Complex terrain karst tunnel disaster identification method and system based on advanced drilling

Abstract

According to the advanced drilling-based complex terrain karst tunnel disaster identification method and system provided by the invention, the target karst tunnel risk level is determined by integrating geological conditions, hydrological conditions and structural mechanical response, and then the target karst tunnel advanced prediction detection scheme is determined, so that the target karst tunnel disaster identification method and system can be realized according to the target karst tunnel risk condition. An advanced forecasting detection scheme is reasonably selected; according to the method, dynamic weights of different advanced prediction and detection methods are determined based on matching degrees of different advanced prediction and detection methods for target karst tunnel types, historical accuracy rates and quality of target karst tunnel related data currently acquired by the advanced prediction and detection methods, so that the more accurate advanced prediction and detection methods can be effectively utilized; and the subsequent identification accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field related to tunnel identification, and in particular relates to a method and system for identifying karst tunnel disasters in complex terrain based on advance drilling. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Disaster prevention and control in tunnels in complex karst terrain has always been a major engineering challenge. The complex geological characteristics of karst have posed significant challenges to tunnel construction. The diverse nature of karst development, including gullies, rock buds, stone forests, karst depressions, sinkholes, caves, karst fissures, underground flow outlets, and underground river pipes, makes tunnel disaster prevention and control difficult. The difficulties lie in: (1) Accurate identification of karst disaster types.

[0004] (2) Form an effective and feasible karst disaster management plan based on the identification results. Traditional tunnel karst disaster management methods often have difficulty in accurately detecting and identifying karst disasters, and thus cannot accurately depict the scope and morphology of karst disasters. Secondly, after obtaining the geological information ahead, there is a lack of a method to analyze and judge different karst disaster risks, making it difficult to form a targeted treatment plan.

[0005] Therefore, how to accurately identify the types of karst disasters and form targeted disposal plans based on the identification results are problems that need to be solved at present. Summary of the Invention

[0006] To overcome the above-mentioned deficiencies of the prior art, the present invention provides a method and system for identifying karst tunnel disasters in complex terrain based on advance drilling, which can effectively utilize more accurate advance prediction detection methods to improve the accuracy of subsequent identification.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for identifying karst tunnel hazards in complex terrain based on advance drilling, comprising: Determining the risk level of the target karst tunnel based on the geological conditions, hydrological conditions, and structural mechanical response of the target karst tunnel, and determining an advanced prediction and detection scheme for the target karst tunnel based on the risk level of the target karst tunnel; the advanced prediction and detection scheme for the target karst tunnel includes at least one advanced prediction and detection method; Determine the dynamic weights of different advanced prediction detection methods based on their matching degree to the target karst tunnel type, their historical accuracy, and the quality of the target karst tunnel-related data currently collected by the advanced prediction detection methods; Based on the data collected from the target karst tunnel using the advance prediction detection method in the target karst tunnel advance prediction detection scheme, the corresponding target karst tunnel hazard identification results are obtained using a deep learning model; The different target karst tunnel hazard identification results are fused with the corresponding dynamic weights to obtain the final target karst tunnel hazard identification result.

[0008] In a second aspect, the present invention provides a method for identifying karst tunnel hazards in complex terrain based on advance drilling, comprising: a detection scheme determination module configured to: determine a risk level of a target karst tunnel based on the geological conditions, hydrological conditions, and structural mechanical response of the target karst tunnel; and determine an advanced prediction detection scheme for the target karst tunnel based on the risk level of the target karst tunnel; the advanced prediction detection scheme for the target karst tunnel including at least one advanced prediction detection method; a weight determination module configured to determine dynamic weights of different advanced prediction detection methods based on the matching degree of different advanced prediction detection methods to the target karst tunnel type, historical accuracy, and quality of target karst tunnel related data currently collected by the advanced prediction detection methods; The initial identification module is configured to: obtain corresponding target karst tunnel hazard identification results using a deep learning model based on the data collected from the target karst tunnel using the advance prediction detection method in the target karst tunnel advance prediction detection scheme; The final identification module is configured to: fuse different target karst tunnel disaster identification results in combination with corresponding dynamic weights to obtain the final target karst tunnel disaster identification result.

