A method and system for predicting rock fracture disasters in tunnel engineering

Abstract

The present invention proposes a prediction method and system for rock fracture disasters in tunnel engineering, which belongs to the field of tunnel engineering. It includes the steps of multi-modal monitoring data acquisition and synchronization, data preprocessing and feature calculation, multi-modal information entropy evolution analysis, disaster point detection and precursor identification, risk assessment and early warning output. Multiple types of sensors are deployed in key parts of the tunnel, and connected to the edge computing gateway through the industrial-grade field bus to achieve the time and space benchmark unification of multi-source heterogeneous data; the collected raw data are preprocessed and feature calculated; the sliding window probability distribution model is performed on each modal feature, and the information theory index is calculated to analyze the nonlinear correlation between multi-modal features; a variety of methods are used to determine whether there is a disaster precursor event; finally, based on the disaster point detection results, a multi-dimensional risk quantification model is used to conduct risk assessment on the tunnel area, and early warning output is issued for high-risk and emergency conditions.

Description

Technical Field

[0001] The present invention belongs to the field of tunnel engineering, and more particularly relates to a method and system for predicting rock fracture disasters in tunnel engineering. Background Art

[0002] Tunnel projects run through various rock and soil bodies. Due to the influence of various factors such as rock properties, working conditions, and construction technology, tunnel rock fracture disasters such as rock fracture, landslides, water outflow, and geological disasters are prone to occur, which have a serious impact on the safety and economic benefits of tunnel project construction.

[0003] Traditional methods for predicting tunnel rock fracture hazards are typically based on geological survey reports and field experience. However, due to complex geological structures, highly variable working conditions, and the influence of human factors, their prediction accuracy is not ideal. Once a fracture hazard occurs, it often results in serious consequences such as casualties, equipment loss, and project delays. Therefore, developing a method that can accurately and timely predict tunnel rock fracture hazards has significant practical value and broad market prospects.

[0004] While recent advances in tunnel engineering monitoring and prediction technologies have been made through the use of sensors and data analysis methods, existing prediction technologies for rock fracture hazards in tunnels often rely on a single monitoring modality, such as acoustic emissions or microseismic observations. These methods often fail to fully and accurately reflect the true state of the tunnel's rock formations, leading to false alarms and missed warnings. Furthermore, most prediction systems only provide prediction results, without specific warnings or protective measures. Furthermore, their mechanisms for identifying precursor events are relatively simplistic, making them unsuitable for disaster prediction in complex geological conditions.

[0005] In the existing technology, CN110671153A - Monitoring and early warning system for sudden water disasters in tunnels and underground projects focuses on geological disasters caused by tunnel construction, covering various disasters caused by poor geology, and collecting tunnel profiles, geological survey data and advance forecast data.

[0006] CN116092269A - A tunnel engineering rock disaster early warning method, device and electronic equipment are mainly aimed at geological disasters in rock and soil engineering construction, and obtain relevant data of the construction area through a multi-source monitoring system.

[0007] CN116562656A - Tunnel construction geological disaster early warning and prevention and control intelligent decision-making method and auxiliary platform for tunnel water inrush disasters, based on geological information near the tunnel axis, geological exposure of the excavated tunnel section and information obtained from advanced geological forecasts, monitors high-risk sections for sudden water inrush.

[0008] CN117711140A - Method and system for timely warning of tunnel water inrush disasters based on multi-source data fusion does not mention the type of disaster, but it can be inferred from the content that it focuses on a certain type of engineering disaster and collects data through multiple monitoring methods.

[0009] Furthermore, existing tunnel rock hazard prediction technologies often neglect correlation analysis between multiple monitoring data sets and fail to integrate multimodal data, which reduces prediction accuracy. The acquisition, synchronization, processing, and analysis of multimodal data are complex, making existing monitoring systems unable to effectively address this challenge.

[0010] Therefore, developing a system and method that can comprehensively process multimodal monitoring data, accurately predict and timely warn of rock fracture disasters in tunnel projects is of great significance for improving the safety of tunnel projects and protecting the lives and property of personnel. Summary of the Invention

[0011] The present invention aims to address the problem that existing methods for predicting rock fracture disasters in tunnel engineering fail to fully and accurately reflect the true state of the tunnel rock, suffer from false alarms and missed reports, and lack warning and protective measures. Furthermore, the existing techniques are difficult to collect, synchronize, process, and analyze multimodal data, lack effective methods for comprehensive processing, and thus reduce prediction accuracy. The present invention proposes a system and method that can comprehensively process multimodal monitoring data, accurately predict, and promptly warn of rock fracture disasters in tunnel engineering, thereby improving the safety of tunnel engineering and protecting the lives and property of personnel.

