Tunnel mud burst and water burst intelligent early warning method and system based on multi-modal data

Abstract

The invention discloses a tunnel mud burst and water burst intelligent early warning method and system based on multi-modal data, relates to the technical field of tunnel engineering safety monitoring, and solves the problems that in the prior art, the early warning precision is low, the false alarm rate is high, and prevention and control measures are difficult to dynamically match risk positions. According to the technical scheme, the method is characterized in that geological and hydrological multi-modal data are fused to construct a three-dimensional geological and water flow model, and a space-time registration engine is adopted to eliminate positioning errors and guarantee data timeliness; dynamically calculating an inrush risk value in combination with a construction disturbance factor, and quantifying a disaster chain triggering mechanism; according to the risk grade and the space coordinate, automatically mapping a grading prevention and control instruction to realize accurate resource allocation; and the lithology identification rule and the permeability coefficient calculation model are optimized in real time through an event feedback mechanism so as to improve the model precision. The effects that the false alarm rate is remarkably reduced through global accurate early warning, the response speed is increased through dynamic cooperation of prevention and control measures and risks, and construction safety is guaranteed are achieved.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering safety monitoring technology, and more specifically, to a method and system for intelligent early warning of tunnel mudslides and water inrushes based on multimodal data. Background Technology

[0002] Water and mud inrush are typical geological hazards during tunnel construction, mainly caused by the disturbance of rock masses and disruption of groundwater balance during tunnel excavation. They are characterized by their suddenness and destructiveness, with few or no warning signs before they occur. Large amounts of water or mud rush into the tunnel in a short period, leading to rock collapse, equipment damage, and even casualties. Therefore, accurate prediction of water and mud inrush is crucial for tunnel safety. Existing prediction technologies use geological analysis to identify fault fracture zones and qualitatively assess risk based on rock mass integrity classification; they employ mathematical models combined with physical simulations and seepage mechanics principles to construct prediction models; monitoring and early warning technologies rely on microseismic monitoring systems to capture rock fracture signals and trigger graded responses by linking multiple parameters such as water pressure and displacement; and machine learning methods train intelligent models based on historical data. However, existing technologies have the following core shortcomings: 1) Reliance on a single data source leads to blurred fault boundary identification and inaccurate risk area location in rock strata analysis; 2) In hydrological analysis, the use of a static model for permeability coefficients fails to dynamically respond to construction disturbances or geological changes, resulting in errors in seepage direction; 3) Traditional linear weighted summation risk models ignore the chain-like triggering mechanism of geological, hydrodynamic, and construction disturbance factors, leading to missed risk reports; 4) Prevention and control measures lack spatial matching, making it impossible to dynamically allocate resources according to risk levels, resulting in resource waste or response delays. Therefore, researching and designing an intelligent early warning method and system for tunnel mudslides and water inrushes based on multimodal data that can overcome the above-mentioned shortcomings is an urgent problem we need to solve. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent early warning method and system for tunnel mudslides and water inrushes based on multimodal data. This method integrates geological and hydrological multimodal data, generates a three-dimensional geological distribution map using a geological analysis model, and generates a three-dimensional water flow distribution map using a seepage inversion model. These are then coupled into a comprehensive three-dimensional model through spatiotemporal registration. Furthermore, the method dynamically calculates the inrush risk value based on construction data, and generates spatially graded prevention and control instructions accordingly. This achieves globally accurate early warning, significantly reduces the false alarm rate, and improves the accuracy of seepage velocity and direction. It also enables the matching of prevention and control measures with risk locations, optimizes resource utilization, and improves model maintenance efficiency and response speed through a local update mechanism, thus ensuring construction safety.

[0004] The above-mentioned technical objective of the present invention is achieved through the following technical solution: Firstly, a smart early warning method for tunnel mudslides and water inrushes based on multimodal data is provided, including the following steps: Based on historical geological data and measured resistivity data, geological structure inversion and parameter calculation are performed through geological analysis models to obtain a three-dimensional geological distribution map containing rock layer distribution and permeability coefficient. Based on hydrological monitoring data, the hydrodynamic field is reconstructed using a seepage inversion model to obtain a three-dimensional water flow distribution map that includes the distribution of water pressure and seepage velocity. Based on the three-dimensional geological distribution map and the three-dimensional water flow distribution map, a comprehensive three-dimensional model is obtained by coupling the two through a time-space registration engine. Based on the comprehensive three-dimensional model and construction plan data, dynamic risk calculation is performed to obtain the surge risk value; Based on the surge risk value and the corresponding spatial coordinates, a tiered prevention and control instruction is generated through strategy mapping.

[0005] Furthermore, the geological analysis model includes: Data interpolation layer: Based on the historical geological data and the measured resistivity data, a three-dimensional resistivity distribution map covering the entire construction space area is generated using the GPU parallel RBF interpolation algorithm; Lithology identification layer: Based on the three-dimensional resistivity distribution map, complete rock strata, fault zones, weak interlayers and fractured layers are identified through a preset mapping rule library, and a three-dimensional rock strata distribution map is output; Permeability coefficient generation layer: Based on the three-dimensional rock layer distribution map, a corresponding permeability coefficient is generated for each spatial coordinate point through the lithology-permeability characteristic conversion rule; Boundary extraction layer: Based on the three-dimensional rock strata distribution map, the boundary of the fault zone is extracted using the Marching Cubes boundary extraction algorithm to obtain the triangular mesh surface of the fault zone; Parameter calculation layer: Based on the triangular mesh, the fault width and fault thickness are calculated, and the three-dimensional geological distribution map is output.

