Tunnel engineering monitoring method and system based on multiple risk identification

Abstract

The invention discloses a tunnel engineering monitoring method and system based on multiple risk identification, and relates to the technical field of safety monitoring, and the method comprises the steps: binding multi-source monitoring data with working condition information, then carrying out the preprocessing, carrying out the prediction of the future state of each monitoring index based on the preprocessed monitoring data, and carrying out the prediction of the future state of each monitoring index. Obtaining an abnormal risk level and an uncertainty measurement value; based on the preprocessed monitoring data, performing extraction and weighted combination on multiple evolution characteristics of each monitoring index to obtain an evolution risk level; based on the preprocessed monitoring data and the corresponding working condition label, performing logic arbitration through an engineering state machine to obtain an engineering risk level; and based on the uncertainty measurement value, performing fusion decision on the abnormal risk level, the evolution risk level and the engineering risk level, generating a final risk level and interpretable reason chain information, and avoiding missed judgment or misjudgment caused by lack of working condition information or single view angle.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology, specifically to a tunnel engineering monitoring method and system based on multiple risk identification. Background Technology

[0002] During the construction and operation phases, tunnel engineering typically faces challenges such as complex surrounding rock conditions, intense construction disturbances, and uncertain stress states of the support structure. To ensure project safety, engineering practice usually involves deploying sensor systems to continuously monitor the tunnel structure's condition, including those for displacement, settlement, convergence, support internal forces, vibration, and environmental parameters.

[0003] Existing tunnel engineering monitoring and early warning methods mostly employ fixed threshold identification mechanisms, triggering an alarm when a monitored quantity exceeds a preset control value. However, fixed thresholds are difficult to adapt to different construction stages, surrounding rock conditions, and working conditions, easily leading to false alarms. Furthermore, they rely solely on a binary "yes or no" judgment based on the instantaneous state of the current monitored value, resulting in a limited perspective on risk perception. This makes it impossible for the system to predict the probability of future exceedances or to identify adverse evolution trends (such as continuous acceleration or trend drift) inherent in the data. Therefore, for potential risks that accumulate slowly or accelerate rapidly, the system often only issues an alarm when the risk actually materializes (value exceeds the limit), missing a valuable early warning window and posing a serious risk of underreporting.

[0004] In summary, existing technologies suffer from high false alarm rates and significant risks of underreporting due to a lack of operational information and a limited perspective on risk perception. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that the existing fixed threshold identification method has a high false alarm rate and a high risk of missed alarm. The purpose is to provide a tunnel engineering monitoring method and system based on multiple risk identification, which solves the above problems.

[0006] This invention is achieved through the following technical solution:

[0007] In a first aspect, the present invention provides a tunnel engineering monitoring method based on multiple risk identification, comprising:

[0008] Acquire multi-source monitoring data for tunnel engineering, bind each monitoring data point with corresponding working condition information, and obtain monitoring data with working condition labels;

[0009] The monitoring data with the working condition label is preprocessed to obtain preprocessed monitoring data;

[0010] Based on the preprocessed monitoring data, the future state of each monitoring indicator is predicted to obtain the abnormal risk level and uncertainty measure value of each monitoring indicator.

[0011] Based on the preprocessed monitoring data, multiple evolutionary features of each monitoring indicator are extracted and weighted to obtain the evolutionary risk level of each monitoring indicator.

[0012] Based on the preprocessed monitoring data and the corresponding operating condition labels, the engineering risk level of each monitoring indicator is obtained through logical arbitration by the engineering state machine.

[0013] Based on the uncertainty metric of each monitoring indicator, the abnormal risk level, evolutionary risk level and engineering risk level of each monitoring indicator are integrated and a decision is made to generate the final risk level and explainable cause chain information.

[0014] Optionally, the preprocessing of the monitoring data with operating condition labels to obtain preprocessed monitoring data includes:

[0015] The monitoring data with working condition labels are time-aligned and missing data are filled in to obtain a time-consistent and continuous multidimensional monitoring data sequence.

[0016] Determine whether the multidimensional monitoring data sequence satisfies the preset physical feasible region constraint;

[0017] Anomalies that do not meet the physical feasible domain constraints are projected and corrected to obtain preprocessed monitoring data.

[0018] Optionally, the physical feasible domain constraints include: value range constraints, rate of change constraints, and consistency constraints. The value range constraints are used to limit the monitored values ​​to be within a preset value range. The rate of change constraints are used to limit the change in monitored values ​​between adjacent time steps to not exceed a preset rate threshold. The consistency constraints are used to limit the differences in monitored values ​​between adjacent measuring points or multiple measuring points within the same cross-section to meet preset geometric consistency conditions and force consistency conditions.

[0019] Optionally, the projection correction for anomalous data that does not satisfy the physical feasible region constraint includes:

[0020] Identify outlier data points that do not satisfy the physical feasible region constraints;

[0021] The original monitoring value of the abnormal data point is replaced with a value on the corresponding physical feasible domain boundary, or the original monitoring value of the abnormal data is replaced with a feasible value that satisfies the physical feasible domain constraint and has the smallest correction amount between it and the original monitoring value.

[0022] Optionally, based on the preprocessed monitoring data, the future state of each monitoring indicator is predicted to obtain the abnormal risk level and uncertainty measure value of each monitoring indicator, including:

[0023] Based on the preprocessed monitoring data, the future state of each monitoring indicator is predicted using a time series prediction model to obtain the predicted probability distribution of each monitoring indicator.

[0024] Based on the predicted probability distribution and the preset engineering threshold, the probability of each monitoring indicator exceeding the preset engineering threshold in the future is calculated; the preset engineering threshold is determined according to the working condition label carried by each monitoring indicator.

[0025] The uncertainty metric is calculated based on the predicted probability distribution;

[0026] The abnormal risk level is obtained by combining the probability of exceeding the limit with the uncertainty metric.

[0027] Optionally, the multiple evolutionary features include trend features, drift features, and acceleration features; the extraction and weighted combination of multiple evolutionary features for each monitoring indicator based on the preprocessed monitoring data to obtain the evolutionary risk level of each monitoring indicator includes:

[0028] The preprocessed monitoring data is subjected to robust scale normalization to obtain normalized monitoring data and robust scale factor;

[0029] Based on the normalized monitoring data, the long-term change slope of each monitoring indicator within a preset long time window is extracted, and normalization is performed based on the scale factor to obtain trend characteristics.

[0030] Based on the normalized monitoring data, the offset of the current monitoring value relative to the historical baseline value is calculated within a preset drift analysis window, and normalization is performed based on the scale factor to obtain the drift characteristics.

[0031] Based on the normalized monitoring data, the short-term change slope of each monitoring indicator within a preset short time window is extracted, the difference between the short-term change slope and the long-term change slope is calculated, and the difference is normalized based on the scaling factor to obtain the acceleration feature.

[0032] The evolution risk level of each monitoring indicator is obtained by weighting and combining the trend characteristics, drift characteristics, and acceleration characteristics of each monitoring indicator.