[0012] In order to achieve the above object, the present invention is implemented by adopting the following technical solutions: the method comprises:

[0013] Multimodal monitoring data acquisition and synchronization: Multiple types of sensors, including acoustic emission sensors, microseismic sensors, stress sensors, displacement sensors, and ground temperature sensors, are deployed in key locations in the tunnel. Multimodal sensor arrays are deployed in high-risk areas based on geological survey reports and numerical simulation results. All sensors are connected to the edge computing gateway via an industrial-grade fieldbus. Clock synchronization is achieved using a precise time protocol. Time synchronization is achieved through a master clock source and a fiber-optic transmission delay compensation algorithm to eliminate link delay differences. Data acquisition terminals are equipped with multi-channel synchronous sampling cards. Hardware trigger signals ensure the time axis alignment of each modal data. After preprocessing, the raw data stream is encapsulated into data packets with unified timestamps and uploaded to the central server via a dedicated network, achieving unified spatiotemporal benchmarks for multi-source heterogeneous data.

[0014] Data preprocessing and feature calculation: The collected raw monitoring data are preprocessed by denoising, standardization, and drift compensation. Features such as acoustic emission energy, number of microseismic events, stress fluctuation rate, displacement accumulation, and geothermal gradient are extracted using a sliding window statistical method to generate a feature time series matrix.

[0015] Multimodal Information Entropy Evolution Analysis: Sliding window probability distribution modeling is used for each modal feature. Information theory indicators such as Shannon entropy, joint entropy, and mutual information are calculated. Entropy gradients are extracted through entropy evolution curves and smoothing. Multi-scale entropy correlation matrices and maximum information coefficients are used to analyze the nonlinear correlations between multimodal features.

[0016] Disaster point detection and precursor identification: Based on single-modal mutation detection, multi-modal collaborative judgment, joint entropy extreme value detection and causal testing methods, it determines whether there are disaster precursor events and outputs the spatiotemporal label of the disaster point, the precursor confidence and the dominant modal characteristics;

[0017] Risk assessment and early warning output: Based on the results of disaster point detection, combined with parameters such as the confidence level of precursor events, entropy change gradient, and historical case similarity, a multi-dimensional risk quantification model is used to assess the risk level of the tunnel area. The risk level is divided into five levels. The system automatically associates the 3D geological model with the BIM model to achieve spatial visualization and positioning of risk events. In high-risk and emergency situations, it simultaneously triggers multi-channel early warnings, including information push, sound and light alarms, automatic generation of protection support plans, and feedback learning mechanisms to dynamically optimize model parameters.

[0018] In one solution, the multimodal sensor array is deployed by installing acoustic emission sensors every 10 meters on the tunnel vault and side walls, using three-component geophones as microseismic sensors arranged in a grid pattern at intervals of no more than 20 meters, using fiber optic Bragg grating sensors as stress sensors to form circumferential and longitudinal cross-measurement points, using laser rangefinders and total stations as displacement sensors to form a three-dimensional monitoring network, and using a distributed fiber optic temperature measurement system to deploy ground temperature sensors in the tunnel surrounding rock to achieve all-round, multi-dimensional monitoring of high-risk areas.

[0019] In one solution, the data synchronization uses an industrial-grade field bus to connect to the edge computing gateway, and uses the precise time protocol for master-slave clock synchronization. The master clock source uses a satellite timing module deployed at the tunnel entrance. Each sub-node eliminates delays through an optical fiber link delay compensation algorithm. All monitoring data acquisition terminals use multi-channel synchronous sampling cards, and hardware triggering is used to ensure the time consistency of various data collections. Finally, the data is encapsulated into data packets with unified timestamps for remote transmission and storage.

[0020] In one scheme, in the risk assessment and early warning output step, after detecting high risk and emergency conditions, the system can automatically push structured alarm information containing risk coordinates, dominant modal characteristics and recommended disposal measures to the monitoring center, and activate the on-site sound and light alarm device through the Internet of Things relay equipment. At the same time, for areas where there is a risk of chain collapse, a protective support plan is automatically generated and sent to the construction robot queue. The system also has a feedback learning mechanism to automatically optimize model parameters according to the frequency of subsequent microseismic events and energy release.

[0021] In one scheme, in the multimodal information entropy evolution analysis step, a sliding window is used to model the probability distribution of each modal feature, and information theory indicators including Shannon entropy, joint entropy and mutual information are calculated. The entropy curve is smoothed by cubic spline interpolation, and the entropy change gradient and maximum information coefficient are used to analyze the coupling relationship between each monitored mode, and this is used as an important basis for identifying disaster precursors.