[0006] Furthermore, the lithology-permeability conversion rule includes: For the intact rock strata, the baseline permeability coefficient obtained from historical permeability tests is used as the permeability coefficient. For the weak interlayer, the permeability coefficient is the product of the reduction factor and the reference permeability coefficient; For the fractured layer, the permeability coefficient is the product of the increase factor and the reference permeability coefficient; For the fault zone, the permeability coefficient is the product of the dynamic correction coefficient and the baseline permeability coefficient.

[0007] Furthermore, the seepage inversion model includes: The data preprocessing layer, based on the water level, water pressure and flow data in the hydrological monitoring data, obtains preprocessed multi-source hydrological data through time series alignment and noise filtering; The inversion solution layer obtains the water potential gradient field by constructing a three-dimensional unsteady seepage field control equation and iteratively solving it using the finite element method, based on the preprocessed multi-source hydrological data and hydraulic laws. The vector generation layer, based on the water potential gradient field and the permeability coefficient, obtains the seepage velocity vector distribution by dynamically calculating the seepage velocity vector; The visualization output layer overlays the water pressure and the seepage velocity vector distribution onto a three-dimensional grid space to obtain the three-dimensional water flow distribution map.

[0008] Furthermore, the dynamic calculation process of the vector generation layer includes: First, based on the water potential gradient field, the initial value of the seepage velocity vector is calculated using Darcy's law; The permeability coefficient is then introduced to compensate for the differences in permeability of the rock strata in different directions, so as to correct the direction of the seepage velocity and obtain the vector distribution of the seepage velocity.

[0009] Furthermore, the process of constructing the integrated 3D model includes: The coordinate systems of the three-dimensional geological distribution map and the three-dimensional water flow distribution map are converted into a unified engineering coordinate system through an affine transformation matrix. For areas lacking geological and hydrological parameters, an adaptive Kriging interpolation algorithm is used to complete the spatial data. Based on the construction progress sequence, a sliding time window matching is performed on the geological change data and the seepage field update data, and the data is integrated into a unified whole through a voxel fusion engine to generate the comprehensive three-dimensional model.

[0010] Furthermore, the calculation process for the surge risk value includes: Based on the distance from the current construction point to the nearest fault and the thickness of the fault, the geological fault factor is obtained by multiplying the distance inverse relationship with the thickness saturation growth function. The hydrodynamic factor is generated by dividing the potential gradient modulus by a preset gradient saturation value and adding it to the corrected seepage velocity modulus divided by a preset velocity saturation value, and then processing it through a saturated compression function. The average of three factors—the ratio of the real-time thrust of the tunnel boring machine to its rated thrust, the ratio of the cutter torque to the material yield strength, and the ratio of the vibration acceleration to the critical value of rock instability—is used as the construction disturbance factor. The confidence correction coefficient is obtained by correcting the error through the error sensitivity coefficient based on the ratio of the absolute value of the measurement error of the positioning device to the reference positioning error. The geological fault factor, the hydrodynamic factor, the construction disturbance factor, and the confidence correction coefficient are multiplied consecutively to obtain the final surge risk value.

[0011] Furthermore, the tiered prevention and control instructions include: When the surge risk value exceeds the first threshold, the entire section is shut down, high-pressure grouting reinforcement is initiated, and the drainage system operates at maximum power. When the surge risk value is between the second threshold and the first threshold, the excavation advance rate is limited and local radial grouting is initiated; When the surge risk value is between the third threshold and the second threshold, increase the power of the drainage system; When the surge risk value is less than the third threshold, normal construction shall continue.

[0012] Furthermore, when a surge risk value is below a set threshold but a surge event occurs, a feedback correction process is executed: Collect real-time microseismic data, water pressure change data, and slag particle size distribution data around the event point; If the density of microseismic events exceeds the set threshold or the proportion of large-diameter slag flakes exceeds the standard, the lithology identification is determined to be incorrect and the fault zone resistivity threshold in the mapping rule library is adjusted proportionally. If the water pressure drops more than the set threshold or the proportion of small-diameter slag flakes does not exceed the predetermined threshold, the permeability coefficient is determined to be incorrect and the dynamic correction coefficient is increased. Meanwhile, a time decay factor is introduced into the calculation of the surge risk value to compensate for the time effect.