[0033] Optionally, the step of obtaining the engineering risk level of each monitoring indicator by performing logical arbitration through an engineering state machine based on the preprocessed monitoring data and corresponding working condition labels includes:

[0034] Configure an independent engineering state machine instance for each monitoring indicator; each engineering state machine instance includes three states: idle state, active state, and upgrade state.

[0035] Within each monitoring cycle, based on the comparison between the preprocessed monitoring data and the preset engineering threshold, it is determined whether the hit condition is met; if yes, it is determined as a hit and the hit count is increased; if no, the hit count is decreased; the preset engineering threshold is determined according to the working condition label carried by each monitoring indicator.

[0036] When the duration since the first hit reaches a preset duration threshold and the hit count reaches a preset minimum number of consecutive hits, the engineering state machine instance is switched from the idle state to the active state, and the engineering risk level is determined based on the comparison result.

[0037] When the duration of the active state of the engineering state machine instance reaches a preset upgrade time threshold, the engineering state machine instance is switched from the active state to the upgrade state, and the engineering risk level is increased.

[0038] When the monitoring period ends, the engineering state machine instance outputs the engineering risk level.

[0039] Optionally, after switching the engineering state machine instance from the idle state to the active state or from the active state to the upgraded state, the method further includes:

[0040] When the engineering state machine instance is in the active state, the engineering risk level is lower than the preset level, and the duration of continuous misses reaches the preset hysteresis time threshold, the engineering state machine instance is switched from the active state back to the idle state.

[0041] When the engineering state machine instance is in the upgraded state and the duration of consecutive misses reaches a preset hysteresis time threshold, the engineering state machine instance is switched from the upgraded state back to the idle state.

[0042] Optionally, the step of fusing the anomaly risk level, evolutionary risk level, and engineering risk level of each monitoring indicator based on the uncertainty metric of each monitoring indicator to generate a final risk level and explainable causal chain information includes:

[0043] Determine whether the engineering risk level of each monitoring indicator has reached the preset risk level;

[0044] If the engineering risk level of any monitoring indicator reaches the preset risk level, then the engineering risk level of that monitoring indicator shall be taken as the final risk level.

[0045] If the engineering risk level of the monitoring indicator does not reach the preset risk level, then determine whether the uncertainty measure of the monitoring indicator is less than or equal to the preset uncertainty threshold and whether the evolution risk level of the monitoring indicator reaches the preset risk level.

[0046] If the uncertainty measure of the monitoring indicator is less than or equal to the preset gate threshold, and the evolution risk level of the monitoring indicator reaches the preset risk level, then the evolution risk level of the monitoring indicator is taken as the final risk level.

[0047] If the uncertainty measure of the monitoring indicator is greater than the preset threshold or the evolution risk level of the monitoring indicator does not reach the preset risk level, then the maximum value between the abnormal risk level and the engineering risk level of the monitoring indicator shall be taken as the final risk level.

[0048] While generating the final risk level, explainable cause chain information is recorded simultaneously.

[0049] Secondly, the present invention provides a tunnel engineering monitoring system based on multiple risk identification, comprising:

[0050] The data acquisition module is used to acquire multi-source monitoring data of tunnel engineering, bind each monitoring data with the corresponding working condition information, and obtain monitoring data with working condition tags;

[0051] The preprocessing module is used to preprocess the monitoring data with working condition labels to obtain preprocessed monitoring data;

[0052] An anomaly risk calculation module is used to predict the future state of each monitoring indicator based on the preprocessed monitoring data, and obtain the anomaly risk level and uncertainty measure value of each monitoring indicator.

[0053] The evolution risk calculation module is used to extract and weight and combine multiple evolution features of each monitoring indicator based on the preprocessed monitoring data to obtain the evolution risk level of each monitoring indicator.

[0054] The engineering state machine module is used to obtain the engineering risk level of each monitoring indicator by performing logical arbitration through the engineering state machine based on the preprocessed monitoring data and the corresponding working condition labels.

[0055] The fusion decision module is used to make fusion decisions on the abnormal risk level, evolution risk level and engineering risk level of each monitoring indicator based on the uncertainty metric value of each monitoring indicator, and generate the final risk level and explainable cause chain information.

[0056] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0057] This application provides a tunnel engineering monitoring method based on multiple risk identification. First, each monitoring data point is bound to working condition information. The bound working condition tags can be used for differentiated selection of rules / thresholds in subsequent processes. That is, the control threshold for the same monitoring indicator varies at different stages, in different surrounding rock conditions, and at different locations, enabling the selection of thresholds and identification rules according to different working conditions and avoiding false alarms caused by using a uniform threshold. Furthermore, abnormal risk identification, evolving risk identification, and engineering risk identification are performed simultaneously. Finally, using uncertainty measurement as a gating signal, the abnormal risk level, evolving risk level, and engineering risk level are intelligently fused to output a final risk level with an explanation of the causal chain, avoiding missed judgments due to a single perspective and significantly improving the accuracy of risk identification. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0059] Figure 1 A schematic flowchart of a tunnel engineering monitoring method based on multiple risk identification provided in an embodiment of this application;

[0060] Figure 2 This is a schematic diagram of the state transition of the engineering state machine provided in an embodiment of this application;

[0061] Figure 3 A flowchart illustrating the dual-risk fusion and entropy gating provided in this application embodiment;

[0062] Figure 4 Another flowchart illustrating the tunnel engineering monitoring method based on multiple risk identification provided in this application embodiment;

[0063] Figure 5 A schematic diagram of the structure of a tunnel engineering monitoring system based on multiple risk identification provided in an embodiment of this application. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0065] Existing fixed threshold early warning methods have the following shortcomings: fixed thresholds are difficult to adapt to different construction stages, different surrounding rock conditions and working conditions, and are prone to false alarms or missed alarms; judging based solely on the current monitoring value makes it difficult to identify potential risks that have not yet exceeded the limit but have shown an unfavorable evolution trend; under conditions of high construction noise and environmental interference, alarm levels are prone to frequent fluctuations, affecting the engineering reliability of the monitoring system; some methods that introduce prediction models directly use the model output as the basis for alarms, lacking stable control at the engineering decision-making level, and are difficult to meet the needs of on-site engineering use.

[0066] Therefore, this application provides a tunnel engineering monitoring method based on multiple risk identification. Please refer to... Figure 1 This is a schematic flowchart of a tunnel engineering monitoring method based on multiple risk identification provided in an embodiment of this application. The following is a description of... Figure 1 The tunnel engineering monitoring method based on multiple risk identification is introduced.

[0067] S1. Obtain multi-source monitoring data for tunnel engineering, bind each monitoring data with the corresponding working condition information, and obtain monitoring data with working condition labels.