[0009] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0010] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0011] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which implements the method described in the first aspect when executed by a processor.

[0012] One or more of the above technical solutions have the following beneficial effects: In the present invention, the risk level of the target karst tunnel is determined by comprehensively considering the geological conditions, hydrological conditions and structural mechanical responses, and then the advance prediction detection scheme of the target karst tunnel is determined, which can reasonably select the advance prediction detection scheme according to the risk situation of the target karst tunnel; based on the matching degree of different advance prediction detection methods to the target karst tunnel type, the historical accuracy rate and the quality of the target karst tunnel related data currently collected by the advance prediction detection method, the dynamic weights of different advance prediction detection methods are determined, which can effectively utilize more accurate advance prediction detection methods and improve the accuracy of subsequent identification.

[0013] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0015] Figure 1 This is a flow chart of a method for identifying karst tunnel disasters in complex terrain based on advance drilling in Example 1 of the present invention. DETAILED DESCRIPTION

[0016] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0017] It should be noted that the terms used herein are for describing particular embodiments only and are not intended to limit the exemplary embodiments according to the present invention.

[0018] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0019] Example 1 This embodiment discloses a method for identifying karst tunnel hazards in complex terrain based on advance drilling, comprising: Determining the risk level of the target karst tunnel based on the geological conditions, hydrological conditions, and structural mechanical response of the target karst tunnel, and determining an advanced prediction and detection scheme for the target karst tunnel based on the risk level of the target karst tunnel; the advanced prediction and detection scheme for the target karst tunnel includes at least one advanced prediction and detection method; Determine the dynamic weights of different advanced prediction detection methods based on their matching degree to the target karst tunnel type, their historical accuracy, and the quality of the target karst tunnel-related data currently collected by the advanced prediction detection methods; Based on the data collected from the target karst tunnel using the advance prediction detection method in the target karst tunnel advance prediction detection scheme, the corresponding target karst tunnel hazard identification results are obtained using a deep learning model; The different target karst tunnel disaster identification results are fused with the corresponding dynamic weights to obtain the final target karst tunnel disaster identification result and generate a disaster disposal measure plan.

[0020] The following is a detailed description of a method for identifying karst tunnel disasters in complex terrain based on advance drilling proposed in this embodiment: Step 1: Determine the risk level of the target karst tunnel based on the geological conditions, hydrological conditions, and structural mechanical response of the target karst tunnel, and determine an advanced prediction detection plan for the target karst tunnel based on the risk level of the target karst tunnel; the advanced prediction detection plan for the target karst tunnel includes at least one advanced prediction detection method.

[0021] Adaptive detection method selection and survey line planning system are used to collect electrical geological information and layered geological information ahead of the tunnel.

[0022] Adaptive detection method, which uses a weighted scoring mechanism to select an advanced prediction detection scheme based on the karst risk level:

[0023] in, Represents geological conditions, Represents the hydrological conditions, represents the structural response, 、 are the coefficients of each term respectively.

[0024] In this embodiment, 、 The values ​​are 0.5, 0.3 and 0.2 respectively.

[0025] when ≥0.8, defined as high risk, using direct current + transient electromagnetic + ground penetrating radar detection + advance drilling; 0.8> ≥0.5, defined as medium risk, and detection using TEM+GPR+advance drilling combination; 0.5> , defined as low risk, using GPR rapid scanning + advance drilling.

[0026] This embodiment selects different detection schemes according to risk levels. When the risk is relatively low, detection can be performed in a rapid scanning manner to reduce costs.

[0027] For geological conditions, the parameters are as follows:

[0028] in, Representative lithology score: obtained based on geological survey reports and drill core data, such as limestone / dolomite = 1.0; sandstone = 0.3; shale = 0.1; Representative karst density: Based on historical exploration data and regional geological map data, the number of caves / 100m: 0-1 → 0.2, 2-5 → 0.6, >5 → 1.0; Representative weathering degree: determined based on on-site appraisal and point load tests, unweathered: 0; slightly weathered: 0.2; strongly weathered: 0.8; d, e, and f are the coefficients of each item respectively.

[0029] In this embodiment, the values ​​of d, e, and f are 0.4, 0.3, and 0.3, respectively.