[0013] Secondly, a tunnel mudslide and water inrush intelligent early warning system based on multimodal data is provided. This system is used to implement the tunnel mudslide and water inrush intelligent early warning method based on multimodal data as described in any one of the first aspects, including: Geological distribution generation module: Based on historical geological data and measured resistivity data, the geological structure is inverted and parameters are calculated through a geological analysis model to obtain a three-dimensional geological distribution map containing rock layer distribution and permeability coefficient; Water flow distribution generation module: Based on hydrological monitoring data, the hydrodynamic field is reconstructed through a seepage inversion model to obtain a three-dimensional water flow distribution map that includes the distribution of water pressure and seepage velocity; Integrated model generation module: Based on the three-dimensional geological distribution map and the three-dimensional water flow distribution map, a time-space registration engine is used to couple the data to obtain an integrated three-dimensional model; Risk value calculation module: Based on the comprehensive three-dimensional model and construction plan data, dynamic risk calculation is performed to obtain the surge risk value; Prevention and control strategy generation module: Based on the surge risk value and the corresponding spatial coordinates, it generates hierarchical prevention and control instructions through strategy mapping.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention integrates geological and hydrological multimodal data, generates a three-dimensional geological distribution map using a geological analysis model, generates a three-dimensional water flow distribution map through a seepage inversion model, and couples these into a comprehensive three-dimensional model through spatiotemporal registration. Then, it dynamically calculates the inrush risk value based on construction data, and generates spatially graded prevention and control instructions accordingly. This achieves globally accurate early warning, significantly reduces the false alarm rate, and improves the accuracy of seepage velocity and direction. It also enables the matching of prevention and control measures with risk locations, optimizes resource utilization, and improves model maintenance efficiency and response speed through a local update mechanism, ensuring construction safety. 2. This invention employs multimodal data fusion and dynamic lithology identification technology. By integrating borehole and resistivity data through GPU parallel RBF interpolation algorithm, a continuous three-dimensional resistivity distribution map is generated. Combined with a preset mapping rule base, rock strata types are dynamically identified, which solves the problems of inaccurate static calculation of geological parameters and ambiguous fault boundaries, reduces the false alarm rate, and achieves accurate correlation of risk spatial coordinates. 3. This invention adopts dynamic correction technology for seepage vectors. By introducing anisotropic correction coefficients to correct the initial value of Darcy's law, and combining permeability coefficients to compensate for the directional seepage differences in rock strata, it solves the directional distortion defect of traditional hydrological models in fractured rock masses and achieves accurate spatial partitioning of seepage velocity vectors. 4. This invention integrates geological fault factors, hydrodynamic factors, construction disturbance factors, and confidence correction coefficients by reconstructing a product-type risk calculation formula. Through the product structure, it realistically simulates the coordinated process from geological weakening to water pressure drive and then to construction disturbance, accurately quantifies the disaster chain triggering mechanism, significantly reduces the false alarm rate in low-risk areas, and improves the accuracy of early warning. 5. This invention adopts a four-level threshold strategy mapping technology, which triggers instructions such as full-section shutdown, speed-limited excavation combined with local grouting, and drainage power increase by risk value partitioning. Combined with spatial coordinate positioning, it solves the defect of the decoupling between prevention and control measures and risk intensity, realizes targeted resource allocation, and reduces data waste. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1 This is a system flowchart from Embodiment 1 of the present invention; Figure 2 This is the three-dimensional geological distribution map in Embodiment 1 of the present invention; Figure 3 This is a three-dimensional water flow distribution diagram from Embodiment 1 of the present invention; Figure 4This is the integrated three-dimensional model diagram in Embodiment 1 of the present invention; Figure 5 This is a system block diagram in Embodiment 2 of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0017] Example 1: Intelligent early warning method for tunnel mudslide and water inrush based on multimodal data, such as Figure 1 As shown, it includes the following steps: S1: Based on historical geological data and measured resistivity data, geological structure inversion and parameter calculation are performed through geological analysis models to obtain a three-dimensional geological distribution map containing rock layer distribution and permeability coefficient; S2: Based on hydrological monitoring data, the hydrodynamic field is reconstructed through a seepage inversion model to obtain a three-dimensional water flow distribution map that includes the distribution of water pressure and seepage velocity; S3: Based on the three-dimensional geological distribution map and the three-dimensional water flow distribution map, a comprehensive three-dimensional model is obtained by coupling the time-space registration engine. S4: Based on the comprehensive three-dimensional model and construction plan data, dynamic risk calculation is performed to obtain the surge risk value; S5: Generate tiered prevention and control instructions based on the surge risk value and the corresponding spatial coordinates through strategy mapping.

[0018] In step S1, considering the shortcomings of existing technologies such as reliance on a single data source leading to blurred fault boundaries, static calculation of permeability coefficients, and inability to correlate risk spatial coordinates in two-dimensional planar analysis, this invention constructs a three-dimensional geological distribution map using borehole coordinates, measured resistivity values, and historical core permeability test data, such as... Figure 2 The diagram shows the spatial geological structure of a tunnel construction area. Each coordinate point contains two numbers: the first number represents the rock stratum type (e.g., 1 for intact rock strata, 2 for weak interlayers, 3 for fractured layers, and 4 for fault zones), and the second number represents the permeability coefficient at that point. The established 3D graphics provide a more intuitive view of the spatial variations in rock strata distribution, fault thickness, and permeability coefficient near the tunnel, enabling precise quantification of geological risks.

[0019] This geological analysis model includes: Data interpolation layer: Transforms discrete resistivity measurement points into a continuous three-dimensional resistivity distribution map. Considering that resistivity data is discrete and non-uniformly distributed, and that RBF (Radial Basis Function) interpolation is highly adaptable to non-uniform data, and that GPU parallel computing solves the efficiency bottleneck of large-scale grid generation, this invention employs a GPU-parallel RBF (Radial Basis Function) interpolation algorithm to transform discrete resistivity measurement data into a continuous three-dimensional resistivity map. The steps include: First, based on historical borehole coordinate information and corresponding measured resistivity data, a spatial function relationship is constructed using the radial basis function (RBF) interpolation algorithm based on these discrete data points to calculate the resistivity value of the unsampled area. The GPU parallel computing architecture accelerates the interpolation process and improves the speed of large-scale grid computing. Finally, a high-resolution three-dimensional resistivity distribution map covering the entire tunnel construction area is generated.

[0020] Lithology Identification Layer: This layer converts the three-dimensional resistivity distribution map into lithology classification results. Because resistivity is extremely sensitive to changes in rock porosity / water content and is more adaptable to the confined space of tunnels than seismic wave methods, this invention chooses the resistivity method for rock mass identification. The principle is to utilize the physical characteristic that different lithologies have significant resistivity differences, and to establish a correspondence between resistivity thresholds and lithology types through a pre-defined mapping rule base.

[0021] In some examples, the mapping rule between resistivity and rock mass is as follows: When resistivity is greater than 520 When the resistivity is between 70-200, it is determined to be an intact rock stratum. When the resistivity is between 200-400, it is determined to be a weak interlayer. When the resistivity is between 50-600, it is determined to be a fractured layer. And the resistivity gradient is greater than 180 At that time, it was determined to be a fault zone.

