the following two key indicators: predicted inflow rate. Its 95% confidence interval and landslide risk index , This is a combined function of the plastic zone penetration and the damage threshold, expressed as:

[0084] ;

[0085] in The radius of the plastic zone of the surrounding rock (obtained through stress field calculation). Where is the tunnel radius. This threshold is based on literature experience, but this invention incorporates dynamic correction using microseismic data: when the microseismic event rate suddenly increases, Automatically reduce to achieve early warning.

[0086] when Issued a landslide warning; when A landslide warning has been issued.

[0087] In addition, a joint early warning rule is set: if the predicted water inflow... Exceeding the design allowable drainage capacity and The risk is determined to be dominated by water inrush; if and If the risk is below the alarm threshold, it is determined to be the primary risk of landslide; if both exceed the threshold, it is determined to be a warning of a chain disaster of water inrush and landslide (the highest level).

[0088] Step 6: Determine the risk type based on the predicted water inflow and landslide risk index; based on the risk type and the selected optimal coupling model, automatically match and output the treatment plan from the preset treatment rule base;

[0089] Based on the dominant risk type and the currently selected optimal coupling model, the treatment plan is automatically output from the preset rule base in Table 2:

[0090] Table 2 Treatment Rule Base

[0091]

[0092] In addition, if the temperature field data is significantly abnormal (e.g., local temperature rise > 2°C), temperature tracing tests are additionally recommended to identify the seepage channels.

[0093] Step 7: After implementing the treatment plan, continuously collect new multi-source data and repeat the subsequent steps to achieve a closed-loop dynamic update of prediction and treatment.

[0094] After treatment, new monitoring data continues to be collected, and steps three through six are automatically repeated to achieve dynamic correction of the prediction results. Transfer learning: The four coupling parameters (including coupling coefficients), the final dominant model type, and the geological feature vectors (lithology, burial depth, fault fracture zone width, and microseismic baseline level) obtained from each successful inversion are stored in a knowledge base. The knowledge base is a database used to automatically store the above data, and similar cases are called for calculations in subsequent applications.

[0095] When a new tunnel segment begins, the three most similar cases are retrieved from the knowledge base, and their posterior probabilities are used as the model priors for the new segment. The posterior distribution of its parameters is used as the prior parameters (fitted by a Gaussian mixture model). This can significantly reduce the amount of data required to identify the correct model for new segments and accelerate inversion convergence.

[0096] Application Case: To verify the feasibility of a method for joint prediction and treatment of tunnel water inrush and collapse based on four-field coupling and multi-source data fusion, the KCT12+375~355 section of the Kuncheng Tunnel, Section 5 of the Kunming section of the Dianzhong Water Diversion Project in Yunnan Province, China, is used as an example. This section traverses a fault fracture zone, where the rock mass is extremely fractured, and the filling material is mainly composed of gravel, breccia, and soft plastic clay, which is a typical fractured rock mass medium. The overlying layer is a gravel layer (2~3m thick, with a permeability coefficient of approximately 5×10⁻⁶). -4 m / s) and the underlying soft plastic clay layer (thickness > 5m, permeability coefficient approximately 1×10) -7 The tunnel has a 5.5m excavation radius, a groundwater level depth of 45m, and an overlying gravel layer of 2.5m thickness.

[0097] On August 21, 2023, a sudden water inrush occurred at excavation face KCT12+364, with the water volume surging from 50 m³ / d to 120 m³ / d. The water was turbid and contained mud, accompanied by collapse and deformation of the initial support. On-site construction of a C30 concrete grout-stopping wall was forced, along with full-section advanced curtain grouting and large pipe roof emergency repairs. This case study utilizes monitoring data from the first 72 hours of construction for this section to validate the methodology.

[0098] The following monitoring systems were deployed during the construction of this section:

[0099] Routine hydrological monitoring: Automatic water inflow recorder (collects data hourly), interstitial water pressure gauges at 2m and 5m behind the working face, and fiber optic temperature sensors (attached to the surface of the initial support, with one measuring point every 10m).

[0100] Microseismic monitoring: 6-channel microseismic sensor array (sampling frequency 10 kHz) to extract event count, cumulative energy release rate and source location.

[0101] Advanced geological prediction: TSP detection (showing a 25%~35% reduction in wave velocity) and transient electromagnetic (TEM) detection (showing a low-resistivity anomaly zone where the apparent resistivity drops from 200Ω·m to 50Ω·m) were completed before excavation.