[0030] For hydrological conditions, the parameters are as follows:

[0031] in, Represents groundwater pressure (MPa), determined based on data obtained from hydrological monitoring wells, such as <0.1MPa→0.1; 0.1-0.3MPa→0.5; >0.3MPa→1.0; Represents the water output per unit time of the drilling hole, determined according to the advanced drilling data, such as <5L / min→0.2; 5-10L / min→0.6; >10L / min→1.0; Represents groundwater corrosiveness (SO4 2- Concentration) is determined based on laboratory analysis data of water samples, e.g. <100 mg / L → 0.1; 100-500 mg / L → 0.5; >500 mg / L → 1.0. g, h, and k are the coefficients of each term.

[0032] In this embodiment, the values ​​of g, h, and k are 0.5, 0.3, and 0.2, respectively.

[0033] For structural response, the parameters are as follows:

[0034] Represents the deformation rate, i.e. the surrounding rock convergence rate (mm / d), determined according to engineering measurements: <0.1mm / d→0.1; 0.1-0.5mm / d→0.5; >0.5mm / d→1.0; Represents the support stress, i.e., initial support stress (MPa). <30% of design value → 0.1; 30-70% → 0.5; >70% → 1.0. m and n are the coefficients of each term.

[0035] In this embodiment, the values ​​of m and n are 0.6 and 0.4 respectively.

[0036] The DC detection method uses the cross-hole resistivity CT method, which includes electrodes, cables, current generators, potential measuring instruments, and data acquisition systems. A three-hole triangular array is arranged on the tunnel face, with a hole spacing of 2m and a hole depth of 30m. 20 electrodes are deployed in each hole with a spacing of 1.5m. A rotating excitation mode is used, with one hole selected as the current injection hole each time and the other two holes as potential measurement holes, for a total of six data acquisition modes. The traditional constant current of 100mA causes severe polarization. An adaptive power supply system is used to adjust the injection current in real time according to the resistivity of the rock formation, which can reduce the data distortion rate in the water-rich area by 40%. Specifically,

[0037] The transient electromagnetic method uses a multi-turn square coil with 14 turns, 2 meters on a side, and 5 milliwatts of inductance, fixed to the tunnel face. It supports 10-100 Hz frequency switching and receives response signals in the borehole. The detection depth is increased from 40 meters to 60 meters at the same power level.

[0038] The transient electromagnetic data acquisition module places a receiving coil in the advance borehole and uses a three-component magnetic probe with a sensitivity of 0.1nT and a bandwidth of 1Hz-10kHz. Data is transmitted in real time via optical fiber. The in-hole magnetic probe is isolated from the influence of metal noise in the tunnel, improving the signal-to-noise ratio: the SNR of a traditional integrated coil at a depth of 30m is 8dB, while the SNR of a split coil is 22dB.

[0039] Ground-penetrating radar uses a transmitting antenna to emit high-frequency electromagnetic waves and a receiving antenna to detect the electromagnetic characteristics of the underground medium. The transmitting antenna, positioned on the tunnel face, must be capable of emitting electromagnetic fields forward at a frequency range of 50 MHz to 200 MHz. Detection is performed using a spiral scanning path, with variable-density spiral scanning performed across the tunnel section. Low-resolution, rapid coverage is performed at the edges, with line spacing of 0.5 m; high-resolution, detailed scanning is performed in the center, with line spacing of 0.2 m. The scanning speed is dynamically adjusted based on signal quality, ranging from 0.1 m to 0.5 m.

[0040] The ground-penetrating radar's acquisition module utilizes an orthogonal dipole antenna array with 16 elements each for H and V polarization. The center frequency is adjustable from 50 to 200 MHz, with a dynamic range of >120 dB. For depths greater than 20 m, 50 MHz is used; for depths between 10 and 20 m, 100 MHz is used; and for depths less than 10 m, 200 MHz is used. Depending on the operating mode, the polarization mode can be adjusted: V polarization is used for horizontal fractures (those with an inclination angle of less than 30°), H polarization is used for vertical fractures, and H+V polarization is used for fine delineation of cave boundaries in anisotropic rock formations.