[0022] Permeability coefficient generation layer: This layer generates accurate permeability coefficient values ​​for each spatial point on the 3D rock strata distribution map. Considering the dynamic changes in rock mass permeability during tunnel mudslide and water inrush early warning and the inability of traditional static permeability coefficient models to respond to real-time disturbances, this invention constructs a permeability coefficient generation layer and achieves dynamic correction through lithology-permeability characteristic conversion rules.

[0023] The principle is to dynamically correct resistivity and microseismic data through a differentiated correction strategy, thereby solving the problem of overestimation of permeability coefficient in the fault zone by the static model.

[0024] Because the mineral particles in intact rock strata are densely cemented and have poor pore connectivity, groundwater can only slowly seep through micro-fractures. Therefore, historical core permeability test statistics are directly used. As a benchmark value for permeability coefficient.

[0025] For weak interlayers, which contain clay minerals such as montmorillonite, the permeability drops sharply when exposed to water as these minerals swell and block the pores. Therefore, the permeability coefficient is obtained by combining the reduction factor with the benchmark permeability coefficient. ;in, , This is the reduction factor. This is the baseline value for the permeability coefficient.

[0026] Considering that the dense fracture network in the fractured layer forms dominant seepage channels, and that permeability increases exponentially with a greater rate of decrease in resistivity and a higher degree of fracture development, an increase factor is used in conjunction with a permeability coefficient benchmark to obtain the permeability coefficient. ;in, , To increase the coefficient.

[0027] Because the fault zone contains both fault gouge and breccia, and the degree of fracturing reflects water conductivity, the proportion of fault gouge and the frequency of microseismic events activating fractures both lead to dynamic changes in permeability. Therefore, a dynamic correction coefficient is used to obtain the permeability coefficient. , , This is a dynamic correction coefficient.

[0028] In some examples, the reduction factor is related to the properties and resistivity of the rock strata themselves, and its calculation formula is as follows: ; in, The clay swelling sensitivity coefficient is calibrated in the laboratory. This is the resistivity correction factor fitted to the experiment, used to quantify the effect of resistivity differences on permeability; The content of clay minerals, For reference purposes, a smaller fixed value is used when the clay mineral content is low, which is obtained from historical geological data; The resistivity value at the current point. The standard resistivity of the intact rock strata.

[0029] In some examples, the amplification factor is related to the water conductivity of the rock strata and the fractures, and its calculation formula is as follows: ; in, The hydraulic conductivity coefficient is the one calibrated in the laboratory. The fracture development index is obtained from historical geological data.

[0030] In some examples, the dynamic correction factor is adjusted based on monitoring data and historical data, and its calculation formula is as follows: ; in, The fault activation coefficient obtained from the experiment; To obtain the fragmentation index in the experiment; The frequency of monitored microseismic activity (in times per second). For reference microseismic frequencies, historical average values ​​are generally used; The percentage of fault gouge can be obtained from historical geological data or surveys. For reference, the proportion of fault gouge, To prevent the parameter from being divided by zero, and to ensure the balance of the formula.

[0031] Boundary Extraction Layer: Considering that the accurate quantification of the geometric parameters of the fault zone in the early warning of tunnel mudslide and water inrush risk is crucial for risk calculation, and that traditional manual boundary delineation methods suffer from low efficiency and large subjective errors, this invention constructs a boundary extraction layer. The Marching Cubes algorithm is used to process the fault zones in the 3D rock strata distribution map into triangular mesh panels. The steps are as follows: First, input three-dimensional rock strata distribution data containing fault markers, set a resistivity gradient threshold as a mutation criterion, then execute the Marching Cubes algorithm to traverse the three-dimensional grid space, construct isosurfaces in the gradient exceeding the threshold region and generate triangular grid patches of fault zone boundaries, and finally output a fault geometric model that can quantify the dip angle and thickness, providing accurate spatial constraints for subsequent geological risk assessment.

[0032] Parameter calculation layer: This layer calculates fault dip angle and thickness in real time based on triangular mesh patches, and outputs a 3D geological distribution map including strata distribution and permeability coefficients. The implementation steps are as follows: The normal vector of the patch is calculated using the vector cross product of the coordinates of the vertices of adjacent triangular patches; the dip angle is inverted based on the angle between the normal vector and the horizontal plane. The formula is: ; in, The fault dip angle, Let be the normal vector of the triangular facet. Let be the z-component of the normal vector of the triangular facet.

[0033] Then measure the fault width along the cross-section. Substitute into the thickness formula: ; Obtain the fault center coordinates and dip angle and thickness A structured parameter set is used. These parameter sets are embedded in the 3D geological distribution map, and the final output is a 3D geological distribution map containing rock layer distribution and permeability coefficient.

[0034] This invention integrates historical geological data with measured resistivity data, uses geological analysis models to perform geological structure inversion and dynamic parameter calculation, and generates a three-dimensional geological distribution map containing rock layer distribution and permeability coefficient. This achieves the effect of accurately quantifying geological risks and breaking through the limitations of single data, providing a geological basis for subsequent mudslide and water inrush prediction.

[0035] In step S2, the hydrodynamic field is reconstructed based on hydrological monitoring data using a seepage inversion model.

[0036] Considering that traditional methods for early warning of mudslides and water inrushes in tunnels rely on a single hydrological index, resulting in a high rate of misjudgment of seepage direction and an inability to quantify the spatial hydrodynamic field, this invention constructs a three-dimensional water flow distribution map. By integrating real-time data on water level, water pressure, and flow rate, and combining the rock stratum permeability coefficient to dynamically correct the seepage vector direction, the hydrodynamic field can be accurately reconstructed.