[0068] Specifically, multi-source monitoring data for tunnel engineering is acquired from various sensing subsystems deployed within the tunnel, including data on crown settlement, perimeter convergence, support internal forces, temperature, vibration, and groundwater level. Each monitoring data point also includes a timestamp, a measuring point number, and a monitored value. The monitored value is the sensor reading and unit, the timestamp indicates the specific time the monitored value was generated, and the measuring point number uniquely identifies the sensor. The monitoring data also includes the spatial location of the measuring point, indicating its physical location within the tunnel.

[0069] Using timestamps, spatial locations of measuring points, and measuring point numbers as indexes, the system queries the corresponding database in real time for the project status and environmental conditions at that specific time and location, i.e., the working condition information. Examples include construction stage, surrounding rock grade, construction step distance, and excavation face mileage. Finally, each monitoring data point is bound to its corresponding working condition information to obtain monitoring data with working condition tags.

[0070] S2. Preprocess the monitoring data with operating condition labels to obtain preprocessed monitoring data.

[0071] Specifically, monitoring data with operating condition labels can be preprocessed, including time alignment, missing data completion, physical feasible region constraints, and projection correction, to obtain preprocessed monitoring data. The specific implementation process of S2 will be described below.

[0072] S3. Based on the preprocessed monitoring data, predict the future state of each monitoring indicator to obtain the abnormal risk level and uncertainty measure value of each monitoring indicator.

[0073] Unlike methods that rely solely on comparing instantaneous values ​​with thresholds, this application's embodiments predict and analyze future states to identify abnormal risks. It identifies sudden anomalies or rapid approaches to engineering control thresholds in monitoring indicators within a short timescale, thereby obtaining an anomaly risk level and an uncertainty measure. The anomaly risk level characterizes the likelihood that the monitoring indicator will exceed a preset engineering threshold in the future, while the uncertainty measure quantifies the reliability of the prediction result, reflecting whether the anomaly risk level in this risk assessment has significant bias or instability. The specific implementation process of S3 will be described below.

[0074] S4. Based on the preprocessed monitoring data, multiple evolutionary features of each monitoring indicator are extracted and weighted to obtain the evolutionary risk level of each monitoring indicator.

[0075] Specifically, based on the preprocessed monitoring data, multiple evolutionary features are extracted for each monitoring indicator, including trend features, drift features, and acceleration features. Trend features quantify the overall changing trend of the monitoring indicator over a longer time scale; drift features quantify the continuous deviation of the current value of the monitoring indicator relative to historical levels; and acceleration features quantify whether the changing trend of the monitoring indicator has accelerated. By weighted combination of multiple evolutionary features for each monitoring indicator, evolutionary risk is identified, and the slow deterioration or continuous shift of the structural state over a longer time scale is determined, thereby obtaining the evolutionary risk level of each monitoring indicator. The evolutionary risk level characterizes the relative risk of unfavorable evolution of the structural state. The specific implementation process of S4 will be described below.

[0076] S5. Based on the engineering state machine, the abnormal risk level and evolution risk level of each monitoring indicator are stabilized to obtain the engineering risk level of each monitoring indicator.

[0077] Specifically, the engineering state machine is used to transform continuous risks or engineering indicators into stable alarm levels, preventing alarms from frequently fluctuating around thresholds. Its core functions include: a combination of continuous hit count, duration, hysteresis clearing, and escalation timing, and persistent state writing to a database to support continuous operation across cycles. The engineering state machine is used for hit-related parameters requiring temporal continuity of risk evidence, hysteresis-related parameters delaying state clearing after risk decline, and escalation control parameters controlling the progressive escalation of alarm levels. The specific implementation process of S5 will be described below.

[0078] S6. Based on the uncertainty metric of each monitoring indicator, the abnormal risk level, evolution risk level and engineering risk level of each monitoring indicator are integrated and decided to generate the final risk level and explainable cause chain information.

[0079] Specifically, it simultaneously considers the anomaly risk level, evolutionary risk level, and engineering risk level to form a triple risk perspective. Based on a gating fusion strategy using uncertainty metrics, it outputs the final risk level and cause chain, realizing a paradigm shift from passive, static, and single-point threshold alarms to proactive, dynamic, multi-dimensional, and interpretable intelligent risk warnings. The specific implementation process of S6 will be described below.

[0080] In one possible embodiment, the specific steps of S2 include:

[0081] S2.1 Perform time alignment and missing data completion processing on the monitoring data with working condition labels to obtain a time-consistent and continuous multidimensional monitoring data sequence.

[0082] In the specific implementation process, the monitoring data collected by each monitoring point at different times are mapped to a unified time reference grid. The time reference grid can be set to a fixed time interval according to engineering requirements, such as one minute, five minutes or other preset time intervals, so as to achieve consistent alignment of multi-source monitoring data in the time dimension.

[0083] For missing data segments generated during time alignment, interpolation methods, state estimation-based filtering methods, or data diffusion-based completion methods can be used to reconstruct the missing data. Any one of these methods can be chosen, or multiple methods can be used in parallel for completion processing to improve data continuity and reliability. After time alignment and missing data completion, a time-uniform and continuous multidimensional monitoring data sequence is obtained, which can be used to represent the monitoring status information at each time step, each monitoring point, and each monitoring feature dimension.

[0084] In this embodiment of the application, unified time alignment and missing data completion processing of multi-source monitoring data can effectively reduce the problem of discontinuous monitoring data caused by factors such as communication packet loss and asynchronous sampling, thereby reducing alarm jitter caused by missing data and improving the stability and reliability of engineering monitoring and risk warning results.

[0085] S2.2 Determine whether the multidimensional monitoring data sequence meets the preset physical feasible region constraint.

[0086] In one possible implementation, the physical feasible domain constraint includes the following three constraints:

[0087] 1. Value range constraint: Used to limit the monitored value to a preset value range.

[0088] The preset numerical range can be determined based on at least one of the following factors: sensor range boundary, engineering upper limit of structural stress or deformation, and historical experience statistical range. The sensor range boundary is a rigid physical boundary that cannot be infringed upon. The engineering upper limit of structural stress or deformation is the limit state value calculated based on material structure, design load, and safety factor. The historical experience statistical range is the fluctuation range obtained by calculating the statistical distribution of historical data from this project or similar projects under the same or similar working conditions.

[0089] 2. Rate of change constraint: Used to limit the change in monitored values ​​between adjacent time steps to not exceed a preset rate threshold.

[0090] The preset rate threshold can be determined based on the structural response rate and the sampling time interval. The theoretical structural response rate is based on the theoretical maximum deformation or stress change rate estimated from the material structure.

[0091] 3. Consistency constraint: Used to restrict the differences in monitoring values ​​between adjacent measuring points or multiple measuring points within the same cross section to meet preset geometric consistency conditions and force consistency conditions.

[0092] Geometric consistency conditions are based on spatial geometric relationships, requiring that monitoring data from different locations on the same structure must satisfy specific mathematical relationships. Force consistency conditions are based on mechanical equilibrium principles, requiring that monitoring data reflecting the forces acting on the structure (such as stress, strain, and pressure) satisfy mechanical laws in their spatial distribution.

