A method for dynamically evaluating stability of surrounding rock of a tunnel

Abstract

The application relates to a tunnel surrounding rock stability dynamic evaluation method and relates to the field of tunnel engineering safety monitoring, and comprises the following steps: constructing a multi-dimensional evaluation index system and designing a monitoring scheme; synchronously collecting multi-source original data and performing pretreatment; adopting a three-level fusion architecture to fuse the pretreated multi-source data, extracting core characteristic parameters, wherein the information conflict among multiple indexes is processed based on an improved evidence theory model; adopting a principal component analysis method to perform dimension reduction processing on the core characteristic parameters; constructing a deep learning evaluation model to predict a surrounding rock stability comprehensive evaluation index and divide a grade; predicting a future stability change trend through a stability trend prediction model, identifying a mutation node; and starting a graded early warning response according to the evaluation result and the trend prediction. The scheme provided by the application realizes deep layered fusion of multi-source data, effectively solves the information conflict problem, and can dynamically and accurately evaluate the surrounding rock stability.

Description

Technical Field

[0001] This application relates to the field of tunnel engineering safety monitoring technology, and in particular to a method for dynamic assessment of the stability of tunnel surrounding rock. Background Technology

[0002] Tunnel engineering is widely used in transportation, water conservancy, energy and other fields. The stability of the surrounding rock directly determines the safety of tunnel construction and the reliability of long-term operation. Under the influence of multiple factors such as excavation disturbance, groundwater seepage, geological tectonic activity and external loads, the surrounding rock of tunnels is prone to deformation, collapse and water inrush, which seriously threaten the lives of construction workers and cause huge economic losses.

[0003] Currently, methods for assessing the stability of tunnel surrounding rock mainly include traditional geological analysis, single monitoring index assessment, and numerical simulation. Traditional geological analysis relies on engineers' experience to make qualitative judgments on the geological conditions of the surrounding rock, which is highly subjective, lacks precision, and is difficult to adapt to the dynamic assessment needs in complex geological environments. Single monitoring index assessment can only reflect the local state of the surrounding rock and cannot comprehensively capture the overall stability characteristics of the surrounding rock, which is prone to bias in assessment results due to incomplete data. Although numerical simulation can achieve quantitative analysis of the mechanical behavior of the surrounding rock, the selection of model parameters depends on empirical assumptions, which differ from actual geological conditions, and it is difficult to integrate field monitoring data in real time and dynamically update the assessment results.

[0004] With the development of sensing technology, it has become possible to acquire various types of monitoring data such as strain, displacement, and seepage pressure. However, existing technologies lack effective multi-source data fusion mechanisms, resulting in problems such as data redundancy, information conflicts, and unreasonable weight allocation. This leads to low utilization of monitoring data, making it impossible to fully explore the evolution law of surrounding rock stability behind the data, and making it difficult to achieve accurate and real-time dynamic assessment.

[0005] Therefore, there is an urgent need for a method for assessing the stability of tunnel surrounding rock that can integrate multi-source monitoring data, scientifically fuse information, and dynamically update assessment results. Summary of the Invention

[0006] To address or partially address the problems existing in related technologies, this application provides a dynamic assessment method for the stability of tunnel surrounding rock, aiming to solve the problems of insufficient fusion of multi-source monitoring data and improper handling of information conflicts in existing tunnel surrounding rock stability assessments, which lead to inaccurate assessments.

[0007] This application provides a method for dynamic evaluation of the stability of surrounding rock in tunnels, including: Construct a multi-dimensional evaluation index system that includes mechanical, deformation, environmental, and geological indicators, and design a monitoring scheme based on the tunnel's geological conditions and risk distribution; Multi-source raw data are collected synchronously through deployed monitoring equipment, and the collected raw data is cleaned, standardized, fused, preprocessed, and its quality is assessed. A three-level fusion architecture is adopted to fuse preprocessed multi-source data and extract core feature parameters. This includes fusing data from multiple devices for the same monitoring indicator using a weighted average method; cross-validating geological radar data with borehole sampling analysis data to correct the surrounding rock integrity coefficient and joint and fracture development parameters in the geological indicators; handling information conflicts among multiple indicators based on an improved evidence theory model to extract core feature parameters; and using principal component analysis to reduce the dimensionality of the extracted core feature parameters. Construct a deep learning evaluation model to predict the comprehensive evaluation index of surrounding rock stability based on the core feature parameters after dimensionality reduction, and classify the surrounding rock stability level. Based on the core feature parameters after dimensionality reduction, the stability trend prediction model is used to predict the change trend of the surrounding rock stability level over a certain period of time and identify the nodes of stability abrupt change. Based on the assessed stability level and trend prediction results, the corresponding early warning response is initiated, and targeted engineering management recommendations are generated.

[0008] Optional, improved models of evidence theory include: A reliability coefficient based on a comprehensive data quality index is introduced to modify the original basic probability allocation function; An adaptive conflict coefficient based on the similarity between evaluation metrics is constructed to replace the globally fixed conflict coefficient; Optimize the evidence synthesis rules and add a conflict information redistribution mechanism to allocate conflict information to each proposition according to the reliability coefficient and similarity of the indicators.

[0009] Optionally, information conflicts among multiple indicators can be addressed using an improved evidence theory model, including: Using the five-level stability classification of surrounding rock as the identification framework, and based on the deviation of the monitoring data of each assessment indicator from the preset stability critical value range, the basic probability allocation function and confidence level of each indicator are calculated: Calculate the similarity and local conflict coefficients between each pair of evaluation indicators, and combine them with the indicator reliability coefficients to calculate the global conflict coefficient: When the global conflict coefficient reaches the preset threshold, it is determined to be a high conflict assessment indicator, and the conflict correction synthesis rule is activated; when the local conflict coefficient is lower than the preset threshold, it is synthesized according to the traditional DS rule.

[0010] Optionally, calculate the basic probability assignment function and confidence level for each indicator, including: Determine the critical value range for each indicator, and in conjunction with engineering specifications and tunnel geological conditions, divide the critical value range for each evaluation indicator into five levels of stability. Calculate the deviation of the indicator monitoring values Let the preprocessed data of index i be... The critical value range for the k-th level of stability is Corresponding to levels I-V, calculate Deviation from the k-th level interval : In the formula, The midpoint of the k-th interval. For normalization coefficients, ensure .

[0011] Calculate the original basic probability assignment For single-element propositions of level k in the recognition framework, the original basic probability assignment (BPA) is: empty set Basic probability allocation Global proposition Basic probability allocation ; Based on the reliability coefficient, the basic probability allocation (BPA) is corrected, combined with the indicator reliability coefficient. The original basic probability assignment (BPA) is modified to obtain the final basic probability assignment function: Calculate the confidence level based on the degree of deviation. Indicator i represents the degree of confidence in proposition A, based on the modified basic probability assignment (BPA), and is divided into single-element propositions and compound propositions: The confidence level of a single-element proposition corresponds to a single stability level: The confidence level of a composite proposition corresponds to multiple stability levels. Let's assume a composite proposition... ,but: Global confidence reflects the degree of confidence that an indicator has in the overall stability. The closer the confidence value is to 1, the higher the degree of support that the indicator provides for the corresponding stability level; the smaller the deviation, the higher the confidence level.

