A method for identifying karst tunnel collapse disaster microseismic precursor information

Abstract

The present application relates to the technical field of geological disaster prediction, and more particularly to a method for identifying microseismic precursor information of karst tunnel collapse disaster. The technical solution comprises constructing a microseismic monitoring system integrating multi-source data, processing and analyzing based on machine learning, establishing a multi-factor coupling analysis and early warning model, and real-time early warning and dynamic feedback. The present application can accurately remove noise by wavelet transform, retain the key features of microseismic signals, improve the quality of data, and extract multi-dimensional features including frequency, amplitude, phase, etc. from microseismic signals, achieving multi-dimensional feature extraction and improving the accuracy of capturing precursor information related to collapse disaster. In addition, the present application utilizes machine learning algorithms to comprehensively classify and judge various factors such as geological conditions and construction progress, making the identification results more consistent with actual situations and avoiding misjudgments caused by single factors, thereby improving the accuracy and efficiency of identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological disaster prediction, and particularly relates to a method for identifying microseismic precursor information of karst tunnel collapse disaster. BACKGROUND

[0002] With the continuous development of transportation infrastructure construction, the number of tunnels in karst areas is increasing. Karst geology is complex and uncertain, and there are a large number of karst caves, dissolution fissures and groundwater activities and other special geological phenomena inside. In the process of karst tunnel construction, collapse disaster is one of the most serious engineering accidents, which not only causes delay in construction progress and increase in cost, but also may endanger the safety of construction personnel. Therefore, how to effectively predict and prevent karst tunnel collapse disaster has become a key problem to be solved in the engineering field. Since karst tunnel collapse is often the result of the joint action of multiple factors such as geological structure, groundwater activity, rock mechanical properties and engineering construction, the existing identification methods mostly only focus on the analysis of a single factor or a few factors, lack comprehensive consideration and in-depth research on microseismic precursor information under the coupling action of multiple factors, and are difficult to accurately reflect the incubation process of collapse disaster. Therefore, we propose a method for identifying microseismic precursor information of karst tunnel collapse disaster. SUMMARY

[0003] The purpose of the present application is to solve the problem of the existing identification method lacking comprehensive consideration and in-depth research on microseismic precursor information under the coupling action of multiple factors, and being difficult to accurately reflect the incubation process of collapse disaster in the background art. A method for identifying microseismic precursor information of karst tunnel collapse disaster is proposed.

[0004] The technical scheme of the present application: a method for identifying microseismic precursor information of karst tunnel collapse disaster, comprising the following steps:

[0005] Constructing a microseismic monitoring system integrating multi-source data;

[0006] Processing and analyzing based on machine learning according to the microseismic signals collected by the microseismic detection system, and extracting multi-dimensional features of the microseismic signals;

[0007] Establishing a multi-factor coupling analysis and early warning model based on the multi-dimensional features of the microseismic signals, analyzing the influence mechanism and degree of different factors on the microseismic precursor information, and determining the early warning index and its threshold;

[0008] Real-time early warning and dynamic feedback, dynamically evaluating and warning the collapse risk of karst tunnel according to the early warning index and its threshold.

[0009] Optionally, the construction of the microseismic monitoring system integrating multi-source data comprises the following steps:

[0010] Distributed optical fiber sensors are laid along the tunnel lining, arch top and side wall, and point sensors are arranged in high-risk areas with complex geological structure and broken rock, as well as at the entrance and exit of the tunnel and fault fracture zone, to form a monitoring network;

[0011] Geological data, underground water data and construction data are synchronously collected, wherein the geological data includes rock mechanics parameters and geological structure information, the underground water data includes water level change and water flow velocity, and the construction data includes excavation progress and support parameters.

[0012] Optionally, the processing and analysis based on machine learning include the following steps:

[0013] Wavelet transform is used to denoise and filter the microseismic signals collected by the microseismic detection system for pretreatment;

[0014] The amplitude, frequency, energy and duration characteristics of the signals are extracted to construct a feature data set;

[0015] Support vector machines are used to train and optimize the model with the feature data set as input and whether it is a collapse disaster microseismic precursor information as output label;

[0016] Weighted average method is used to fuse and integrate learning of multiple trained machine learning models.