[0093] In this embodiment, value range constraints can effectively eliminate abnormal data that clearly does not conform to engineering physics due to equipment failure or signal interference. Rate-of-change constraints can suppress non-physical abrupt changes caused by transient noise or communication anomalies, making the data sequence smoother, more continuous, and more realistically reflecting the continuous evolution of the structural state. Consistency constraints further distinguish between local data anomalies and the true structural response.

[0094] S2.3. Projection correction is performed on abnormal data points that do not meet the physical feasible region constraints to obtain preprocessed monitoring data.

[0095] In one possible embodiment, the projection correction is performed as follows:

[0096] Identify outlier data points that do not meet the physical feasible region constraints; replace the original monitoring value of the outlier data point with the value on the corresponding physical feasible region boundary, or replace the original monitoring value of the outlier data point with a feasible value that meets the physical feasible region constraints and has the smallest correction amount between it and the original monitoring value.

[0097] In practical implementation, for outlier data points that do not meet the value range constraints, their original monitoring values ​​are directly corrected to the nearest boundary value of their respective intervals. For outlier data points that do not meet the rate of change and consistency constraints, these outlier data points and their affected related points are treated as a local optimization problem. The objective function is to minimize the correction amount (such as Euclidean distance), and the constraint condition is that all physical feasible region constraints must be satisfied. Solving this optimization problem yields feasible values, and the original monitoring values ​​of the outlier data are replaced with these feasible values. This corrects the outlier data without introducing additional engineering risk assessments.

[0098] In this embodiment, by introducing physical feasible domain constraints and projection corrections before risk identification, it is possible to effectively distinguish between abnormal fluctuations caused by data anomalies and actual structural anomalies. This forms a data quality and safety gate in the engineering process, preventing data problems caused by sensor failures or communication anomalies from being directly transmitted to the risk decision-making level, avoiding false alarms caused by non-engineering factors, and improving the accuracy and stability of risk warning results. Furthermore, by adopting the principle of minimum correction, it avoids simple deletion or crude replacement of abnormal data, maximizing the preservation of effective information in the original monitoring sequence.

[0099] In one possible embodiment, the specific steps of S3 include:

[0100] S3.1 Based on the preprocessed monitoring data, the future state of each monitoring indicator is predicted using a time series prediction model to obtain the predicted probability distribution of each monitoring indicator.

[0101] Specifically, for each monitoring indicator under a specific operating condition label, its corresponding time series prediction model is loaded or trained. For the input sequence of each monitoring indicator, the model outputs the predicted probability distribution for the next h steps (e.g., the next 1 day, 3 days). The predicted probability distribution includes the expected value representing the central position of the prediction result and the dispersion parameters (e.g., variance, standard deviation, etc.) representing the degree of dispersion of the prediction.

[0102] S3.2 Based on the predicted probability distribution and the preset engineering threshold, calculate the probability of each monitoring indicator exceeding the preset engineering threshold in the future.

[0103] Specifically, based on the operating condition label of each monitoring data point, a corresponding preset engineering threshold is obtained. Based on the predicted probability distribution, the probability that the predicted value at each future time step will exceed the preset engineering threshold is calculated. For multiple future time steps, the maximum or average of all probabilities can be taken as the probability of exceeding the limit.

[0104] S3.3 Calculate the uncertainty metric based on the predicted probability distribution.

[0105] Specifically, based on the predicted probability distribution, its risk entropy or information entropy is calculated as the uncertainty metric value. Finally, the uncertainty metric value is normalized to a unified range (such as [0, 1]) to quantify the reliability of the current risk judgment result. The higher the uncertainty metric value, the lower the credibility of the current judged abnormal risk level.

[0106] S3.4. Combine the overrun probability and the uncertainty metric value to obtain the abnormal risk level.

[0107] Specifically, combine the overrun probability and the uncertainty metric value, and adopt a combined form of weighted suppression or enhancement of uncertainty to obtain the abnormal risk score A, so as to comprehensively reflect the overrun possibility and the reliability of risk judgment.

[0108] For example: ; where A is the abnormal risk score, is the overrun probability; is the uncertainty metric value; is the uncertainty penalty coefficient.

[0109] Furthermore, according to the preset abnormal risk level division threshold, map the abnormal risk score A to the abnormal risk level.

[0110] For example, if A < A1, the corresponding abnormal risk level is low level (green level); if A1 ≤ A < A2, the corresponding abnormal risk level is medium level (yellow level); if A2 ≤ A < A3, the corresponding abnormal risk level is medium-high level (orange level); if A3 ≤ A, the corresponding abnormal risk level is high level (red level). Among them, A1, A2, and A3 are all preset abnormal risk level division thresholds.

[0111] In the embodiment of the present application, through the comprehensive modeling of the future state distribution and risk uncertainty, an abnormal risk identification mechanism different from the traditional threshold trigger is formed, which can timely identify potential overrun risks when the monitoring index suddenly becomes abnormal or quickly approaches the engineering threshold; at the same time, when the prediction result is unstable or the uncertainty is high, the risk judgment result is suppressed, so as to avoid false alarms caused by prediction fluctuations and improve the reliability of engineering monitoring and early warning decisions.

[0112] In a possible embodiment, the specific steps of S4 include:

[0113] S4.1. Perform robust scale normalization on the preprocessed monitoring data to obtain the normalized monitoring data and the robust scale factor.

[0114] Specifically, to improve the robustness of handling outliers and noise interference, a scaling estimation method based on median absolute deviation is used to normalize the monitoring data to obtain normalized monitoring data and robust scaling factors. The scaling factors are used for the unified dimensional conversion of various risk characteristics in the subsequent process.

[0115] S4.2 Based on the monitoring data, extract the long-term change slope of each monitoring indicator within a preset long-term window, and perform normalization processing based on the scale factor to obtain trend characteristics.

[0116] Specifically, for each monitoring indicator, normalized monitoring data within a preset long-term window is extracted. Linear fitting is then performed on the normalized monitoring data within the long-term window to obtain the trend slope representing the long-term change trend, i.e., the long-term change slope. The long-term change slope is then normalized based on a scaling factor to obtain the trend characteristics.

[0117] S4.3 Based on normalized monitoring data, calculate the offset of the current monitoring value relative to the historical baseline value within the preset drift analysis window, and perform normalization processing based on the scale factor to obtain drift characteristics.

[0118] Specifically, for each monitoring indicator, the normalized monitoring data within the preset drift analysis window is extracted. The quantile or median value of the historical monitoring data is used as the historical baseline value. The offset of the normalized monitoring data within the drift analysis window relative to the historical baseline value is calculated, and the offset is normalized based on the scaling factor to obtain the drift characteristics.

[0119] S4.4 Based on normalized monitoring data, extract the short-term change slope of each monitoring indicator within a preset short time window, calculate the difference between the short-term change slope and the long-term change slope, and use the difference as an acceleration feature.