[0102] Measured data for the first 72 hours: Water inflow sequence: Day 1: 50 m³ / d, Day 2: 68 m³ / d, Day 3: 95 m³ / d, surging to 120 m³ / d in the afternoon of Day 3; Microseismic energy release rate gradually increased from 0.2 J / h on Day 2 to 8.5 J / h in the afternoon of Day 3; Temperature monitoring points showed a 2.1℃ increase in water temperature near the working face; Pore pressure change rate reached +0.15 MPa / d. After time synchronization alignment, outlier removal, and wavelet transform denoising, the hourly moving average of the microseismic energy release rate was extracted from the above data, and the cross-sectional average was used for temperature data.

[0103] A four-field coupled model of heat, water, force, and damage was established following step one. Based on in-situ crack statistics (box dimension method), the initial crack fractal dimension was determined. Initial fracture permeability Initial matrix permeability Non-Darcy coefficients are calculated using empirical formulas. The shape factor is calculated from the average crack spacing.

[0104] Using the superposition results of the TSP wave velocity anomaly region and the TEM low resistivity anomaly region, the extent of the water-bearing fractured zone (approximately 30m long and 15m wide) was delineated. Within this area, strong constraint priors were established (Step 3, Layer 1):

[0105] fracture permeability Log-normal distribution, mean m / s, standard deviation 1.5. Damage variable Beta distribution, mean 0.3, variance 0.1. Coupling coefficient. , , Temperature sensitivity coefficient Outside the fracture zone, the prior probability of infiltration decreases by 1 to 2 orders of magnitude. The prior probabilities of all models are set to be equal. .

[0106] Step 3, Layer 2 data assimilation, was performed using monitoring data from the previous 72 hours. A likelihood function was constructed (including four errors: inflow rate, pore pressure, microseismic energy release rate, and temperature), with each standard deviation set as follows: , , , We employ MCMC sampling (5000 iterations, with the first 2000 being burn-in) and simultaneously calculate the marginal likelihood of four candidate coupled models (C1~C4).

[0107] Calculation results: C1 (seepage + stress): marginal likelihood = -318.2, C2 (+damage - plastic strain driven): marginal likelihood = -295.7, C3 (four-field full coupling - plastic damage): marginal likelihood = -272.4, C4 (four-field full coupling + microseismic driven damage): marginal likelihood = -261.3. Posterior probabilities: C4 = 0.68, C3 = 0.24, C2 = 0.06, C1 = 0.02. The highest probability model C4 has a posterior probability of 0.68 (close to the 0.7 threshold), and the system selects C4 as the optimal coupling model.

[0108] Figure 4 The posterior probabilities of four candidate models are shown. The evolution curve over 72 hours. As shown in the graph: the probability of C1 is 0.55 on day 1, rises to 0.40 for C2 on day 2, and rises to 0.68 for C4 on day 3. Around day 2.5, the probability of C4 exceeds that of C2 and C3, and the system automatically switches to C4, while the model transition intensity... If the threshold of 0.1 / day is exceeded, a "seepage-mechanical mechanism abrupt change warning" is issued, indicating that the surrounding rock has entered a microseismically driven damage-seepage strong coupling stage. This figure verifies the ability of the method of this invention to dynamically identify the optimal coupling model structure based on real-time monitoring data. The posterior mean of the C4 model parameters is obtained through inversion: m / s, (95% confidence interval) Damage evolution parameters , , . The coefficient of variation of the posterior distribution of the parameters is between 0.08 and 0.18, indicating that the inversion has a high degree of confidence.

[0109] Substituting the parameters from the C4 model inversion into the four-field coupled model, the predicted inflow rate was 110–135 m³ / d (95% confidence interval). After the 7th day, it gradually stabilized at 115 m³ / d, showing good agreement with the measured value of 120 m³ / d (error approximately 8%). Collapse risk index: Calculation of the plastic zone radius. m, tunnel radius m, inverted damage variable Dynamically adjust critical damage based on microseismic event rate The energy rate of the microseismic event decreased to 0.75 on the third day due to a sudden increase (reaching 8.5 J / h). A landslide warning was issued. The water inflow exceeded the design allowable drainage capacity. m³ / d, and It has been determined to be a flood-landslide cascade disaster warning (highest risk level).