[0017] Optionally, when wavelet transform is used to denoise and filter the microseismic signals, the original microseismic signal is f(t), the wavelet base function ψ(t) is selected, the original signal is decomposed by wavelet to obtain wavelet coefficients under different scales a and displacements b:

[0018]

[0019] The wavelet coefficients are processed by setting a threshold λ, and the processed wavelet coefficients are :

[0020] When ,

[0021] When , The processed wavelet coefficients are reconstructed to obtain the denoised signal:

[0022]

[0023] Wherein,

[0024]

[0025] Wherein, is the Fourier transform of ψ(t);

[0026] In the constructed feature dataset, the amplitude feature includes the maximum amplitude of the signal and the average amplitude where T is the duration of the signal

[0027] The frequency feature is obtained by performing Fourier transform on the signal to obtain the frequency spectrum of the signal, and then calculating the dominant frequency where ω p is the angular frequency corresponding to the frequency component with the largest amplitude in the frequency spectrum, and the frequency bandwidth Δf = f h -f t , f h and f t are the high and low boundary frequencies corresponding to a certain energy proportion in the signal frequency spectrum

[0028] The energy feature uses the energy of the signal

[0029] The duration feature is the time interval T0 from the start to the end of the recorded signal

[0030] In the model training and optimization, the feature dataset {x i , y i} (where x i is the feature vector, and y i ∈{-1,1} represents whether it is precursor information) is constructed, and the optimization function is as follows:

[0031]

[0032] where the constraint condition y i (ω·x i +b)≥1-ξ i , ξ i ≥0 (i = 1, 2, …, n), where ω is the normal vector of the hyperplane, b is the bias term, C is the penalty parameter, ξ i is the relaxation variable, and the optimal hyperplane parameters ω * and b * are obtained, so as to determine the classification model f(x) = sign(ω * ·x+b * ). In the training process, the cross-validation method is used to adjust the parameters to improve the generalization ability of the model

[0033] The weighted average method is used to fuse and integrate learning of multiple trained machine learning models. For the prediction results of the test samples of multiple models M1, M2, …, M k , the final prediction result where α i is the model M iThe weights of the multiple models can be determined according to the model performance evaluation index, and the prediction information of the multiple models is comprehensively determined.

[0034] Optionally, the cross-validation method adopts a K-fold cross-validation method, and the verification process includes the following steps.

[0035] Data set division: assuming that the microseismic signal feature data set contains n samples, the n samples are randomly divided into K mutually exclusive subsets F1, F 2, …, F K , satisfying D represents the entire data set, and The number of samples in each subset is

[0036] Training and verification process and performance index calculation: for k = 1, 2, …, K rounds, K-1 subsets except F k are selected as the training set Train k = ∪ i≠k F i , F k is used as the validation set Val k , the training set Train k is used to train the machine learning model, and the trained model is denoted as M k For a classification task, if the prediction result of the model M k on the validation set Val k is i represents the sample number, j represents the category number, and the true label is y ij , then the accuracy calculation formula is as follows:

[0037]

[0038] Where II(·) is an indicator function, which takes the value 1 when the condition in the parentheses is true, and 0 otherwise, and |Val k | represents the number of samples in the validation set Val k For a regression task, if the predicted value is and the true value is y i , the mean square error calculation formula is as follows:

[0039]

[0040] Parameter adjustment and model selection: for the model parameters to be adjusted, multiple values are set, for each parameter value θ, the training and verification process is repeated K times, and the average performance index under the parameter value is obtained, including the average accuracy or the average mean square error, and the calculation formulas are as follows:

[0041]

[0042]

[0043] Selecting the optimal parameter value θ according to the average performance index opt , so that for the classification task For the regression task To improve the generalization ability of the model.

[0044] Optionally, the establishment of the multi-factor coupling analysis and early warning model comprises the following steps:

[0045] Based on the principles of physical mechanics and numerical simulation methods, a multi-factor coupling analysis model is constructed;

[0046] According to the interaction of geological structure, groundwater, rock mechanical properties and engineering construction factors, the stress and strain state of the surrounding rock and soil mass and the microseismic activity law under different working conditions are simulated through data simulation methods;

[0047] The influence degree of different factors on microseismic precursor information is analyzed through the multi-factor coupling analysis model, and the early warning index and its threshold are determined.