[0037] like Figure 3 As shown, the spatial distribution of the hydrodynamic field in a tunnel construction area is displayed. Each coordinate point contains two numbers: the first number represents the water pressure value, and the second number represents the seepage velocity value. This accurately characterizes the groundwater hydrodynamic state, facilitating the identification of areas with high water pressure or high-velocity seepage, reducing misjudgments of water flow, and thus improving the reliability of mudslide and water inrush warnings.

[0038] The seepage inversion model of this invention includes a data preprocessing layer, an inversion solution layer, a vector generation layer, and a visualization output layer.

[0039] The data preprocessing layer obtains time-synchronized and noise-removed multi-source hydrological data, resolving the issues of spatiotemporal asynchrony and noise interference in multi-source hydrological data. Because the sampling frequencies of tunnel hydrological monitoring equipment vary significantly and construction vibrations generate high-frequency noise, the data preprocessing layer of this invention uses cubic spline interpolation to increase the low-frequency sampling frequency, then employs wavelet threshold denoising to remove mechanical vibration noise, retaining the effective hydrological signal, and outputting a time-synchronized and noise-removed multi-source hydrological data matrix.

[0040] Inversion Solution Layer: The preprocessed hydrological data is transformed into a hydropotential gradient field. Considering that traditional hydrological models for tunnel mudslide and water inrush early warning neglect the anisotropy of rock strata permeability and cannot represent the dynamic changes of unsteady seepage, this invention constructs an inversion solution layer. Its technical principle is based on constructing a three-dimensional unsteady seepage control equation according to the fundamental laws of seepage mechanics and using the finite element method for iterative solution. The hydropotential value and its gradient distribution at each node are calculated through spatial discretization and matrix operations. The hydropotential value represents the total mechanical energy per unit weight of water.

[0041] In some examples, the formula for the governing equation of three-dimensional unsteady seepage is: ; in, This represents the water potential value at a spatial location. for Permeability coefficient in direction, for Permeability coefficient in direction, for Permeability coefficient in direction, The water storage capacity of the rock strata. The time variable is used. The permeability coefficients in each direction are obtained through a permeability coefficient generation layer.

[0042] Then, the construction area was discretized into grid cells using the finite element method. Boundary conditions and initial values ​​were defined based on the preprocessed hydrological data. Finally, the water potential value of each node was obtained through iterative solution. Then calculate the water potential gradient. .

[0043] Vector Generation Layer: This layer transforms the hydraulic pressure gradient field into a seepage velocity vector with built-in direction and modulus. To address the distortion of seepage direction in fractured rock masses caused by traditional Darcy's law, this invention's vector generation layer outputs a precise seepage velocity vector with direction and modulus through a dynamic correction mechanism, supporting the calculation of hydrodynamic factors and the location of sudden surge paths.

[0044] Its core technology is to solve the problem of directional distortion of the traditional Darcy's law in fractured rock masses through anisotropic correction. The steps are as follows: First, the initial seepage velocity is calculated based on Darcy's law. Then, the water potential gradient at the spatial point is calculated based on real-time monitoring data and the vertical unit vector of the rock layer is obtained. Next, the correction formula is applied point by point in three-dimensional space to calculate the compensated seepage velocity vector. Finally, the vectors of each spatial point are integrated to generate a precise seepage velocity vector field for each zone.

[0045] In some examples, the correction formula for the seepage velocity vector is: ; in, This is the corrected seepage velocity vector; For reference velocity, the average seepage velocity or a typical value can be used; The permeability coefficient at a spatial point; The anisotropy correction coefficients were determined experimentally. The water potential gradient represents the rate of change of water potential in space. This is a unit vector in the vertical direction of the rock strata.

[0046] The visualization output layer overlays the vector distributions of water pressure and seepage velocity to generate a three-dimensional water flow distribution map.

[0047] In some examples, to enhance the visual appeal of the graphics, a 3D water flow distribution map with directional markers can be generated: first, the vector field is superimposed onto a 3D geological grid, then directional arrows are added to grid points with higher seepage velocities, and a rotatable and scalable 3D water flow distribution map is output for risk point location analysis.

[0048] Based on hydrological monitoring data, this invention reconstructs the hydrodynamic field through a seepage inversion model, generating a three-dimensional water flow distribution map that includes water pressure and seepage velocity. This achieves the effect of accurately characterizing the groundwater dynamic state and improving the accuracy of seepage direction prediction, providing hydrological basis for risk quantification.

[0049] In step S3, considering that the spatial reference difference between the geological model and the water flow model in the tunnel mudslide and water inrush early warning system can lead to deviations in risk point location, and that traditional methods ignore the asynchronous nature of construction progress and data updates, this invention constructs a comprehensive three-dimensional model based on a three-dimensional geological distribution map and a three-dimensional water flow distribution map, and performs coupling processing through a time-space registration engine. For example... Figure 4 As shown, a three-dimensional model that integrates multiple parameters such as rock strata distribution, permeability coefficient, water pressure gradient and seepage velocity is formed to intuitively demonstrate the interaction between the geological structure and water flow field at the location of the tunnel.

[0050] Its design principle lies in constructing a unified spatiotemporal representation framework for geological and hydrological data. It resolves spatial benchmark differences through coordinate system transformation, completes missing data areas through adaptive interpolation, and ensures dynamic consistency between construction progress and model updates through a sliding time window, ultimately generating a spatiotemporally aligned integrated 3D model. The specific layered implementation is as follows: Coordinate system transformation layer: unifying the spatial reference between the geological model (local coordinate system) and the flow model (geocentric coordinate system). To solve the problem of misaligned control commands caused by the difference in spatial references between the geological model and the flow model, this invention maps the local coordinates to geocentric coordinates through an affine transformation matrix, eliminating positioning errors.