[0041] Advance drilling is carried out after geophysical exploration is completed. The diameter of the advance horizontal drilling hole is usually 5cm~10cm, and one hole is drilled directly above the tunnel face, in the lower left corner, and in the lower right corner. Cores are taken using a core tube for composition analysis to determine the geological conditions in front of the tunnel face. Core samples are numbered in sequence, and the sampling depth and location are recorded. Lithology identification is used to determine the type and composition of the rock; physical and mechanical property tests are used to measure the density, porosity, compressive strength, etc. of the rock; and water content tests are used to determine the water content of the rock and determine the groundwater conditions. The results of drilling construction, sampling, data analysis, and verification are compiled into a report. The report should include a drilling trajectory map, geological parameter profile, core and cuttings analysis results, and recommendations for tunnel construction.

[0042] Step 2: Determine the dynamic weights of different advanced prediction detection methods based on their matching degree to the target karst tunnel type, historical accuracy, and the quality of the target karst tunnel related data currently collected by the advanced prediction detection methods.

[0043] Dynamic weight allocation is based on the formula: wi = α·Si + β·Qi + γ·Gi, Computational detection methods i The weight of Si Indicates detection method i The historical accuracy of the target karst type, such as facing the water-rich cave area, transient electromagnetic STEM = 0.82, GPR SGPR = 0.65; α, β, γ These are weight coefficients, which change dynamically according to the increase of historical data. The default values ​​are α=0.4, β=0.3, and γ=0.3.

[0044] Qi Rate real-time quality: ,in, is the effective data volume, is the total amount of data.

[0045] Gi Indicates the regional geological matching degree. According to the analysis of the drill core, the matching degree of the geophysical prospecting method is judged. For example: G in the limestone area DC =0.7, G GPR =0.5; broken zone G TEM =0.9.

[0046] Detection accuracy of different karst types based on different methods Si , establish a database of karst type-detection method correspondence. Establish real-time data quality Qi The data quality of different methods is obtained by calculating the SNR after wavelet transform noise reduction. The geological matching degree is adjusted in real time according to the results of the advanced exploration. Gi , and finally perform weighted processing during data fusion.

[0047] As an optional implementation, this example performs regional fusion. First, a regional division rule is established, dividing the exploration area into a 1m*1m*1m cube grid. Data coverage for each grid is marked. Each time a grid is covered by data from one of the GPR, transient electromagnetic, and DC methods, the data coverage is increased by 1. If the data coverage is > 2, the grid is considered an overlapping area; if the data coverage is 1, it is a non-overlapping area; and if the data coverage is 0, it is an unexplored area. Grid boundaries are then modified based on drillhole lithologic boundaries and geophysical anomalous zones. This is typically determined based on resistivity gradients, such as a resistivity gradient > 50Ω·m / m.

[0048] For the overlapping area, the abnormal area of ​​each method is binarized. If there is a low resistance abnormality in the detection method, then V i =1, normal V i = 0. Based on the above weight w i Perform weighted voting, as follows:

[0049] in, 、 、 Represent the weights of DC, transient electromagnetic, and ground penetrating radar methods, 、 、 They respectively indicate whether the detection results of the DC, transient electromagnetic and ground penetrating radar methods are low resistance anomalies.

[0050] Judgment rule: If V>1.5, it is confirmed as an abnormal area.

[0051] For non-overlapping areas, if an abnormal mutation area (resistivity gradient > 50Ω·m / m) appears, a supplementary detection protocol is triggered and marked as a "suspected abnormal area". It is recommended to start other geophysical exploration methods or drilling methods for confirmation. Compensation algorithms are introduced, such as: the Cole-Cole model is introduced into the DC method, and the complex resistivity model is introduced into the inversion; the transient electromagnetic method introduces the time-frequency joint inversion, and the frequency domain constraint including the short-time Fourier transform (STFT) is added to the original time domain objective function. The frequency domain weight matrix W is used to calculate the frequency domain constraint. f Highlight the 1-10kHz effective frequency band.

[0052] For the area without data coverage, the volume of the area without data coverage is used for judgment. If the area without data coverage is greater than 5m 3 , directional drilling is started to fill holes along the direction of maximum risk; otherwise, it is marked as a "low confidence zone" and a risk warning is output.