[0048] Optionally, the numerical simulation method adopts the finite element method or the discrete element method, and the constitutive relation of the rock and soil mass in the simulation process adopts the Mohr-Coulomb criterion or the Drucker-Prager criterion, and its stress and strain relationship expression is: {σ}=[D]{ε}, wherein {σ} is the stress vector, [D] is the elastic matrix, and {ε} is the strain vector.

[0049] Optionally, the real-time early warning and dynamic feedback comprises:

[0050] A data transmission system is established to transmit multi-source data in the monitoring site to the data processing center in real time;

[0051] In the data processing center, cloud computing or edge computing technology is used to quickly process and analyze the data;

[0052] The karst tunnel collapse risk is dynamically evaluated and warned, and when the microseismic precursor information exceeds the early warning threshold, an early warning signal is issued and fed back to the construction site.

[0053] Optionally, the data transmission system adopts wireless transmission technology or wired transmission technology, and the data transmission protocol complies with specific industrial standard protocols, including Modbus protocol or OPC UA protocol, and the application of cloud computing or edge computing technology in data processing includes data distributed storage, parallel computing task allocation and resource scheduling optimization.

[0054] Optionally, in the process of engineering construction, when the tunnel construction progress and the geological conditions change, the parameters in the early warning model are updated and optimized, including adjusting the early warning index threshold and retraining the machine learning model weight, to adapt to the identification needs of the karst tunnel collapse disaster microseismic precursor information under different construction stages and geological environments.

[0055] In summary, the present application includes at least one of the following beneficial technical effects:

[0056] The present application can accurately remove noise by wavelet transform, retain the key features of microseismic signals, and improve the quality of data, laying a foundation for subsequent accurate identification of precursor information, and extracting features including frequency, amplitude, phase, etc. from microseismic signals, realizing multi-dimensional feature extraction, fully depicting the characteristics of microseismic events, thereby improving the accuracy of capturing precursor information related to collapse disasters and reducing the possibility of false positives and false negatives.

[0057] The present application uses machine learning algorithms to automatically learn and identify different types of collapse disasters from a large amount of microseismic data, and also combines geological conditions, construction progress and other factors for comprehensive classification and judgment, making the identification results more in line with the actual situation and avoiding misjudgment caused by a single factor, improving the accuracy and efficiency of identification.

[0058] The present application processes and analyzes the collected microseismic data in real time, quickly gives the identification result of the precursor information, meets the needs of real-time monitoring and timely warning in the process of karst tunnel construction, and has strong dynamic adaptability, can continuously update and optimize the identification model as the tunnel construction progresses and the geological conditions change, always maintains accurate identification of microseismic precursor information, and effectively deals with complex and variable construction environments. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 A flowchart of a method for identifying microseismic precursor information of karst tunnel collapse disasters is given. DETAILED DESCRIPTION

[0060] The technical solutions of the present application will be described in detail below with reference to the accompanying drawings, obviously, the described embodiments are part of the embodiments of the present application, not all embodiments.

[0061] The components of the embodiments of the present application described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the present application.

[0062] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative labor fall within the scope of the present application.

[0063] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second", "third" are only for the purpose of description, and cannot be understood as indicating or implying relative importance.

[0064] It should be noted that the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. In this specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0065] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mount", "connect", "connect" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the communication between two elements. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0066] As Figure 1 shown, the present application proposes a method for identifying karst tunnel collapse disaster microseismic precursor information, comprising the following implementation steps:

[0067] I. Construct a microseismic monitoring system integrating multi-source data

[0068] Sensor arrangement: Distributed optical fiber sensors are laid along the tunnel lining, vault, and sidewall, with a monitoring node set every 10 meters to ensure continuous monitoring of the full cross-section of the tunnel for microseismicity. Meanwhile, point sensors are arranged in complex geological areas (including near proven faults and areas with dense large caves), high-risk areas with broken rock (including sections with serious joint fissure development identified through previous geological exploration), and key locations at the tunnel entrance and exit, fault fracture zone, totaling 500 distributed optical fiber sensor nodes and 100 point sensors, forming a comprehensive and multi-level monitoring network.

