A space-time self-consistent and multi-factor adaptive risk assessment system and method for a typical scene of a tunnel structure

Abstract

The application discloses a kind of tunnel structure typical scene-oriented space-time self-consistent and multi-factor adaptive risk assessment system and method, the system includes at least one multi-source sensor array integration intelligent perception terminal, data convergence and space-time registration module, scene identification and factor coupling effect module, variable weight dynamic risk assessment engine and man-machine interaction and early warning terminal;Data are collected and structured data are generated by perception terminal;Four-level layered space reference frame is constructed by space-time registration module;Typical scene faced by tunnel is identified by scene identification module and corresponding five types of disaster-causing factor cause-effect coupling effect matrix is called;Variable weight dynamic risk assessment engine generates adaptive weight, and assesses tunnel structure risk;And variable weight dynamic risk assessment engine is built-in rule engine, adjusts sampling strategy to perception terminal with assessment result feedback, forms full-link closed loop, significantly improves the accuracy, adaptability and intelligent level of tunnel risk assessment.

Description

Technical Field

[0001] This invention relates to a spatiotemporally self-consistent and multi-factor adaptive risk assessment system and method for typical tunnel structure scenarios. Background Technology

[0002] During operation, highway tunnels are affected by multiple factors such as geological conditions, climate, and traffic loads, resulting in significant spatiotemporal variability and scenario-specific structural safety risks. Particularly in typical scenarios such as collapsible loess strata, high-altitude cold regions with freeze-thaw cycles, landslide and debris flow impact areas, and high-intensity earthquake zones, the disaster-causing mechanisms of tunnel structures vary, and the evolution patterns of loads and defects differ significantly, posing severe challenges to structural health monitoring and risk assessment.

[0003] Currently, there is a large amount of research and engineering practice both domestically and internationally regarding tunnel structure safety monitoring and risk assessment. For example, Chinese patent application CN120724355A discloses a tunnel environment monitoring method and system based on multi-source data fusion. This method achieves comprehensive perception of multi-source data by performing multi-source monitoring of the target tunnel and synchronizing, preprocessing, and extracting spatiotemporal features from the multi-source monitoring data. Another Chinese patent application CN120706883A discloses a gas tunnel construction risk assessment method based on game theory combination weighting. This method uses the analytic hierarchy process (AHP) to obtain subjective weights, combines them with the anti-entropy weighting method to calculate objective weights, and solves for the optimal combination coefficients of subjective and objective weights through a game theory combination weighting model. At the academic research level, there are also many reports on tunnel risk assessment related to using dynamic Bayesian networks to construct multi-source information fusion analysis models or tunnel deformation early warning based on LSTM-CNN models.

[0004] However, in practice, it has been found that existing tunnel structure safety monitoring schemes only perform spatiotemporal alignment processing at the data preprocessing level when extracting features from multi-source monitoring data. Furthermore, the spatial positioning of monitoring sections largely depends on the initial mileage markers. During the service of a tunnel, factors such as earthquakes, landslides, and surrounding rock creep can cause longitudinal displacement or circumferential deformation of its structure, resulting in the original mileage markers failing to accurately reflect the actual spatial position of the sensors. Consequently, the spatial correspondence between historical and current data is lost, leading to a decrease in the accuracy of risk assessment.

[0005] In addition, the algorithm weight adjustment of existing risk assessment schemes mainly relies on changes in statistical characteristics at the data level. Once the monitoring system is deployed, it remains relatively fixed. When the tunnel faces different typical disaster scenarios, the assessment model still uses the original weight system, resulting in a serious misalignment between the assessment focus and the actual risk situation. Moreover, the tunnel structure has progressive degradation characteristics such as material aging and resistance reduction as the service life increases. If the assessment focus cannot be dynamically adjusted with the evolution of the service stage, the assessment results will also differ greatly from the actual risk value.

[0006] In addition, most existing risk assessment systems are one-way processes. The data collected by the front-end sensors is uploaded to the central platform and then the assessment model is run to output the results. This does not involve the reverse optimization of the front-end perception strategy based on the assessment results. As a result, the monitoring data in high-risk areas is sparse, while the data in low-risk areas is over-collected, causing an imbalance in the allocation of monitoring resources and a lag in emergency response. This makes it difficult to meet the actual needs of intelligent risk assessment throughout the entire life cycle of tunnels.

[0007] Therefore, how to establish a spatiotemporal reference system with adaptive updates to the spatial benchmark for the entire life cycle of a tunnel from the system architecture level, which can adaptively adjust risk weights according to the actual situation of the scenario and tunnel structure, and can reversely optimize the front-end perception strategy based on the assessment results to achieve closed-loop control of the entire link, remains a key technical problem that urgently needs to be solved in the field of tunnel risk assessment. Summary of the Invention

[0008] To address the aforementioned problems in tunnel safety monitoring and risk assessment, and to achieve the aforementioned objectives, this invention provides a spatiotemporally self-consistent and multi-factor adaptive risk assessment system and method for typical tunnel structure scenarios. It achieves spatiotemporal self-consistency by constructing a four-level hierarchical spatial reference system, achieves adaptive weighting through the introduction of confidence constraints and dual adjustment based on scenario and service stage, quantifies effects using a hybrid-driven five-dimensional factor coupling matrix, and forms a closed-loop chain through a built-in event-action rule engine. The combined effect of these multiple methods significantly improves the accuracy, adaptability, and intelligence level of tunnel risk assessment. The specific technical solution is as follows:

[0009] First, this invention proposes a spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios, including four types: collapsible loess strata, high-altitude freeze-thaw zones, landslide and debris flow affected areas, and high-intensity earthquake zones; the system includes:

[0010] At least one integrated intelligent sensing terminal with a multi-source sensor array is used to simultaneously collect various monitoring data and generate structured sensing data with timestamps and preliminary location identifiers after edge preprocessing.

[0011] The data aggregation and spatiotemporal registration module is communicatively connected to the integrated intelligent sensing terminal of the multi-source sensor array. It is used to construct a four-level hierarchical spatial reference system and map the structured sensing data to a five-dimensional spatiotemporal grid composed of time and the four-level hierarchical spatial reference system, thereby generating a spatiotemporally consistent standardized sensing dataset.

[0012] The scene identification and factor coupling effect module is communicatively connected to the data aggregation and spatiotemporal registration module. It is used to identify the typical scene where the tunnel is currently located based on the standardized perception dataset and call the causal coupling effect matrix of five types of disaster-causing factors corresponding to the scene.

[0013] The variable-weight dynamic risk assessment engine is communicatively connected to the data aggregation and spatiotemporal registration module and the scene identification and factor coupling effect module. It includes a variable-weight calculation unit and a comprehensive risk assessment unit. The variable-weight calculation unit is used to dynamically generate adaptive weights for evaluation indicators based on the typical scenario and service stage of the tunnel. The comprehensive risk assessment unit calculates the tunnel structure safety risk index and classifies the risk level based on the standardized perception dataset, the causal coupling effect matrix of five types of disaster-causing factors, and the adaptive weights.

[0014] The human-computer interaction and early warning terminal is connected to the variable weight dynamic risk assessment engine to visualize the risk index and risk level, and to issue a graded early warning when the risk level exceeds the threshold.

[0015] The aforementioned spatiotemporal self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios includes a unified clock synchronization module in the integrated intelligent sensing terminal of the multi-source sensor array. This module uses BeiDou / GPS dual-mode time synchronization when there is a satellite signal, and achieves clock synchronization through a multi-terminal mutual calibration mechanism based on the cross-correlation of vibration signals when there is no satellite signal or communication interruption.

[0016] The multi-terminal mutual calibration mechanism is specifically as follows:

[0017] For vibration signal sequences of similar dynamic sensors in integrated intelligent sensing terminals with different multi-source sensor arrays, calculate the normalized cross-correlation coefficient between signals of any two terminals (terminal pairs); when the peak value of the cross-correlation coefficient is greater than a preset threshold, it is determined that the two terminals have experienced the same sudden excitation event, and the relative time deviation between the two terminals is extracted using the time delay position corresponding to the peak value of the cross-correlation coefficient; traverse all terminals involved in the event to obtain multiple redundant relative time deviations between terminal pairs within the terminal group;

[0018] The terminal with the lowest crystal oscillator aging rate in the terminal group and the first to restore the communication link with the data aggregation and spatiotemporal registration module is used as a temporary benchmark. Based on the measured time deviation and steps of each terminal relative to this temporary benchmark... The redundant relative time deviation between each terminal pair, the crystal oscillator aging rate of each terminal, and the communication interruption duration are extracted to establish an overdetermined system of equations with the global optimal compensation value of each terminal as the unknown. In addition to the direct observation equations from the temporary reference to each terminal, the overdetermined system of equations also includes constraint equations composed of the redundant relative time deviations, so that the total number of equations is greater than the number of compensation values ​​to be determined.

[0019] The overdetermined equations are solved using the least squares method to obtain the global optimal compensation value for each terminal. This is used to uniformly compensate the timestamps of data collected by each terminal during the interruption period, thereby achieving synchronization of the clocks of all terminals in the group.

[0020] The aforementioned spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios includes a four-level hierarchical spatial reference system comprising a logical section number S, a dynamic physical mileage d, a circumferential sector code φ, and a radial depth layer r. The logical section number S is a unique and unchanging identifier throughout its entire lifecycle, while the dynamic physical mileage d is dynamically updated with structural deformation. These two elements constitute a longitudinal dual-identifier decoupling mechanism, used to track the axial displacement field when longitudinal displacement occurs in the tunnel in real time. The circumferential sector code φ is used for circumferential positioning, and the radial depth layer r is used for radial positioning.

[0021] The construction process of the four-level hierarchical spatial reference system is as follows:

[0022] Step 1: Assign a unique and unchanging logical section number S to each monitoring section based on the tunnel design parameters and sensor deployment information;

[0023] Step 2: The associated design mileage is used as the initial dynamic physical mileage d0, which is dynamically updated as the tunnel structure deforms. The mileage of the most recent accurate measurement of the cross-section is used as the station number and recorded as the dynamic physical mileage d.

[0024] Step 3: Divide the discrete sectors along the circumferential direction of the tunnel cross section and assign codes, denoted as circumferential sector codes φ;

[0025] Step 4: Divide the tunnel lining thickness along the inner surface outwards, according to the sensor burial depth, into layers and assign codes, denoted as radial depth layer r.

[0026] The aforementioned spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios uses a matrix of causal coupling effects of five types of disaster-causing factors. This matrix is ​​constructed based on a Bayesian network according to five types of disaster-causing factors, including geological conditions, environmental loads, structural status, maintenance history, and traffic loads. During system operation, the conditional probability parameters of this matrix are iteratively updated using the expectation-maximization algorithm based on the observed tunnel structural status and environmental load data. This update is used to correct the superimposed effects between the risk values ​​of each disaster-causing factor dimension.

[0027] Specifically, the process of constructing and updating the causal coupling effect matrix of the five types of disaster-causing factors is as follows:

[0028] a) Based on the well-known causal relationships in the field of tunnel engineering, a qualitative analysis is conducted on the causal relationships among five disaster-causing factors, including geological conditions, environmental loads, structural conditions, maintenance history, and traffic loads.