[0022] In one solution, in the data preprocessing and feature calculation steps, wavelet denoising and bandpass filtering are used to suppress noise for acoustic emission and microseismic signals, respectively, and multi-order smoothing algorithms and temperature drift compensation technology are used for stress and displacement data to achieve high-precision preprocessing of various monitoring data.

[0023] In one scheme, in the disaster point detection and precursor identification steps, a method combining multimodal mutation detection and historical case comparison is adopted. Only when abnormalities appear in multiple monitoring modes at the same time and are highly similar to historical disaster samples, is it determined to be a valid disaster point, significantly reducing the probability of false alarms and missed alarms.

[0024] In one solution, during the risk assessment and early warning output step, the system can automatically spatially associate risk events with the tunnel's three-dimensional geological model and BIM model, and display the risk level and precursor location in real time on a visualization platform through color grading and dynamic labeling, allowing managers to promptly grasp the risk distribution and take targeted measures.

[0025] On the other hand, a prediction system for rock fracture disasters in tunnel engineering is provided, and the system is applicable to the method described. The system includes: a multimodal monitoring data acquisition and synchronization module, which is used to deploy multiple types of sensors at key locations of the tunnel to achieve the unification of the spatiotemporal benchmarks of multi-source heterogeneous data; a data preprocessing and feature calculation module, which is used to preprocess the various types of collected original monitoring data and generate a feature time series matrix; a multimodal information entropy evolution analysis module, which is used to analyze the nonlinear correlation between multimodal features; a disaster point detection and precursor identification module, which is used to determine whether there are disaster precursor events; a risk assessment and warning output module, which is used to assess the risk level of the tunnel area and output a warning after detecting high risk and emergency conditions.

[0026] Beneficial effects of the present invention:

[0027] 1. The present invention combines multimodal monitoring data acquisition and synchronization, data preprocessing and feature calculation, multimodal information entropy evolution analysis, disaster point detection and precursor identification, risk assessment and early warning output, etc., comprehensively processes multimodal monitoring data, accurately predicts and promptly warns of rock stratum rupture disasters in tunnel engineering, and greatly improves the accuracy and timeliness of early warning of rock stratum rupture disasters in tunnel engineering.

[0028] 2. The present invention realizes the unification of time and space benchmarks for multi-source heterogeneous data, adopts the precise time protocol for master-slave clock synchronization, and combines the fiber optic transmission delay compensation algorithm to eliminate link delay differences, ensuring the time axis alignment of each modal data, thereby improving the accuracy and efficiency of data processing.

[0029] 3. In the process of disaster point detection and precursor identification, the present invention combines multiple methods such as single-modal mutation detection, multi-modal collaborative judgment, joint entropy extreme value detection and causal test to determine whether there are disaster precursor events, significantly reducing the probability of false alarms and missed alarms.

[0030] 4. Compared with existing technologies, the method and system for predicting rock fracture hazards in tunnel projects proposed in this invention offer significantly different technical features and outstanding innovation in terms of monitoring targets, data collection, processing and analysis, and early warning decision-making. Regarding monitoring targets, this invention focuses on rock fracture hazards, collecting multimodal data through multiple types of sensors and achieving a unified spatiotemporal benchmark. Existing technologies focus on different types of hazards separately, resulting in weak synchronization and heterogeneous data processing. Regarding data processing, this invention uses sliding window probability distribution modeling and information theory indicators to analyze the nonlinear correlations of multimodal features, providing more targeted preprocessing of data from different modalities. Existing methods have different focuses but lack the uniqueness of this invention. Regarding early warning decision-making, this invention conducts a five-level risk assessment based on a multidimensional risk quantification model, combines multiple methods to detect disaster points, reduces false alarms and missed alarms, triggers multi-channel warnings when risks are high, and incorporates a feedback learning mechanism. Existing technologies lack the comprehensiveness and accuracy of early warning mechanisms compared to this invention. With its innovative multimodal monitoring and data fusion, precise disaster point detection and precursor identification, and comprehensive and efficient risk assessment and early warning, this invention provides a superior solution for tunnel project safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Flow chart of the method of the present invention;

[0032] Figure 2 This is a system block diagram of the present invention. DETAILED DESCRIPTION

[0033] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate exemplary embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present invention.

[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as those understood by those skilled in the art to which the present invention pertains. The terms used in the present specification are for the purpose of describing specific embodiments only and are not intended to limit the present invention. To facilitate understanding of the present invention, a more comprehensive description of the present invention will be provided below with reference to the accompanying drawings. Typical embodiments of the present invention are shown in the drawings. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present invention.

[0035] like Figure 1 As shown, a method for predicting rock fracture disasters in tunnel engineering specifically includes:

[0036] Step 1. Multimodal monitoring data acquisition and synchronization: Deploy multiple types of sensors (such as acoustic emission, microseismic, stress, displacement, and ground temperature) at key locations in the tunnel to collect multimodal monitoring data. All data are time-stamped to achieve time-series synchronization of multi-source data.