[0053] Step 3: Based on the data collected from the target karst tunnel using the advance prediction detection method in the target karst tunnel advance prediction detection scheme, the corresponding target karst tunnel disaster identification results are obtained using the deep learning model.

[0054] Data collected by different advance prediction detection methods is processed. The collected potential data is denoised, corrected, and inverted to generate resistivity distribution images of the subsurface medium. Transient electromagnetic signals are filtered, denoised, and inverted to extract the electrical parameter distribution of the subsurface medium. Radar waveform data is gain-adjusted, offset-corrected, and imaged in three dimensions to generate electromagnetic reflection characteristic images of the subsurface medium. For advance drilling, features of textual and numerical information are extracted, fused, and a fully connected layer is used to generate the final output features.

[0055] For DC electrical data, the convolution kernel weight W and the bias b are used to extract the characteristics of the underground resistivity distribution and identify the resistivity abnormal area. The potential data is mapped to a two-dimensional or three-dimensional grid to generate an input tensor. ,in and is the height and width of the grid, is the number of channels (such as potential value, current value, etc.). The CNN model includes preprocessed data X, and the convolution layer is used to extract local features. The formula is:

[0056] in: is the convolution kernel weight, is the bias term, is the output sign map; Map the characteristic map exhibits to the output space to generate the resistivity distribution characteristic image

[0057] For transient electromagnetic method data, recurrent neural network (RNN) is used to extract the time series characteristics of transient electromagnetic signals and capture the changes in underground electrical structure. The processed transient electromagnetic time series signals are converted into a format suitable for RNN input, such as tensor , where T is the time step and F is the feature dimension (such as signal amplitude and time derivative). The RNN model for extracting time series features is as follows:

[0058] in: is the hidden state at the current time step; is the hidden state at the previous time step; and is the weight matrix; is the bias term; Is the activation function. Generate the characteristic representation of the underground electrical structure ,in and are the weights and biases of the output layer.

[0059] For GPR data, a two-dimensional or three-dimensional convolutional neural network is used to extract the reflection wave features in the geological radar image and identify underground cavities and cracks. The radar waveform data is converted into a format suitable for CNN input. The two-dimensional image is , three-dimensional data volume , where H is the time (depth) dimension, W is the line position dimension, and D is the number of lines (the third dimension). The above CNN model is used to extract the local feature map, flatten the feature map and map it to the output space to generate the reflection wave feature. , where K is the feature dimension.

[0060] For ultra-strong drilling, we use advanced drilling data to conduct lithology identification, physical and mechanical property analysis, and geological parameter modeling on core, cuttings samples, and borehole measurement data. We also use word segmentation, stop word removal, and stem extraction to pre-process text data. We extract text information from drilling logs using natural language processing technology and combine it with numerical data such as core strength and water content for feature extraction to generate text input data. and numerical input data . Build a BERT model for text feature extraction: , where H is the contextual representation of the text; numerical features are extracted using a fully connected neural network (FCN): ,in, and are weights and biases, Is the activation function; the text features and numerical features Fusion into a unified feature representation , and then a fully connected layer is used to generate the final output ,in and are the weights and biases of the output layer.

[0061] Step 4: The different target karst tunnel disaster identification results are combined with the corresponding dynamic weights to obtain the final target karst tunnel disaster identification result and generate a disaster disposal measure plan.

[0062] Mapping the features of each data type into a unified embedding space is achieved through fully connected layers or attention mechanisms. Using self-attention or cross-attention mechanisms, weighted fusion of features from different modalities is performed. These fused features are then concatenated to form a high-dimensional feature vector, which is then input into subsequent classification or regression models. A high-dimensional feature space is constructed based on the extracted data features to capture the complex spatial distribution and physical characteristics of karst hazards.

[0063] Specifically, DC electrical data is used to identify underground resistivity anomalies, transient electromagnetic data is used to detect underground electrical structures, geological radar data is used for high-resolution imaging, and advanced drilling data is used to verify and supplement the detection results of other methods. The features of different modes are unified into a high-dimensional feature space, and a high-dimensional feature vector is constructed as follows:

[0064] in, is the fused high-dimensional feature vector, is the DC method data feature vector, is the transient electromagnetic method data eigenvector, is the GPR data feature vector, is the feature vector of the advanced horizontal drilling data.