[0012] Optionally, calculate the global conflict coefficient, including: First, calculate the similarity of each pair of evaluation indicators and the adaptive conflict coefficient, then calculate the global conflict coefficient, and finally perform weighted synthesis of evaluation indicators based on the conflict coefficient, while simultaneously redistributing conflict information. Similarity between pairwise evaluation indicators and local conflict coefficient : Set up a recognition framework There are n evaluation indicators, and the basic probability assignments (BPA) for the i-th and j-th evaluation indicators are respectively... Evaluation index similarity : The closer the value is to 1, the higher the consistency between evaluation indicators i and j, and the smaller the conflict. Local conflict coefficient : The closer the value is to 1, the higher the degree of local conflict between evaluation indicators i and j; The global conflict coefficient K is based on the pairwise local conflict coefficients of n evaluation indicators. Combined with the reliability coefficient of the indicators Calculate the global conflict coefficient K, which reflects the overall degree of conflict among all evaluation indicators: .

[0013] Optional conflict-correcting composition rules include: Let the basic probability distribution BPA of the n evaluation indicators to be synthesized after correction be... The synthesized comprehensive basic probability distribution (BPA) is: For any proposition The synthesis rules are as follows: when hour: In the formula, Results of traditional DS evidence synthesis: Assign coefficients to conflicting information based on the reliability coefficient and similarity of the indicators: To evaluate the similarity of support for proposition A to index i; when hour: Composition rule constraints satisfy normalization ; Under high conflict, the proportion of conflict allocation items increases with the increase of K, making full use of conflict information; Under low conflict conditions, the conflict assignment term approaches 0, and the synthesis result is consistent with the DS rule, ensuring method compatibility.

[0014] Optionally, extract core feature parameters, including: The core characteristic parameters after fusion include comprehensive deformation rate, mechanical response coordination degree, comprehensive environmental impact coefficient, and surrounding rock structure integrity index, which are extracted based on the physical meaning of the evaluation index synthesis results and the original monitoring index. Comprehensive deformation rate: The weighted sum of the surrounding rock convergence displacement rate, the crown settlement rate, and the sidewall horizontal displacement rate is allocated according to the importance of the project. Mechanical response coordination degree is calculated based on the normalized surrounding rock stress, support structure strain and anchor bolt axial force data. The deviation rate between the corresponding index and the mechanical coordination benchmark value is calculated and weighted. The weights are allocated according to the stress priority of the support system. The comprehensive environmental impact coefficient is a weighted sum of normalized groundwater seepage pressure, cave temperature and humidity and geological tectonic activity data, with the weights allocated according to the degree of environmental impact. The surrounding rock structural integrity index is a weighted sum of the normalized surrounding rock integrity coefficient, the reverse normalized joint and fissure development density, and the surrounding rock lithology category data, with the weights allocated according to the importance of geological characteristics.

[0015] Optionally, principal component analysis can be used to reduce the dimensionality of the extracted core feature parameters, including: Construct the feature parameter matrix and standardize it: Suppose that after feature layer fusion, we obtain a group of sample data, and the core feature parameters are: Construct a sample matrix : Standardize matrix X to eliminate the influence of dimensions and orders of magnitude, and obtain the standardized matrix. ; Calculate the covariance matrix of the standardized matrix: The covariance matrix reflects the degree of linear correlation between the characteristic parameters, and the elements The covariance of the i-th and j-th feature parameters is calculated using the following formula: In the formula ， Let be the standardized mean of the i-th and j-th feature parameters, and let the covariance matrix be a symmetric matrix. diagonal elements Let be the variance of the i-th feature parameter; Find the eigenvalues ​​and eigenvectors of the covariance matrix: Solve the characteristic equation p eigenvalues ​​are obtained The eigenvalues ​​represent the variance of the corresponding principal components; the larger the value, the more information the principal component contains. For each eigenvalue Solve the homogeneous linear equation system The corresponding feature vectors are obtained. The eigenvectors are normalized, and the normalized eigenvectors are principal component load vectors. Screening principal component factors: Calculate the variance contribution rate of the p-th principal component. : Calculate the cumulative variance contribution rate : Find the smallest p such that The corresponding p principal components are the core factors after dimensionality reduction; the remaining ones are discarded. One principal component is used to reduce the dimensionality of b-dimensional features to p-dimensionality; Calculate the principal component scores after dimensionality reduction: Suppose that p principal components are selected, and the corresponding normalized eigenvectors are... Construct the principal component loading matrix Then the principal component score matrix corresponding to the standardized matrix Z for: The principal component score matrix Y is the result after dimensionality reduction.

[0016] Optionally, a weighted average method can be used to fuse data collected from multiple devices for the same monitoring indicator, including: The weights are determined based on the measurement accuracy of the equipment; the higher the accuracy, the greater the weight. The fusion value is the weighted sum of the data collected by each equipment and its weight. Cross-validation of ground-penetrating radar data and borehole sampling analysis data was used to correct the surrounding rock integrity coefficient and joint and fracture development parameters in the geological indices, including: The correction coefficient is determined based on the complexity of the geological conditions, and the correction coefficient is used to weight and fuse the ground-penetrating radar detection data and borehole sampling analysis data.

[0017] Optionally, a multi-dimensional evaluation index system can be constructed, including mechanical, deformation, environmental, and geological indicators, such as: The mechanical indicators include surrounding rock stress, support structure strain, and anchor bolt axial force; data are collected using fiber optic stress sensors, fiber optic strain sensors, and anchor bolt axial force sensors; the sensors are closely fitted to the surrounding rock / support structure, with 3-5 measuring points per section in key areas and 2-3 measuring points per section in general areas; Deformation indicators include surrounding rock convergence displacement, crown settlement, and sidewall horizontal displacement rate. Monitoring and data collection are carried out using laser displacement gauges, total stations, and convergence meters. One laser displacement gauge is installed on the crown, and one to two laser displacement gauges are installed on each side of the sidewall. Convergence meters are installed at the key convergence sections of the surrounding rock. Environmental indicators include groundwater seepage pressure, temperature and humidity inside the cave, and geological tectonic activity; among which, geological tectonic activity includes fault creep and ground motion; data are collected using seepage pressure sensors, temperature and humidity sensors, ground motion acceleration sensors, and fault creep monitoring instruments; seepage pressure sensors are buried in groundwater-rich areas, and one temperature and humidity sensor and one ground motion acceleration sensor are deployed at each monitoring section. Geological indicators include the lithology of the surrounding rock, integrity coefficient, and density of joints and fissures. Geological exploration and analysis are carried out using ground-penetrating radar, borehole sampling equipment, and rock mechanics testing instruments. Ground-penetrating radar antennas are deployed around the tunnel, and a sampling point is set up every 50m for borehole sampling.

[0018] The technical solution provided in this application may include the following beneficial effects: By constructing a multi-dimensional assessment system encompassing mechanical, deformation, environmental, and geological indicators, and adopting a three-level fusion architecture of data-feature-decision, the system achieves consistent fusion of data from multiple devices for the same indicator through cross-validation using weighted average method and ground-penetrating radar / borehole data. Based on an improved evidence theory model, an indicator reliability coefficient and an adaptive conflict coefficient are introduced to effectively handle information conflicts among multiple indicators and avoid the paradox of synthesis under high conflict. Principal component analysis is used to reduce dimensionality and eliminate data redundancy, fully exploring the value of multi-source heterogeneous data and significantly improving data utilization and the accuracy of assessment results. A deep learning assessment model is constructed to predict the comprehensive assessment index of surrounding rock stability in real time and classify it into levels. A trend prediction model is used to predict future stability changes and abrupt change nodes, overcoming the shortcomings of traditional methods such as reliance on experience and lag in parameter updates. This achieves dynamic and accurate assessment and trend early warning of surrounding rock stability, providing a scientific and quantitative decision-making basis for tunnel construction safety management and proactive disaster prevention.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0020] The above and other objects, features and advantages of this application will become more apparent from the more detailed description of exemplary embodiments thereof in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments thereof.