[0037] The detailed implementation of multimodal monitoring data acquisition and synchronization involves three key aspects: sensor network construction, data synchronization mechanisms, and hardware integration. First, based on geological survey reports and numerical simulation results, a multimodal sensor array was deployed in high-risk areas of the tunnel (such as fault zones, high-stress areas, and weak surrounding rock boundaries). Acoustic emission sensors (using broadband piezoelectric ceramic probes with a frequency range of 1-100kHz) were installed every 10 meters on the vault and side walls. Microseismic monitoring employed three-component geophones (sensitivity >30V / m / s) arranged in a grid pattern (spacing ≤20 meters). Stress monitoring utilized fiber Bragg grating sensors (accuracy ±0.5%FS) embedded in the primary support layer to form circumferential and longitudinal cross-points. Displacement monitoring employed a laser rangefinder (range ±50mm, resolution 0.01mm) and a total station to form a three-dimensional monitoring network. Ground temperature monitoring employed a distributed fiber-optic temperature measurement system (spatial resolution 1m, temperature accuracy ±0.5°C).

[0038] All sensors are connected to the edge computing gateway through industrial-grade fieldbuses (such as CAN or Modbus), and IEEE1588 Precision Time Protocol (PTP) is used to achieve microsecond-level clock synchronization. GPS / Beidou dual-mode timing modules are deployed at the tunnel entrance as the main clock source. Each sub-node uses the optical fiber transmission delay compensation algorithm (compensation formula: , where T1 is the sending timestamp, T2 is the receiving timestamp, T3 is the response sending timestamp, and T4 is the response receiving timestamp) to eliminate the transmission link delay difference, To compensate for the delay, the data acquisition terminal is equipped with a multi-channel synchronous sampling card (such as the NI PXIe-6368, with a sampling rate of ≥1MS / s). Hardware trigger signals ensure the time axis alignment of each modal data. After preprocessing, the raw data stream is encapsulated into an HDF5 format data packet with a unified timestamp (format: UTC + geological time scale, accuracy of ±1ms). Ultimately, it is uploaded to the central server via the tunnel's dedicated 5G slice network, achieving the unification of the spatiotemporal benchmarks of multi-source heterogeneous data and providing strictly synchronized multimodal data streams for subsequent feature extraction.

[0039] Step 2. Data preprocessing and feature calculation: De-noise and standardize all monitoring data. Using a sliding window method, calculate the characteristic parameters of each mode (such as acoustic emission energy, number of microseismic events, and stress fluctuation rate) by time period.

[0040] The implementation process of data preprocessing and feature calculation needs to be customized according to the characteristics of different monitoring modes. For the acoustic emission signal, the wavelet threshold denoising algorithm is first used to eliminate the equipment noise: the Daubechies 8 wavelet basis is selected for 5-layer decomposition, and the improved SUREShrink threshold function (formula: ,in is the noise standard deviation, N is the signal length, Soft threshold processing is performed on each scale coefficient (with as threshold) to obtain a denoised waveform after reconstruction; microseismic data is bandpass filtered (1-200 Hz) combined with the STA / LTA (short-time average / long-time average) algorithm to achieve event detection (trigger condition: Lasts 0.5 seconds. Short-term average, Long-term average). Stress monitoring data is smoothed using Kalman filtering, and the state equation is: , The state vector at time t, State transition matrix, Process noise. The observation equation is , The observation vector at time t, is the observation matrix, is the observation noise. The covariance matrix of process noise w and observation noise v is obtained through historical data training. Displacement data needs to compensate for temperature drift and establish a temperature-displacement compensation model ,in is the coefficient of thermal expansion, determined by laboratory calibration; Current temperature, Reference temperature.

[0041] The standardization process uses the improved Z-score method: ,in is the median value, Raw data, After normalization, The sliding window method adopts an overlapping design, and the window length w is dynamically adjusted according to the sampling rate (for example, a 10-second window corresponds to 1000 sample points at a 100Hz sampling rate), and the sliding step size is set to . Feature calculation: The acoustic emission energy passes through the root mean square of the signal in the window characterization, Acoustic emission energy, The number of sample points in the window, The signal value of the i-th sample point in the window; the number of microseismic events is the number of triggers in the statistical window , The number of microseismic events, Window length, , indicator function, if the condition is met, it is 1, if not, it is 0;; Stress fluctuation rate calculation standard deviation , Stress fluctuation rate, The number of sample points in the window, The stress value of the i-th sample point in the window, The average value of stress in the window; the displacement accumulation is calculated by window integration ,in is the instantaneous speed, Cumulative displacement, is the window length, time interval, Instantaneous velocity; linear regression coefficient for geothermal gradient calculation , geothermal gradient, is the window length, is the ground temperature value of the i-th sample point, is the time of the i-th sample point; all characteristic parameters are calculated and stored in the time series characteristic matrix to provide input for subsequent entropy analysis.