[0029] b) Using the five types of disaster-causing factors as nodes, each node is discretized into three states: good, deteriorated, and dangerous. The causal edges between each node are defined as: geological conditions → environmental load, geological conditions → structural state, environmental load → structural state, maintenance history → structural state, and traffic load → structural state. A Bayesian network is drawn accordingly.

[0030] c) Define the factor coupling effect matrix as a 5×5 square matrix, whose elements (i, j=1…5) represents the coupling influence strength of factor j on factor i among the five types of disaster-causing factors; where the diagonal elements are 0, it indicates that the factor has no coupling with itself; the off-diagonal elements are... ( The value range of )) is [0, 1]. The larger the value, the stronger the driving effect of the deterioration of the state of factor j on the deterioration of the state of factor i.

[0031] d) Based on historical disease cases and known engineering experience, a prior condition probability table is assigned to each node of the Bayesian network; according to the prior condition probability table, the probability increment of the related nodes in the dangerous state when each node changes from a good state to a dangerous state is extracted, and the weighted average is calculated according to the frequency of each state in historical cases to obtain the influence intensity; all influence intensities are normalized, and the resulting value is the initial coupling coefficient of the matrix.

[0032] e) During system operation, the parameters of the conditional probability table of the Bayesian network are iteratively optimized using the expected maximization algorithm based on the observed structural state and environmental load data. In the iterative process, a historical data weight exponential decay mechanism is introduced to assign weights λ=exp(-Δt / τ) to samples in different time windows. Where: Δt is the time difference between the sample and the current time, and τ is a preset time constant.

[0033] f) After iterating alternately between the expectation step and the maximization step until convergence, the coupling coefficients are re-extracted based on the updated conditional probability table, and the causal coupling effect matrix of the five types of disaster-causing factors is updated.

[0034] The aforementioned spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios, wherein the variable weight calculation unit dynamically generates adaptive weights for evaluation indicators based on the current scenario and service stage, involves the following steps: determining the initial subjective weights using the standard fuzzy hierarchical analysis method; calculating the objective weights using the improved entropy weight method; establishing a combined weight formula and a set of game equilibrium equations based on a game theory-based combined weighting method, solving the equations, and normalizing the solution vectors to obtain the final basic combined weights; then introducing the scenario type adjustment vector and the service stage decay coefficient vector, multiplying the basic combined weights element-wise with the scenario type adjustment vector and the service stage decay coefficient vector, and normalizing the result to generate the final adaptive weights. Wherein:

[0035] The improved entropy weight method introduces a data confidence factor. Select high-confidence sample points, only using The information entropy and difference coefficient of the samples are calculated, and the learning rate is dynamically adjusted according to the proportion of high-confidence samples to suppress the magnitude of weight updates when the data quality is poor, so as to obtain robust objective weights.

[0036] The formula for calculating the combined weights is:

[0037] ;

[0038] In the formula: For combined weights; Subjective weighting, For objective weighting; The coefficients are linear combination coefficients of subjective weights. Let be the linear combination coefficients of the objective weights, and satisfy . .

[0039] The adaptive weights are:

[0040] ;

[0041] In the formula: For adaptive weights, Adjust the vector for scene type. This is the service phase degradation coefficient vector.

[0042] The aforementioned spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios, wherein the comprehensive risk assessment unit calculates the tunnel structure safety risk index based on a standardized sensing dataset combined with a causal coupling effect matrix of five types of disaster-causing factors and adaptive weights, is as follows: Using the standardized sensing dataset as input, the system first aggregates the risks of the five types of disaster-causing factors to obtain five-dimensional risk values. Then, it uses the causal coupling effect matrix of the five types of disaster-causing factors corresponding to the typical scenario in which the tunnel is located to perform a chain correction on the five-dimensional risk values. Finally, it uses adaptive weights to perform a weighted summation of the corrected five-dimensional risk values ​​to obtain the final comprehensive structural safety risk index; wherein:

[0043] The five categories of disaster-causing factors are aggregated as follows: based on a preset indicator-factor mapping matrix. The risk value of a single index is obtained by substituting each monitored physical quantity at the bottom layer into the membership function. The initial risk values ​​of the five hazard-causing factors are weighted and aggregated. ,in These correspond to geological conditions, environmental loads, structural status, maintenance history, and traffic loads, respectively.

[0044] The chain correction formula is:

[0045] ;

[0046] In the formula: Let k be the initial risk value for dimension k. Let j be the coupling coefficient of factor j to factor k (which resides in the matrix C corresponding to the current scene, and...) ), This is the risk value after coupling correction, with an upper limit of 1.0.

[0047] The formula for calculating the tunnel structure safety risk index is as follows:

[0048] ;

[0049] In the formula: The tunnel structure safety risk index. For the adaptive weights of dimension k.

[0050] The risk level classification is based on the calculated tunnel structure safety risk index. Mapped to the range of 0 to 100; Low risk Medium risk. High risk, This is extremely high risk.

[0051] The aforementioned spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios also includes a built-in event-action rule engine in its variable-weight dynamic risk assessment engine. This engine automatically generates closed-loop control commands based on the risk assessment results. These commands are precisely addressed by the data aggregation and spatiotemporal registration module according to the logical section number S and circumferential sector number φ of the target device, and then forwarded to the corresponding integrated intelligent sensing terminal with multi-source sensor array. This allows the terminal to adjust its sampling strategy, forming a complete closed-loop process from assessment to sensing optimization.

[0052] Secondly, this invention also provides a spatiotemporally consistent and multi-factor adaptive risk assessment method for typical tunnel structure scenarios, specifically by applying the aforementioned system for assessment; including the following steps:

[0053] S1: Multiple monitoring data are collected simultaneously through an integrated intelligent sensing terminal with a multi-source sensor array, and after edge preprocessing, structured sensing data with timestamps and preliminary location identifiers are generated;

[0054] S2: A unified four-level hierarchical spatial reference system for the entire tunnel is constructed by the data aggregation and spatiotemporal registration module, which maps the structured sensing data to a five-dimensional spatiotemporal grid composed of time and the four spatial dimensions, thereby obtaining a standardized sensing dataset.

[0055] S3: The scene identification and factor coupling effect module identifies the current typical scene of the tunnel based on the standardized perception dataset, and calls the five types of disaster-causing factor causal coupling effect matrix corresponding to the scene;

[0056] S4: The variable-weight dynamic risk assessment engine adaptively adjusts the weights of evaluation indicators based on the current typical scenario and the tunnel service stage, and uses the causal coupling effect matrix and the standardized perception dataset to calculate the structural safety risk index and classify the risk level;

[0057] S5: Visualized display and tiered early warning provided through human-computer interaction and early warning terminals;

[0058] S6: When the risk level exceeds the preset threshold, the event-action rule engine built into the variable weight dynamic risk assessment engine automatically generates control instructions. Through the data aggregation and spatiotemporal registration module, the instructions are accurately addressed according to the logical section number S and the circumferential sector number φ of the target device, and forwarded to the corresponding multi-source sensor array integrated intelligent sensing terminal to adjust its sampling strategy.

[0059] The beneficial effects of this invention are as follows:

[0060] 1) This invention utilizes the collaborative work of an integrated intelligent sensing terminal with a multi-source sensor array and a data aggregation and spatiotemporal registration module. The former outputs structured data with preliminary location identifiers, while the latter constructs a four-level hierarchical spatial reference system consisting of logical section number S, dynamic physical mileage d, circumferential sector code φ, and radial depth layer r, and maps the data to a five-dimensional spatiotemporal grid. This enables the monitoring data of the entire tunnel to obtain a unified benchmark in both time and space dimensions, solves the problem of data alignment from heterogeneous sensors, and lays a high-fidelity data foundation for subsequent risk assessment.

[0061] 2) This invention utilizes the synergistic effect of the five-dimensional spatiotemporal grid data output by the data aggregation and spatiotemporal registration module and the scene identification and factor coupling effect module. The former provides refined spatial data including circumferential sectors and radial depth layers, while the latter identifies the current typical scene based on these data and calls the corresponding five types of disaster-causing factor causal coupling effect matrices, enabling scene identification to distinguish the differentiated responses of different circumferential positions on the same cross section, thus greatly improving the identification accuracy.

[0062] 3) This invention utilizes the collaborative work of the scenario identification and factor coupling effect module and the variable weight dynamic risk assessment engine. The former outputs scenario labels and factor coupling matrices, while the latter introduces scenario type adjustment vectors and service stage degradation coefficient vectors to dual-adjust the basic weights. The coupling matrix is ​​then used to couple and correct the risk values ​​of each dimension before weighted aggregation. This enables the assessment model to dynamically adjust its focus as typical scenarios change and structural service degradation occurs, while quantifying the chain evolution effect between disaster-causing factors, thus significantly reducing assessment errors.

[0063] 4) This invention utilizes the event-action rule engine built into the variable weight dynamic risk assessment engine and the closed-loop feedback of the integrated intelligent sensing terminal with multi-source sensor array. When the risk level exceeds the standard, the former automatically generates control commands, which are then accurately addressed by the data aggregation and spatiotemporal registration module according to the logical section number S and the circumferential sector φ before being forwarded to the latter to adjust the sampling strategy. This enables automatic encryption of monitoring in high-risk areas and avoids over-collection in low-risk areas, forming a closed-loop chain from assessment to perception optimization, which significantly shortens the emergency response time.

[0064] Overall, this invention, through the step-by-step driving and feedback coordination of the above four stages, from raw data acquisition, spatiotemporal benchmark unification, scene adaptive identification, factor coupling quantitative evaluation to perception strategy closed-loop optimization, realizes the system property change of tunnel risk assessment from heterogeneous scattered points, fixed weights, and unidirectional open loop to spatiotemporal self-consistency, factor coupling, and adaptive closed loop, significantly improving the accuracy, adaptability, and intelligence level of tunnel structure safety assessment. Attached Figure Description

[0065] Figure 1 This is a block diagram of the overall composition of the spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios of the present invention.

[0066] Figure 2 This is a system deployment architecture diagram for the present invention;

[0067] Figure 3 This is a schematic diagram of the structure of the integrated intelligent sensing terminal with multi-source sensor array of the present invention;

[0068] Figure 4 This is a schematic diagram of the multi-terminal mutual calibration mechanism of the present invention;

[0069] Figure 5 This is a schematic diagram illustrating the spatial grid division and spatiotemporal registration of the data aggregation and spatiotemporal registration module in the system of the present invention;

[0070] Figure 6 This is a flowchart of the scene identification and factor coupling effect module of the system of the present invention;

[0071] Figure 7This is a flowchart illustrating the dynamic evaluation process performed by the variable-weight dynamic risk assessment engine of the present invention. Detailed Implementation

[0072] To address the shortcomings of existing tunnel structure safety monitoring and risk assessment systems, such as the lack of a refined spatiotemporal unified benchmark for heterogeneous sensor data, reliance on initial mileage markers for monitoring sections leading to inaccurate localization of the same structural entity after longitudinal displacement or circumferential deformation, relatively fixed risk assessment weighting systems that cannot dynamically adjust assessment priorities based on typical scenarios (collapsible loess, high-altitude freeze-thaw cycles, landslides and debris flows, high-intensity earthquakes) and service stage evolution, systematic neglect of causal coupling effects among multiple disaster-causing factors, and the use of linear weighting in assessment models to aggregate risks resulting in systematic bias, and the disconnect between the perception and assessment layers preventing back-end optimization of front-end perception strategies based on assessment results, leading to a unidirectional open-loop process and delayed emergency response, this invention provides a spatiotemporally self-consistent and multi-factor adaptive risk assessment system and method for typical tunnel structure scenarios. Figure 1 As shown, the system includes an integrated intelligent sensing terminal with multi-source sensor arrays, a data aggregation and spatiotemporal registration module, a scene identification and factor coupling effect module, a variable weight dynamic risk assessment engine, and a human-computer interaction and early warning terminal.