[0069] Data collection: Geological data, groundwater data, and construction data are collected simultaneously. For geological data, rock cores are drilled at different locations along the tunnel to test rock mechanics parameters, including compressive strength, elastic modulus, and Poisson's ratio, with a total of 100 sets of rock mechanics parameter data collected. Geological structure information is mapped using geological radar detection to determine the location and scale of faults and large caves. Groundwater data is collected by setting 20 water level sensors and 15 water flow velocity sensors at different depths within the tunnel to monitor real-time changes in groundwater level and water flow velocity, with data collected every 30 minutes. Construction data is automatically recorded by the construction record system, including the progress of each excavation, support parameters (including anchor rod length, installation time, sprayed concrete thickness, and strength grade).

[0070] II. Processing and analysis based on machine learning using microseismic signals collected by the microseismic detection system to extract multi-dimensional features of the microseismic signals.

[0071] Data preprocessing: Wavelet transform is used to denoise and filter the collected microseismic signals. Let the original microseismic signal be f(t), select the wavelet basis function ψ(t), and perform wavelet decomposition on the original signal to obtain wavelet coefficients at different scales a and displacements b:

[0072]

[0073] By setting the threshold value λ = 0.3, the processed wavelet coefficients are:

[0074] When ,

[0075] When , The denoised signal is reconstructed from the processed wavelet coefficients:

[0076]

[0077] where

[0078] ​

[0079] wherein, is the Fourier transform of ψ(t);

[0080] Feature extraction: extract the amplitude, frequency, energy, duration features of the signal to construct the feature dataset, the amplitude features include the maximum amplitude and average amplitude of the signal wherein T is the duration of the signal;

[0081] The frequency feature is obtained by performing Fourier transform on the signal to obtain the frequency spectrum of the signal, and then calculate the dominant frequency wherein ω p is the angular frequency corresponding to the frequency component with the largest amplitude in the frequency spectrum, and the frequency bandwidth Δf = f h -f t , f h and f t are the high and low boundary frequencies corresponding to a certain energy proportion in the signal frequency spectrum, respectively;

[0082] The energy feature uses the energy of the signal

[0083] The duration feature is the time interval T0 recording the start and end of the signal;

[0084] Model training and optimization: select support vector machine (SVM) algorithm for model training, in the model training and optimization, the feature dataset {x i , y i} (wherein x i is the feature vector, y i ∈{-1,1} represents whether it is precursor information), construct the optimization function as follows:

[0085]

[0086] wherein the constraint condition y i (ω·x i +b)≥1-ξ i , ξ i ≥0 (i = 1, 2, …, n), wherein ω is the normal vector of the hyperplane, b is the bias term, C is the penalty parameter, ξ i is the relaxation variable, and the optimal hyperplane parameters ω * and b * are obtained, so as to determine the classification model f(x) = sign(ω * ·x+b * ), in the training process, cross-validation method is used to adjust the parameters to improve the generalization ability of the model;​

[0087] The cross-validation method uses 5-fold cross-validation, randomly dividing the dataset into 5 mutually exclusive subsets F1, F2, F3, F4, F5, F6, F7, F8, F9, F1, F1, F1, F2, F1, F2, F3, F4, F1, F2, F3, F4, F5, F6, F7, F8, F9, F1, F1, F1, F2 ...2, F3, F1, F2, F2, F3, F1, F2, F2, F3, F1, F2, F2, F3, F1, F2, F2, F3, F2, F2, 2, F 3, ,F 4, F5, for rounds k = 1, 2, ..., 5, select the rounds excluding F. k The other four subsets are used as the training set. k =∪ i≠k F i F k As a verification set Val k Using the training set Train k Train an SVM model, and denote the trained model as M. k For classification tasks, if model M k In the validation set Val k The above prediction result is i represents the sample number, j represents the category number, and y represents the true label. ij The accuracy calculation formula is as follows:

[0088]

[0089] For different C values, repeat the 5-fold cross-validation process described above to obtain the average validation set performance index for each C value. The calculation formula is:

[0090]

[0091] Compare the average accuracy at different C values;

[0092] Model fusion: A weighted average method is used to fuse and integrate multiple trained machine learning models. For multiple models M1, M2, ..., M k Prediction results for the test samples, final prediction results Where α i Model M i The weights can also be determined based on model performance evaluation metrics, combining the prediction information from multiple models.

[0093] III. Based on the multi-dimensional characteristics of microseismic signals, establish a multi-factor coupled analysis and early warning model to analyze the influence mechanism and degree of different factors on microseismic precursor information, and determine the early warning indicators and their thresholds.