[0021] Figure 1 This is a schematic flowchart illustrating the dynamic evaluation method for the stability of surrounding rock in tunnels, as shown in the embodiments of this application. Figure 2 This is a schematic diagram of the tunnel monitoring section and monitoring point layout for the dynamic assessment method of tunnel surrounding rock stability shown in the embodiments of this application.

[0022] Attached reference numerals: 1-Tunnel body, 2-Key monitoring section, 3-General monitoring section, 4-Monitoring point, 5-Fiber optic grating sensor, 6-Laser displacement meter, 7-Pressure sensor, 8-Ground radar detection antenna. Detailed Implementation

[0023] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0024] The technical solutions of the embodiments of this application are described in detail below with reference to the accompanying drawings.

[0025] In some implementations, see Figure 1 A dynamic assessment method for the stability of surrounding rock in tunnels, comprising: S101. Construct a multi-dimensional evaluation index system that includes mechanical indicators, deformation indicators, environmental indicators and geological indicators, and design a monitoring scheme based on the tunnel's geological conditions and risk distribution; Specifically, the constructed multi-dimensional evaluation index system is divided into four major categories: mechanical, deformation, environmental, and geological indicators. Each indicator is equipped with a dedicated monitoring system and data acquisition equipment, and a unified wireless real-time transmission system is established. The overall system and data transmission process are shown in Table 1 below: A three-tiered monitoring system is adopted, consisting of a field acquisition layer, a data transmission layer, and a cloud processing layer. The field acquisition layer completes real-time acquisition of multi-source data, the data transmission layer enables wireless uploading and synchronization of data, and the cloud processing layer receives, stores, and preprocesses data, providing a foundation for subsequent fusion analysis.

[0026] Table 1. Specific Monitoring Configuration and Data Transmission for Each Indicator To ensure unified data transmission requirements, all monitoring devices are equipped with time synchronization functionality. Based on the NTP network time protocol, the time error is ≤1s. The cloud processing layer is equipped with a data receiving interface to uniformly receive data from each system and classify and store it according to monitoring section, device number, and acquisition time, ensuring that the data is traceable and aligned.

[0027] Based on the tunnel length, the complexity of geological conditions, and the distribution of risk areas, the monitoring point layout principle of "intensified monitoring in key areas + routine monitoring in general areas" is adopted: Monitoring scheme design: Monitoring sections are set up in key monitoring areas such as fault fracture zones, weak surrounding rock sections, and shallow buried sections of tunnel entrances, with a section spacing of 5-8m and 3-5 monitoring points set up in each section; monitoring sections are set up in general monitoring areas with a section spacing of 15-20m and 2-3 monitoring points set up in each section.

[0028] We select various monitoring devices, including fiber optic grating sensors, laser displacement gauges, pressure sensors, and ground-penetrating radar (for detecting within the surrounding rock), and all devices have real-time data transmission capabilities.

[0029] Set the monitoring frequency: 1-2 times / day for key areas and 3-5 times / day for general areas during the construction period; 1 time / 3 days for key areas and 1 time / 7 days for general areas during the operation period; increase the monitoring frequency during periods of high incidence of geological disasters or under abnormal working conditions.

[0030] S102. Simultaneously collect multi-source raw data through deployed monitoring equipment, and perform data cleaning, data standardization, data fusion preprocessing and data quality assessment on the collected raw data; Specifically, monitoring equipment synchronously collects raw data from multiple sources, including: Mechanical data: surrounding rock stress, support structure strain, anchor bolt axial force; Deformation data: surrounding rock convergence displacement, crown settlement, and sidewall horizontal displacement rate; Environmental data: groundwater seepage pressure, cavern temperature, and seismic acceleration; Geological data: The width and density of internal fractures in the surrounding rock are obtained through ground-penetrating radar detection, and the integrity coefficient of the surrounding rock is obtained by combining borehole sampling analysis.

[0031] During the data collection process, metadata such as data collection time, equipment number, and monitoring section location are recorded simultaneously to ensure data traceability.

[0032] Data preprocessing: The process involves four steps: data cleaning, data standardization, data fusion preprocessing, and data quality assessment. Different data types employ appropriate processing methods, as detailed below: Data cleaning: The Laida criterion (3σ criterion) is used to identify outliers, and linear interpolation is used to fill in missing data. The calculation process is consistent across different data sets. Single-index cleaning steps are as follows: Outlier removal: Suppose the original data sequence of a certain indicator is... Calculate the mean Sum of standard deviations σ: (1) (2) If data points satisfy These are identified as outliers and removed. Missing data imputation, let the missing data point be . Its adjacent valid data points are The collection times were respectively ,but: (3) Data cleaning adaptability: Mechanical / deformation data are continuous time-series data, and are processed directly according to the above methods; Environmental data, temperature and humidity are high-frequency continuous data, and ground motion is sudden data, so only the continuously acquired segments are cleaned; Geological data are static / quasi-static data, so no interpolation is needed if there are no missing data, outliers are directly removed, and the data is re-detected.

[0033] Data standardization: The Z-score standardization method was used to eliminate the influence of different units of measurement, and all indicator data were uniformly normalized to [the standard value]. Single indicator calculation process: Suppose that the data sequence after cleaning a certain indicator is... mean Standard deviation σ, standardized data : (4) Standardization parameters for each data set: The mean and standard deviation of mechanical data, deformation data, and environmental data are calculated independently; in geological data, the integrity coefficient is a dimensionless value of 0-1 and is directly normalized, and the fracture parameters are processed according to the Z-score.

[0034] Data fusion preprocessing: The core task is to achieve time synchronization and Kalman filtering to remove interference, which is carried out on all monitoring data. The specific process is as follows: Time synchronization and alignment are based on cloud-based NTP time, with each device collecting a timestamp. The cloud-based base time is The time deviation is The acquisition time for each data point is adjusted. This ensures that the timestamps of multi-source data from the same monitoring section at the same time are completely consistent.

[0035] Kalman filtering reduces measurement errors: The Kalman filter state equation and observation equation are constructed to filter the mechanical, deformation, and environmental data of time-series monitoring data. Taking the surrounding rock stress data as an example, the filtering process is as follows: Equations of state: Where A is the state transition matrix and B is the control matrix. To control the quantity, This is process noise; Observation equation: (where H is the observation matrix). To observe noise, the accuracy is determined by the equipment, such as R=0.02 for a fiber Bragg grating sensor; Recursive calculation: predict: Kalman gain: renew: final The data is the filtered effective data. Geological data is static and does not require Kalman filtering.

[0036] Data quality assessment: The calculation process for each indicator is as follows: Data integrity and signal-to-noise ratio are the two core metrics. Data that does not meet quality standards is removed. Data Integrity I ,in Indicates the amount of valid data. This indicates the theoretically required amount of data to be collected. Requirements. If the data is below the threshold, the monitoring section / indicator data will be collected again.

[0037] Signal-to-noise ratio (SNR): (5) In the formula, The signal power is the variance of the filtered data. The noise power is the difference in variance between the data before and after filtering. If the reading is below the threshold, the monitoring equipment or its location should be re-optimized.

[0038] S103. A three-level fusion architecture is adopted to fuse the preprocessed multi-source data and extract core feature parameters. This includes fusing data from multiple devices for the same monitoring indicator using a weighted average method, cross-validating geological radar detection data with borehole sampling analysis data, and correcting the surrounding rock integrity coefficient and joint and fracture development parameters in the geological indicators; handling information conflicts between multiple indicators based on an improved evidence theory model and extracting core feature parameters; and using principal component analysis to reduce the dimensionality of the extracted core feature parameters. Specifically, a weighted average method is used to fuse data collected from multiple devices for the same monitoring indicator. This data is then cross-validated using ground-penetrating radar data and borehole sampling analysis data to correct the surrounding rock integrity coefficient and joint / fracture development parameters in the geological indicators, including: Data fusion is conducted on two types of data: multi-device collected data and geological exploration and sampling data for the same monitoring indicator. The core methods used are weighted average and cross-validation correction. The specific calculation process is as follows: Data fusion from multiple devices for the same indicator: Weights are assigned based on the measurement accuracy of the equipment, with higher accuracy resulting in greater weights. Let a certain indicator be weighted by... Data was collected by multiple devices, with the following accuracy levels: The accuracy is the allowable error of the equipment, calibrated by the manufacturer, then the weight of a single device. : (6) Satisfy the weight normalization condition: .