[0065] For example: a water-rich area in a karst tunnel: 1, , ; After weighted fusion: , , Input karst tunnel hazard identification model:

[0066] Secondly, a training set for the karst tunnel hazard identification model was constructed and trained. The training set was synthesized from 3D lidar scanning data of karst hazards, detection data of various karst hazards, and karst treatment plans. Karst types were classified into karst gullies, rock buds, stone forests, karst depressions, sinkholes, caves, karst fissures, underground vents, and underground river pipelines; detection methods included direct current, transient electromagnetic, ground penetrating radar, and advanced horizontal drilling; karst disasters were classified into small, medium, and large types based on their size; and treatment methods included filling, grouting, spanning, pile foundation, drainage, and plugging.

[0067] Select a suitable regression model as the karst tunnel disaster identification model and train it based on the constructed training set. Input multi-source features , and finally output the karst cavity volume V or area A. The loss function is defined as follows:

[0068] in: is the volume or area predicted by the karst tunnel hazard identification model, is the true value, and N is the number of samples.

[0069] Based on the trained karst tunnel hazard identification model and the fused data, the volume of the karst cavity is predicted by the regression model. or area The three-dimensional coordinates of the disaster are determined and a three-dimensional spatial distribution map is generated. The scale and location information of the karst cavity is input into the classification model softmax classifier to intelligently identify the scale, level and location of the karst disaster.

[0070] The classification rules are as follows: Classification by scale: karst hazards are divided into small, medium, and large scales; Classification by level: karst hazards are divided into low risk, medium risk, and high risk; Classification by location: karst hazards are divided into the front, left, right, above, and below the tunnel.

[0071] The output of the Softmax classifier is the probability distribution of each classification category:

[0072] Where Z is the input feature vector, and is the weight and bias of the kth category, k is the total number of categories, where small / medium / large, low / medium / high distributions correspond to k=0,1,2 respectively, and front / left / right / above / below distributions correspond to k=0,1,2,3,4 respectively, and finally the probability distribution of size categories is obtained. , probability distribution of risk level categories , probability distribution of location categories Output the classification results of karst hazards in a structured data format such as CSV format.

[0073] The judgment rules are as follows: Scale level determination: Small scale: karst cavity volume V<10m 3 and (0)>0.7; Medium scale: 10m 3 ≤V≤50m 3 ; Large scale: V ≥ 50m 3 ; Risk level determination: Low risk: >0.6; Medium risk: >0.5; High risk: >0.4; Position category determination: select the maximum probability max location.

[0074] Finally, based on the identification results, the scale, level, location and three-dimensional spatial distribution map of the karst disaster are input. Based on the disposal plans corresponding to different karst disaster scales, different karst disaster levels and different karst disaster locations stored in the system, the system automatically generates a disposal plan for the current karst disaster.

[0075] The corresponding disposal plans for different karst disaster scales, different karst disaster levels, and different karst disaster locations stored in the system are as follows:

[0076] This embodiment integrates three-dimensional parameters—geology, hydrology, and structural response—to achieve real-time karst risk classification and early warning based on real-time detection data. Based on tunnel karst hazard detection data provided by direct current (DC) methods, transient electromagnetic (TEM) methods, GPR, and advanced horizontal drilling, a geologically adaptive weighting model is proposed to generate weight coefficients for different detection methods under different geological conditions. Furthermore, weighted data fusion and compensation algorithms are proposed for overlapping and non-overlapping areas. A supplementary measurement mechanism is implemented for areas without data coverage.

[0077] A large-scale model construction and training method for intelligent identification and disposal of karst hazards. A multimodal deep learning model is used to fuse features from different sources. The features of each data type are mapped to a unified embedding space through a fully connected layer or attention mechanism. Self-attention or cross-attention mechanisms are used to weightedly fuse features from different modalities to form a high-dimensional feature vector, which is then input into subsequent classification or regression models to capture the complex spatial distribution and physical characteristics of karst hazards. A dataset is constructed for training based on karst type, detection method, and karst scale. Stochastic gradient descent and the Adam optimizer are used to train the model. Karst hazards are identified through regression and classification models. Ultimately, the optimal disposal measures are recommended through a rule engine and optimization algorithm.