[0094] Model construction: Based on the finite element method, a multi-factor coupling analysis model is constructed, considering the interaction of geological structure, groundwater, rock mechanical properties and engineering construction factors. In the model, the constitutive relation of rock-soil mass adopts Mohr-Coulomb criterion, and its stress-strain relationship expression is: {σ} = [D] {ε}, where {σ} is the stress vector, [D] is the elastic matrix, and {ε} is the strain vector. According to the actual size of the tunnel, geological structure and material properties, etc. Parameters, the finite element model is established, and the tunnel and surrounding rock-soil mass are divided into 10,000 units.

[0095] Simulation analysis and early warning index determination: Simulate the stress-strain state and microseismic activity law of the surrounding rock-soil mass of the tunnel under different working conditions: set the groundwater level to gradually change from low to high (a total of 10 water level change levels, from 1 meter to 10 meters deep), change the elastic modulus of the rock (set 5 different elastic modulus values, from 10 GPa to 50 GPa) and the excavation progress of the tunnel (simulate different stages from the initial to the middle and late stages of excavation, including the initial excavation of 1000 meters, the middle excavation of 2000 meters, and the late excavation of 2000 meters), through simulation, it is found that when the groundwater level rises at a speed of more than 0.5 m / d and the energy change rate of microseismic signals in a specific frequency band (80-120 Hz) is more than 30%, the risk of collapse increases significantly. A warning model is constructed by combining multiple early warning indicators, and a weighted comprehensive evaluation method is used to give different weights according to the importance of each early warning indicator to the collapse disaster. The actual monitoring value of each indicator is compared with the corresponding threshold value and normalized, and then weighted summation is performed according to the weight to obtain a comprehensive early warning index;

[0096] Four, real-time early warning and dynamic feedback, based on early warning indicators and their thresholds, dynamically assess and warn the collapse risk of karst tunnels;

[0097] Data transmission and processing: Establish a data transmission system to transmit multi-source data from the monitoring site to the data processing center in real time. The data transmission system uses wired Ethernet transmission technology and follows the Modbus protocol. Data acquisition terminals are set up in the tunnel to collect and preliminarily process the data collected by the sensors, and then transmit the data to the data processing center located outside the tunnel through Ethernet. The data processing center uses cloud computing technology to quickly process and analyze the data. The data is distributed and stored in 5 cloud storage nodes, and parallel computing task allocation is used to decompose the data processing tasks to 10 computing nodes for simultaneous processing. Resource scheduling optimization dynamically allocates computing resources, storage resources and network resources according to task requirements and resource usage;

[0098] Early warning and feedback: Based on the real-time monitoring data and the calculation results of the early warning model, the karst tunnel collapse risk is dynamically evaluated and warned. At a certain moment, when the monitoring underground water level rising speed reaches 0.6 m / d and the energy change rate of the microseismic signal in a specific frequency band is 35%, the data is input into the early warning model to calculate the comprehensive early warning index, which is 0.7, exceeding the set early warning threshold 0.6. The data processing center immediately issues an early warning signal, and through the sound and light alarm device in the local data processing center, a sound and light alarm is sounded, and the early warning information is sent to the mobile phones of the construction site managers through the short message notification system. The short message content includes the early warning position, the danger degree, the possible collapse reason, in addition, an early warning pop-up window is popped up on the monitoring terminal at the construction site, which displays the detailed early warning information and the response suggestion. The construction site personnel suspend the construction according to the early warning information, organize the personnel to evacuate to the safe area, and carry out emergency support to the key parts of the tunnel, including increasing the number of anchor rods and improving the strength of the sprayed concrete. With the progress of the engineering construction and the change of the geological conditions, the parameters in the early warning model are constantly updated and optimized. When the tunnel excavation enters a new geological area, the values of the rock mechanics parameters in the model are adjusted according to the geological exploration data of the new area, the model parameters related to the support are adjusted according to the actual effect of the support structure in the construction process, and the machine learning model is retrained according to the newly collected microseismic data and the actually occurred collapse examples, so as to optimize the model weights, so that the early warning model can always adapt to the needs of identifying the microseismic precursor information of the karst tunnel collapse disaster in different construction stages and geological environments, and continuously provide accurate and reliable early warning services.