[0073] The integrated intelligent sensing terminal with multi-source sensor array (hereinafter, "intelligent sensing terminal," "sensing terminal," or "terminal," etc., are all the same concept) can be one or more, such as Figure 2 As shown, it communicates with the data aggregation and spatiotemporal registration module via fieldbus, LoRa, 4G / 5G, or Ethernet; the data aggregation and spatiotemporal registration module, the scene identification and factor coupling effect module, and the variable weight dynamic risk assessment engine are deployed on a local server or cloud platform, and exchange data with each other via bus or network. The human-computer interaction and early warning terminal obtains the assessment results from the variable weight dynamic risk assessment engine through the network, and can directly send the user-configured closed-loop control rules to the variable weight dynamic risk assessment engine, thereby forming a full-link closed loop from front-end perception to back-end assessment and early warning, and then feedback to front-end perception strategy adjustment.

[0074] The integrated intelligent sensing terminal with multi-source sensor array, such as Figure 3 The diagram shows the front-end data source and edge processing core of the system of this invention, which includes a sensor integration module, a unified clock synchronization module, an edge intelligent data processing unit, a unified communication module, a unified power supply module, and a modular sealed housing. Specifically, the sensor integration module, through modular slots and configured acquisition channels, allows for the selection of various sensors based on the typical scenario of the tunnel. These sensors include strain sensors, displacement sensors, crack sensors, temperature and humidity sensors, and seepage pressure sensors;

[0075] The strain sensor is a vibrating wire or fiber optic grating surface strain gauge with built-in temperature compensation. It is installed on the inner surface of the lining or embedded in the structure to monitor the micro-strain of the lining caused by the surrounding rock pressure.

[0076] The displacement sensor includes a laser rangefinder and a wire displacement meter, which are deployed at key displacement sections such as construction joints, expansion joints, arches, and sidewalls to acquire convergence deformation and joint misalignment in real time.

[0077] The crack sensor is a vibrating wire crack gauge or an LVDT crack gauge, which is installed across the crack to monitor changes in the width of existing cracks and the trend of new cracks.

[0078] The temperature and humidity sensor integrates a digital temperature and humidity probe and an internally embedded thermometer, and is deployed at different depths on the lining surface and inside. Combined with atmospheric temperature and humidity comparison, it is used for environmental load analysis such as freeze-thaw cycles.

[0079] The pressure sensor is a vibrating wire piezometer or a piezoresistive pressure transmitter, which is installed in the drainage blind ditch behind the lining and under the invert arch to monitor pore water pressure and seepage status.

[0080] All of the above sensors convert their outputs into standard 4-20mA current or 0-10V voltage signals through a unified signal conditioning circuit. These signals are then connected to the signal conditioning and acquisition board within the integrated intelligent sensing terminal of the multi-source sensor array via an aviation connector, thus achieving a unified physical interface.

[0081] The unified clock synchronization module incorporates a BeiDou / GPS dual-mode timing module and a highly stable temperature-controlled crystal oscillator that is rigorously calibrated before leaving the factory. This invention preferentially uses the OCXO-2522 temperature-controlled crystal oscillator, which has a short-term stability better than 1×10⁻⁶. -12 The aging rate is better than ±0.5 ppb / day. When satellite signal is available, the timing module provides hardware-level timestamps for all sensor channels and simultaneously calibrates the frequency of the temperature-controlled crystal oscillator, locally recording clock deviation parameters to ensure strict synchronization of sampling data from dissimilar sensors.

[0082] In extreme conditions where there are no satellite signals and communication networks are interrupted deep in tunnels, the system of this invention achieves time self-consistency and uniformity through a multi-terminal mutual calibration mechanism. For example... Figure 4 As shown, the multi-terminal mutual calibration mechanism is as follows:

[0083] Each integrated intelligent sensing terminal of multi-source sensor arrays relies on a constant-temperature crystal oscillator to keep time independently; at this time, the edge intelligent data processing unit marks each piece of collected data with a local timestamp based on the local crystal oscillator.

[0084] After communication is restored, the data aggregation and spatiotemporal registration module receives data with local timestamps. Combining this with pre-stored crystal oscillator aging rate and interruption duration, it first uses a simplified compensation model. Perform preliminary calibration.

[0085] Because the same sudden excitation event is recorded simultaneously by multiple terminals (spatial redundancy), and each terminal performs continuous high-frequency sampling of the event (temporal redundancy), a far greater number of observation equations than unknowns can be obtained, forming an overdetermined equation set. This overdetermined equation set includes, in addition to the direct observation equations from the temporary reference to each terminal, constraint equations formed by the redundant relative time deviations, making the total number of equations greater than the number of compensation values ​​to be determined. Specifically:

[0086] After initial calibration, the spatiotemporal redundancy of data collected by multiple terminals in the tunnel can be used to calculate the normalized cross-correlation coefficient between any two terminal (terminal pair) signals for vibration signal sequences of the same dynamic sensors in different sensing terminals. When the peak value of the cross-correlation coefficient is greater than 0.7 (preset threshold), it is determined that the two terminals have experienced the same sudden excitation event.

[0087] Specifically, taking a vibration signal sampled at 100Hz as an example (the actual sampling rate depends on the terminal configuration), the signal sequence is first normalized, and a 256-point sliding window is used to calculate the normalized cross-correlation coefficient between any two terminal signals. When the peak value of the cross-correlation coefficient is greater than 0.7, it is determined that the two terminals have experienced the same sudden excitation event, and the time delay position corresponding to this peak value is the relative time deviation between the two terminals. (Subscripts p and q are terminal numbers), definition That is, the time difference between terminal q and terminal p.

[0088] For a group of terminals that can be associated in this way (e.g., a total of M terminals), the temporary reference terminal is determined according to the following priority rules:

[0089] (1) The terminal with the lowest crystal aging rate in the last system self-test before the interruption occurs should be selected as the reference terminal;

[0090] (2) If the terminal fails to establish a connection during this communication recovery process, the terminal with the lowest crystal oscillator aging rate shall be selected from the remaining terminals.

[0091] (3) If there are multiple lowest terminals with the same crystal oscillator aging rate, the first terminal that establishes a stable heartbeat connection with the data aggregation and spatiotemporal registration module (successfully sends and receives 3 heartbeat messages in a row, and the maximum round-trip delay is less than 100ms) is selected as the temporary reference terminal.

[0092] Let the global optimal compensation value for each terminal be... (i=1...M), which are the unknowns to be solved. Let the sequence number of the temporary reference terminal be 1, and for the temporary reference terminal, its compensation value be set as... For any other terminal tag, a direct observation equation can be constructed: ,in The measured relative time delay between the temporary baseline and the terminal tag (as defined above) Simultaneously, for any two non-reference terminals a and b, the redundant relative deviation is... Constraint equations can be constructed: Therefore, a total of more than There are M-1 equations with M unknowns. When M ≥ 3, the number of equations exceeds the number of unknowns, forming an overdetermined system of equations. The least squares method is used to solve this system of equations, obtaining a set of globally optimal compensation values. These values ​​are then used to compensate for the timestamps of data collected by all terminals during the interruption period, achieving mathematical self-consistency and unification of the clock.

[0093] For terminals that only have static sensors such as temperature and osmotic pressure deployed but not connected to dynamic sensors, they cannot participate in the cross-calibration and rely on aging model compensation. This joint calibration ensures the validity of multi-source data in the time dimension.

[0094] In this invention, the edge intelligent data processing unit adopts a heterogeneous computing core of ARM and FPGA, runs an embedded system, and internally integrates data cleaning algorithms, anomaly detection algorithms, and feature extraction algorithms.

[0095] The data cleaning algorithm eliminates electromagnetic interference and isolated outliers by improving sliding window midpoint filtering and adaptive outlier removal based on median absolute deviation.

[0096] The anomaly detection algorithm combines an adaptive threshold and a one-dimensional convolutional autoencoder model to identify anomalies in slow variables and dynamic quantities respectively. A continuous period of data (usually no less than 7 days, covering the entire operation and environmental cycle) after system deployment, manually confirmed to be in normal structural condition, is used as normal historical data to train the one-dimensional convolutional autoencoder, obtaining the mean reconstruction error μ. error and standard deviation 3σ error During online detection, if the reconstruction error is greater than μ... error +3σ error If the value is not specified, it is considered abnormal. Among them, slow variables refer to physical quantities such as temperature and osmotic pressure with a rate of change of less than 0.01% / min, while dynamic variables refer to physical quantities such as vibration and displacement rates with a higher rate of change. Different autoencoder models are trained for each type, and the feature extraction algorithm calculates features such as peak value, mean, and rate of change according to a set granularity.

[0097] The structured sensing data generated by the system of this invention is encapsulated in JSON or Protobuf format. Each message contains device ID, timestamp, spatial location code, sensor type, physical quantity value and data quality identifier.

[0098] The legal entity spatial location code consists of a quadruple: logical section number S, dynamic physical mileage d, circumferential sector code φ, and radial depth layer r. This spatial reference system is constructed according to the following steps:

[0099] Step 1: Establish Logical Section (S): The system assigns a unique, incrementing integer starting from 1 as its logical section number S to each monitoring section marked in the design drawings or the section where the sensor is actually installed. Once assigned, this number remains unchanged throughout the entire lifecycle.

[0100] Step 2: Associate the initial dynamic physical mileage (d0): The system reads the "Mileage Station Number and Cross-section Relationship Table" in the tunnel design file and associates the design mileage of each logical cross-section S with the initial dynamic physical mileage d0.

[0101] Step 3: Define the circumferential sector (φ): Using the center of the tunnel cross-section as the center and the top of the arch as the 0-degree starting point, divide the 360° circumference evenly into 8 sectors in a clockwise direction. Each sector covers a 45° arc segment, and assign a code φ to each sector: φ0—arch top, φ1—left arch shoulder, φ2—left arch waist, φ3—left arch foot, φ4—inverted arch, φ5—right arch foot, φ6—right arch waist, φ7—right arch shoulder. When the sensor is installed, its circumferential angle is mapped to the nearest sector code.