[0120] Specifically, for each monitoring indicator, normalized monitoring data within a preset short time window (smaller than a long time window) is extracted. Linear fitting is performed on the normalized monitoring data within the short time window to obtain the trend slope representing the short-term change trend, i.e., the short-term change slope. The difference between the short-term and long-term change slopes is calculated, and this difference is normalized based on a scaling factor to obtain acceleration characteristics. Only when this difference is positive is an unfavorable acceleration trend considered to exist, thus avoiding misjudgments of slowing or declining trends.

[0121] S4.5. Weighted combination of trend characteristics, drift characteristics and acceleration characteristics of each monitoring indicator to obtain the evolution risk level of each monitoring indicator.

[0122] Specifically, the trend characteristics, drift characteristics, and acceleration characteristics of each monitoring index are weighted and combined to obtain the comprehensive evolution risk characteristics of each monitoring index. The comprehensive evolution risk characteristics are converted into normalized evolution risk scores through a preset non - linear mapping function. When a monitoring point contains multiple monitoring characteristics, the evolution risk scores corresponding to each characteristic are aggregated, and the maximum value is taken as the evolution risk score of the monitoring point to preferentially reflect the characteristic dimension that is most unfavorable to engineering safety. According to the preset evolution risk level division threshold, the evolution risk score E is mapped to the evolution risk level.

[0123] For example, if E < E1, the corresponding evolution risk level is low level (green level); if E1 ≤ E < E2, the corresponding evolution risk level is medium level (yellow level); if E2 ≤ E < E3, the corresponding evolution risk level is medium - high level (orange level); if E3 ≤ E, the corresponding evolution risk level is high level (red level). Among them, E1, E2, and E3 are all preset evolution risk level division thresholds.

[0124] In the embodiment of the present application, through the evolution risk identification method, it is possible to issue a warning about the adverse evolution trend of the structural state before the engineering monitoring index reaches the preset over - limit threshold, thereby increasing the lead time for engineering risk disposal and enhancing the initiative of engineering safety management.

[0125] In a possible embodiment, the specific steps of S5 include:

[0126] Configure an independent engineering state machine instance for each monitoring index; among them, each engineering state machine instance includes three states: idle state, active state, and upgrade state. In each monitoring period, based on the comparison result between the pre - processed monitoring data and the preset engineering threshold, it is judged whether the hit condition is satisfied; if so, it is determined as a hit, and the hit count is incremented; if not, the hit count is decremented; the preset engineering threshold is determined according to the working condition label carried by each monitoring index.

[0127] When the duration since the first hit reaches the preset duration threshold and the hit count reaches the preset minimum continuous hit times, the engineering state machine instance is switched from the idle state to the active state, and the engineering risk level is determined according to the comparison result. When the duration of the engineering state machine instance in the active state reaches the preset upgrade time threshold, the engineering state machine instance is switched from the active state to the upgrade state, and the engineering risk level is increased.

[0128] When the engineering state machine instance is in the active state, the engineering risk level is lower than the preset level, and the duration of consecutive misses reaches the preset hysteresis time threshold, the engineering state machine instance is switched back from the active state to the idle state. When the engineering state machine instance is in the upgrade state and the duration of consecutive misses reaches the preset hysteresis time threshold, the engineering state machine instance is switched back from the upgrade state to the idle state.

[0129] When the monitoring period ends, the engineering state machine instance outputs the engineering risk level; the engineering risk level is used to characterize the impact of anomalies on engineering safety.

[0130] It should be noted that when the hit count decays to zero, the corresponding first hit time record is cleared to avoid interference of historical stale hits on the current risk identification process. According to engineering needs, the hit condition can be set to be greater than or equal to the preset engineering threshold, less than the preset engineering threshold, or fall within the range of the preset engineering threshold.

[0131] The following example illustrates how to determine the engineering risk level based on the comparison results.

[0132] For example, the preset engineering threshold corresponding to the operating condition label carried by a certain monitoring index is M. When the monitoring value X of a certain monitoring index is greater than the preset engineering threshold M, a hit is determined, and the single - time over - limit amplitude is X - M. When the hit count reaches the preset minimum continuous hit count, the single - time over - limit amplitudes of all hit events are accumulated to obtain the cumulative over - limit amplitude N. If N < N1, the engineering risk level corresponding to the active state is low level (green level); if N1 ≤ N < N2, the engineering risk level corresponding to the active state is high level (yellow level); if N2 ≤ N < N3, the engineering risk level corresponding to the active state is medium - high level (orange level); if N3 ≤ N, the engineering risk level corresponding to the active state is high level (red level). Here, N1, N2, and N3 are all preset cumulative threshold parameters.

[0133] Please refer to Figure 2 , for the state transition schematic diagram of the engineering state machine instance provided by the embodiment of this application. Each engineering state machine instance includes three states: idle state, active state, and upgrade state.

[0134] Idle state: The system is in the normal monitoring stage, and no persistent risk events are confirmed. All internal counters and timers are in the reference or zero - reset state.

[0135] Active state: The system has confirmed the occurrence of an abnormal event and entered the formal tracking and alarm phase. This state indicates that the anomaly has met the initial persistence and severity conditions. To transition from the idle state to the active state, two conditions must be met simultaneously: hit count and duration. A hit count reaching a preset minimum number of consecutive hits can prevent single-point spikes, and the duration from the first hit reaching a preset duration threshold can prevent instantaneous spikes.

[0136] Escalation Stage: Confirmed anomalies persist without mitigation, and risks accumulate further over time, requiring a higher level of attention and response. In this case, the corresponding engineering risk level is raised to a higher level to reflect the continued accumulation and escalation of risks over time.

[0137] After the engineering state machine enters the active state, the active state is only allowed to be cleared if the duration of consecutive misses exceeds a preset hysteresis time threshold, in order to avoid frequent switching of risk states near the threshold and causing alarm jitter. Moreover, it does not roll back immediately after a miss, but only rolls back to the idle state after a period of continuous misses, to avoid alarm interruptions.

[0138] When the monitoring period ends, the current status, hit count, first time information, and corresponding engineering risk level will be persistently stored to support subsequent operation audits, historical playback, and accountability.

[0139] In one possible embodiment, the specific steps of S6 include:

[0140] Determine whether the engineering risk level of each monitoring indicator reaches the preset risk level; if the engineering risk level of any monitoring indicator reaches the preset risk level, then the engineering risk level of that monitoring indicator shall be taken as the final risk level.

[0141] If the engineering risk level of the monitoring indicator does not reach the preset risk level, then it is determined whether the uncertainty measure of the monitoring indicator is less than or equal to the preset uncertainty threshold and whether the evolution risk level of the monitoring indicator reaches the preset risk level.

[0142] If the uncertainty measure of the monitoring indicator is less than or equal to the preset threshold, and the evolution risk level of the monitoring indicator reaches the preset risk level, then the evolution risk level of the monitoring indicator is taken as the final risk level.

[0143] If the uncertainty measure of the monitoring indicator is greater than the preset threshold or the evolution risk level of the monitoring indicator does not reach the preset risk level, then the maximum value between the abnormal risk level and the engineering risk level of the monitoring indicator shall be taken as the final risk level.