[0042] Step 3. Multimodal information entropy evolution analysis; Information entropy calculation: Within each time window, the Shannon entropy of each modal feature is calculated to reflect the system's degree of disorder and uncertainty. Joint entropy and mutual information analysis: The joint entropy and mutual information between different modalities are calculated to quantify the coupling and correlation changes between the various monitored variables. Entropy evolution curve construction: The information entropy, joint entropy, and mutual information are plotted over the time series.

[0043] The implementation process of multimodal information entropy evolution analysis relies on the time series characteristic matrix constructed in the early stage, and reveals the evolution law of the system state through entropy theory. , number of microseismic events , stress fluctuation rate etc.), firstly, dynamic probability distribution modeling is adopted: using sliding window as the basic unit, the range of each feature parameter is discretized into equal-width intervals, and count the frequencies of each interval within the window and calculate the probability mass function ; The probability of the i-th interval, The frequency of the jth interval, Total number of intervals.

[0044] Shannon entropy calculation uses improved weight formula ,in, Shannon entropy, Weight coefficient Used to enhance the entropy contribution of abnormal intervals. Joint entropy calculation is achieved by constructing a two-dimensional joint probability distribution: for any two modal features X and Y, their joint value range is divided into Grid, statistical joint frequency Get the joint probability , then the joint entropy , Joint entropy, Joint probability. Mutual information analysis introduces standardized mutual information index , Normalized mutual information index, which eliminates the dimension difference of entropy value, Shannon entropy of mode X, Shannon entropy of mode Y, The joint entropy of mode X and mode Y effectively characterizes the coupling strength between modes. For high-dimensional modal spaces (such as five-mode systems), the maximum information coefficient (MIC) is used to evaluate the nonlinear correlation of multiple variables: ,in, The maximum information coefficient, Mutual information, Gateway partitioning parameters, grid partitioning parameter related functions , the optimal partitioning method is searched through dynamic programming algorithm. In the entropy evolution curve construction stage, the entropy value sequence of the discrete time window is smoothed by using cubic spline interpolation method to generate a continuous evolution trajectory. For mutation point detection, the entropy gradient index is designed , Entropy gradient index, The entropy value at time t, The entropy value at time t−1, Time interval, when ( The innovation of the algorithm is reflected in two aspects: first, the introduction of dynamic weight factor Enhance the entropy representation ability of non-stationary signals; secondly, propose a multi-scale entropy correlation matrix (k is the number of modes, Multiscale entropy correlation matrix, The elements in the matrix represent the modes and modal The standardized mutual information metric (MHI) between the two systems is used to reveal the evolution of system-level stability through matrix eigenvalue spectrum analysis. The final output includes multi-dimensional visualizations such as Shannon entropy curves, mutual information heat maps, and joint entropy surfaces, providing a quantitative basis for tunnel stability assessment.

[0045] Step 4. Detection of catastrophic points and identification of precursors; Design a catastrophic point detection algorithm based on mutation theory: Utilize the entropy mutation criterion: When the information entropy or joint entropy experiences a sharp jump in a short period of time (i.e., the first-order derivative or second-order derivative exceeds the adaptive threshold), it is determined to be a catastrophic critical point; Introduce a multimodal collaborative mutation criterion: Only when the entropy changes of multiple modes mutate simultaneously can it be determined as an effective rupture precursor, reducing the false alarm rate.

[0046] Adopting the sliding window adaptive threshold algorithm, the sensitivity of the mutation criterion is automatically adjusted in combination with historical data.

[0047] The core of the disaster point detection and precursor identification algorithm is to build a multi-level mutation criterion system to capture the precursors of system instability through the spatiotemporal coupling characteristics of entropy dynamic characteristics. First, based on the Shannon entropy sequence generated in step 3 (m represents the mth mode), using the improved derivative mutation detection algorithm: calculate the first-order derivative for each mode With the second derivative ,in is the sliding window step size, The first-order derivative of the mth mode at time t is, The entropy value of the mth mode at time t+1, The entropy value of the mth mode at time t−1, Time interval. The second-order derivative of the mth mode at time t. The adaptive threshold setting uses the dynamic quantile method: based on the past L=24 hours of data, the moving quantile of the absolute value of the derivative is calculated As the critical threshold, when and A single-mode mutation alarm is triggered when 、 is the sensitivity adjustment coefficient, The first derivative of the mth mode at time t. The second derivative of the mth mode at time t.