[0039] Fusion value calculation: Suppose that at a certain monitoring point and at a certain time, the data collected and preprocessed by each device is as follows: Corresponding weight Then the data layer fusion value of this indicator : (7) Engineering example calculations: Taking displacement monitoring as an example, laser displacement gauges (accuracy) ), total station (accuracy) Displacement data were collected from the same measuring point and preprocessed as follows: ,but: Cross-validation correction of geological index data: Regarding the surrounding rock integrity coefficient and joint and fracture development density Using ground-penetrating radar data With borehole sampling analysis data Cross-validation, correction value calculate: (8) In the formula, This is a correction factor, with values ​​determined based on the complexity of geological conditions, assuming homogeneous lithology and no faults in simple geological formations. Medium-complex geology Complex geology of fault fracture zones and weak surrounding rocks ; like ,in, If the problem persists, re-exploration and sampling should be conducted to ensure data consistency.

[0040] Specifically, based on an improved evidence theory model, information conflicts among multiple indicators are addressed, and core feature parameters are extracted, including: Traditional Data Synthesis (DS) evidence theory is prone to paradoxes when dealing with highly conflicting data across multiple indicators. Specifically, the synthesized results under highly conflicting evidence do not conform to actual engineering logic. To address this issue, an improved evidence theory is constructed, with three core improvements, detailed below: This paper introduces a reliability coefficient based on a comprehensive data quality index to modify the original basic probability allocation function. Traditional evidence theory uses equal weights for each piece of evidence (evaluation index) in the synthesis process, failing to consider differences in the quality of the monitored data, such as data integrity and signal-to-noise ratio. This application assigns a reliability coefficient to each evaluation index based on the comprehensive data quality index after data preprocessing. The original Basic Probability Assignment (BPA) is revised to increase the evidence weight of high-quality indicators and reduce the interference of low-quality indicators. Data Quality Composite Index , For data integrity, Signal-to-noise ratio and reliability coefficient .

[0041] This application constructs an adaptive conflict coefficient based on the similarity between evaluation indicators to replace the globally fixed conflict coefficient. The traditional conflict coefficient K in evidence theory is a globally fixed value and cannot reflect the degree of local conflict between pieces of evidence. This application constructs an adaptive conflict coefficient by calculating the similarity between pairwise pieces of evidence (evaluation indicators). Furthermore, it calculates the global conflict coefficient K based on the local conflict coefficient, thereby achieving accurate quantification of the conflict level and avoiding overestimation or underestimation of the conflict.

[0042] This application optimizes the evidence synthesis rules by adding a conflict information redistribution mechanism, allocating conflict information to each proposition based on the reliability coefficient and similarity of the indicators. Traditional evidence theory discards all conflicting evidence, resulting in information loss. Instead of directly discarding conflicting information, this optimized synthesis rules allocate conflicting information to each proposition based on the evidence reliability coefficient and similarity, thus resolving the synthesis paradox under high conflict and fully utilizing the information from all evidence, improving the rationality and engineering adaptability of the fusion results.

[0043] An improved evidence theory model addresses information conflicts among multiple indicators, including: Using the five-level stability classification of surrounding rock as the identification framework, and based on the deviation of the monitoring data of each assessment indicator from the preset stability critical value range, the basic probability allocation function and confidence level of each indicator are calculated: The Basic Probability Assignment Function (BPA) is constructed, using a five-level system of surrounding rock stability as the identification framework. For each evaluation index i, a basic probability allocation function is constructed based on the degree of deviation between the monitoring data and the critical value. This represents the basic probability that the proposition regarding the stability of surrounding rock belongs to class A, specifically including: Determine the stable critical value range for each indicator: Based on engineering specifications and tunnel geological conditions, critical value ranges corresponding to five stability levels are defined for each assessment indicator. Taking the crown settlement S as an example, the critical value range is as follows: , , , , .

[0044] Calculate the deviation of the indicator monitoring values : Let the preprocessed data of index i be... The critical value range for the k-th level of stability is Corresponding to levels I-V, calculate Deviation from the k-th level interval : (9) In the formula, The midpoint of the k-th interval. For normalization coefficients, ensure .

[0045] Calculate the original basic probability assignment : For single-element propositions of level k in the recognition framework, the original basic probability assignment (BPA) is: (10) In the formula, the empty set Basic probability allocation Global proposition Basic probability allocation .

[0046] BPA based on reliability coefficient correction: Combining the reliability coefficient of the indicators The original basic probability assignment (BPA) is modified to obtain the final basic probability assignment function: (11) (12) Confidence level calculation based on the degree of deviation: Confidence Indicator i represents the degree of confidence in proposition A, based on the modified basic probability assignment (BPA), and is divided into single-element propositions and compound propositions: The confidence level of a single-element proposition corresponds to a single stability level: (13) The confidence level of a composite proposition corresponds to multiple stability levels. Let's assume a composite proposition... ,but: (14) Global confidence reflects the degree of confidence that an indicator has in the overall stability. (15) Engineering significance: The closer the confidence value is to 1, the higher the degree of support of the indicator for the corresponding stability level; the smaller the deviation, the higher the confidence, and vice versa.

[0047] A conflict coefficient-corrected evidence synthesis rule is introduced, which consists of three steps: first, calculating the pairwise similarity of evidence and the adaptive conflict coefficient; second, calculating the global conflict coefficient; and finally, performing weighted evidence synthesis based on the conflict coefficient, while simultaneously redistributing conflict information. The specific rules are as follows: Similarity between pairwise evaluation indicators and local conflict coefficient : Set up a recognition framework There are n evaluation indicators, and the basic probability assignments (BPA) for the i-th and j-th evaluation indicators are respectively... Evaluation index similarity : (16) The closer the value is to 1, the higher the consistency between evaluation indicators i and j, and the smaller the conflict. Local conflict coefficient : (17) The closer the value is to 1, the higher the degree of local conflict between evaluation indicators i and j; The global conflict coefficient K is based on the pairwise local conflict coefficients of n evaluation indicators. Combined with the reliability coefficient of the indicators Calculate the global conflict coefficient K, which reflects the overall degree of conflict among all evaluation indicators: (18) ,when When it is determined to be high-conflict evidence, conflict correction needs to be initiated; when At that time, it is synthesized according to the traditional DS rules.

[0048] Conflict-corrected rules for evidence synthesis: Let the basic probability distribution BPA of the n evaluation indicators to be synthesized after correction be... The synthesized comprehensive basic probability distribution (BPA) is: For any proposition The synthesis rules are as follows: when hour: (19) In the formula, Results of traditional DS evidence synthesis: (20) Assign coefficients to conflicting information based on the reliability coefficient and similarity of the indicators: (twenty one) To evaluate the similarity of support for proposition A to index i; when hour: (twenty two) Composition rule constraints satisfy normalization ; Under high conflict ( The proportion of conflict allocation items increases as K increases, making full use of conflict information; Under low conflict ( The conflict assignment term approaches 0, and the synthesis result is consistent with the traditional DS rule, ensuring method compatibility.