[0144] While generating the final risk level, explainable cause chain information is recorded simultaneously.

[0145] Please refer to Figure 3 This is a flowchart illustrating a dual-risk fusion and entropy gating method provided in an embodiment of this application. The preset risk level can be either orange or red. Figure 3 This is an example of a risk level that is set to orange.

[0146] First, input the abnormal risk level, evolution risk level, engineering risk level, and uncertainty measure; second, determine whether the engineering risk level is lower than the orange level.

[0147] When the engineering risk level is not lower than orange level, the final risk level is: Engineering Risk Level; the cause chain is: the final risk level is dominated by the engineering state machine result. This ensures that in situations with clear engineering risks, the engineering safety judgment result is prioritized. Depending on different implementation needs, a hard safety floor or safety ceiling strategy can also be configured to limit the upper and lower boundaries of the final risk level.

[0148] When the engineering risk level does not reach the orange level, determine whether the uncertainty metric does not exceed the threshold and whether the evolution risk level is not lower than the orange level.

[0149] When the uncertainty metric does not exceed the gating threshold (indicating that the current risk assessment result has high reliability) and the evolutionary risk level is not lower than the orange level, the final risk level is the evolutionary risk level, thereby achieving early warning of potential structural deterioration trends. The causal chain is the evolutionary risk trigger level upgrade.

[0150] When the uncertainty metric exceeds the threshold or the evolutionary risk level falls below the orange level, the risk level is not automatically increased based on model predictions or evolutionary risk results to avoid false alarms caused by unstable predictions or model fluctuations. In this case, the final risk level is the higher of the abnormal risk level and the engineering risk level to ensure the robustness of the alarm decision. The cause chain is insufficient uncertainty threshold control or evolutionary risk, failing to trigger an upgrade.

[0151] Finally, the final risk level, i.e., the corresponding cause chain information, is written into the alarm record table and the risk snapshot table. The cause chain includes at least the abnormal risk level, the evolving risk level, the engineering risk level, the uncertainty metric, and the gating threshold triggering status, which are used to support the interpretation, review, and auditing of engineering alarm results.

[0152] In this embodiment, by establishing the output of the engineering state machine as the highest priority decision-making basis, it is ensured that in critical risk events, engineering experience and logical judgment always take precedence over any algorithmic prediction, thus guaranteeing the engineering rationality and inherent safety of risk decisions from a mechanism perspective. The entropy gating design, which uses the uncertainty metric as the core control signal, enables the system's adaptive adjustment capability. When uncertainty is low and the evolutionary risk is significant, it can proactively adopt the early warning signals of intelligent algorithms; when uncertainty is high, it automatically switches to a conservative strategy to avoid false alarms caused by model bias.

[0153] In one possible embodiment, after executing S6, all key tables involved in the entire risk assessment process are written into the database to form a chain of evidence. These key tables include a monitoring data table, a risk snapshot table, an alarm record table, an engineering state machine rule definition table, an engineering state machine operating status table, a threshold change audit table, and an operating product index table. The key fields, uses, and audit value of each key table are shown in Table 1.

[0154] Table 1

[0155]

[0156] The parameters involved in the tunnel engineering monitoring method based on multiple risk identification provided in this application are shown in Table 2.

[0157] Table 2

[0158]

[0159] The comparison between the method provided in this application and the prior art is shown in Table 3.

[0160] Table 3

[0161]

[0162] Please refer to Figure 4 This is another flowchart illustrating a tunnel engineering monitoring method based on multiple risk identification provided in this application. First, data is collected; second, preprocessing and physical constraints are applied; then, based on the preprocessed monitoring data, abnormal risk calculation, evolving risk calculation, and engineering state machine processing are performed to obtain the abnormal risk level, evolving risk level, and engineering risk level. Next, these three are integrated for decision-making, outputting the final risk level and cause chain. Finally, an alarm is issued based on the risk level, while the synchronously output cause chain information provides a transparent and reliable basis for risk tracing, handling decisions, and post-event auditing, achieving closed-loop management of early warning.

[0163] The tunnel engineering monitoring method based on multiple risk identification provided in this application has the following beneficial effects: Through dual-risk fusion identification, it can simultaneously identify sudden hazardous events and long-term adverse structural evolution processes, improving the ability to identify hidden risks; through rule engines and state machine mechanisms, it improves the engineering stability and interpretability of early warning results; and through traceability and auditability design, it meets the needs of engineering safety management and responsibility determination.

[0164] Based on the same inventive concept, please refer to Figure 5 This application also provides a tunnel engineering monitoring system based on multiple risk identification, the system comprising:

[0165] The data acquisition module is used to acquire multi-source monitoring data of tunnel engineering, bind each monitoring data with the corresponding working condition information, and obtain monitoring data with working condition tags;

[0166] The preprocessing module is used to preprocess the monitoring data with operating condition labels to obtain preprocessed monitoring data;

[0167] The anomaly risk calculation module is used to predict the future state of each monitoring indicator based on the preprocessed monitoring data, and obtain the anomaly risk level and uncertainty measure value of each monitoring indicator.

[0168] The evolution risk calculation module is used to extract and weight a combination of multiple evolutionary features of each monitoring indicator based on the preprocessed monitoring data to obtain the evolution risk level of each monitoring indicator.

[0169] The engineering state machine module is used to obtain the engineering risk level of each monitoring indicator by performing logical arbitration based on the preprocessed monitoring data and the corresponding working condition labels through the engineering state machine.

[0170] The fusion decision module is used to make fusion decisions on the abnormal risk level, evolution risk level and engineering risk level of each monitoring indicator based on the uncertainty metric value of each monitoring indicator, and generate the final risk level and explainable cause chain information.

[0171] Optionally, the preprocessing module is specifically used for:

[0172] Time alignment and missing data completion processing are performed on the monitoring data with working condition labels to obtain a time-consistent and continuous multidimensional monitoring data sequence;

[0173] Determine whether the multidimensional monitoring data sequence meets the preset physical feasible region constraints;

[0174] Anomalies that do not meet the physical feasible region constraints are projected and corrected to obtain preprocessed monitoring data.

[0175] Optionally, the physical feasible domain constraints include: value range constraints, rate of change constraints, and consistency constraints. Value range constraints are used to limit the monitored values ​​to be within a preset value range; rate of change constraints are used to limit the change in monitored values ​​between adjacent time steps to not exceed a preset rate threshold; consistency constraints are used to limit the differences in monitored values ​​between adjacent measuring points or multiple measuring points within the same cross section to meet preset geometric consistency conditions and force consistency conditions.

[0176] Optionally, the preprocessing module is specifically used for:

[0177] Identify outlier data points that do not meet the physical feasible region constraints;

[0178] Replace the original monitoring value of the abnormal data point with the value at the corresponding physical feasible region boundary, or replace the original monitoring value of the abnormal data with a feasible value that satisfies the physical feasible region constraint and has the smallest correction amount between it and the original monitoring value.