[0110] Figure 5 This figure displays the combined output of the water inflow prediction curve and the landslide risk index for this case. The horizontal axis represents time (the prediction period of the first 72 hours and the following 48 hours), the vertical axis shows the water inflow (m³ / d) on the left and the landslide risk index on the right. (Dimensionless). The black solid line represents the measured inflow; the blue solid line represents the predicted median of the method of this invention (C4 model), and the blue shaded area represents the 95% confidence interval; the red dashed line represents the design allowable drainage capacity. The green solid line represents the calculated result. It exceeded the 1.0 threshold (horizontal dashed line) on the afternoon of the 3rd day, issuing a landslide warning; the light yellow area in the background color partition is the attention zone ( The light red area is the warning zone. ).from Figure 4 It is evident that the method of this invention not only accurately predicts the water inflow, but also provides an early warning of landslide risks approximately 6 hours in advance, allowing valuable time for emergency decision-making.

[0111] Based on the cascading disaster early warning and optimal coupling model C4, the system outputs from the rule base (step six): Immediately stop work, seal the working face, and construct a grout-stopping wall. Perform full-section high-pressure curtain grouting, with a grouting pressure ≥ 5 MPa, borehole spacing ≤ 1.5 m, and a radial thickness of 5 m for the grouting ring. Use cement-water glass dual-liquid grout (volume ratio 1:1). Use a large pipe shed, 15 m long, with a circumferential spacing of 0.4 m. Install a pressure relief hole at the bottom, 12 m long, controlling the drainage volume to not exceed 30 m³ / h. Increase the microseismic sampling frequency to 20 kHz, increase the temperature measuring point spacing to 5 m, and add two more borehole pressure gauge sections. After implementing this plan on-site, the borehole water output was < 3 m³ / m·d, the acoustic velocity increased by 28%, the permeability decreased to 3.5 Lu, the water inflow was stably controlled below 65 m³ / d, the microseismic event rate decreased to 0.5 J / h, the collapse risk index decreased to 0.48, and the risk level decreased to a safe level.

[0112] The geological feature vectors of this section (fault fracture zone, burial depth 45m, surrounding rock consisting of gravel and breccia interbedded with soft plastic clay, microseismic baseline energy rate 0.2 J / h, fracture fractal dimension 2.68), inversion parameters, and the final dominant model C4 were stored in the knowledge base. Before the excavation of the subsequent adjacent tunnel section (KCT12+335~315, which also crosses a fault fracture zone but at a burial depth of 52m), the system retrieved a case with a similarity of 0.82 and automatically set the model prior to it. , , , The prior parameters were also transferred accordingly. The new section converged correctly to C4 using only 2 days of data, saving about 50% of the initial data requirement compared to the case without transfer, and the inflow prediction error was controlled within 7%.

[0113] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for joint prediction and treatment of tunnel water inrush and collapse based on four-field coupling and multi-source data fusion, characterized in that, Includes the following steps: A four-field coupled numerical model of the tunnel is constructed, and multiple candidate coupled models are built based on the four-field coupled numerical model. Collect multi-source data during tunnel construction and preprocess the multi-source data to obtain preprocessed multi-source data; Based on the preprocessed multi-source data, hierarchical Bayesian inversion is performed to calibrate the posterior distribution of parameters of each candidate coupled model and to calculate the Bayesian model evidence of each candidate coupled model. Based on Bayesian model evidence, the current optimal coupling model is adaptively selected from multiple candidate coupling models. Using the optimal coupling model, we can perform joint prediction of water inrush and landslide, and output the predicted water inrush volume, confidence interval, and landslide risk index. The risk type is determined based on the predicted water inflow and the landslide risk index; Based on the risk type and the selected optimal coupling model, the system automatically matches and outputs treatment plans from the preset treatment rule base. After implementing the treatment plan, new multi-source data is continuously collected, and subsequent steps are repeated to achieve a closed-loop dynamic update of prediction and treatment.

2. The method according to claim 1, characterized in that: The four-field coupled numerical model includes the seepage field equation, stress field equation, damage field equation, and temperature field equation.

3. The method according to claim 2, characterized in that: Several candidate coupling models of varying complexity were constructed, including: Model C1: Contains only the seepage field equations and stress field equations; Model C2: Includes the seepage field equation, stress field equation, and damage field equation; Model C3: It contains a fully coupled four-field equation, including thermal field equation, seepage field equation, stress field equation and damage field equation, where damage is driven by plastic strain; Model C4: It consists of four fully coupled fields: thermal field equation, seepage field equation, stress field equation, and damage field equation. Damage is jointly driven by the energy release rate monitored by microseismic monitoring and plastic strain.