[0049] Extract the core feature parameters after fusion: The core characteristic parameters after fusion include comprehensive deformation rate, mechanical response compatibility, comprehensive environmental impact coefficient, and surrounding rock structure integrity index. These are all extracted based on the evidence synthesis results after feature layer fusion and the physical meaning of the original monitoring indicators, using a weighted synthesis method combined with engineering feature normalization. The specific extraction methods for each parameter are as follows: Overall deformation rate : Reflecting the overall deformation rate of the surrounding rock, the extraction steps involve integrating the results of three deformation indices: the convergence displacement rate of the surrounding rock, the settlement rate of the arch, and the horizontal displacement rate of the sidewalls. Let the combined value of the three deformation indices be the surrounding rock convergence displacement rate. , rate of arch settlement Horizontal displacement rate of sidewall ; Weighting based on project importance: Crown settlement rate Surrounding rock convergence rate Horizontal displacement rate of sidewall ,satisfy ; Calculate the overall deformation rate: (twenty three) Normalization process: , The maximum allowable deformation rate of the surrounding rock of the tunnel is determined by engineering specifications. .

[0050] Mechanical response coordination : This reflects the coordination of the mechanical response of the surrounding rock-support system. It integrates the results of three mechanical indices—surround rock stress, support structure strain, and anchor bolt axial force—to characterize the degree of matching between these indices. The extraction steps are as follows: Let the normalized value of the three mechanical indices be the surrounding rock stress. Support strain Anchor bolt axial force ,all ; Calculate the mechanical compatibility benchmark values ​​for each index. The deviation rate This indicates that the mechanical response is in an optimal state of coordination. ; Calculate the mechanical response compatibility: (twenty four) The weights are allocated according to the stress priority of the support system: support strain weight. Anchor bolt axial force weight Surrounding rock stress weight ; The closer the value is to 1, the more coordinated the mechanical response and the better the stability of the surrounding rock.

[0051] Comprehensive environmental impact coefficient : This reflects the comprehensive impact of the external environment on the stability of the surrounding rock. It integrates the results of three environmental indicators: groundwater seepage pressure, temperature and humidity inside the tunnel, and geological tectonic activity. The extraction steps are as follows: Let the normalized combined value of the three environmental indicators be the groundwater seepage pressure. Temperature and humidity inside the cave Geological tectonic activity All values ​​are ∈ [0,1], and the larger the value, the more significant the environmental impact. Weights are allocated based on the degree of environmental impact: Groundwater seepage pressure Geological tectonic activity Temperature and humidity inside the cave ; Calculate the comprehensive environmental impact coefficient: (25) The closer the value is to 1, the greater the adverse impact of the environment on the stability of the surrounding rock.

[0052] Surrounding rock structural integrity index : Reflecting the integrity of the surrounding rock's structure, the extraction steps involve integrating three geological indicators: surrounding rock integrity coefficient, joint and fracture development density, and surrounding rock lithology. Let the normalized combined value of the three geological indicators be the rock integrity coefficient. The higher the value, the more complete the joints and fissures, indicating a higher density of joint development. The smaller the value, the more complete the lithology of the surrounding rock. The values ​​are assigned according to the degree of hardness, with larger values ​​indicating better lithology, and all values ​​are ∈[0,1]. Inverse normalization of fracture density: The larger the value of all geological indicators, the more complete the surrounding rock structure. Weights are assigned based on the importance of geological features: Rock Integrity Coefficient lithological categories Crack density ; Calculate the structural integrity index of the surrounding rock: (26) The closer the value is to 1, the more complete the structure of the surrounding rock and the better its stability.

[0053] Core extraction principle: All parameters are normalized to... This ensures consistency between the input and the comprehensive evaluation index dimension used in subsequent stability assessments.

[0054] Specifically, principal component analysis is used to reduce the dimensionality of the extracted core feature parameters, including: Principal component analysis (PCA) was used to reduce the dimensionality of the four core feature parameters after feature layer fusion: comprehensive deformation rate, mechanical response compatibility, comprehensive environmental impact coefficient, and surrounding rock structure integrity index. The core objective was to eliminate data redundancy, select principal component factors with a contribution rate ≥85%, and improve the computational efficiency of the subsequent evaluation model. The specific processing steps included data standardization, covariance matrix calculation, eigenvalue and eigenvector solving, principal component selection, and principal component score calculation, as follows: Construct the feature parameter matrix and standardize it: Suppose that after feature layer fusion, we obtain a group of sample data, and the four core feature parameters are the comprehensive deformation rate. Mechanical response coordination Comprehensive environmental impact coefficient Rock structure integrity index Construct a sample matrix : (27) Standardize matrix X to eliminate the influence of dimensions and orders of magnitude, and obtain the standardized matrix. The standardization method is the Z-score method.

[0055] Calculate the covariance matrix of the normalized matrix. : The covariance matrix reflects the degree of linear correlation between the characteristic parameters, and the elements The covariance of the i-th and j-th feature parameters is calculated using the following formula: (28) In the formula ， Let be the standardized mean of the i-th and j-th feature parameters, and let the covariance matrix be a symmetric matrix. diagonal elements Let be the variance of the i-th feature parameter.

[0056] Find the eigenvalues ​​and eigenvectors of the covariance matrix: Solve the characteristic equation Four eigenvalues ​​were obtained. The eigenvalues ​​represent the variance of the corresponding principal components; the larger the value, the more information the principal component contains. For each eigenvalue Solve the homogeneous linear equation system The corresponding feature vectors are obtained. and normalize the feature vectors (satisfying) The normalized eigenvectors are principal component load vectors.

[0057] Screening principal component factors: Calculate the variance contribution rate of the p-th principal component. : (29) Calculate the cumulative variance contribution rate : (30) Filtering principle: Find the smallest p such that The corresponding p principal components are the core factors after dimensionality reduction. Discarding the remaining 4-p principal components results in low information contribution and high redundancy.

[0058] Engineering example: After PCA processing of the four feature parameters, the cumulative contribution rate of the first two principal components can usually reach more than 90%, meeting the requirement of ≥85%, thus reducing the dimensionality of the 4-dimensional features to 2-dimensional.

[0059] Calculate the principal component scores of the dimensionality-reduced data: Suppose that p principal components are selected, and the corresponding normalized eigenvectors are... Construct the principal component loading matrix Then the standardized matrix Z corresponds to the principal component score matrix of the data after dimensionality reduction. for: (31) The principal component score matrix Y is the result of PCA dimensionality reduction, which serves as the input data for subsequent deep learning evaluation models, thereby eliminating data redundancy and improving the efficiency of model training and evaluation.

[0060] The core requirements for PCA dimensionality reduction are: the cumulative variance contribution rate must be ≥85% to ensure that the data after dimensionality reduction retains most of the information of the original features; the principal component factors are linear combinations of the original feature parameters, which are orthogonal and eliminate the linear correlation between the original features; the principal component factors after dimensionality reduction have no actual physical meaning and are only used as feature vectors for model input, which does not affect the accuracy of the results of subsequent stability assessment.

[0061] S104. Construct a deep learning evaluation model, predict the comprehensive evaluation index of surrounding rock stability based on the core feature parameters after dimensionality reduction, and classify the surrounding rock stability level. Specifically, the constructed deep learning evaluation model is a deep feedforward neural network (DNN), which takes the principal component factors after PCA dimensionality reduction as input and the five levels of surrounding rock stability as output. The model has three hidden layers and uses the ReLU activation function and Adam optimizer to achieve accurate evaluation of surrounding rock stability, as detailed below: Model structure diagram: The model is a five-layer feedforward neural network structure consisting of an input layer, hidden layer 1, hidden layer 2, hidden layer 3, and output layer. The number of neurons and activation functions in each layer are configured as follows. The overall structure is fully connected, meaning that all neurons in adjacent layers are connected pairwise: Table 2 Core Functions and Configurations of Each Layer Activation function formula: ReLU activation function: This solves the gradient vanishing problem and improves model training efficiency. Softmax activation function: This transforms the output value into a probability value of 0-1, corresponding to the attribution probability of level 5 stability.