[0048] Multimodal synergy criterion introduces modal coupling coefficient matrix , is the normalized mutual information between modalities i and j, when It is judged as an effective precursor when is the number of single-mode alarms triggered, k is the number of modes), this criterion quantifies the correlation strength between modes through mutual information and dynamically adjusts the minimum number of alarms required for collaborative mutation. Joint entropy mutation detection uses wavelet modulus maximum theory: for the joint entropy sequence Perform Mallat wavelet decomposition to the 4th layer, when the detail coefficient The local maximum value exceeds the historical mean If the mutation lasts for three sampling points, it is considered a system-level mutation.

[0049] The false alarm suppression algorithm designs a double verification mechanism: first, the KL divergence test is performed Verify the authenticity of the probability distribution mutation (the threshold is set to 0.4), where KL divergence of the probability distribution at time t and time t−1, The probability distribution at time t, The probability at time t-1 is respectively expressed, and the Granger causality test is then used to verify the temporal relationship of multimodal mutations (F statistic > 4.0). The algorithm's innovation is reflected in the integration of the triple criteria of derivative mutation, wavelet singularity detection, and probability distribution migration, and the establishment of a dynamic coordination mechanism based on mutual information weights. Its mathematical expression is:

[0050] ;

[0051] in is the standard normal distribution function, is the collaborative penalty factor, Probability of early warning, The first-order derivative of the mth mode at time t is, The mean of the first-order derivative of the mth mode, The standard deviation of the first-order derivative of the mth mode, Collaborative penalty factor, The normalized mutual information between modes i and j is calculated. The final output includes a warning signal with the disaster point timestamp, precursor confidence (0-1 range), and a list of dominant modes, providing quantitative support for engineering emergency decision-making.

[0052] Step 5. Risk Assessment and Warning Output: Based on the catastrophic point detection results, the rupture risk level for the current tunnel area is calculated. If a catastrophic point is detected, an alert is automatically triggered, and the corresponding spatiotemporal location and risk level are output.

[0053] The implementation process of risk assessment and early warning output is based on the spatiotemporal markers and confidence parameters output during the disaster point detection phase, and a multi-dimensional risk quantification model is used to implement tunnel stability graded early warning. The system first constructs a dynamic risk index , The dynamic risk index at time t, , , is the weight coefficient, is the confidence level of the precursor event, represents the entropy change gradient of each mode, is the KL divergence similarity between the current state and the historical disaster case library, and the weight coefficient Dynamic adjustment is made through the analytic hierarchy process. Risk levels are divided into five levels: When it is level Ⅰ (safe state), Level II (low risk), Grade III (moderate risk), Level IV (high risk), When a warning message is generated, the system automatically associates the spatial topology data in the 3D geological model, maps the risk event to the corresponding lining section of the tunnel BIM model, and visualizes the risk through color gradient rendering.

[0054] For risks above Level IV, a multi-channel early warning is triggered simultaneously: structured alarm information containing risk coordinates, dominant modal characteristics, and recommended treatment measures is pushed to the monitoring center; on-site audio and visual alarms are activated via IoT relay devices; and for areas at risk of cascading collapses, a protective support plan is automatically generated and sent to the construction robot fleet. The system has a built-in feedback learning mechanism that continuously tracks the frequency of microseismic events and energy release curves within 72 hours of the early warning. If the actual rupture intensity does not meet the predicted value, the sensitivity parameters for similar scenarios are automatically reduced; otherwise, the weight coefficients of the associated modes are increased, achieving dynamic optimization of the risk assessment model.

[0055] like Figure 2As shown, a prediction system for rock fracture disasters in tunnel engineering includes: a multimodal monitoring data acquisition and synchronization module, which is used to deploy multiple types of sensors at key locations in the tunnel to achieve the unification of the spatiotemporal benchmarks of multi-source heterogeneous data; a data preprocessing and feature calculation module, which is used to preprocess the various types of collected original monitoring data and generate a feature time series matrix; a multimodal information entropy evolution analysis module, which is used to analyze the nonlinear correlation between multimodal features; a disaster point detection and precursor identification module, which is used to determine whether there are disaster precursor events; a risk assessment and warning output module, which is used to assess the risk level of the tunnel area and output a warning after detecting high risk and emergency conditions.

[0056] The present invention focuses on rock fracture disasters in tunnel engineering, such as rock fracture and landslide, by deploying multiple types of sensors such as acoustic emission, microseismic, stress, displacement and ground temperature at key locations in the tunnel to collect multimodal monitoring data.