[0102] Step 4: Define the radial depth layer (r): Based on the lining structure design and sensor installation specifications, three depth layers are pre-defined: r0—surface layer (inner surface of the lining, 0-2cm), primarily controlling the risk of spalling and spalling; r1—shallow layer (burial depth 2-10cm, protective layer range), primarily controlling the risk of steel reinforcement rust expansion and cracking; r2—deep lining layer (burial depth >10cm), primarily controlling the risk of structural crushing due to surrounding rock pressure. Each sensor is labeled with a corresponding r based on its installation depth.

[0103] The unified communication module of this invention supports multiple communication methods such as RS485 bus, Ethernet, LoRa, and 4G / 5G. The protocol layer adopts MQTT or CoAP IoT protocols, supporting offline caching and offline data transmission to ensure that the integrated intelligent sensing terminal with multi-source sensor array can still operate independently in communication dead zones and automatically upload data after network recovery. The unified power supply module supports AC power supply, DC power supply, and PoE power supply, and has built-in overcurrent and overvoltage protection and a lithium iron phosphate backup battery, enabling the integrated intelligent sensing terminal with multi-source sensor array to operate for no less than 8 hours after a power outage. The modular sealed housing has an overall protection level of IP68 and is fixed to different tunnel lining surfaces via an adaptive mounting base. This base has a universal adjustment function to ensure that the sensor axis is aligned with the required monitoring direction, and the installation process does not damage the lining structure.

[0104] The data aggregation and spatiotemporal registration module of the system of this invention, such as Figure 5 As shown, an edge server / industrial control computer deployed at the tunnel site or a nearby monitoring center receives structured sensing data uploaded by the integrated intelligent sensing terminal of the multi-source sensor array and performs unified spatiotemporal normalization processing. The data aggregation and spatiotemporal registration module includes a data aggregation gateway and a spatiotemporal dual-dimensional calibration unit. The data aggregation gateway acts as a multi-protocol parsing engine, simultaneously accessing protocol data such as MQTT, CoAP, and ModbusTCP, authenticating and verifying the integrity of each sensing terminal, and distributing the data to the corresponding tunnel segment queue according to spatial location encoding.

[0105] The spatiotemporal dual-dimensional calibration unit performs time and space alignment operations to establish a unified four-level hierarchical spatial reference system for the entire tunnel. The specific definition of this spatial reference system is as follows:

[0106] 1) Longitudinal Positioning: Decoupling of Logical Section and Physical Mileage. Along the tunnel axis, the system employs a dual-identification positioning mechanism combining logical identifiers and physical coordinates. Each monitoring section with deployed sensors is assigned a unique and unchanging section number S during installation, serving as its logical identity throughout its entire lifecycle and remaining unchanged despite structural deformation. Simultaneously, the system records the mileage marker d (accurate to 0.1 meters) of this section in the most recent precise measurement, serving as its current physical positioning coordinates. This decoupling design ensures that even after longitudinal displacement due to earthquakes, landslides, or other events, the system can still accurately locate the same structural entity through S for historical data comparison. Furthermore, the continuously updated d value reflects and tracks the tunnel's axial displacement field under surrounding rock deformation in real time.

[0107] 2) Circumferential Positioning: Dual-scale representation of engineering zoning sectors and continuous angles. Within a monitoring section, to balance on-site installation efficiency, maintenance zoning habits, and model calculation accuracy, the system adopts a dual-scale circumferential positioning scheme. For operation and maintenance and sensor deployment, the lining circumferentially is divided into 8 main orientation sectors and discretely encoded φ, with the mapping relationship as follows: φ(0) - arch top, φ(1) - left arch shoulder, φ(2) - left arch waist, φ(3) - left arch foot, φ(4) - inverted arch, φ(5) - right arch foot, φ(6) - right arch waist, φ(7) - right arch shoulder. Each sector covers an arc segment of approximately 45°. At the same time, for refined damage modeling, the system synchronously stores the precise radial angle at the time of installation in the sensor configuration metadata. (From 0° to 360° clockwise from the arch) This supports precise calculations of details such as the direction of the crack and the angle between the principal stresses during advanced analysis.

[0108] 3) Radial Positioning: Layered Engineering Decision-Making Based on Disaster-Causing Mechanisms. Based on the sensor's embedment depth in the lining and the typical failure modes it controls, three radial depth layers r are defined, allowing the assessment results to directly guide the corresponding maintenance decisions: r0—surface layer (inner surface of the lining), primarily controlling the risk of spalling and spalling; r1—shallow layer (embedding depth ≤ 10cm, within the protective layer range), primarily controlling the risk of steel reinforcement rust expansion and cracking; r2—deep lining layer (embedding depth > 10cm), primarily controlling the risk of structural crushing due to surrounding rock pressure.

[0109] Based on the above definition, any sensor is mapped to a quadruple coordinate (S, d, φ, r) to calibrate its unique spatial location. Here, S ensures the consistency of the structural entity throughout its entire lifecycle, d provides a dynamically updatable physical location reference (updated periodically by measurement when the tunnel undergoes longitudinal displacement), and φ and r together characterize the partitions and depth within the cross-section, thereby constructing a spatiotemporal grid system that is physically and logically unified and capable of adapting to structural deformation.

[0110] In the time dimension, the spatiotemporal dual-dimensional calibration unit sets a uniform time grid step Δt = 1 hour. For the time series of each monitored physical quantity, third-order Lagrange interpolation or nearest-neighbor resampling is used to ensure that all data points have a uniform time interval. Simultaneously, the spatial coordinates of each data point are mapped to the nearest quadruple spatial grid node. For multiple data points of the same type falling into the same spatiotemporal grid node (t, S, d, φ, r), inverse distance weighted averaging or arithmetic averaging is used for fusion processing. Through this process, a multi-dimensional standardized sensing dataset covering strain field, displacement field, crack aperture field, temperature field, humidity field, and seepage pressure field is generated, internally organized into a high-dimensional spatiotemporal cube structure with (t, S, d, φ, r) as the index.

[0111] The scene recognition and factor coupling effect module of the present invention, such as Figure 6As shown, this is used to automatically identify the typical scenario currently in which the tunnel is located and match the corresponding disaster-causing factor coupling model. The scenario identification and factor coupling effect module includes a typical scenario parameter library, a scenario identification engine, and a factor coupling effect matrix construction unit.

[0112] The typical scenario parameter database stores key environmental and structural characteristic parameters for four major categories of typical scenarios: collapsible loess strata, freeze-thaw cycles in high-altitude cold regions, landslide and debris flow affected areas, and high-intensity earthquake zones, including the collapsibility coefficient. Annual freeze-thaw cycles Inclination angle of the tunnel entrance slope Design peak ground acceleration The system includes parameters and corresponding discrimination rules. These key parameters are derived from tunnel geological survey reports, design documents, and a database of standards and specifications.

[0113] The scene identification engine is configured to use a fusion of rule matching and machine learning classifiers to determine the current scene category based on tunnel geological information, environmental load characteristics, and dynamic features of monitoring data. Rule matching refers to scene discrimination based on preset threshold rules; for example, if the annual freeze-thaw cycle count in the tunnel entrance section is greater than 50, it is initially identified as a high-altitude cold-region freeze-thaw scene. The machine learning classifier is an XGBoost classification model trained on historical scene data. Input features include nine continuous features such as section geological type, quarterly average temperature and humidity, maximum displacement rate, 24-hour seepage pressure change rate, and daily temperature difference, as well as discrete features such as surrounding rock grade. The output is the probability of the scene category. The training data for this model comes from 24 consecutive months of monitoring data from three existing operational tunnels, along with corresponding geological survey reports and maintenance records.

[0114] Specifically, the process of determining the current scene category based on tunnel geological information, environmental load characteristics, and dynamic characteristics of monitoring data is as follows:

[0115] First, based on the tunnel engineering geological survey report and design drawings, a static initial scene label is assigned to each section. Then, temperature and humidity fields, seepage pressure fields, and external meteorological data from the standardized sensing dataset are read in real time, and scene determinations are activated or corrected based on the dynamic characteristics of environmental loads.

[0116] For example, suppose there are N main rules for scene identification (the specific value of N is determined by the number of rules configured for each tunnel, usually 10 to 20, covering threshold conditions for indicators such as collapsibility coefficient, annual freeze-thaw cycle count, tunnel entrance slope angle, design peak ground acceleration, daily temperature difference, and displacement rate), and currently all N rules are met, then the rule confidence level is... Simultaneously, an XGBoost classification model trained on historical scene data was deployed. This model takes nine continuous features, including section geological type, quarterly average temperature and humidity, maximum displacement rate, 24-hour seepage pressure change rate, and daily temperature difference, as well as discrete features such as surrounding rock grade, as input, and outputs the probability of scene category. Fusion weights Initially set to 0.6, then set the rule confidence level. With model probability The overall confidence level is obtained by weighted summation. The system accumulates K manually confirmed valid scene transition events, which are then used to calculate the accuracy of the recent rules and models. and ,according to Update the fusion weights and limit them to [0.3, 0.8], where the learning rate is... The value is set to 0.05, and experiments have verified that it can converge stably in the range of 0.02 to 0.1.

[0117] If manual confirmation fails to elicit a response within the set timeout window, the current weight is automatically retained, and the update request is marked as "overdue for confirmation" and moved to the end of the manual confirmation queue. If the request remains unconfirmed after the timeout window expires again in the queue, the system automatically discards the request and logs it. At that time, the scenario with the highest confidence level is established as the active scenario for the current segment.

[0118] Specifically, for high-intensity seismic zones, the system operates a rapid triggering channel in parallel. This channel directly subscribes to real-time acceleration or displacement time history data from the data aggregation and spatiotemporal registration module, calculating the Arias intensity or peak acceleration within a short time window of 1–3 seconds. When the calculated value within the short time window exceeds a preset threshold (e.g., PGA > 0.2g or Arias intensity > 0.1m / s), the system disregards the confidence level of the current slowly changing scene identification process, forcibly switching the active scene of the current segment to a high-intensity seismic zone and locking it for at least 1 hour. This mechanism ensures the system's immediate response capability to sudden seismic events.

[0119] The factor coupling effect matrix construction unit of the system of this invention is configured to construct a Bayesian network based on well-known causal relationships in the field of tunnel engineering to obtain an initial causal coupling effect matrix of five types of disaster-causing factors. During system operation, the coupling coefficients of the causal coupling effect matrix of the five types of disaster-causing factors are updated online using only observable structural state and environmental load data. The original information of the five types of disaster-causing factors comes from: geological survey reports, on-site meteorological and traffic monitoring equipment, the multi-source sensor array of this system, and maintenance operation records imported by the user. After processing, the information is stored in the database associated with the data aggregation and spatiotemporal registration module.