[0179] Optionally, the anomaly risk calculation module is specifically used for:

[0180] Based on the preprocessed monitoring data, the future state of each monitoring indicator is predicted by a time series prediction model to obtain the predicted probability distribution of each monitoring indicator.

[0181] Based on the predicted probability distribution and the preset engineering threshold, the probability of each monitoring indicator exceeding the preset engineering threshold in the future is calculated; the preset engineering threshold is determined according to the working condition label carried by each monitoring indicator.

[0182] Calculate the uncertainty measure based on the predicted probability distribution;

[0183] By combining the probability of exceeding limits with the uncertainty metric, an abnormal risk level is obtained.

[0184] Optional, multiple evolutionary features include trend features, drift features, and acceleration features; the evolutionary risk calculation module is specifically used for:

[0185] Robust scaling normalization is performed on the preprocessed monitoring data to obtain normalized monitoring data and robust scaling factors;

[0186] Based on normalized monitoring data, the long-term change slope of each monitoring indicator within a preset long time window is extracted, and normalized based on the scale factor to obtain trend characteristics.

[0187] Based on normalized monitoring data, the offset of the current monitoring value relative to the historical baseline value is calculated within a preset drift analysis window, and normalization is performed based on the scale factor to obtain drift characteristics.

[0188] Based on normalized monitoring data, the short-term change slope of each monitoring indicator within a preset short time window is extracted, the difference between the short-term change slope and the long-term change slope is calculated, and the difference is normalized based on the scaling factor to obtain acceleration characteristics.

[0189] The evolution risk level of each monitoring indicator is obtained by weighting and combining the trend characteristics, drift characteristics, and acceleration characteristics of each monitoring indicator.

[0190] Optionally, the engineering state machine module is specifically used for:

[0191] Configure an independent engineering state machine instance for each monitoring indicator; each engineering state machine instance includes three states: idle state, active state, and upgrade state.

[0192] Within each monitoring cycle, based on the comparison between the preprocessed monitoring data and the preset engineering threshold, it is determined whether the hit condition is met; if yes, it is determined as a hit and the hit count is increased; if no, the hit count is decreased; the preset engineering threshold is determined based on the working condition label carried by each monitoring indicator.

[0193] When the duration since the first hit reaches a preset duration threshold and the hit count reaches a preset minimum number of consecutive hits, the engineering state machine instance is switched from the idle state to the active state, and the engineering risk level is determined based on the comparison results.

[0194] When the duration of the active state of the engineering state machine instance reaches the preset upgrade time threshold, the engineering state machine instance will be switched from the active state to the upgrade state, and the engineering risk level will be increased.

[0195] When the monitoring period ends, the engineering state machine instance outputs the engineering risk level.

[0196] Optionally, the engineering state machine module is specifically used to: after switching the engineering state machine instance from the idle state to the active state or from the active state to the upgraded state, when the engineering state machine instance is in the active state, the engineering risk level is lower than the preset level, and the duration of continuous misses reaches the preset hysteresis time threshold, switch the engineering state machine instance back from the active state to the idle state.

[0197] When an engineering state machine instance is in the upgrade state and the duration of consecutive misses reaches a preset hysteresis time threshold, the engineering state machine instance is switched from the upgrade state back to the idle state.

[0198] Optionally, the fusion decision module is used for:

[0199] Determine whether the engineering risk level of each monitoring indicator has reached the preset risk level;

[0200] If the engineering risk level of any monitoring indicator reaches the preset risk level, then the engineering risk level of that monitoring indicator shall be taken as the final risk level.

[0201] If the engineering risk level of the monitoring indicator does not reach the preset risk level, then determine whether the uncertainty measure of the monitoring indicator is less than or equal to the preset uncertainty threshold and whether the evolution risk level of the monitoring indicator reaches the preset risk level.

[0202] If the uncertainty measure of the monitoring indicator is less than or equal to the preset threshold, and the evolution risk level of the monitoring indicator reaches the preset risk level, then the evolution risk level of the monitoring indicator is taken as the final risk level.

[0203] If the uncertainty measure of the monitoring indicator is greater than the preset threshold or the evolution risk level of the monitoring indicator does not reach the preset risk level, then the maximum value between the abnormal risk level and the engineering risk level of the monitoring indicator shall be taken as the final risk level.

[0204] While generating the final risk level, explainable cause chain information is recorded simultaneously.

[0205] It should be noted that each module in the tunnel engineering monitoring device based on multiple risk identification in this embodiment corresponds one-to-one with each step in the tunnel engineering monitoring method based on multiple risk identification in the aforementioned embodiment. Therefore, the specific implementation of this embodiment can refer to the implementation of the tunnel engineering monitoring method based on multiple risk identification mentioned above, and will not be repeated here.

[0206] Based on the same inventive concept, this application also provides a computer device, which includes a processor, a memory, and a computer program stored in the memory. The computer program is executed by the processor to implement the aforementioned tunnel engineering monitoring method based on multiple risk identification.

[0207] Based on the same inventive concept, this application also provides a computer storage medium storing a computer program, which is executed by a processor to implement the aforementioned tunnel engineering monitoring method based on multiple risk identification.

[0208] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a device including one or any combination of the above-mentioned memories. The computer may be a variety of computing devices, including smart terminals and servers.

[0209] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0210] As an example, executable instructions may, but do not necessarily, correspond to files in the file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborative files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0211] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0212] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0213] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0214] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A tunnel engineering monitoring method based on multiple risk identification, characterized in that, include: Acquire multi-source monitoring data for tunnel engineering, bind each monitoring data point with corresponding working condition information, and obtain monitoring data with working condition labels; The monitoring data with the working condition label is preprocessed to obtain preprocessed monitoring data; Based on the preprocessed monitoring data, the future state of each monitoring indicator is predicted to obtain the abnormal risk level and uncertainty measure value of each monitoring indicator. Based on the preprocessed monitoring data, multiple evolutionary features of each monitoring indicator are extracted and weighted to obtain the evolutionary risk level of each monitoring indicator. Based on the preprocessed monitoring data and the corresponding operating condition labels, the engineering risk level of each monitoring indicator is obtained through logical arbitration by the engineering state machine. Based on the uncertainty metric of each monitoring indicator, the abnormal risk level, evolutionary risk level and engineering risk level of each monitoring indicator are integrated and a decision is made to generate the final risk level and explainable causal chain information.

2. The tunnel engineering monitoring method based on multiple risk identification according to claim 1, characterized in that, The preprocessing of the monitoring data with operating condition labels to obtain preprocessed monitoring data includes: The monitoring data with working condition labels are time-aligned and missing data are filled in to obtain a time-consistent and continuous multidimensional monitoring data sequence. Determine whether the multidimensional monitoring data sequence satisfies the preset physical feasible region constraint; Anomalies that do not meet the physical feasible domain constraints are projected and corrected to obtain preprocessed monitoring data.