[0057] During data acquisition, this invention focuses on achieving a unified spatiotemporal basis for heterogeneous multi-source data. It employs a precise time protocol and a fiber transmission delay compensation algorithm to ensure the time axis alignment of each modality. Existing technologies have not yet achieved the level of this invention in terms of data acquisition synchronization and handling heterogeneous data types.

[0058] This method uses sliding window probability distribution modeling for each modal feature, calculates information theory metrics such as Shannon entropy, joint entropy, and mutual information, and analyzes the nonlinear correlations between multimodal features. Prior Art 1 uses multi-source information fusion technology to comprehensively identify adverse geology and constructs a three-dimensional geological model using multi-scale geological model theory and multi-scale topological reconstruction technology. Prior Art 2 constructs a data feature space for feature extraction and cluster analysis. Prior Art 3 predicts future time series monitoring data based on historical disaster warning monitoring information and calculates risk probabilities using a risk prediction model. Prior Art 4 does not explicitly mention methods such as information entropy analysis.

[0059] In data preprocessing, the present invention employs specific denoising and compensation techniques for different monitoring modalities, such as a wavelet threshold denoising algorithm for acoustic emission signals and a multi-order smoothing algorithm and temperature drift compensation for stress and displacement data. Existing technologies are relatively weak in terms of targeted data preprocessing and method diversity.

[0060] This invention utilizes a variety of sensors for multimodal monitoring, unifying the temporal and spatial benchmarks of heterogeneous data from multiple sources, enabling a comprehensive and accurate reflection of the true state of the tunnel's rock formations. The multimodal information entropy evolution analysis method exploits the nonlinear correlations between modal features, providing a more precise basis for identifying disaster precursors—an innovation not previously addressed.

[0061] This invention employs multiple methods to identify disaster precursor events, combining single-modal mutation detection, multimodal collaborative criteria, joint entropy extreme value detection, and causal testing to form a comprehensive and accurate disaster point detection and precursor identification system, significantly reducing the probability of false positives and missed negatives. Existing technologies for disaster detection and precursor identification lack the multidimensionality and comprehensiveness of this invention.

[0062] The multidimensional risk quantification model of this invention combines multiple parameters to assess risk levels. It not only issues timely and accurate warnings, but also automatically pushes structured alarm information, generates protective support plans, and incorporates a feedback learning mechanism, improving the efficiency and safety of responding to rock fracture disasters in tunnel projects. In terms of disaster response, the early warning and decision-making mechanism of this invention is more comprehensive and efficient, better ensuring the safety of tunnel projects.

[0063] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0064] It should be understood that the detailed description of the technical solutions of the present invention using the preferred embodiments above is illustrative and not restrictive. A person skilled in the art, after reading the present specification, may modify the technical solutions described in the embodiments or replace some of the technical features therein with equivalents; such modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting rock fracture disasters in tunnel engineering, characterized in that: The method includes: Multimodal monitoring data acquisition and synchronization: Multiple types of sensors, including acoustic emission sensors, microseismic sensors, stress sensors, displacement sensors, and ground temperature sensors, are deployed in key locations in the tunnel. Multimodal sensor arrays are deployed in high-risk areas based on geological survey reports and numerical simulation results. All sensors are connected to the edge computing gateway via an industrial-grade fieldbus. Clock synchronization is achieved using a precise time protocol. Time synchronization is achieved through a master clock source and a fiber-optic transmission delay compensation algorithm to eliminate link delay differences. Data acquisition terminals are equipped with multi-channel synchronous sampling cards. Hardware trigger signals ensure the time axis alignment of each modal data. After preprocessing, the raw data stream is encapsulated into data packets with unified timestamps and uploaded to the central server via a dedicated network, achieving unified spatiotemporal benchmarks for multi-source heterogeneous data. Data preprocessing and feature calculation: The collected raw monitoring data are preprocessed by denoising, standardization, and drift compensation. The characteristics of acoustic emission energy, number of microseismic events, stress fluctuation rate, displacement accumulation, and geothermal gradient are extracted using a sliding window statistical method to generate a feature time series matrix. Multimodal Information Entropy Evolution Analysis: We use a sliding window probability distribution model to model each modal feature, calculate information theory indicators including Shannon entropy, joint entropy, and mutual information, extract the entropy gradient through entropy evolution curves and smoothing, and use a multi-scale entropy correlation matrix and maximum information coefficient method to analyze the nonlinear correlation between multimodal features. Disaster point detection and precursor identification: Based on single-modal mutation detection, multi-modal collaborative judgment, joint entropy extreme value detection and causal test methods, it determines whether there is a disaster precursor event and outputs the spatiotemporal label of the disaster point, the precursor confidence and the dominant modal characteristics; Risk assessment and early warning output: Based on the results of disaster point detection, combined with the confidence level of precursor events, entropy change gradient and historical case similarity parameters, a multi-dimensional risk quantification model is used to assess the risk level of the tunnel area. The risk level is divided into five levels, and the 3D geological model and BIM model are automatically associated to achieve spatial visualization and positioning of risk events. In high-risk and emergency situations, multi-channel early warnings are triggered simultaneously, including information push, sound and light alarms, automatic generation of protection support plans and feedback learning mechanisms, and dynamic optimization of model parameters.