[0099] The present application can accurately remove noise by wavelet transform, retain the key features of microseismic signals, and improve the quality of data, laying a foundation for subsequent accurate identification of precursor information. Moreover, multi-dimensional features including frequency, amplitude, phase and the like are extracted from the microseismic signals, multi-dimensional feature extraction is realized, and the characteristics of the microseismic event are comprehensively described, so as to improve the accuracy of capturing the precursor information related to the collapse disaster and reduce the possibility of false negatives and false positives.

[0100] The present application can automatically learn and identify different types of collapse disasters from a large amount of microseismic data by using machine learning algorithms, and comprehensively classifies and judges by combining factors such as geological conditions and construction progress, so that the identification result is more in line with the actual situation, avoids misjudgment caused by a single factor, and improves the accuracy and efficiency of identification.

[0101] The present application processes and analyzes the collected microseismic data in real time, quickly gives the identification result of the precursor information, meets the needs of real-time monitoring and timely warning in the process of karst tunnel construction, and has strong dynamic adaptability. With the progress of the tunnel construction and the change of the geological conditions, the identification model is constantly updated and optimized, the accurate identification of the microseismic precursor information is always maintained, and the complex and changeable construction environment is effectively coped with.

[0102] The above specific embodiments are only several optional embodiments of the present application, and based on the technical solutions of the present application and the related inspiration of the above embodiments, the person skilled in the art can make various alternative improvements and combinations on the above specific embodiments.

Claims

1. A method for identifying karst tunnel collapse disaster microseismic precursor information, characterized in that, Includes the following steps: Construct a microseismic monitoring system that integrates multi-source data; Based on machine learning, the microseismic signals collected by the microseismic monitoring system are processed and analyzed to extract multi-dimensional features of the microseismic signals. Based on the multi-dimensional characteristics of microseismic signals, a multi-factor coupled analysis and early warning model is established to analyze the influence mechanism and degree of different factors on microseismic precursor information and determine the early warning indicators and their thresholds. Real-time early warning and dynamic feedback: Based on the aforementioned early warning indicators and their thresholds, the collapse risk of karst tunnels is dynamically assessed and early warning is provided. The machine learning-based processing and analysis includes the following steps: Wavelet transform is used to perform noise reduction and filtering preprocessing on the microseismic signals collected by the microseismic monitoring system; Extract the amplitude, frequency, energy, and duration features of the signal to construct a feature dataset; Support vector machine is used, with the feature dataset as input and whether it is micro-earthquake precursor information of landslide disaster as output label for model training and optimization; A weighted average method is used to fuse and ensemble learn multiple trained machine learning models. include: When wavelet transform is used to denoise and filter the microseismic signal, the original microseismic signal is , a wavelet base function is selected , the original signal is decomposed by wavelet to obtain wavelet coefficients at different scales and displacements . ； By setting a threshold The wavelet coefficients are processed, the processed wavelet coefficients are: when hour, = ; when hour, =0, reconstruct the denoised signal from the processed wavelet coefficients: ; in, ; in, yes Fourier transform; In constructing the feature dataset, the amplitude feature includes the maximum amplitude of the signal. and average amplitude ,in The duration of the signal; The frequency characteristics are obtained by performing a Fourier transform on the signal. The spectrum of the signal is obtained, and then the dominant frequency is calculated. ,in It is the angular frequency corresponding to the frequency component with the largest amplitude in the spectrum, and the frequency bandwidth. , and These are the high-frequency and low-frequency boundary frequencies corresponding to a certain proportion of energy in the signal spectrum; The energy characteristic uses the energy of the signal. ; The duration characteristic is the time interval from the start to the end of the recorded signal. ; In model training and optimization, the feature dataset The optimization function is constructed as follows: ； Among them, constraints , 0 ( n), where Let be the normal vector of the hyperplane. For bias terms, For penalty parameters, Using slack variables, the optimal hyperplane parameters are obtained by solving. and Thus, the classification model is determined. During training, cross-validation was used to adjust the parameters; A weighted average method is used to fuse and ensemble multiple trained machine learning models. Prediction results for the test samples, final prediction results ,in It is a model The weights are determined based on model performance evaluation metrics, and the prediction information from multiple models is integrated.