[0078] For the direct current method, a flexible electrode array for tunnel advance detection is used, comprising a conductive rubber core layer, a hydraulic expansion chamber, and a contact pressure feedback module. The contact pressure between the electrode and the borehole wall is dynamically maintained within the range of 0.3 to 1.2 MPa. The direct current current generator described is characterized by dynamically adjusting the injection current based on real-time inversion resistivity and adopting a pulsed power supply mode in water-rich rock formations. The electrode arrangement method for cross-hole resistivity CT is characterized by arranging three boreholes in a triangular shape on the tunnel face, with an electrode spacing of 1.5 m within each borehole. Six sets of potential distribution data are acquired through a rotating excitation mode.

[0079] For the transient electromagnetic method, the lane-hole transient electromagnetic method is adopted: the transmitting coil is fixed on the tunnel face, and the receiving coil is a three-component magnetic probe built into the borehole. The two are synchronized by optical fiber, and the spacing can be dynamically adjusted in the range of 0.5-5m.

[0080] For GPR, a dual-polarization spiral scanning method is adopted, using an H / V orthogonal antenna and a variable-density spiral path detection method. The antenna frequency is adjusted according to different detection depth requirements, and the orthogonal method is adjusted according to the anomaly direction.

[0081] Example 2 The purpose of this embodiment is to provide a method for identifying karst tunnel hazards in complex terrain based on advance drilling, including: a detection scheme determination module configured to: determine a risk level of a target karst tunnel based on the geological conditions, hydrological conditions, and structural mechanical response of the target karst tunnel; and determine an advanced prediction detection scheme for the target karst tunnel based on the risk level of the target karst tunnel; the advanced prediction detection scheme for the target karst tunnel including at least one advanced prediction detection method; a weight determination module configured to determine dynamic weights of different advanced prediction detection methods based on the matching degree of different advanced prediction detection methods to the target karst tunnel type, historical accuracy, and quality of target karst tunnel related data currently collected by the advanced prediction detection methods; The initial identification module is configured to: obtain corresponding target karst tunnel hazard identification results using a deep learning model based on the data collected from the target karst tunnel using the advance prediction detection method in the target karst tunnel advance prediction detection scheme; The final identification module is configured to: fuse different target karst tunnel disaster identification results with the corresponding dynamic weights to obtain the final target karst tunnel disaster identification result In further embodiments, there is also provided: An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed by the processor. When the computer instructions are executed by the processor, the method described in Example 1 is performed. For the sake of brevity, no further details are given here.

[0082] It should be understood that in this embodiment, the processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), off-the-shelf field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0083] The memory may include a read-only memory and a random access memory, and provides instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0084] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in embodiment 1 is performed.

[0085] The method in Example 1 can be directly implemented as being executed by a hardware processor, or by a combination of hardware and software modules within the processor. The software module can be located in a storage medium well-established in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not given here.

[0086] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in embodiment 1 is implemented.

[0087] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions contained in program modules, which are executed in a device on a real or virtual processor of a target to perform the process / method described above. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided between program modules as needed. The machine-executable instructions for the program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0088] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the computer or other programmable data processing device, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0089] In the context of the present invention, computer program code or related data can be carried by any appropriate carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals include electrical, optical, radio, acoustic, or other forms of propagated signals, such as carrier waves, infrared signals, and the like.

[0090] Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0091] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A method for identifying karst tunnel hazards in complex terrain based on advanced drilling, characterized in that: include: Determine the risk level of the target karst tunnel based on its geological conditions, hydrological conditions, and structural mechanical response, and determine the advanced prediction and detection plan for the target karst tunnel based on the risk level of the target karst tunnel; The target karst tunnel advance prediction detection scheme includes at least one advance prediction detection method; Determine the dynamic weights of different advanced prediction detection methods based on their matching degree to the target karst tunnel type, their historical accuracy, and the quality of the target karst tunnel-related data currently collected by the advanced prediction detection methods; Based on the data collected from the target karst tunnel using the advance prediction detection method in the target karst tunnel advance prediction detection scheme, the corresponding target karst tunnel hazard identification results are obtained using a deep learning model; The different target karst tunnel hazard identification results are fused with the corresponding dynamic weights to obtain the final target karst tunnel hazard identification result.