2. The method for predicting rock fracture disasters in tunnel engineering according to claim 1, characterized in that: The layout of the multimodal sensor array includes: installing acoustic emission sensors every 10 meters on the tunnel vault and side walls, using three-component detectors for microseismic sensors arranged in a grid pattern at intervals of no more than 20 meters, using fiber optic Bragg grating sensors for stress sensors to form circumferential and longitudinal cross-measurement points, using laser rangefinders and total stations to form a three-dimensional monitoring network for displacement sensors, and using a distributed fiber optic temperature measurement system for ground temperature sensors arranged in the tunnel surrounding rock to achieve all-round and multi-dimensional monitoring of high-risk areas.

3. The method for predicting rock fracture disasters in tunnel engineering according to claim 1, characterized in that: The data synchronization adopts an industrial-grade field bus to connect with the edge computing gateway, and uses the precise time protocol for master-slave clock synchronization. The master clock source adopts the satellite timing module deployed at the tunnel entrance. Each sub-node eliminates the delay through the optical fiber link delay compensation algorithm. All monitoring data acquisition terminals use multi-channel synchronous sampling cards, and hardware triggering is used to ensure the time consistency of various data collections. Finally, the data is encapsulated into data packets with unified timestamps for remote transmission and storage.

4. The method for predicting rock fracture disasters in tunnel engineering according to claim 1, characterized in that: In the risk assessment and early warning output step, after detecting high-risk and emergency conditions, structured alarm information containing risk coordinates, dominant modal characteristics and recommended disposal measures can be automatically pushed to the monitoring center, and the on-site sound and light alarm device can be activated through the Internet of Things relay equipment. At the same time, for areas where there is a risk of chain collapse, a protective support plan is automatically generated and sent to the construction robot queue. It also has a feedback learning mechanism to automatically optimize model parameters according to the frequency of subsequent microseismic events and energy release.

5. The method for predicting rock fracture disasters in tunnel engineering according to claim 1, characterized in that: In the multimodal information entropy evolution analysis step, a sliding window is used to model the probability distribution of each modal feature, and information theory indicators including Shannon entropy, joint entropy and mutual information are calculated. The entropy curve is smoothed by cubic spline interpolation, and the entropy change gradient and maximum information coefficient are used to analyze the coupling relationship between each monitoring mode, which is used as an important basis for identifying disaster precursors.

6. The method for predicting rock fracture disasters in tunnel engineering according to claim 1, characterized in that: In the data preprocessing and feature calculation steps, wavelet denoising and bandpass filtering are used to suppress noise for acoustic emission and microseismic signals, respectively, and a multi-order smoothing algorithm and temperature drift compensation technology are used for stress and displacement data to achieve high-precision preprocessing of various monitoring data.

7. The method for predicting rock fracture disasters in tunnel engineering according to claim 1, characterized in that: In the disaster point detection and precursor identification steps, a method combining multimodal mutation detection and historical case comparison is adopted. Only when abnormalities appear simultaneously in multiple monitoring modes and are highly similar to historical disaster samples, is it determined to be a valid disaster point, which significantly reduces the probability of false alarms and missed alarms.

8. The method for predicting rock fracture disasters in tunnel engineering according to claim 1, characterized in that: In the risk assessment and early warning output step, risk events can be automatically spatially associated with the tunnel's three-dimensional geological model and BIM model, and risk levels and precursor locations can be displayed in real time on a visualization platform through color grading and dynamic labeling, making it easier for managers to grasp risk distribution in a timely manner and take targeted measures.

9. A prediction system for rock fracture disasters in tunnel engineering, wherein the system is applicable to the method according to any one of claims 1 to 8, characterized in that: The system includes: a multimodal monitoring data acquisition and synchronization module, which is used to deploy multiple types of sensors in key parts of the tunnel to achieve the unification of the spatiotemporal benchmarks of multi-source heterogeneous data; a data preprocessing and feature calculation module, which is used to preprocess the various types of collected original monitoring data and generate a feature time series matrix; a multimodal information entropy evolution analysis module, which is used to analyze the nonlinear correlation between multimodal features; a disaster point detection and precursor identification module, which is used to determine whether there are disaster precursor events; and a risk assessment and warning output module, which is used to assess the risk level of the tunnel area and output a warning after detecting high risk and emergency conditions.

Images