[0120] The factor coupling effect matrix building unit is based on a data- and knowledge-driven hybrid approach and possesses online evolution capabilities. The specific construction process is as follows:

[0121] 1) Initial Construction and Matrix Definition

[0122] Based on well-known causal relationships in tunnel engineering, this invention qualitatively analyzes the causal relationships among five disaster-causing factors: geological conditions, environmental load, structural condition, maintenance history, and traffic load. A directed acyclic graph (DAG) is then drawn as the structure of a Bayesian network. In this invention, the network is defined as follows: nodes represent geological conditions, environmental load, structural condition, maintenance history, and traffic load. Each node is discretely represented by three states: {good, deteriorated, dangerous}. Causal edges are: geological conditions → environmental load, geological conditions → structural condition, environmental load → structural condition, maintenance history → structural condition, and traffic load → structural condition. Based on desensitized historical defect cases and well-known engineering experience, a priori conditional probability table (CPT) is assigned to each node.

[0123] Define the factor coupling effect matrix C as a 5×5 square matrix, whose elements... (i, j = 1…5, corresponding to the five factors mentioned above) represents the coupling influence strength of factor j on factor i. The diagonal elements... (The factor is not coupled to itself). Off-diagonal elements. ( The value range of ) is limited to [0, 1]. The larger the value, the stronger the driving effect of the deterioration of state j on the deterioration of state i.

[0124] The initial values ​​are extracted from the conditional probability table of the Bayesian network, and the specific algorithm is as follows:

[0125] For each factor j (parent node) and each factor i (child node), with other parent nodes fixed as the baseline state (e.g., "good"), consider the probability increment of factor i being in the "dangerous" state when factor j changes from "good" to "dangerous". A weighted average is then applied across different baseline states and child node states to obtain a single influence strength. Then, for each column (i.e., when each factor i is a child node), the influence strength of all its incoming edges is normalized, such that... (Usually set to 1; if the sum of influence strengths is less than 1, the original value is retained). The normalized value is the initial coupling coefficient. , stored in matrix C. For factor pairs that do not have causal edges, .

[0126] 2) Matrix adaptation in different scenarios

[0127] The four typical scenarios (collapsible loess strata, high-altitude cold region freeze-thaw, landslide and debris flow affected area, and high-intensity earthquake zone) share the same Bayesian network topology (i.e., the same causal graph), but the conditional probability table parameters of the nodes differ in different scenarios. Therefore, the factor coupling effect matrix C varies with the scenario. The system is pre-set with a set of initial CPT parameters for each scenario at the factory, and the corresponding initial matrix is ​​calculated accordingly. When the scene recognition engine outputs a new scene category, the comprehensive risk assessment unit automatically calls the corresponding category for that scene. .

[0128] 3) Online iterative learning based on partially observable data

[0129] During system operation, the observable data mainly consists of direct or indirect measurements of structural condition (such as crack width, displacement rate, and seepage pressure) and environmental load measurements. However, the actual geological conditions, maintenance history, and traffic loads are latent variables that are difficult to observe directly. This system employs the Expectation-Maximization (EM) algorithm for parameter learning.

[0130] Let the model parameter set be θ (including all entries in the conditional probability table). E-step: Based on the current parameters and all observable data Calculate the posterior probability distribution of the latent variable Z. M-step: Maximize the expected log-likelihood function Thus, the updated parameters are obtained. Historical data. The weights decay exponentially over time; specifically, when calculating the cumulative sufficient statistic, weights are assigned to samples from different time windows. , Let τ be the time difference between the sample and the current time, and let τ be a time constant whose value is set according to the tunnel structure degradation rate. For example, for a regularly operating tunnel, it can be set to 80–120 days to balance the retention of historical information and the adaptability to new data on a quarterly scale. After the EM algorithm converges iteratively, the coupling coefficient is recalculated according to the updated Bayesian network conditional probability table and the same extraction rules as the initial construction. This allows for the online evolution of the coupling effect matrix. This mechanism enables the model to indirectly learn and update the parameters of the entire causal network, including latent variables, using readily available structural disease data, overcoming the challenge of simultaneously observing the states of all factors.

[0131] The variable-weight dynamic risk assessment engine described in this invention, such as... Figure 7 As shown, this is the core of the system's evaluation, which includes a variable weight calculation unit and a comprehensive risk assessment unit.

[0132] The variable weight calculation unit is used to dynamically generate adaptive weights for evaluation indicators based on the current scenario and service stage. It establishes an evaluation hierarchy from the top-level tunnel structure safety risk index to the bottom-level various monitoring characteristic indicators. The criterion layer includes five dimensions: geological conditions, environmental load, structural status, maintenance history, and traffic load, which are consistent with the dimensions of the coupling matrix.

[0133] Subjective initial weights ( The determination of the criteria layer and each indicator layer elements is performed pairwise using a standard fuzzy hierarchical analysis method. A fuzzy complementary judgment matrix is ​​constructed by comparing the elements of the criterion layer and each indicator layer pairwise using a scale of 0.1 to 0.9 based on a pre-set judgment matrix based on general engineering experience. This matrix is ​​then subjected to row-wise summation and mathematical transformation to obtain a fuzzy consistent matrix that meets the consistency requirements. Finally, the subjective weight vector of each element is directly calculated using analytical formulas. It should be noted that subjective weighting Characterization is based on the long-term experience of experts and is updated annually or according to the tunnel's scheduled maintenance cycle.

[0134] Objective weight ( The calculation of entropy value employs an improved entropy weighting method that incorporates data confidence constraints. First, standardized sensing data from the most recent evaluation period is selected, and extreme value normalization is performed on multiple sample values ​​for each underlying index j. To overcome the impact of anomalous jumps in single sensor data on the entropy value, a data confidence factor is introduced. . The condition is met only if the following conditions are simultaneously satisfied: (a) the acquisition quality of the data point is normal; (b) the instantaneous rate of change of the current sampled value compared to the previous time series value does not exceed the physical limit threshold of the indicator; and (c) the current value falls within the historical normal fluctuation range of the sensor. The historical normal fluctuation range is defined as: the median ± 3 times the median absolute deviation calculated based on valid data from the sensor over the past 30 days (excluding marked abnormal events). For the system cold start phase, if there is insufficient historical data, only conditions (a) and (b) are temporarily used, and a more stringent rate of change threshold is set. If condition (c) cannot be evaluated due to lack of data, it is assumed to be satisfied by default. The number of samples accounts for the total number of samples. If the proportion is below the threshold of 0.8, the data quality for the current period is considered poor, and the update learning rate of the objective weights is adjusted. The dynamic reduction is specifically as follows: The system simultaneously records the "data consistency anomaly" event. Conversely, it uses high-confidence sample points to calculate the information entropy. and coefficient of difference And update the objective weights. Simultaneously reset Initialize to 1.0. The larger the difference coefficient, the greater its objective weight. The higher the objective weight, the more it is calculated dynamically based on the fresh data acquired in each assessment cycle. The assessment cycle can be flexibly set according to the tunnel monitoring needs. During normal operation, it is usually set to 1 to 4 hours to balance the timeliness of risk assessment with the actual sampling frequency of sensor data. In emergency monitoring or high-risk early warning states, it can be shortened to 15 to 30 minutes to achieve rapid response.

[0135] In obtaining (Subjective weight) and After (objective weights), the variable weight calculation unit adopts a game theory-based combinatorial weighting method, treating subjective weights and objective weights as two parties in a game, and combining the weights. Represented as a linear combination of the two: In the formula: This is the combined weight vector; This is the subjective weight vector; For objective weight vectors; and These are the linear combination coefficients corresponding to the subjective weights and the objective weights, respectively, and satisfy the following conditions: ;in, and The optimal value is determined using game theory, that is, by treating subjective weights and objective weights as the two sides of a game, and solving for the combined weight vector. and and When the deviation reaches the equilibrium coefficient, the resulting combined weight vector is the optimal basic combined weight.

[0136] The above game equilibrium is equivalent to solving the following system of linear equations:

[0137] ;

[0138] The analytical solution is obtained by directly solving this system of two linear equations in two variables, and the solution vector is... Normalization is performed so that .make The final basic combination weight is .

[0139] This method does not introduce any subjective preference parameters; its weight in the game is determined entirely by the inherent magnitude and inner product structure of the weight vector itself, truly achieving data-driven adaptive optimal combination weighting. The innovation of this invention lies not in the combination weighting algorithm itself, but in combining it with the aforementioned objective weights of confidence constraints, as well as the scenario adjustment coefficients and service stage parameters described later, to form a full-chain adaptive weighting process.

[0140] Subsequently, a scene type adjustment vector was introduced. and service phase degradation coefficient vector .in, For example, in collapsible loess scenarios, the adjustment coefficients for the environmental load dimension and the structural state dimension can be appropriately increased (typically ranging from 1.3 to 1.8, with specific values ​​determined based on the actual geological conditions of the tunnel and expert experience), while the adjustment coefficients for other dimensions remain at 1. The settings switch in real time based on the scene recognition results. (Elements in Γ) The calculation formula is Where t represents the tunnel's service life. This is the cumulative equivalent traffic load factor (initially 1). The design baseline period is 100 years. This formula introduces the influence of cumulative traffic load into a simplified logarithmic model; when the cumulative load is large, the decay accelerates. It should be noted that this formula is only one implementation method; in practical applications, it can be configured as a piecewise linear or other functional form depending on the specific tunnel's material and load characteristics. This coefficient mainly affects dimensions related to structural resistance (such as structural state) and is updated annually.

[0141] Furthermore, the variable weight calculation unit obtains the basic combined weights. Scene type adjustment vector and service phase degradation coefficient vector Then, the final adaptive weights are synthesized using the following formula:

[0142] ;

[0143] In the formula, The basic weighting incorporates subjective weights that are updated annually or on a regular inspection cycle. With objective weights that are dynamically updated according to the evaluation cycle The update frequency is of a mixed type; The scene type adjustment vector is updated in real time as the typical scene of the current tunnel changes, and the update frequency is event-driven. The service phase degradation coefficient vector is updated annually based on the tunnel's service life, with an update frequency exhibiting a long-term gradual variation. Therefore, by synthesizing these three parameters with different update frequencies in real time, the final adaptive weights can respond to changes in short-term data statistical characteristics, adapt to switching between typical scenarios, and track the impact of long-term gradual degradation of the tunnel structure. It should be noted that the above element-wise product adjustment method is an engineering simplification approximation, implicitly assuming that the influence of scenario and degradation on each dimension of the weights is independent. This approximation is engineeringly reasonable when the adjustment coefficient deviates slightly from 1.

[0144] The comprehensive risk assessment unit takes a standardized perception dataset as input. First, it performs factor risk aggregation: based on a pre-defined indicator-factor mapping matrix. The risk value of a single index is obtained by substituting each monitored physical quantity at the bottom layer into the membership function. The initial risk values ​​are weighted and aggregated into five disaster factor dimensions. ,in These correspond to geological conditions, environmental loads, structural status, maintenance history, and traffic loads, respectively.

[0145] For example, in a simplified embodiment, The calibration method is as follows: For each physical quantity measured by the sensor, the dimension of the most directly affecting disaster-causing factor is determined according to the principles of structural mechanics. The system developers then pre-set weights based on the "Highway Tunnel Design Code" and engineering experience, ensuring that the sum of the total risk allocation weights for each physical quantity is 1. In this embodiment, These are fixed values ​​and do not change with the scene. Specifically: the risk value of the strain sensor is assigned to the structural state dimension with a weight of 1.0; the risk value of the displacement sensor is assigned to the structural state dimension with a weight of 0.7 and to the geological condition dimension with a weight of 0.3; the risk value of the crack sensor is assigned to the structural state dimension with a weight of 0.6 and to the maintenance history dimension with a weight of 0.4; and the risk value of the pressure sensor is assigned to the environmental load dimension with a weight of 0.5 and to the structural state dimension with a weight of 0.5.