2. The method for identifying karst tunnel disasters in complex terrain based on advance drilling according to claim 1, characterized in that: The risk levels of the target karst tunnel include high risk, medium risk and low risk; among them, the advanced prediction detection plan for the target karst tunnel corresponding to high risk is: detection using direct current method, transient electromagnetic method, ground penetrating radar and advance drilling; the advanced prediction detection plan for the target karst tunnel corresponding to medium risk is: detection using transient electromagnetic method, ground penetrating radar and advance drilling; the advanced prediction detection plan for the target karst tunnel corresponding to low risk is: detection using ground penetrating radar and advance drilling.

3. The method for identifying karst tunnel disasters in complex terrain based on advance drilling according to claim 1, characterized in that: The geological conditions include lithology score, karst density and weathering degree; the hydrological conditions include groundwater pressure, borehole water output per unit time and groundwater corrosivity; and the structural mechanical response includes surrounding rock convergence rate and support stress.

4. The method for identifying karst tunnel disasters in complex terrain based on advance drilling according to claim 1, characterized in that: Also includes: The detection area is divided, and whether it is an abnormal area is determined based on the coverage degree of different advance prediction detection methods in the divided area and the dynamic weights of different advance prediction detection methods.

5. The method for identifying karst tunnel disasters in complex terrain based on advance drilling according to claim 2, characterized in that: The potential data of the target karst tunnel collected by the direct current method is mapped and processed, and the features of the mapped potential data are extracted using the CNN model to obtain the resistivity distribution feature image; The transient electromagnetic data of the target karst tunnel collected by the transient electromagnetic method are used to extract the transient electromagnetic time series features of the transient electromagnetic data through a recursive neural network; The geological radar image of the target karst tunnel collected by the ground penetrating radar is used, and the convolutional neural network is used to extract the reflection wave characteristics of the geological radar image.

6. The method for identifying karst tunnel disasters in complex terrain based on advance drilling according to claim 1, characterized in that: The different target karst tunnel hazard identification results are combined with the corresponding dynamic weights to obtain the final target karst tunnel hazard identification results, which are as follows: The identification results of different target karst tunnel hazards are combined with the corresponding dynamic weights to obtain the karst cavity volume or area through the regression model; Generate a three-dimensional spatial distribution map based on the volume or area of ​​the karst cavity and the three-dimensional coordinates of the disaster; Based on the three-dimensional spatial distribution map, the scale, level and location of the karst disaster are obtained through a classification model.

7. A method for identifying karst tunnel disasters in complex terrain based on advanced drilling as described in any one of claim 6, which generates a treatment plan corresponding to the current karst disaster based on the obtained karst disaster scale, level and location and according to the stored treatment plans corresponding to different karst disaster scales, different karst disaster levels and different karst disaster locations.

8. A method for identifying karst tunnel hazards in complex terrain based on advance drilling, characterized in that: include: a detection scheme determination module configured to: determine a risk level of the target karst tunnel based on the geological conditions, hydrological conditions, and structural mechanical response of the target karst tunnel, and determine an advanced prediction detection scheme for the target karst tunnel based on the risk level of the target karst tunnel; The target karst tunnel advance prediction detection scheme includes at least one advance prediction detection method; a weight determination module configured to determine dynamic weights of different advanced prediction detection methods based on the matching degree of different advanced prediction detection methods to the target karst tunnel type, historical accuracy, and quality of target karst tunnel related data currently collected by the advanced prediction detection methods; The initial identification module is configured to: obtain corresponding target karst tunnel hazard identification results using a deep learning model based on the data collected from the target karst tunnel using the advance prediction detection method in the target karst tunnel advance prediction detection scheme; The final identification module is configured to fuse different target karst tunnel disaster identification results with corresponding dynamic weights to obtain a final target karst tunnel disaster identification result.

9. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 7 is completed.

10. A computer-readable storage medium, characterized in that Used to store computer instructions, which, when executed by a processor, complete the method according to any one of claims 1 to 7.

Images