A Service Status Prediction System and Method for Tunnel Structures Based on State-Space Model

Abstract

This application provides a system and method for predicting the service status of tunnel structures based on a state-space model, relating to the field of structural health monitoring and intelligent early warning technology during tunnel operation. The method includes: an observation data preprocessing module for generating observation vectors; a state variable construction module for constructing implicit state vectors encompassing five states: deformation, stress, cracking, leakage, and environmental disturbance; a state transition modeling module for constructing a discrete-time state transition model; an observation mapping module for constructing an observation mapping model; a time-series prediction module for generating future observation vectors; an anomaly identification module for calculating a comprehensive anomaly score and anomaly type; and a service status assessment module for determining the service status of the tunnel structure. This application achieves dynamic prediction of structural service status and early anomaly identification, improving the accuracy of tunnel operation status prediction and operational decision-making capabilities.

Description

Technical Field

[0001] This application relates to the field of tunnel operation period structural health monitoring and intelligent early warning technology, and in particular to a tunnel structure service status prediction system and method based on state space model. Background Technology

[0002] With the continuous expansion of tunnel engineering projects in my country, including railways, highways, urban rail transit, and hydraulic tunnels, a large number of tunnels have entered the long-term operation phase. Under the influence of factors such as vehicle loads, surrounding rock pressure, groundwater changes, temperature and humidity cycles, and aging of lining materials, tunnel structures are prone to problems such as convergence deformation, crown settlement, abnormal lining stress, crack propagation, and leakage development. Because the service status of tunnels is characterized by long-term evolution, concealed development, and multi-factor coupling, continuous analysis of multi-source monitoring data is necessary in operation and management to promptly grasp the trends of structural changes and provide a basis for safety early warning, maintenance decisions, and life assessment.

[0003] In related technologies, tunnel operation monitoring data is typically collected using convergence meters, settlement monitoring points, stress gauges, crack gauges, leakage monitoring devices, temperature and humidity sensors, and groundwater pressure sensors. This data is then combined with manual inspections, threshold alarms, statistical analysis, empirical regression models, or general machine learning models to assess the structural condition and analyze trends. These methods can reflect changes in local observation indicators to a certain extent and assist operators in identifying risks exceeding limits.

[0004] However, existing technologies mostly remain at the level of observable data, making it difficult to describe the true service condition of tunnel structures that cannot be directly observed. Monitoring data is also susceptible to noise, missing measurements, equipment drift, and occasional disturbances, and fixed threshold judgments are prone to false alarms or missed alarms. At the same time, traditional prediction methods cannot uniformly express the dynamic relationship between structural state transitions, external loads, hydrological environment, and maintenance and treatment. Anomaly identification often lags behind the state deterioration process, making it difficult to achieve early trend warnings.

[0005] To address the aforementioned issues, a system for predicting the service status of tunnel structures based on a state-space model is designed. Summary of the Invention

[0006] The purpose of this invention is to provide a tunnel structure service status prediction system and method based on a state-space model to address the problems of difficulty in modeling the long-term evolution of tunnel structure status during operation, high noise in monitoring data, delayed identification of abnormal states, insufficient accuracy in service trend prediction, and lack of time-series basis for maintenance decisions.

[0007] The above-mentioned objective of this application is achieved through the following technical solution: The observation data preprocessing module is used to collect and preprocess the observation data of the monitoring section to obtain the observation vector. The state variable construction module is used to construct implicit state vectors based on observation vectors, including five types of states: deformation state, stress state, crack damage state, leakage state, and environmental disturbance state. The state transition modeling module is used to construct discrete-time state transition models; The observation mapping module is used to build observation mapping models. The time series prediction module is used to generate future observation vectors based on implicit state vectors, discrete-time state transition models, and observation mapping models. An anomaly identification module is used to obtain a comprehensive anomaly score and anomaly type based on future observation vectors and actual observation vectors; The service status assessment module is used to determine the service status of the tunnel structure based on the implicit state vector, future observation vector, comprehensive anomaly score, and anomaly type.

[0008] Optionally, the observation data includes: convergence deformation, crown settlement, lining stress, crack width, leakage, temperature and humidity, and groundwater pressure; The state variable construction module uses a single monitoring section or a group of adjacent monitoring sections as the basic calculation unit to convert the observation vector into an implicit state vector:

[0009] in, Represents the deformation state components. Indicates the force state components, Indicates the component representing the damage state. Indicates the leakage status component. Represents the environmental disturbance state component; T represents transpose; Hidden state vector The state components include: deformation state components, stress state components, crack damage state components, leakage state components, and environmental disturbance state components.

[0010] Optionally, the deformation state components consist of standardized convergent deformation, lining stress, crown settlement and its rate of change; The crack damage state component consists of crack width, crack growth rate, and continuous growth duration. The leakage status component consists of leakage amount, temperature and humidity, and groundwater pressure. The weighting coefficients of each state component in the implicit state vector are determined by historical stable period observation data, engineering preset limits, and tunnel operation data of the same type, and are updated according to a set cycle. Among them, the engineering preset limits refer to the allowable or warning values ​​of indicators such as convergence deformation, crown settlement, lining stress, crack width, and leakage, determined according to design documents, current specifications, operation management requirements, or project alarm rules; tunnel operation data of the same type refers to the monitoring, inspection, and maintenance data of existing tunnels that are similar to the current tunnel in terms of structural form, surrounding rock conditions, operating load, monitoring indicators, or types of defects.