[0051] Data completion layer: High-precision interpolation is performed on missing areas of geological permeability coefficient and hydraulic pressure gradient field. Because geological and hydrodynamic parameters are missing at fault edges, this invention ensures model continuity through spatial interpolation. The specific process is as follows: Spatial topological relationships are constructed; an adaptive Kriging interpolation algorithm is adopted to calculate spatial correlation weights through a variogram; and interpolation parameters are adaptively adjusted based on semivariogram analysis to ensure data continuity.

[0052] Time-matching layer: To address the time-asynchrony between geological change data and seepage field update data, this invention ensures the real-time performance of the model through a dynamic time window. Based on the construction progress timeline, a dynamic time window is set; geological change data and seepage field update data are matched within the window; when the data timestamp deviation exceeds a threshold, re-registration is triggered. Finally, the data is integrated using a voxel fusion engine to generate a comprehensive 3D model.

[0053] This invention couples geological and hydrological models through a time-space registration engine to achieve coordinate system transformation, data completion, and timeliness matching, generating a spatiotemporally unified three-dimensional integrated model. This achieves the effects of eliminating positioning errors, ensuring data timeliness consistency, and building a foundation for global risk assessment.

[0054] In step S4, considering that the chain-triggered nature of tunnel mudslides and water inrushes requires the simultaneous fulfillment of three critical conditions: weakened geological structure, hydrodynamic drive, and construction disturbance triggering, and that traditional linear weighted models cannot characterize the multiplier effect between factors, such as construction vibrations near faults that may cause a sudden increase in permeability, this invention constructs a multi-factor coupled quantitative risk assessment model. By integrating geological structural features, hydrodynamic conditions, and construction disturbance effects, and combining a positioning accuracy correction mechanism, the spatial dynamic calculation of risk values ​​is achieved.

[0055] The construction plan data includes: real-time thrust value of the tunnel boring machine, cutter torque monitoring value, vibration acceleration sensor data, tunneling speed setting value, and equipment positioning coordinates; these parameters are updated in real time at a frequency of 10Hz and are connected to the risk calculation engine through the industrial Internet of Things platform.

[0056] Furthermore, considering that existing methods ignore false alarms caused by monitoring errors and dynamic weight changes between factors, a confidence correction coefficient is then used. To compensate for monitoring errors, since a surge requires the simultaneous fulfillment of three critical conditions, a product-based approach better aligns with the chain reaction mechanism of disasters. Therefore, a product-based approach rather than a linear superposition is used to avoid false alarms in low-risk areas using the traditional weighted summation method. The formula for calculating the surge risk value is: ;in, For surge risk value, Geological fault factors, As a hydrodynamic factor, Construction disturbance factor, This is the confidence correction coefficient.

[0057] In some examples, the formula for calculating the geological fault factor is: ; in, The distance from the current point to the nearest fault; For reference distance, historical average distance or typical value can be used; The thickness of the most recent fault; The preset maximum effective thickness; To prevent the removal of the zero constant.

[0058] In some examples, the hydrodynamic factor is calculated using the following formula: ; in, This represents the water potential gradient value. This represents the saturation value of the water potential gradient. This is the corrected seepage velocity value. This represents the saturation value of the seepage velocity.

[0059] In some examples, the formula for calculating the construction disturbance factor is: ; in, For the current tunnel boring machine thrust, The rated thrust of the tunnel boring machine, The current tool torque, The yield strength of the tool material. The current machine vibration acceleration, This refers to the acceleration due to rock mass instability.

[0060] In some examples, the confidence correction coefficient is calculated using the following formula: ; in, This is the error sensitivity coefficient; This is the positioning error; For reference positioning error, the equipment accuracy standard or historical average error is generally used.

[0061] Based on a comprehensive three-dimensional model and construction data, this invention calculates the surge risk value using a product-type risk formula, thereby quantifying the chain disaster triggering mechanism, reducing the false alarm rate, and achieving minute-level risk warning.

[0062] In step S5, to address the resource waste and delayed response issues caused by the extensive nature of traditional prevention and control measures, this invention generates tiered prevention and control instructions based on the surge risk value and its corresponding spatial coordinates through strategy mapping, thereby achieving dynamic matching of prevention and control measures with risk intensity and location.

[0063] Its design principle is based on the chain propagation characteristics of sudden surge disasters, constructing an intelligent instruction library of risk value, spatial location and prevention and control actions, and realizing dynamic matching of prevention and control instructions and risk intensity through a four-level threshold division and coordinate-driven precise response mechanism.

[0064] In some examples, the hierarchical instruction is: Level 1 prevention and control: When the risk value of sudden surge is greater than 0.8, the entire section is shut down, high-pressure grouting reinforcement is started, the pressure is ≥25MPa, and the drainage system power is 100%.

[0065] Level 2 control: When the risk value of the sudden surge is between 0.6 and 0.8, the excavation advance rate is limited to less than 30% of the rated value, and local radial grouting is initiated near the risk point.

[0066] Level 3 prevention and control: When the surge risk value is between 0.4 and 0.6, increase the power of the drainage system.

[0067] Normal construction: When the risk value of sudden surge is less than 0.4, normal construction shall be maintained.

[0068] Considering the prediction bias of the mudslide and water inrush early warning model under complex geological conditions and the inability of traditional static models to adaptively correct, this invention constructs a feedback mechanism. Furthermore, considering the causal correlation between multi-source characteristic data of mudslide events and geological and hydrological parameters, this invention chooses to construct the feedback mechanism based on geological and hydrological parameters.