[0146] Subsequently, a chain correction is performed using the factor coupling effect matrix C of the current scenario. The correction formula is:

[0147] ;

[0148] In the formula, Let k be the initial risk value for dimension k. Let j be the coupling coefficient of factor j to factor k (which resides in the matrix C corresponding to the current scene, and...) ), This represents the risk value after coupling correction, with an upper limit of 1.0. This formula indicates that the final risk of each dimension is a linear superposition of its own state and the additional quantities affected by coupling with other dimensions, with a limited amplitude.

[0149] It should be noted that this linear superposition model is an engineering simplification approximation of complex nonlinear coupling effects. Its rationality is based on: (a) the coupling coefficient in practical engineering is usually much less than 1 (ensuring this through normalization, with typical values ​​between 0.1 and 0.3), therefore the additional terms are relatively small, and the linear approximation is effective within a local range; (b) the system continuously updates the C matrix through an online EM algorithm, enabling the coupling coefficient to reflect the implicit nonlinear trends in the actual monitoring data. That is, when the coupling effect strengthens, the corresponding c value will adaptively increase, thus macroscopically approximating the true coupling relationship. For special scenarios requiring high-precision nonlinear modeling, those skilled in the art can employ more complex coupling functions based on this invention, but this does not affect the core concept of this invention.

[0150] After the corrections are completed, the comprehensive risk assessment unit uses the final adaptive weights. The corrected dimensional risk values ​​are weighted and summed, i.e.

[0151] ;

[0152] The final comprehensive structural safety risk index is obtained. This is mapped to a range of 0 to 100. The risk levels are categorized as follows: Low risk Medium risk. High risk, This is extremely high risk.

[0153] The end-to-end closed-loop control function of this invention is implemented through the event-action rule engine built into the variable-weight dynamic risk assessment engine. This rule engine is deeply integrated with the unique components of this invention: its conditions can reference the invention's unique five-dimensional factor risk score (e.g., "environmental load dimension risk score > 0.8") and four-level spatiotemporal grid coordinates (e.g., "section S5~S8"); its actions can be precisely addressed to any sensing terminal through the four-level spatial reference system. Users can configure specific control rules through human-computer interaction and early warning terminals, for example: "WHEN the [comprehensive risk level] of section S5~S8 jumps to [high risk] AND [environmental load] dimension risk score > 0.8 THEN switch the sampling frequency of all [pressure sensors] and [displacement sensors] in this section from [1 time / hour] to [1 time / 10 minutes]". When the risk assessment result meets the "event" condition in the rule, the engine automatically generates a structured control instruction containing the target device ID and new sampling parameters. The instruction is sent to the data aggregation and spatiotemporal registration module via the system's internal message channel. This module acts as a unified instruction gateway, determining the target device's registered section number S and circumferential sector. The system performs addressing and forwards the commands to the unified communication module of the corresponding sensing terminal. After parsing the commands, the terminal communication module hands them over to the edge intelligent data processing unit to perform sampling mode switching and return an acknowledgment message, completing the closed loop. This mechanism enables the system to automatically encrypt the monitoring of high-risk areas based on the risk situation, achieving deep integration between the general rule engine and the dedicated tunnel structure model.

[0154] The human-computer interaction and early warning terminal described in this invention is used to receive assessment results and perform multi-dimensional display and graded early warning. This terminal constructs a tunnel digital twin visualization platform based on BIM and GIS, overlaying a real-time risk field onto a 3D model. It supports viewing the risk index distribution along the line using heat maps, viewing multi-directional sensor time history curves by clicking on cross-sections, and displaying the evolution of temperature and seepage pressure fields through spatiotemporal cube slicing. The early warning function is divided into three levels: yellow, orange, and red, corresponding to APP notifications, telephone SMS, and linked to on-site audible and visual alarms and traffic signal systems, respectively. All early warning information includes suggested handling measures. After feedback from on-site personnel, a closed-loop record is generated, realizing full-process management of early warning, dispatch, handling, and review.

[0155] Based on the above system, this invention also provides an integrated intelligent risk assessment method for multi-source sensor arrays in typical tunnel scenarios. First, a multi-source sensor array integrated intelligent sensing terminal synchronously collects various monitoring data such as strain, displacement, cracks, temperature, humidity, and seepage pressure. After edge data cleaning, anomaly detection, and feature extraction, structured sensing data carrying timestamps and spatial location codes is uploaded. Second, a data aggregation and spatiotemporal registration module receives the data and uniformly maps it to a five-dimensional spatiotemporal grid composed of time t, logical cross-section S, dynamic physical mileage d, circumferential sector φ, and radial depth layer r. After time interpolation and spatial gridding processing, a standardized sensing dataset with spatiotemporal consistency is obtained. Subsequently, a scene identification and factor coupling effect module identifies the current typical scenario based on the tunnel's basic information and the standardized sensing data, and retrieves the factor coupling effect matrix corresponding to that scenario, driven by expert knowledge and online learning using an EM algorithm based on partially observable data. Next, the variable-weight dynamic risk assessment engine, based on the current scenario and service stage, dynamically generates adaptive weights through a combination of standard fuzzy hierarchical analysis, improved entropy weighting with confidence constraints, and game-theoretic Nash equilibrium-based weighting. It first aggregates the underlying indicator risks into five factor dimensions, then corrects them using a coupling matrix before weighted aggregation, deriving the structural safety risk index and risk level. When the risk level reaches a threshold, the system automatically increases the sampling frequency of sensors in high-risk areas according to preset "event-action" rules, achieving a closed loop from perception to assessment to perception optimization. Finally, the human-computer interaction and early warning terminal provide visual display and tiered early warnings.

[0156] In summary, through the above-described systems and methods, this invention provides a full-link, evolvable intelligent management and control solution for tunnel structure safety, encompassing hardware integration, spatiotemporal registration, scene identification, dynamic evaluation, and closed-loop control.

[0157] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings. Obviously, the described embodiments are only preferred embodiments of the present invention, and not all embodiments, nor are they intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications or equivalent variations based on the disclosed technical content. However, any simple modifications, equivalent variations, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the technical solution of the present invention shall still fall within the protection scope of the technical solution of the present invention.

[0158] Example 1

[0159] This example uses a highway tunnel section (from chainage K35+200 to K36+000) located in a high-altitude, cold region as an example. The average annual temperature in this region is -2℃, the extreme minimum temperature is -35℃, the number of freeze-thaw cycles is about 120 per year, the maximum burial depth of the tunnel is 150m, and the surrounding rock is mainly of Class III and IV.

[0160] System Deployment: Within this 800m section, a total of 16 integrated intelligent sensing terminals with multi-source sensor arrays are deployed, spaced approximately 50m apart along the tunnel axis. The terminals are numbered NT01 to NT16, corresponding to logical section numbers S01 to S16. Each terminal features a modular, sealed housing structure (IP68 protection rating) and connects to the system via a unified power supply module (supporting dual redundant power supply of AC 220V and PoE, with a built-in lithium iron phosphate backup battery) and a unified communication module (supporting RS485 / Ethernet / LoRa / 4G multi-mode communication, using the MQTT IoT protocol). Each terminal's internal sensor integration module can be equipped with the following sensors via modular slots: one vibrating wire strain gauge each on the lining surface and deep layer (32 in total); one laser rangefinder each on the arch crown and both sides of the arch waist (48 in total); vibrating wire crack gauges installed across key construction joints and existing cracks (24 in total); one digital temperature and humidity probe each on the lining surface and shallow layer (5cm depth) (32 in total); and one vibrating wire piezometer each in the drainage blind ditch behind the lining and under the invert arch (32 in total). Additionally, one traffic load statistics device is installed at the tunnel entrance and intermediate location in each section, using a geomagnetic vehicle detector and a dynamic weighing system to calculate the daily traffic volume of the section. All terminals in the system are calibrated with a unified clock synchronization module before leaving the factory and contain a rigorously calibrated OCXO-2522 temperature-controlled crystal oscillator (short-term stability better than 1×10⁻⁶). -12 With an aging rate better than ±0.5ppb / day, it achieves clock synchronization through BeiDou / GPS dual-mode time synchronization when satellite signals are available, and through a multi-terminal mutual calibration mechanism when there are no satellite signals or communication interruptions.

[0161] Based on this, the data aggregation and spatiotemporal registration module constructs a four-level hierarchical spatial reference system for the entire tunnel: longitudinally, logical section numbers S01 to S16 are assigned to 16 monitoring sections (unique and unchanging throughout the entire life cycle), and each is associated with its design mileage as the initial dynamic physical mileage d0 (dynamically updated with structural deformation); circumferentially, each section is divided into 8 45° sectors and assigned codes φ0 (arch crown) to φ7 (right arch shoulder), with each sensor mapped to the nearest sector according to its actual installation angle; radially, three depth layers are divided according to the sensor burial depth: r0—surface layer (0-2cm, mainly controlling the risk of spalling and falling off), r1—shallow layer (2-10cm, mainly controlling the risk of steel reinforcement rust expansion and cracking), and r2—deep lining layer (>10cm, mainly controlling the risk of structural crushing). Thus, each sensor is assigned a unique quadruple spatial coordinate (S, d, φ, r), which, combined with the time dimension, constitutes a five-dimensional spatiotemporal grid system. The edge intelligent data processing unit performs improved sliding window mid-value filtering and adaptive outlier removal, anomaly detection and feature extraction on the raw data collected by each sensor. It then generates structured sensing data carrying device ID, timestamp, spatial location code, physical quantity value and data quality identifier. After time interpolation (step size 1h) and spatial gridding processing by the spatiotemporal dual-dimensional calibration unit, a standardized sensing dataset covering strain field, displacement field, crack aperture field, temperature field, humidity field and seepage pressure field is formed.

[0162] The risk assessment process of the present invention in this scenario is described in detail below based on the above system.

[0163] (1) Scene recognition

[0164] The system reads the geological survey report and design documents for this section, obtaining key parameters: an average annual freeze-thaw cycle count of 120, a design freezing depth of 3.2m, and a cover layer thickness of 3.2m at the tunnel entrance. These parameters indicate that the freezing depth has reached the same level as the cover layer thickness, and the surrounding rock outside the lining is within the freezing influence range, indicating a significant freeze-thaw risk. Simultaneously, the system obtains real-time monitoring data (last 72 hours) from the standardized sensing dataset:

[0165] Cave temperature: Daytime high +5℃, nighttime low -12℃, daily temperature difference 17℃;

[0166] Internal temperature of the lining (radial depth r1 layer, burial depth 5cm): +1℃ during the day, -5℃ at night, daily temperature difference 6℃;

[0167] Humidity of lining surface and shallow layer: relative humidity 82%–95%;

[0168] The settlement rate of the arch is 0.25 mm / d.