3. The tunnel engineering monitoring method based on multiple risk identification according to claim 2, characterized in that, The physical feasible domain constraints include: value range constraints, rate of change constraints, and consistency constraints. The value range constraints are used to limit the monitored values ​​to be within a preset value range. The rate of change constraints are used to limit the change in monitored values ​​between adjacent time steps to not exceed a preset rate threshold. The consistency constraints are used to limit the differences in monitored values ​​between adjacent measuring points or multiple measuring points within the same cross-section to meet preset geometric consistency conditions and force consistency conditions.

4. The tunnel engineering monitoring method based on multiple risk identification according to claim 2, characterized in that, The projection correction for anomalous data that does not satisfy the physical feasible region constraint includes: Identify outlier data points that do not satisfy the physical feasible region constraints; The original monitoring value of the abnormal data point is replaced with a value on the corresponding physical feasible domain boundary, or the original monitoring value of the abnormal data is replaced with a feasible value that satisfies the physical feasible domain constraint and has the smallest correction amount between it and the original monitoring value.

5. The tunnel engineering monitoring method based on multiple risk identification according to claim 1, characterized in that, Based on the preprocessed monitoring data, the future state of each monitoring indicator is predicted to obtain the abnormal risk level and uncertainty measure value of each monitoring indicator, including: Based on the preprocessed monitoring data, the future state of each monitoring indicator is predicted using a time series prediction model to obtain the predicted probability distribution of each monitoring indicator. Based on the predicted probability distribution and the preset engineering threshold, the probability of each monitoring indicator exceeding the preset engineering threshold in the future is calculated; the preset engineering threshold is determined according to the working condition label carried by each monitoring indicator. The uncertainty metric is calculated based on the predicted probability distribution; The abnormal risk level is obtained by combining the probability of exceeding the limit with the uncertainty metric.

6. The tunnel engineering monitoring method based on multiple risk identification according to claim 1, characterized in that, The aforementioned multiple evolutionary features include trend features, drift features, and acceleration features; Based on the preprocessed monitoring data, multiple evolutionary features of each monitoring indicator are extracted and weighted to obtain the evolutionary risk level of each monitoring indicator, including: The preprocessed monitoring data is subjected to robust scale normalization to obtain normalized monitoring data and robust scale factor; Based on the normalized monitoring data, the long-term change slope of each monitoring indicator within a preset long time window is extracted, and normalization is performed based on the scale factor to obtain trend characteristics. Based on the normalized monitoring data, the offset of the current monitoring value relative to the historical baseline value is calculated within a preset drift analysis window, and normalization is performed based on the scale factor to obtain the drift characteristics. Based on the normalized monitoring data, the short-term change slope of each monitoring indicator within a preset short time window is extracted, the difference between the short-term change slope and the long-term change slope is calculated, and the difference is normalized based on the scaling factor to obtain the acceleration feature. The evolution risk level of each monitoring indicator is obtained by weighting and combining the trend characteristics, drift characteristics, and acceleration characteristics of each monitoring indicator.

7. The tunnel engineering monitoring method based on multiple risk identification according to claim 1, characterized in that, Based on the preprocessed monitoring data and corresponding operating condition labels, the engineering state machine is used for logical arbitration to obtain the engineering risk level of each monitoring indicator, including: Configure an independent engineering state machine instance for each monitoring indicator; each engineering state machine instance includes three states: idle state, active state, and upgrade state. Within each monitoring cycle, based on the comparison between the preprocessed monitoring data and the preset engineering threshold, it is determined whether the hit condition is met; if yes, it is determined as a hit and the hit count is increased; if no, the hit count is decreased; the preset engineering threshold is determined according to the working condition label carried by each monitoring indicator. When the duration since the first hit reaches a preset duration threshold and the hit count reaches a preset minimum number of consecutive hits, the engineering state machine instance is switched from the idle state to the active state, and the engineering risk level is determined based on the comparison result. When the duration of the active state of the engineering state machine instance reaches a preset upgrade time threshold, the engineering state machine instance is switched from the active state to the upgrade state, and the engineering risk level is increased. When the monitoring period ends, the engineering state machine instance outputs the engineering risk level.

8. The tunnel engineering monitoring method based on multiple risk identification according to claim 7, characterized in that, After switching the engineering state machine instance from the idle state to the active state or from the active state to the upgraded state, the method further includes: When the engineering state machine instance is in the active state, the engineering risk level is lower than the preset level, and the duration of continuous misses reaches the preset hysteresis time threshold, the engineering state machine instance is switched from the active state back to the idle state. When the engineering state machine instance is in the upgraded state and the duration of consecutive misses reaches a preset hysteresis time threshold, the engineering state machine instance is switched from the upgraded state back to the idle state.

9. The tunnel engineering monitoring method based on multiple risk identification according to claim 1, characterized in that, The process involves fusing the anomaly risk level, evolutionary risk level, and engineering risk level of each monitoring indicator based on its uncertainty metric value, generating a final risk level and explainable causal chain information, including: Determine whether the engineering risk level of each monitoring indicator has reached the preset risk level; If the engineering risk level of any monitoring indicator reaches the preset risk level, then the engineering risk level of that monitoring indicator shall be taken as the final risk level. If the engineering risk level of the monitoring indicator does not reach the preset risk level, then determine whether the uncertainty measure of the monitoring indicator is less than or equal to the preset uncertainty threshold and whether the evolution risk level of the monitoring indicator reaches the preset risk level. If the uncertainty measure of the monitoring indicator is less than or equal to the preset gate threshold, and the evolution risk level of the monitoring indicator reaches the preset risk level, then the evolution risk level of the monitoring indicator is taken as the final risk level. If the uncertainty measure of the monitoring indicator is greater than the preset threshold or the evolution risk level of the monitoring indicator does not reach the preset risk level, then the maximum value between the abnormal risk level and the engineering risk level of the monitoring indicator shall be taken as the final risk level. While generating the final risk level, explainable cause chain information is recorded simultaneously.

10. A tunnel engineering monitoring system based on multiple risk identification, characterized in that, include: The data acquisition module is used to acquire multi-source monitoring data of tunnel engineering, bind each monitoring data with the corresponding working condition information, and obtain monitoring data with working condition tags; The preprocessing module is used to preprocess the monitoring data with working condition labels to obtain preprocessed monitoring data; An anomaly risk calculation module is used to predict the future state of each monitoring indicator based on the preprocessed monitoring data, and obtain the anomaly risk level and uncertainty measure value of each monitoring indicator. The evolution risk calculation module is used to extract and weight and combine multiple evolution features of each monitoring indicator based on the preprocessed monitoring data to obtain the evolution risk level of each monitoring indicator. The engineering state machine module is used to obtain the engineering risk level of each monitoring indicator by performing logical arbitration through the engineering state machine based on the preprocessed monitoring data and the corresponding working condition labels. The fusion decision module is used to make fusion decisions on the abnormal risk level, evolution risk level and engineering risk level of each monitoring indicator based on the uncertainty metric value of each monitoring indicator, and generate the final risk level and explainable cause chain information.

Images