[0069] In some examples, when a surge event occurs despite a surge risk value below 0.4, a feedback correction process is executed: Data acquisition: Data is collected within a 10m radius sphere centered on the event point, including the density and energy of microseismic events 5 minutes before the surge, water pressure change curves, and particle size distribution of the ejected material.

[0070] Fault diagnosis: If the density of microseismic events is greater than 20 per cubic meter or the proportion of coarse particles larger than 50 mm in the slag exceeds 30%, it means that the actual degree of rock mass fragmentation is far greater than the model prediction. It may misclassify faults or fragmented rock masses as intact rock strata, which is judged as lithology identification error. The fault zone resistivity threshold in the mapping rule base is dynamically adjusted to 70%~80% of the original value. If the water pressure drop exceeds 1.0 MPa or the proportion of fine particles smaller than 5 mm in the slag does not exceed 30%, it indicates that the actual permeability coefficient is significantly greater than the model prediction. This is deemed an error in the permeability coefficient calculation, and the dynamic correction coefficient is increased. In some examples, the update formula for the dynamic correction coefficient is: ; in, The updated dynamic correction coefficients; This is the original dynamic correction coefficient; This represents the drop in water pressure. For reference water pressure changes; The percentage of fine particles in the effluent material.

[0071] Time-related compensation: A time decay factor is introduced into the calculation of surge risk value to compensate for the impact of time-related factors. In some examples, the formula for calculating the time decay factor is: ; in, The time decay factor, The data collection time interval This represents the decay rate.

[0072] This invention generates graded prevention and control instructions based on surge risk values ​​and spatial coordinates through a four-level threshold strategy, achieving dynamic matching between prevention and control measures and risk locations, thereby improving the efficiency of accident handling. At the same time, it continuously optimizes the model accuracy through a feedback correction mechanism.

[0073] Example 2: Intelligent Early Warning System for Tunnel Mud Inrush and Water Inrush Based on Multimodal Data. This system is used to implement the intelligent early warning method for tunnel mud inrush and water inrush based on multimodal data as described in Example 1. Figure 5 As shown, it includes a geological distribution generation module, a water flow distribution generation module, a comprehensive model generation module, a risk value calculation module, and a prevention and control strategy generation module.

[0074] The system comprises the following modules: a geological distribution generation module, a water flow distribution generation module, and a comprehensive 3D geological distribution map. The geological distribution generation module is used to perform geological structure inversion and parameter calculation based on historical geological data and measured resistivity data using a geological analysis model. The water flow distribution generation module is used to reconstruct the hydrodynamic field based on hydrological monitoring data using a seepage inversion model. The comprehensive model generation module is used to couple the 3D geological distribution map and the 3D water flow distribution map using a time-space registration engine to obtain a comprehensive 3D model. The risk value calculation module is used to perform dynamic risk calculation based on the comprehensive 3D model and construction plan data to obtain the surge risk value. The prevention and control strategy generation module is used to generate graded prevention and control instructions based on the surge risk value and the corresponding spatial coordinates through strategy mapping.

[0075] Working principle: This method achieves accurate early warning and graded prevention and control of tunnel inrush risk through multimodal data fusion and dynamic model updating. First, the collaborative processing of geological and hydrological data generates a high-precision three-dimensional model; second, the real-time integration of construction disturbance factors enhances the model's dynamism; finally, a spatialized risk mapping mechanism ensures that prevention and control instructions match the risk location, significantly improving construction safety and resource utilization efficiency.

[0076] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0077] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0078] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0079] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0080] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent early warning of tunnel mudslides and water inrushes based on multimodal data, characterized in that, Includes the following steps: Based on historical geological data and measured resistivity data, geological structure inversion and parameter calculation are performed through geological analysis models to obtain a three-dimensional geological distribution map containing rock layer distribution and permeability coefficient. Based on hydrological monitoring data, the hydrodynamic field is reconstructed using a seepage inversion model to obtain a three-dimensional water flow distribution map that includes the distribution of water pressure and seepage velocity. Based on the three-dimensional geological distribution map and the three-dimensional water flow distribution map, a comprehensive three-dimensional model is obtained by coupling the two through a time-space registration engine. Based on the comprehensive three-dimensional model and construction plan data, dynamic risk calculation is performed to obtain the surge risk value; Based on the surge risk value and the corresponding spatial coordinates, a tiered prevention and control instruction is generated through strategy mapping.

2. The intelligent early warning method for tunnel mudslides and water inrushes based on multimodal data according to claim 1, characterized in that, The geological analysis model includes: Data interpolation layer: Based on the historical geological data and the measured resistivity data, a three-dimensional resistivity distribution map covering the entire construction space area is generated using the GPU parallel RBF interpolation algorithm; Lithology identification layer: Based on the three-dimensional resistivity distribution map, complete rock strata, fault zones, weak interlayers and fractured layers are identified through a preset mapping rule library, and a three-dimensional rock strata distribution map is output; Permeability coefficient generation layer: Based on the three-dimensional rock layer distribution map, a corresponding permeability coefficient is generated for each spatial coordinate point through the lithology-permeability characteristic conversion rule; Boundary extraction layer: Based on the three-dimensional rock strata distribution map, the boundary of the fault zone is extracted using the Marching Cubes boundary extraction algorithm to obtain the triangular mesh surface of the fault zone; Parameter calculation layer: Based on the triangular mesh, the fault width and fault thickness are calculated, and the three-dimensional geological distribution map is output.

3. The intelligent early warning method for tunnel mudslides and water inrushes based on multimodal data according to claim 2, characterized in that, The lithology-permeability conversion rules include: For the intact rock strata, the baseline permeability coefficient obtained from historical permeability tests is used as the permeability coefficient. For the weak interlayer, the permeability coefficient is the product of the reduction factor and the reference permeability coefficient; For the fractured layer, the permeability coefficient is the product of the increase factor and the reference permeability coefficient; For the fault zone, the permeability coefficient is the product of the dynamic correction coefficient and the baseline permeability coefficient.