[0169] Lining surface cracks: A longitudinal crack with a width of 0.28 mm at the arch crown has not expanded in the last 72 hours;

[0170] Piezometer reading: Pore water pressure behind the lining is 8 kPa, with no significant fluctuations;

[0171] Rule matching layer: The system pre-sets 8 main rules for freeze-thaw scenarios in high-altitude cold regions, including "annual freeze-thaw cycle count > 80", "daily temperature difference > 15℃", "daily temperature fluctuation inside the lining > 5℃", and "lining humidity > 80% and temperature below 0℃". Currently, 6 rules are met, and the rule confidence level is... .

[0172] Machine learning model layer: The XGBoost classification model takes features such as geological type of the section, quarterly average temperature and humidity, maximum displacement rate, 24-hour seepage pressure change rate, and daily temperature difference as input, and outputs the probability of freeze-thaw scenarios in high-altitude cold regions. .

[0173] Fusion determination: Take the fusion weight Then the overall confidence level .because The system determines that the current active scenario is "freezing and thawing in high-altitude cold regions".

[0174] (2) Standardized sensing dataset

[0175] The system retrieves standardized sensing data from a multi-source sensor array within the most recent evaluation period (1 hour) for this section, after spatiotemporal registration processing. This dataset is presented in five-dimensional spatiotemporal coordinates. The data is organized, with the time stamp 't' representing the timestamp of this data entry: 2026-01-15 06:00:00 (the exact time of the evaluation cycle, resampled and aligned). Regarding the dynamic physical mileage 'd', the latest measured mileage of logical section S12 in the current scenario is K35+430.05 (due to surrounding rock creep or uneven settlement, an axial displacement of 5cm has occurred compared to the initial design mileage K35+430.00. This displacement is within the system's preset normal fluctuation range; the system only records and updates the mileage without triggering additional alarms). Some key data are as follows:

[0176] Surface strain gauge: spatial position The measured value is microstrain −120 με, and the normalized risk value is 0.25.

[0177] Deep strain gauge: spatial location The measured value is microstrain −45 με, and the normalized risk value is 0.10.

[0178] Displacement gauge (crest settlement): Spatial position The measured value was a cumulative value of 12 mm, the rate was 0.25 mm / d, and the normalized risk value was 0.30.

[0179] Crack gauge: Spatial location The measured value is 0.28 mm in width, with a variation of 0.00 mm and a normalized risk value of 0.35.

[0180] Temperature sensor (inside the cave): Spatial location The measured value is the current temperature -3ºC, and the normalized risk value is 0.65.

[0181] Temperature sensor (shallow layer): spatial location The measured value is the current temperature -2ºC, and the normalized risk value is 0.55.

[0182] Humidity sensor: spatial location The measured value was 88% RH (relative humidity), and the normalized risk value was 0.50.

[0183] Piezometer: Spatial Position The measured value was 8 kPa, and the normalized risk value was 0.20.

[0184] Traffic load counter: Spatial location is section K35+200~K36+000 (section-level statistical sensor, its spatial location is identified according to the section range, not attached to a single cross-section grid node, its risk value is calculated based on the average contribution of each cross-section within the section), the measured value is 9800 vehicles per day (design value 12000 vehicles), and the normalized risk value is 0.25.

[0185] Except for the traffic load counter, all the aforementioned sensors have undergone time interpolation alignment from the original sampling time to the hour t, as well as alignment from the actual installation coordinates to the nearest quadruple mesh node. Spatial mapping. The normalized risk value uses a preset membership function (such as a triangular or trapezoidal distribution) to map the physical quantity to the interval [0, 1], and the larger the value, the higher the risk.

[0186] (3) Call the factor coupling effect matrix

[0187] After the scene recognition engine outputs "high-altitude cold region freeze-thaw," the comprehensive risk assessment unit automatically calls the preset five-dimensional disaster-causing factor causal coupling effect matrix C corresponding to that scene. This matrix has been iteratively optimized through online learning (EM algorithm), and its current parameters are as follows:

[0188] ;

[0189] The order of rows and columns is: geological conditions, environmental loads, structural condition, maintenance history, and traffic loads.

[0190] Example Explanation: (Deteriorating geological conditions will exacerbate the freeze-thaw environmental effects). ; (Freeze-thaw cycles have a significant impact on structural state). ; .

[0191] (4) Risk aggregation of underlying indicators

[0192] By using a preset membership function, each monitored physical quantity is transformed into a single-indicator risk value, and then based on the indicator-factor mapping matrix... The initial risk values ​​of the five hazard-causing factors are weighted and aggregated. .

[0193] Calculation of the contribution of each sensor to the five-dimensional risk (partial examples):

[0194] Strain sensor: Surface strain risk 0.25, deep strain risk 0.10, both only mapped to "structural state" (weight 1.0), contribution: 0.25 + 0.10 = 0.35;

[0195] Displacement sensor: Settlement risk 0.30, mapped by 0.7 (structure) / 0.3 (geology): structural contribution 0.21, geological contribution 0.09;

[0196] Crack sensor: Risk 0.35, mapped by 0.6 (structure) / 0.4 (maintenance): Structure contribution 0.21, maintenance contribution 0.14;

[0197] Temperature sensor: Temperature risk inside the cave is 0.65, and temperature risk in the shallow layer is 0.55. Both are only mapped to "Environmental Load" (weight 1.0). Contribution: 0.65 + 0.55 = 1.20 → Limitation is 1.0.

[0198] Humidity sensor: Risk 0.50, mapped to "Environmental Load", contribution: 0.50;

[0199] Pressure sensor: Risk 0.20, mapped at 0.5 (environment) / 0.5 (structure): Environmental contribution 0.10, structural contribution 0.10;

[0200] Traffic load: Risk 0.25, mapped to "Traffic load" (weight 1.0), contribution: 0.25;

[0201] Aggregation results:

[0202] Geological conditions: ;

[0203] Environmental loads: ;

[0204] Structural status: ;

[0205] Maintenance history: ;

[0206] Traffic load: .

[0207] Initial five-dimensional risk vector:

[0208] (5) Factor Coupling Correction

[0209] Chain correction is performed using the coupling matrix C of the current scene, as shown in the formula:

[0210] ;

[0211] Calculation process:

[0212] ;

[0213] ;

[0214] ;

[0215] ;

[0216] ;

[0217] Corrected risk vector:

[0218] ;

[0219] (6) Adaptive weight calculation

[0220] Subjective weights (determined by fuzzy analytic hierarchy process, updated annually):

[0221] ;

[0222] The vector elements correspond to five disaster-causing factor dimensions: geological conditions (0.20), environmental load (0.25), structural status (0.35), maintenance history (0.10), and traffic load (0.10).

[0223] Objective weights (calculated using the improved entropy weighting method, based on 95% of high-confidence data points within the current period):

[0224] ;

[0225] The vector elements correspond to the following in order: geological conditions (0.15), environmental load (0.35), structural state (0.40), maintenance history (0.05), and traffic load (0.05).

[0226] Game theory combinatorial weighting: Solving the game equilibrium equation to obtain the proportion of subjective weights. Objective weighting percentage .

[0227] Basic portfolio weights:

[0228] ;

[0229] The calculation yields:

[0230] ;

[0231] Scene Adjustment: High-Altitude Cold Region Freeze-Thaw Scene Adjustment Vector (The environmental load dimension and structural state dimension are significantly affected by freeze-thaw cycles and are therefore magnified.)

[0232] Service stage adjustment: The tunnel has been in service for 6 years (t=6), design reference period. Annual cumulative equivalent traffic load factor Calculate the service phase degradation coefficient vector. The decay coefficient of the structural state dimension is:

[0233] ;

[0234] ;

[0235] Final adaptive weights:

[0236] ;

[0237] ;

[0238] ;

[0239] The final weights are obtained after normalization:

[0240] ;

[0241] (7) Calculation and rating of comprehensive risk index

[0242] Comprehensive structural safety risk index:

[0243] ;

[0244] ;

[0245] Will Mapped to the range of 0 to 100, the final comprehensive structural safety risk index is 78.54.

[0246] Risk level determination (based on system preset thresholds: Low risk; Medium risk; High risk; (Extremely high risk), the current risk level of section S12 is extremely high risk.

[0247] (8) Closed-loop control triggering

[0248] The risk assessment engine detected an "extremely high risk" level, and the risk score in the "environmental load" dimension was also high. Risk scoring in the "structural status" dimension The built-in "event-action" rule engine automatically matches and triggers preset rules:

[0249] When the overall risk level of section [S12] is [extremely high risk] AND the risk score of the [environmental load] dimension is >0.8, the sampling frequency of all [temperature sensors], [humidity sensors] and [displacement sensors] in this section is switched from [1 time / hour] to [1 time / 10 minutes], and the continuous monitoring mode of the [crack gauge] is started (from timed acquisition to real-time triggering).

[0250] Subsequently, the engine generates a structured control command containing the target device ID and new sampling parameters. Through the data aggregation and spatiotemporal registration module, it accurately addresses the integrated intelligent sensing terminal with multi-source sensor array corresponding to section S12, adjusting its sampling strategy. Simultaneously, a red alert is issued to the monitoring center through human-machine interaction and the early warning terminal, triggering on-site audible and visual alarms. A dispatch order with suggested handling measures (such as strengthening insulation measures, checking the drainage system, and increasing the monitoring frequency) is generated, completing the closed-loop management of the entire chain from perception to assessment to optimized perception and on-site handling.

[0251] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered exemplary and not restrictive in all respects. Furthermore, it should be understood that although this specification describes embodiments, it does not encompass only one technical solution. This descriptive method is merely for clarity, and those skilled in the art should consider the specification as a whole. The technical solutions in the embodiments can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios, wherein the typical scenarios include four types: collapsible loess strata, high-altitude freeze-thaw zones, landslide and debris flow affected areas, and high-intensity earthquake zones; characterized in that: The system includes: At least one integrated intelligent sensing terminal with a multi-source sensor array is used to simultaneously collect various monitoring data and generate structured sensing data with timestamps and preliminary location identifiers after edge preprocessing. The data aggregation and spatiotemporal registration module is communicatively connected to the integrated intelligent sensing terminal of the multi-source sensor array. It is used to construct a four-level hierarchical spatial reference system and map the structured sensing data to a five-dimensional spatiotemporal grid composed of time and the four-level hierarchical spatial reference system, thereby generating a spatiotemporally consistent standardized sensing dataset. The scene identification and factor coupling effect module is communicatively connected to the data aggregation and spatiotemporal registration module. It is used to identify the typical scene where the tunnel is currently located based on the standardized perception dataset and call the causal coupling effect matrix of five types of disaster-causing factors corresponding to the scene. The variable-weight dynamic risk assessment engine is communicatively connected to the data aggregation and spatiotemporal registration module and the scene identification and factor coupling effect module. It includes a variable-weight calculation unit and a comprehensive risk assessment unit. The variable-weight calculation unit is used to dynamically generate adaptive weights for evaluation indicators based on the typical scenario and service stage of the tunnel. The comprehensive risk assessment unit is used to calculate the tunnel structure safety risk index and classify the safety risk level based on the standardized perception dataset, the causal coupling effect matrix of five types of disaster-causing factors, and the adaptive weights. The human-computer interaction and early warning terminal is connected to the variable weight dynamic risk assessment engine to visualize the security risk index and security risk level, and to issue a graded early warning when the security risk level exceeds the threshold.