2. The method for identifying microseismic precursor information of karst tunnel collapse disasters according to claim 1, characterized in that, The construction of the microseismic monitoring system integrating multi-source data includes the following steps: Distributed fiber optic sensors are laid along the tunnel lining, arch, and sidewalls, and point sensors are deployed in areas with complex geological structures, high risk of rock fracture, as well as at tunnel entrances and exits and fault fracture zones to form a monitoring network. Geological data, groundwater data, and construction data are collected simultaneously. The geological data includes rock mechanics parameters and geological structure information, the groundwater data includes water level changes and water flow velocity, and the construction data includes excavation progress and support parameters.

3. The method for identifying microseismic precursor information of karst tunnel collapse disasters according to claim 1, characterized in that, The cross-validation method uses K-fold cross-validation, and the validation process includes the following steps: Dataset partitioning: Suppose the microseismic signal feature dataset contains The samples were randomly divided into... Mutually exclusive subsets ,satisfy , Represents the entire dataset, and ( The number of samples in each subset is ; Training and validation process and performance metric calculation: For Rounds, select except Other A subset as the training set , As a verification set Using the training set Train a machine learning model, and denote the trained model as... For classification tasks, if the model In the validation set The above prediction result is , Indicates the sample number. Indicates the category number, the actual label is The accuracy calculation formula is as follows: ; in This is an indicator function that takes the value 1 when the condition within the parentheses is true, and 0 otherwise. Represents the verification set The number of samples, for a regression task, if the predicted value is... The actual value is The formula for calculating the mean square error is as follows: ； Parameter tuning and model selection: Multiple values ​​are set for the model parameters to be tuned. For each parameter value... ,repeat The training and validation process described above yields average performance metrics for this parameter value, including average accuracy or average mean squared error, calculated using the following formulas: ； ； Select the optimal parameter value based on the average performance index. This makes classification tasks more feasible. For the return mission This is to improve the model's generalization ability.

4. The method for identifying microseismic precursor information of karst tunnel collapse disasters according to claim 1, characterized in that, The establishment of the multi-factor coupling analysis and early warning model includes the following steps: A multi-factor coupled analysis model is constructed based on physical and mechanical principles and numerical simulation methods; Based on the interaction of geological structure, groundwater, rock mechanical properties and engineering construction factors, the stress-strain state and microseismic activity of the rock and soil around the tunnel under different working conditions are simulated using data simulation methods. The influence of different factors on microseismic precursor information is analyzed by using a multi-factor coupling analysis model to determine the early warning indicators and their thresholds.

5. The method for identifying microseismic precursor information of karst tunnel collapse disasters according to claim 3, characterized in that, The numerical simulation method employs either the finite element method or the discrete element method. During the simulation, the constitutive relationship of the soil and rock mass is determined using the Mohr-Coulomb criterion or the Drucker-Prager criterion, and its stress-strain relationship is expressed as follows: ,in For stress vectors, For the elasticity matrix, This is the strain vector.

6. The method for identifying microseismic precursor information of karst tunnel collapse disasters according to claim 1, characterized in that, The real-time early warning and dynamic feedback include: A data transmission system will be established to transmit multi-source data from the monitoring site to the data processing center in real time. Data processing centers utilize cloud computing or edge computing technologies to rapidly process and analyze data; Dynamic assessment and early warning of karst tunnel collapse risk are conducted. When the microseismic precursor information exceeds the early warning threshold, an early warning signal is issued and fed back to the construction site.

7. The method for identifying microseismic precursor information of karst tunnel collapse disasters according to claim 6, characterized in that, The data transmission system employs wireless or wired transmission technology, and the data transmission protocol follows specific industry standard protocols, including Modbus or OPC UA protocols. The application of cloud computing or edge computing technology in data processing includes distributed data storage, parallel computing task allocation, and resource scheduling optimization.

8. A method for identifying microseismic precursor information of karst tunnel collapse disasters according to any one of claims 1-7, characterized in that, During the construction process, as the tunnel construction progresses and geological conditions change, the parameters in the early warning model are updated and optimized. This includes adjusting the threshold values ​​of the early warning indicators and retraining the weights of the machine learning model to adapt to the needs of identifying microseismic precursor information of karst tunnel collapse disasters under different construction stages and geological environments.

Images