4. The intelligent early warning method for tunnel mudslides and water inrushes based on multimodal data according to claim 1, characterized in that, The seepage inversion model includes: The data preprocessing layer, based on the water level, water pressure and flow data in the hydrological monitoring data, obtains preprocessed multi-source hydrological data through time series alignment and noise filtering; The inversion solution layer obtains the water potential gradient field by constructing a three-dimensional unsteady seepage field control equation and iteratively solving it using the finite element method, based on the preprocessed multi-source hydrological data and hydraulic laws. The vector generation layer, based on the water potential gradient field and the permeability coefficient, obtains the seepage velocity vector distribution by dynamically calculating the seepage velocity vector; The visualization output layer overlays the water pressure and the seepage velocity vector distribution onto a three-dimensional grid space to obtain the three-dimensional water flow distribution map.

5. The intelligent early warning method for tunnel mudslides and water inrushes based on multimodal data according to claim 4, characterized in that, The dynamic calculation process of the vector generation layer includes: First, based on the water potential gradient field, the initial value of the seepage velocity vector is calculated using Darcy's law; The permeability coefficient is then introduced to compensate for the differences in permeability of the rock strata in different directions, so as to correct the direction of the seepage velocity and obtain the vector distribution of the seepage velocity.

6. The intelligent early warning method for tunnel mudslides and water inrushes based on multimodal data according to claim 1, characterized in that, The process of constructing the comprehensive 3D model includes: The coordinate systems of the three-dimensional geological distribution map and the three-dimensional water flow distribution map are converted into a unified engineering coordinate system through an affine transformation matrix. For areas lacking geological and hydrological parameters, an adaptive Kriging interpolation algorithm is used to complete the spatial data. Based on the construction progress sequence, a sliding time window matching is performed on the geological change data and the seepage field update data, and the data is integrated into a unified whole through a voxel fusion engine to generate the comprehensive three-dimensional model.

7. The intelligent early warning method for tunnel mudslides and water inrushes based on multimodal data according to claim 1, characterized in that, The calculation process for the surge risk value includes: Based on the distance from the current construction point to the nearest fault and the thickness of the fault, the geological fault factor is obtained by multiplying the distance inverse relationship with the thickness saturation growth function. The hydrodynamic factor is generated by dividing the potential gradient modulus by a preset gradient saturation value and adding it to the corrected seepage velocity modulus divided by a preset velocity saturation value, and then processing it through a saturated compression function. The average of three factors—the ratio of the real-time thrust of the tunnel boring machine to its rated thrust, the ratio of the cutter torque to the material yield strength, and the ratio of the vibration acceleration to the critical value of rock instability—is used as the construction disturbance factor. The confidence correction coefficient is obtained by correcting the error through the error sensitivity coefficient based on the ratio of the absolute value of the measurement error of the positioning device to the reference positioning error. The geological fault factor, the hydrodynamic factor, the construction disturbance factor, and the confidence correction coefficient are multiplied consecutively to obtain the final surge risk value.

8. The intelligent early warning method for tunnel mudslides and water inrushes based on multimodal data according to claim 1, characterized in that, The tiered prevention and control instructions include: When the surge risk value exceeds the first threshold, the entire section is shut down, high-pressure grouting reinforcement is initiated, and the drainage system operates at maximum power. When the surge risk value is between the second threshold and the first threshold, the excavation advance rate is limited and local radial grouting is initiated; When the surge risk value is between the third threshold and the second threshold, increase the power of the drainage system; When the surge risk value is less than the third threshold, normal construction shall continue.

9. The intelligent early warning method for tunnel mudslides and water inrushes based on multimodal data according to claim 1, characterized in that, When a surge risk value is below a set threshold but a surge event occurs, a feedback correction process is executed: Collect real-time microseismic data, water pressure change data, and slag particle size distribution data around the event point; If the density of microseismic events exceeds the set threshold or the proportion of large-diameter slag flakes exceeds the standard, the lithology identification is determined to be incorrect and the fault zone resistivity threshold in the mapping rule library is adjusted proportionally. If the water pressure drops more than the set threshold or the proportion of small-diameter slag flakes does not exceed the predetermined threshold, the permeability coefficient is determined to be incorrect and the dynamic correction coefficient is increased. Meanwhile, a time decay factor is introduced into the calculation of the surge risk value to compensate for the time effect.

10. A tunnel mudslide and water inrush intelligent early warning system based on multimodal data, characterized in that, The system is used to implement the intelligent early warning method for tunnel mudslides and water inrushes based on multimodal data as described in any one of claims 1-9, including: Geological distribution generation module: Based on historical geological data and measured resistivity data, the geological structure is inverted and parameters are calculated through a geological analysis model to obtain a three-dimensional geological distribution map containing rock layer distribution and permeability coefficient; Water flow distribution generation module: Based on hydrological monitoring data, the hydrodynamic field is reconstructed through a seepage inversion model to obtain a three-dimensional water flow distribution map that includes the distribution of water pressure and seepage velocity; Integrated model generation module: Based on the three-dimensional geological distribution map and the three-dimensional water flow distribution map, a time-space registration engine is used to couple the data to obtain an integrated three-dimensional model; Risk value calculation module: Based on the comprehensive three-dimensional model and construction plan data, dynamic risk calculation is performed to obtain the surge risk value; Prevention and control strategy generation module: Based on the surge risk value and the corresponding spatial coordinates, it generates hierarchical prevention and control instructions through strategy mapping.

Images