[0062] Model building process: Model building is carried out in five steps: dataset preparation, network structure construction, model training, model validation, and model optimization. The entire process is implemented using Python and the TensorFlow / PyTorch framework. The specific process is as follows: Dataset preparation: Data collection: Collect tunnel monitoring data under different geological conditions and construction stages, with a sample size of ≥5000 sets. Each set of data includes 4 core feature parameters, principal component factors after PCA dimensionality reduction, and the corresponding actual level of surrounding rock stability, which are determined by engineering experts and used as labels. Data partitioning: The dataset is randomly partitioned into 70% training set, 15% validation set, and 15% test set. The training set is used for learning model parameters, the validation set is used for tuning model hyperparameters, and the test set is used for final model performance evaluation. Tag coding: The five levels of surrounding rock stability, I-V, are individually coded, and the tag vectors are: I=[1,0,0,0,0], II=[0,1,0,0,0], III=[0,0,1,0,0], IV=[0,0,0,1,0], V=[0,0,0,0,1].

[0063] Network architecture setup: A five-layer fully connected neural network built on the TensorFlow / PyTorch framework, core configuration: Fully connected layer: A dense layer is used, with weights W and biases b set between adjacent layers, and linear calculation is performed. ; Activation functions: Add ReLU activation to the hidden layer and Softmax activation to the output layer; Loss function: Cross-entropy loss function is selected, formula: (32) In the formula, For the true label of the k-th sample, The probability value predicted by the model; Optimizer: The Adam optimizer is selected, with a learning rate of 0.001 and a weight decay coefficient of 0.0001. The learning rate is adaptively adjusted to improve the training convergence speed. Regularization: L2 regularization is added to the hidden layer to prevent overfitting of the model. The regularization coefficient is... .

[0064] Model training: Training parameter settings: Batch size = 32, Epochs = 100, Early Stopping mechanism: If the validation set loss value does not decrease for 10 consecutive epochs, stop training to avoid overfitting; Training process: Input the training set data into the model in batches, obtain the prediction results through forward propagation, calculate the loss function, calculate the gradient through backpropagation, update the model weights W and biases b using the Adam optimizer, and iterate repeatedly until training converges.

[0065] Model validation: The model's performance is validated using a training set and a validation set. The core evaluation metrics are accuracy, recall, and mean squared error. The formulas for calculating these metrics are: Accuracy (Acc): Number of correctly predicted samples / Total number of samples × 100%, with a requirement of ; Recall (Rec): Number of correctly predicted samples for a given level / Actual number of samples for that level × 100%. This is the percentage of each level that can be correctly predicted. ; Mean Square Error (MSE): ,Require .

[0066] Model optimization: If the model performance does not meet the above requirements, optimize it using the following methods: Adjust hyperparameters: Modify learning rate (0.0001-0.01), batch size (16 / 64), and number of hidden layer neurons (128-256). Add data augmentation: Slightly perturb the training set data, such as adding Gaussian white noise, to improve the model's generalization ability; Optimize network structure: Add a Dropout layer (dropout=0.2) to randomly drop some neurons and prevent overfitting.

[0067] The process of model prediction of the comprehensive assessment index of surrounding rock stability: After training, the deep learning model assesses the surrounding rock stability of new tunnel monitoring data. The calculation process involves six steps: data preprocessing → PCA dimensionality reduction → model forward propagation → probability value analysis → comprehensive evaluation index calculation → stability level determination. The specific process is as follows: The new data preprocessing involves cleaning, standardizing, fusion preprocessing, and quality assessment of newly collected multi-source monitoring data to obtain four qualified core characteristic parameters.

[0068] PCA dimensionality reduction involves substituting four core feature parameters into the trained PCA dimensionality reduction model, retaining the mean, standard deviation, covariance matrix, and eigenvectors of the training set, calculating the principal component scores, and obtaining the p-dimensional feature vectors of the model input.

[0069] Forward propagation of the model involves inputting a p-dimensional feature vector into the trained deep learning model and performing forward propagation calculations. The process is as follows: Input layer: ; Hidden layer 1: , for Weight matrix, It is a 256-dimensional bias vector; Hidden layer 2: , for Weight matrix, It is a 192-dimensional bias vector; Hidden layer 3: , for Weight matrix, It is a 128-dimensional bias vector; Output layer: , for Weight matrix, Given a 5-dimensional bias vector, we obtain a 5-dimensional probability vector. The probability of attribution corresponding to stability levels I-V. .

[0070] Probability value analysis involves analyzing the output probability vector P to find the maximum probability value. The corresponding level is the preliminary assessment level.

[0071] The comprehensive evaluation index is calculated based on the probability vector P, combined with the level coefficients of the five levels of stability: Level I = 0.1, Level II = 0.3, Level III = 0.5, Level IV = 0.7, and Level V = 0.9. The comprehensive evaluation index E is calculated using the following formula: (33) In the formula, Let be the rank coefficient for the i-th level. It is compatible with the comprehensive evaluation index range for stability level classification.

[0072] The final determination of the surrounding rock stability level is based on the comprehensive assessment index E and the five-level classification standard of this method. Level I (Extremely Stable): ; Level II (Stable): ; Level III (Basically Stable): ; Level IV (Unstable): ; Level V (Extremely Unstable): .

[0073] Output the evaluation results.

[0074] The model ultimately outputs two core results: a five-level rating of surrounding rock stability and a comprehensive evaluation index E, providing a direct basis for generating subsequent graded early warning and control recommendations.

[0075] S105. Based on the core feature parameters after dimensionality reduction, predict the trend of change in the stability level of the surrounding rock within a certain period of time through the stability trend prediction model, and identify the nodes of stability abrupt change. Specifically, a stability trend prediction model is constructed based on the Long Short-Term Memory (LSTM) network. By inputting historical fusion feature data, the model predicts the trend of changes in the stability level of the surrounding rock within the next 7-30 days and identifies nodes where stability changes abruptly.

[0076] S106. Based on the assessed stability level and trend prediction results, initiate the corresponding early warning response and generate targeted engineering control recommendations.

[0077] Specifically, the tiered early warning mechanism activates a five-level early warning response based on stability assessment results and trend forecasts: Level I warning (blue): Corresponds to stability levels I-II. Continue routine monitoring and submit assessment reports periodically. Level II warning (yellow): Corresponds to stability level III. Increase monitoring frequency: once a day during construction period, once every two days during operation period. Strengthen on-site inspections. Level III Warning (Orange): Corresponds to Stability Level IV. Activate the emergency monitoring plan, collect data once every 6 hours, suspend construction in dangerous areas, and reinforce the support structure. Level IV Warning (Red): Corresponds to Stability Level V. Immediately activate the emergency response plan, evacuate on-site personnel and equipment, close off the danger zone, and organize expert consultation. The assessment results and warning levels are updated every 24 hours based on real-time monitoring data to ensure the timeliness of warnings.

[0078] Based on the assessment results and surrounding rock characteristics, differentiated management and control recommendations are generated: Areas with abnormal mechanical response: It is recommended to use measures such as anchor cable reinforcement and shotcrete thickening to improve the bearing capacity of the support; For areas where deformation exceeds limits: it is recommended to add temporary supports, optimize the excavation step distance, shorten it to 1-2m, and control the unloading disturbance of the surrounding rock. In areas with abundant groundwater, it is recommended to adopt measures such as curtain grouting for water blocking and drainage hole diversion to reduce the impact of seepage pressure. For areas with developed fissures, it is recommended to use grouting to reinforce and fill the fissures, thereby improving the integrity of the surrounding rock.