2. The spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios as described in claim 1, characterized in that: The integrated intelligent sensing terminal with multi-source sensor array includes a unified clock synchronization module. When there is a satellite signal, the module uses BeiDou / GPS dual-mode time synchronization. When there is no satellite signal or communication is interrupted, clock synchronization is achieved through a multi-terminal mutual calibration mechanism based on the cross-correlation of vibration signals.

3. The spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios according to claim 2, characterized in that: The multi-terminal mutual calibration mechanism is specifically as follows: For vibration signal sequences of similar dynamic sensors in integrated intelligent sensing terminals with different multi-source sensor arrays, calculate the normalized cross-correlation coefficient between any two terminal signals; when the peak value of the cross-correlation coefficient is greater than a preset threshold, determine that the two terminals have experienced the same sudden excitation event, and use the time delay position corresponding to the peak value of the cross-correlation coefficient to extract the relative time deviation between the two terminals; traverse all terminals involved in the event to obtain multiple redundant relative time deviations between each pair of terminals in the terminal group. The terminal with the lowest crystal oscillator aging rate in the terminal group and the first to restore the communication link with the data aggregation and spatiotemporal registration module is used as a temporary benchmark. Based on the measured time deviation and steps of each terminal relative to this temporary benchmark... The redundant relative time deviation between each terminal pair, the crystal oscillator aging rate of each terminal, and the communication interruption duration are extracted to establish an overdetermined system of equations with the global optimal compensation value of each terminal as the unknown. In addition to the direct observation equations from the temporary reference to each terminal, the overdetermined system of equations also includes constraint equations composed of the redundant relative time deviations, so that the total number of equations is greater than the number of compensation values ​​to be determined. The overdetermined equations are solved using the least squares method to obtain the global optimal compensation value for each terminal. This is used to uniformly compensate the timestamps of data collected by each terminal during the interruption period, thereby achieving synchronization of the clocks of all terminals in the group.

4. The spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios according to claim 1, characterized in that: The four-level hierarchical spatial reference system includes logical section number S, dynamic physical mileage d, circumferential sector code φ, and radial depth layer r; Among them, the logical section number S is a unique and unchanging identifier throughout the entire life cycle, and the dynamic physical mileage d is dynamically updated with structural deformation. The two constitute a longitudinal dual-identifier decoupling mechanism, which is used to track the axial displacement field when the tunnel undergoes longitudinal displacement in real time; the circumferential sector code φ is used for circumferential positioning; and the radial depth layer r is used for radial positioning.

5. The spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios according to claim 4, characterized in that: The construction process of the four-level hierarchical spatial reference system is as follows: Step 1: Assign a unique and unchanging logical section number S to each monitoring section based on the tunnel design parameters and sensor deployment information; Step 2: The associated design mileage is used as the initial dynamic physical mileage d0, which is dynamically updated as the tunnel structure deforms. The mileage of the most recent accurate measurement of the cross-section is used as the station number and recorded as the dynamic physical mileage d. Step 3: Divide the discrete sectors along the circumferential direction of the tunnel cross section and assign codes, denoted as circumferential sector codes φ; Step 4: Divide the tunnel lining thickness along the inner surface outwards, according to the sensor burial depth, into layers and assign codes, denoted as radial depth layer r.

6. The spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios according to claim 1, characterized in that: The five-category disaster-causing factor causal coupling effect matrix is ​​constructed based on a Bayesian network according to five categories of disaster-causing factors, including geological conditions, environmental load, structural state, maintenance history, and traffic load. During system operation, the conditional probability parameters of the matrix are iteratively updated based on the observed tunnel structural state and environmental load data through the expectation-maximization algorithm, which is used to correct the superimposed influence between the risk values ​​of each disaster-causing factor dimension. Its construction and update process is as follows: a) Based on the well-known causal relationships in the field of tunnel engineering, a qualitative analysis is conducted on the causal relationships among five disaster-causing factors, including geological conditions, environmental loads, structural conditions, maintenance history, and traffic loads. b) Using the five types of disaster-causing factors as nodes, each node is discretized into three states: good, deteriorated, and dangerous. The causal edges between each node are defined as: geological conditions → environmental load, geological conditions → structural state, environmental load → structural state, maintenance history → structural state, and traffic load → structural state. A Bayesian network is drawn accordingly. c) Define the factor coupling effect matrix as a 5×5 square matrix, whose elements (i, j=1…5) represents the coupling influence strength of factor j on factor i among the five types of disaster-causing factors; where the diagonal elements are 0, it indicates that the factor has no coupling with itself; the off-diagonal elements are... ( The value range of )) is [0, 1]. The larger the value, the stronger the driving effect of the deterioration of the state of factor j on the deterioration of the state of factor i. d) Based on historical disease cases and known engineering experience, a prior condition probability table is assigned to each node of the Bayesian network; according to the prior condition probability table, the probability increment of the related nodes in the dangerous state when each node changes from a good state to a dangerous state is extracted, and the weighted average is calculated according to the frequency of each state in historical cases to obtain the influence intensity; all influence intensities are normalized, and the resulting value is the initial coupling coefficient of the matrix. e) During system operation, the parameters of the conditional probability table of the Bayesian network are iteratively optimized using the expected maximization algorithm based on the observed structural state and environmental load data. In the iterative process, a historical data weight exponential decay mechanism is introduced to assign weights λ=exp(-Δt / τ) to samples in different time windows. Where: Δt is the time difference between the sample and the current time, and τ is a preset time constant. f) After iterating alternately between the expectation step and the maximization step until convergence, the coupling coefficients are re-extracted based on the updated conditional probability table, and the causal coupling effect matrix of the five types of disaster-causing factors is updated.

7. The spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios according to claim 1, characterized in that: The process by which the variable weight calculation unit dynamically generates adaptive weights for evaluation indicators based on the current scenario and service stage is as follows: the standard fuzzy hierarchical analysis method is used to determine the subjective initial weights. An improved entropy weight method is used to calculate objective weights; a game theory-based combined weighting method is used to treat subjective weights and objective weights as two sides of a game, establish a combined weight formula and a set of game equilibrium equations, solve the set of equations, and normalize the solution vectors to obtain the final basic combined weights. Then, the scenario type adjustment vector and the service stage decay coefficient vector are introduced. The basic combined weights are multiplied element by element with the scenario type adjustment vector and the service stage decay coefficient vector and then normalized to generate the final adaptive weights. The improved entropy weight method introduces a data confidence factor. Select high-confidence sample points, only using The information entropy and difference coefficient of the samples are calculated, and the learning rate is dynamically adjusted according to the proportion of high-confidence samples to suppress the magnitude of weight updates when the data quality is poor and obtain robust objective weights. The formula for calculating the combined weights is: ； In the formula: For combined weights; Subjective weighting, For objective weighting; The coefficients are linear combination coefficients of subjective weights. Let be the linear combination coefficients of the objective weights, and satisfy . ; The adaptive weights are: ； In the formula: For adaptive weights, Adjust the vector for scene type. This is the service phase degradation coefficient vector.

8. The spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios according to claim 7, characterized in that: The process by which the comprehensive risk assessment unit calculates the tunnel structure safety risk index based on the standardized perception dataset, the causal coupling effect matrix of five types of disaster-causing factors, and adaptive weights is as follows: taking the standardized perception dataset as input, the risk of the five types of disaster-causing factors is first aggregated to obtain five-dimensional risk values. Then, the causal coupling effect matrix of the five types of disaster-causing factors corresponding to the typical scenario in which the tunnel is located is used to perform a chain correction on the five-dimensional risk values. Finally, the adaptive weights are used to perform a weighted summation of the corrected five-dimensional risk values ​​to obtain the final tunnel structure safety risk index. in: The five categories of disaster-causing factors are aggregated as follows: based on a preset indicator-factor mapping matrix. The risk value of a single index is obtained by substituting each monitored physical quantity at the bottom layer into the membership function. The initial risk values ​​of the five hazard-causing factors are weighted and aggregated. ,in These correspond to geological conditions, environmental loads, structural status, maintenance history, and traffic loads, respectively. The chain correction formula is: ； In the formula: Let k be the initial risk value for dimension k. Let be the coupling coefficient of factor j with respect to factor k. This is the risk value after coupling correction, with an upper limit of 1.0; The formula for calculating the tunnel structure safety risk index is as follows: ； In the formula: The tunnel structure safety risk index. For the adaptive weights of dimension k; The risk level classification is based on the calculated tunnel structure safety risk index. Mapped to the range of 0 to 100; Low risk Medium risk. High risk, This is extremely high risk.

9. The spatiotemporally self-consistent and multi-factor adaptive risk assessment system for typical tunnel structure scenarios according to claim 1, characterized in that: The variable weight dynamic risk assessment engine also has a built-in event-action rule engine, which is used to automatically generate closed-loop control instructions based on the risk assessment results. The instructions are precisely addressed by the data aggregation and spatiotemporal registration module according to the logical section number S and circumferential sector number φ of the target device, and forwarded to the corresponding multi-source sensor array integrated intelligent sensing terminal to adjust the sampling strategy of the terminal, forming a full-link closed loop from assessment to sensing optimization.

10. A spatiotemporally self-consistent and multi-factor adaptive risk assessment method for typical tunnel structure scenarios, characterized in that: The evaluation is performed using the system described in any one of claims 1-9, comprising the following steps: S1: Multiple monitoring data are collected simultaneously through an integrated intelligent sensing terminal with a multi-source sensor array, and after edge preprocessing, structured sensing data with timestamps and preliminary location identifiers are generated; S2: A unified four-level hierarchical spatial reference system for the entire tunnel is constructed by the data aggregation and spatiotemporal registration module, and the structured sensing data is mapped to a five-dimensional spatiotemporal grid composed of time and the four-level hierarchical spatial reference system to obtain a standardized sensing dataset. S3: The scene identification and factor coupling effect module identifies the current typical scene of the tunnel based on the standardized perception dataset, and calls the five types of disaster-causing factor causal coupling effect matrix corresponding to the scene; S4: The variable-weight dynamic risk assessment engine adaptively adjusts the weights of the evaluation indicators based on the current typical scenarios and the tunnel service stage, and uses the causal coupling effect matrix and the standardized perception dataset to calculate the tunnel structure safety risk index and classify the risk level. S5: Visualized display and tiered early warning provided through human-computer interaction and early warning terminals; S6: When the risk level exceeds the preset threshold, the event-action rule engine built into the variable weight dynamic risk assessment engine automatically generates control instructions. Through the data aggregation and spatiotemporal registration module, the instructions are accurately addressed according to the logical section number S and the circumferential sector number φ of the target device, and forwarded to the corresponding multi-source sensor array integrated intelligent sensing terminal to adjust its sampling strategy.

Images