4. The method according to claim 3, characterized in that: The multi-source data includes conventional hydrological monitoring data, microseismic monitoring data, TSP advanced geological prediction data, and transient electromagnetic data.

5. The method according to claim 4, characterized in that: Based on the preprocessed multi-source data, hierarchical Bayesian inversion is performed to calibrate the posterior distribution of parameters for each candidate coupled model, and the Bayesian model evidence for each candidate coupled model is calculated, specifically including: A priori distribution is generated using TSP advanced geological prediction data and transient electromagnetic data; Using conventional hydrological monitoring data, a posterior distribution is generated using the Markov chain Monte Carlo method. Constructing a marginal likelihood function based on multi-source data; Using the marginal likelihood function and prior and posterior distributions, the marginal likelihood of each model in multiple candidate coupled models is calculated.

6. The method according to claim 5, characterized in that: Based on Bayesian model evidence, the current optimal coupling model is adaptively selected from multiple candidate coupling models, specifically as follows: The posterior probability is obtained based on the marginal likelihood of each model, and the model with the largest posterior probability is selected as the optimal coupling model.

7. The method according to claim 6, characterized in that: Based on the predicted water inflow, landslide risk index, and optimal coupling model, the warning level is determined and a warning is issued, specifically including: When the predicted water inflow exceeds the design allowable drainage capacity and the collapse risk index exceeds the first threshold, it is determined to be a water inflow-dominated risk. When the predicted water inflow is lower than the alarm threshold and the collapse risk index exceeds the second threshold, it is determined to be a collapse-dominant risk. When the predicted water inflow exceeds the design allowable drainage capacity and the collapse risk index exceeds the second threshold, it is determined to be a water inflow-collapse cascading disaster warning.

8. The method according to claim 7, characterized in that: The preset treatment rule base includes: When the risk type is water inrush-dominant risk and the optimal coupling model is C1 or C2, output the local grouting or radial grouting scheme, and output the upper limit of grouting pressure and grout type; When the risk type is water inrush-dominated risk and the current optimal coupling model is C3 or C4, output the full-section curtain grouting supplemented by drainage hole scheme, and output the grouting pressure range and the two-liquid grout ratio; When the risk type is collapse-dominant risk, output the advanced pipe roof, shotcrete and anchor support and steel arch frame scheme, and output the pipe roof length and steel frame spacing; When the risk type is a chain reaction of water inrush and collapse, output a combined solution of work stoppage and high-pressure grouting reinforcement, temporary support and pressure relief drainage, and output the minimum grouting pressure.

9. A joint prediction and treatment system for tunnel water inrush and collapse based on four-field coupling and multi-source data fusion, characterized in that, The system for performing the method according to any one of claims 1 to 8 comprises: The model building module is used to construct a numerical model of the thermal-hydraulic-mechanical-damage four-field coupling of the tunnel, and to build multiple candidate coupling models of different complexities based on this model. The data acquisition and preprocessing module is used to acquire multi-source data during tunnel construction and preprocess the multi-source data to obtain preprocessed multi-source data. The Bayesian inversion module is used to perform hierarchical Bayesian inversion based on preprocessed multi-source data, calibrate the posterior distribution of parameters of each candidate coupled model, and calculate the Bayesian model evidence of each candidate coupled model. The model adaptive selection module is used to adaptively select the current optimal coupling model from multiple candidate coupling models based on Bayesian model evidence. The joint prediction module uses the optimal coupling model to perform joint prediction of water inrush and landslide, and outputs the predicted water inrush volume and its confidence interval, as well as the landslide risk index. The treatment recommendation module is used to determine the risk type based on the predicted water inflow and the collapse risk index; and automatically matches and outputs a treatment plan from the preset treatment rule base according to the risk type and the selected optimal coupling model. The closed-loop update and transfer learning module is used to trigger the iterative operation of the Bayesian inversion module, model adaptive selection module, joint prediction module and treatment recommendation module using newly collected monitoring data after the treatment plan is implemented, and to perform transfer learning of model parameters and cases.

10. An electronic device, characterized in that, include: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor. When the computer program is executed by the at least one processor, the at least one processor performs the method for joint prediction and treatment of tunnel water inrush and collapse based on four-field coupling and multi-source data fusion as described in any one of claims 1 to 8.

Images