[0079] In some implementations, corresponding to the aforementioned application function implementation method embodiments, this application also provides an embodiment of a dynamic evaluation method for the stability of surrounding rock in tunnels.

[0080] Taking a water conveyance tunnel project as the research object, the tunnel is 5km long and passes through complex geological areas such as fault fracture zones and weak surrounding rock sections.

[0081] S201. Construction of Evaluation Indicator System and Design of Monitoring Scheme: An evaluation index system was constructed, covering eight core indicators including surrounding rock stress, support strain, arch settlement, groundwater seepage pressure, and surrounding rock integrity coefficient. Key monitoring areas and general monitoring areas were divided. The key monitoring areas were the fault fracture zone and the shallow buried section of the tunnel entrance, with 20 monitoring sections spaced 6m apart and 4 monitoring points per section. The general monitoring areas had 30 monitoring sections spaced 18m apart and 2 monitoring points per section. Equipment such as fiber optic stress sensors, laser displacement gauges, seepage pressure sensors, and ground-penetrating radar antennas were selected. The monitoring frequency for key areas during the construction period was set at 2 times / day, and for general areas at 3 times / day.

[0082] S202. Multi-source monitoring data acquisition and preprocessing: Multi-source data were collected synchronously over three consecutive months using monitoring equipment, including surrounding rock stress (0.5-3.2 MPa), support structure strain (120-850 με), arch crown settlement (2-15 mm), groundwater seepage pressure (50-350 kPa), and surrounding rock fissure width (0.2-5 mm). The Laida criterion was used to remove three outliers from the stress data, and linear interpolation was used to fill in 1.2% of the missing data. Z-score standardization was applied to all index data, with a unified value range of [-1, 1]. Environmental interference was reduced using a Kalman filter algorithm, improving the signal-to-noise ratio from 18 dB to 25 dB, ensuring data quality.

[0083] S203, Multi-source data fusion and feature extraction: The weighted average method was used to fuse displacement data from laser displacement gauges and total stations at the same monitoring point, with weights of 0.65 and 0.35, respectively, to obtain the fused displacement value. Based on the improved evidence theory, a feature fusion model was constructed to correct information conflicts between indicators, reducing the conflict coefficient from 0.82 to 0.35. Four core feature parameters were extracted: comprehensive deformation rate, mechanical response coordination degree, comprehensive environmental impact coefficient, and surrounding rock structure integrity index. Through principal component analysis, two principal component factors were selected, with a cumulative contribution rate of 92%.

[0084] S204. Stability Dynamic Assessment and Trend Prediction: A deep learning evaluation model was constructed, and 3000 sets of training data were input for model training. The accuracy of the validation set reached 95.8%, and the mean square error was 2.3%. The comprehensive evaluation index of a certain monitoring section was calculated to be 0.68, which was judged as Level IV (unstable) based on the input of preprocessed fused feature data. Based on the LSTM model, it was predicted that the comprehensive evaluation index of the section would rise to 0.75 within the next 15 days, and the stability would continue to deteriorate, with a risk of collapse.

[0085] S205, Generation of Tiered Early Warning and Control Recommendations: Based on the assessment results and predicted trends, a Level III (Orange) warning was issued, and an emergency monitoring plan was activated. The monitoring frequency for this section was increased to once every 6 hours, and construction in the area was suspended. Targeted control recommendations were generated: adopt anchor cable reinforcement for the support structure, add temporary steel supports, optimize the excavation step distance to 1.5m, and simultaneously deploy drainage holes to reduce groundwater seepage pressure. The warning information and control recommendations were synchronized to all participating units through the data management platform, experts were organized to conduct on-site inspections, and special disposal plans were formulated. Seven days later, based on updated monitoring data, a reassessment was conducted, and the comprehensive assessment index of this section dropped to 0.52, the stability level was improved to Level III (basically stable), the warning level was adjusted to Level II (Yellow), and restricted construction was resumed.

[0086] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for dynamic evaluation of the stability of surrounding rock in tunnels, characterized in that, include: Construct a multi-dimensional evaluation index system that includes mechanical, deformation, environmental, and geological indicators, and design a monitoring scheme based on the tunnel's geological conditions and risk distribution; Multi-source raw data are collected synchronously through deployed monitoring equipment, and the collected raw data is cleaned, standardized, fused, preprocessed, and its quality is assessed. A three-level fusion architecture is adopted to fuse the preprocessed multi-source data and extract core feature parameters; This includes: using a weighted average method to fuse data collected from multiple devices for the same monitoring indicator; cross-validating ground-penetrating radar data with borehole sampling analysis data to correct the surrounding rock integrity coefficient and joint and fracture development parameters in the geological indicators; handling information conflicts among multiple indicators based on an improved evidence theory model to extract core feature parameters; and using principal component analysis to reduce the dimensionality of the extracted core feature parameters. Construct a deep learning evaluation model to predict the comprehensive evaluation index of surrounding rock stability based on the core feature parameters after dimensionality reduction, and classify the surrounding rock stability level. Based on the core feature parameters after dimensionality reduction, the stability trend prediction model is used to predict the change trend of the surrounding rock stability level over a certain period of time and identify the nodes of stability abrupt change. Based on the assessed stability level and trend prediction results, the corresponding early warning response is initiated, and targeted engineering management recommendations are generated.

2. The method for dynamic evaluation of tunnel surrounding rock stability according to claim 1, characterized in that, The improved evidence theory model includes: A reliability coefficient based on a comprehensive data quality index is introduced to modify the original basic probability allocation function; An adaptive conflict coefficient based on the similarity between evaluation metrics is constructed to replace the globally fixed conflict coefficient; Optimize the evidence synthesis rules and add a conflict information redistribution mechanism to allocate conflict information to each proposition according to the reliability coefficient and similarity of the indicators.

3. The method for dynamic evaluation of tunnel surrounding rock stability according to claim 1, characterized in that, The improved evidence theory model addresses information conflicts among multiple indicators, including: Using the five-level stability classification of surrounding rock as the identification framework, and based on the deviation of the monitoring data of each assessment indicator from the preset stability critical value range, the basic probability allocation function and confidence level of each indicator are calculated: Calculate the similarity and local conflict coefficients between each pair of evaluation indicators, and combine them with the indicator reliability coefficients to calculate the global conflict coefficient: When the global conflict coefficient reaches the preset threshold, it is determined to be a high conflict assessment indicator, and the conflict correction synthesis rule is activated; when the global conflict coefficient is lower than the preset threshold, it is synthesized according to the traditional DS rule.

4. The method for dynamic evaluation of tunnel surrounding rock stability according to claim 3, characterized in that, The calculation of the basic probability allocation function and confidence level for each indicator includes: Determine the critical value range for each indicator, and in conjunction with engineering specifications and tunnel geological conditions, divide the critical value range for each evaluation indicator into five levels of stability. Calculate the deviation of the indicator monitoring value Let the preprocessed data of index i be... The critical value range for the k-th level of stability is Corresponding to levels I-V, calculate Deviation from the k-th level interval : In the formula, For the first Interval midpoint, For normalization coefficients, ensure ; Calculate the original basic probability assignment For single-element propositions of level k in the recognition framework, the original basic probability assignment (BPA) is: empty set Basic probability allocation Global proposition Basic probability allocation ; Based on the reliability coefficient, the basic probability allocation (BPA) is corrected, combined with the indicator reliability coefficient. The original basic probability assignment (BPA) is modified to obtain the final basic probability assignment function: Calculate the confidence level based on the degree of deviation. Indicator i represents the degree of confidence in proposition A, based on the modified basic probability assignment (BPA), and is divided into single-element propositions and compound propositions: The confidence level of a single-element proposition corresponds to a single stability level: The confidence level of a composite proposition corresponds to multiple stability levels. Let's assume a composite proposition... ,but: Global confidence reflects the degree of confidence that an indicator has in the overall stability. The closer the confidence value is to 1, the higher the degree of support that the indicator provides for the corresponding stability level; the smaller the deviation, the higher the confidence level.

5. The method for dynamic evaluation of tunnel surrounding rock stability according to claim 3, characterized in that, The calculation of the global conflict coefficient includes: First, calculate the similarity of each pair of evaluation indicators and the adaptive conflict coefficient, then calculate the global conflict coefficient, and finally perform weighted synthesis of evaluation indicators based on the conflict coefficient, while simultaneously redistributing conflict information. Similarity between pairwise evaluation indicators and local conflict coefficient : Set up a recognition framework There are n evaluation indicators, and the basic probability assignments (BPA) for the i-th and j-th evaluation indicators are respectively... Evaluation index similarity : The closer the value is to 1, the higher the consistency between evaluation indicators i and j, and the smaller the conflict. Local conflict coefficient : The closer the value is to 1, the higher the degree of local conflict between evaluation indicators i and j; The global conflict coefficient K is based on the pairwise local conflict coefficients of n evaluation indicators. Combined with the reliability coefficient of the indicators Calculate the global conflict coefficient K, which reflects the overall degree of conflict among all evaluation indicators: 。 6. The method for dynamic evaluation of tunnel surrounding rock stability according to claim 3, characterized in that, The conflict correction synthesis rules include: Let the basic probability distribution BPA of the n evaluation indicators to be synthesized after correction be... The synthesized comprehensive basic probability distribution (BPA) is: For any proposition The synthesis rules are as follows: when hour: In the formula, Results of traditional DS evidence synthesis: Assign coefficients to conflicting information based on the reliability coefficient and similarity of the indicators: To evaluate the similarity of support for proposition A to index i; when hour: Composition rule constraints satisfy normalization ; Under high conflict, the proportion of conflict allocation items increases with the increase of K, making full use of conflict information; Under low conflict conditions, the conflict assignment term approaches 0, and the synthesis result is consistent with the DS rule, ensuring method compatibility.

7. The method for dynamic evaluation of tunnel surrounding rock stability according to claim 1, characterized in that, The extraction of core feature parameters includes: The core characteristic parameters after fusion include comprehensive deformation rate, mechanical response coordination degree, comprehensive environmental impact coefficient, and surrounding rock structure integrity index, which are extracted based on the physical meaning of the evaluation index synthesis results and the original monitoring index. Comprehensive deformation rate: The weighted sum of the surrounding rock convergence displacement rate, the crown settlement rate, and the sidewall horizontal displacement rate is allocated according to the importance of the project. Mechanical response coordination degree is calculated based on the normalized surrounding rock stress, support structure strain and anchor bolt axial force data. The deviation rate between the corresponding index and the mechanical coordination benchmark value is calculated and weighted. The weights are allocated according to the stress priority of the support system. The comprehensive environmental impact coefficient is a weighted sum of normalized groundwater seepage pressure, cave temperature and humidity and geological tectonic activity data, with the weights allocated according to the degree of environmental impact. The surrounding rock structural integrity index is a weighted sum of the normalized surrounding rock integrity coefficient, the reverse normalized joint and fissure development density, and the surrounding rock lithology category data, with the weights allocated according to the importance of geological characteristics.

8. The method for dynamic evaluation of tunnel surrounding rock stability according to claim 1, characterized in that, The process of using principal component analysis to reduce the dimensionality of the extracted core feature parameters includes: Construct the feature parameter matrix and standardize it: Suppose that after feature layer fusion, we obtain a group of sample data, and the core feature parameters are: Construct a sample matrix : Standardize matrix X to eliminate the influence of dimensions and orders of magnitude, and obtain the standardized matrix. ; Calculate the covariance matrix of the standardized matrix: The covariance matrix reflects the degree of linear correlation between the characteristic parameters, and the elements The covariance of the i-th and j-th feature parameters is calculated using the following formula: In the formula ， Let be the standardized mean of the i-th and j-th feature parameters, and let the covariance matrix be a symmetric matrix. diagonal elements Let be the variance of the i-th feature parameter; Find the eigenvalues ​​and eigenvectors of the covariance matrix: Solve the characteristic equation p eigenvalues ​​are obtained The eigenvalues ​​represent the variance of the corresponding principal components; the larger the value, the more information the principal component contains. For each eigenvalue Solve the homogeneous linear equation system The corresponding feature vectors are obtained. The eigenvectors are normalized, and the normalized eigenvectors are principal component load vectors. Screening principal component factors: Calculate the variance contribution rate of the p-th principal component. : Calculate the cumulative variance contribution rate : Find the smallest p such that The corresponding p principal components are the core factors after dimensionality reduction; the remaining ones are discarded. One principal component is used to reduce the dimensionality of b-dimensional features to p-dimensionality; Calculate the principal component scores after dimensionality reduction: Suppose that p principal components are selected, and the corresponding normalized eigenvectors are... Construct the principal component loading matrix Then the principal component score matrix corresponding to the standardized matrix Z for: The principal component score matrix Y is the result after dimensionality reduction.

9. The method for dynamic evaluation of tunnel surrounding rock stability according to claim 1, characterized in that: The method of merging multi-device data for the same monitoring indicator using a weighted average method includes: The weights are determined based on the measurement accuracy of the equipment; the higher the accuracy, the greater the weight. The fusion value is the weighted sum of the data collected by each equipment and its weight. Cross-validation of ground-penetrating radar data and borehole sampling analysis data was used to correct the surrounding rock integrity coefficient and joint and fracture development parameters in the geological indices, including: The correction coefficient is determined based on the complexity of the geological conditions, and the correction coefficient is used to weight and fuse the ground-penetrating radar detection data and borehole sampling analysis data.

10. The method for dynamic evaluation of tunnel surrounding rock stability according to claim 1, characterized in that, The construction of a multi-dimensional evaluation index system, including mechanical, deformation, environmental, and geological indicators, includes: The mechanical parameters include surrounding rock stress, support structure strain, and anchor bolt axial force; data are collected using fiber optic stress sensors, fiber optic strain sensors, and anchor bolt axial force sensors; the sensors are closely fitted to the surrounding rock / support structure, with 3-5 measuring points per cross-section in key areas and 2-3 measuring points per cross-section in general areas. The deformation indices include the surrounding rock convergence displacement, the crown settlement, and the sidewall horizontal displacement rate. The data are collected using laser displacement gauges, total stations, and convergence meters. One laser displacement gauge is installed on the crown, and one to two laser displacement gauges are installed on each side of the sidewall. The convergence meters are installed at the key convergence sections of the surrounding rock. The environmental indicators include groundwater seepage pressure, temperature and humidity inside the cave, and geological tectonic activity; among which, geological tectonic activity includes fault creep and ground motion; data are collected using seepage pressure sensors, temperature and humidity sensors, ground motion acceleration sensors, and fault creep monitoring instruments; seepage pressure sensors are buried in groundwater-rich areas, and one temperature and humidity sensor and one ground motion acceleration sensor are deployed at each monitoring section. Geological indicators include the lithology of the surrounding rock, integrity coefficient, and density of joints and fissures. Geological exploration and analysis are carried out using ground-penetrating radar, borehole sampling equipment, and rock mechanics testing instruments. Ground-penetrating radar antennas are deployed around the tunnel, and a sampling point is set up every 50m for borehole sampling.

Images