[0011] Optionally, the state transition modeling module is used to establish the following discrete-time state transition model:

[0012] in: This represents the hidden state vector at the current moment. Let be the hidden state vector of the previous time step. To standardize the time step, This is the state transition matrix, used to characterize the continuity of each state component and the influence relationships between them. For external driving vectors, For external driving input matrix, For maintenance and treatment vectors, In order to handle the impact matrix, For state disturbance terms; The external driving vectors include: vehicle or train load intensity, groundwater pressure change, temperature change, humidity change, and surrounding construction disturbance intensity. The state transition matrix is ​​updated using a sliding window method, and the length of the sliding window is determined based on a uniform sampling period. The main diagonal elements of the state transition matrix are subject to stability constraints, with the values ​​of the main diagonal elements corresponding to the deformation state, cracking state, and leakage state limited to between 0.90 and 1.05. The maintenance and treatment vectors are coded separately for drainage, grouting, crack sealing, lining reinforcement and speed-limited operation. The treatment impact matrix records the treatment start time, duration of effect, and state fallback coefficient.

[0013] Optionally, the observation mapping module establishes the following observation mapping model:

[0014] in, This refers to the actual monitored observation vector. For the observation mapping matrix, The hidden state vector, For the observed noise term; The elements in the observation mapping matrix are jointly determined by the historical correlation coefficient between the observation index and the state components in the implicit state vector, the sensor reliability coefficient, and the engineering sensitivity coefficient. Among them, the historical correlation coefficient refers to the correlation parameter between the observation index and the state components calculated based on historical data; the engineering sensitivity coefficient refers to the response strength coefficient of the observation index to changes in service status, which is determined based on the location of the monitoring point, structural stress characteristics, type of defects, and engineering experience. The observation indicators include: convergence deformation, crown settlement, lining stress, crack width, leakage, temperature and humidity, and groundwater pressure. The sensor reliability coefficient is determined by the sensor online rate, missing rate, equipment status code, and historical noise variance. The observation mapping module is equipped with a missing measurement mask matrix. When a certain observation index is missing, the observation vector corresponding to the observation index and the data corresponding to the row of the observation mapping matrix are set to 0 through the missing measurement mask matrix, and the remaining valid observation vectors continue to participate in the state estimation. The observation noise term is estimated using the observation residual vector during the stable operation phase of the tunnel, and the noise variance is recorded separately for convergence deformation, crown settlement, lining stress, crack width, leakage, temperature, humidity, and groundwater pressure. The observation noise term refers to the observation bias caused by sensor errors, environmental interference, and short-term random disturbances; the observation residual vector is the difference vector between the actual observation vector and the model-predicted observation vector.

[0015] Optionally, the time series prediction module uses the current time... Observation vector results Starting from the predicted step size, a sequence of future states is generated step by step to obtain the future observation vector:

[0016]

[0017] in: For the prediction step number, This is the hidden state vector for the k-th step in the future, predicted based on information from the current moment. Let be the state transition matrix for the k-th step in the future. Let be the external driving vector for the k-th future step. Let be the external driving input matrix for the k-th future step. Let this be the maintenance treatment vector for the k-th step in the future. This is the impact matrix for the action taken at the k-th step in the future; Indicates that by the future... The future observation vector is obtained by mapping the hidden state vector step by step. Indicates the future number The observation mapping matrix of the step; Current moment State estimation results The latent state vector after data correction is at time [time]. Utilize time Actual observation data The predicted values ​​were corrected. Then, the optimal estimate is obtained; The external driving vectors for the k-th future step include: the future load sequence, the future temperature and humidity sequence, and the groundwater pressure sequence; The future load sequence is generated by the vehicle or train operation plan; Future temperature and humidity sequences and groundwater pressure sequences are generated from recent averages, weather forecasts, or historical data from the same period.

[0018] Optionally, the anomaly identification module calculates a comprehensive anomaly score based on the future observation vector and the actual observation vector of the tunnel:

[0019] in: For comprehensive anomaly scoring, This is the observation residual vector between the actual observation vector and the future observation vector. The residual covariance matrix is... Indicates matrix transpose; When the comprehensive anomaly score When the threshold is exceeded, a transient anomaly candidate is generated; When the comprehensive anomaly score of multiple consecutive time steps An abnormal trend is generated when the first threshold is exceeded. The observed residual vectors include: convergence deformation residual vector, crown settlement residual vector, lining stress residual vector, crack width residual vector, leakage residual vector, temperature and humidity residual vector, and groundwater pressure residual vector. Anomaly types include: structural deformation anomalies, crack propagation anomalies, and hydrologically driven leakage anomalies; When the residual vectors of convergence deformation, crown settlement, and lining stress increase simultaneously, it is marked as structural deformation anomaly; When the residual vectors of crack width and lining stress increase simultaneously, it is marked as an abnormal crack propagation. When the residual vectors of leakage, temperature and humidity, and groundwater pressure increase synchronously, it is marked as a hydrologically driven leakage anomaly.

[0020] Optionally, the service status assessment module writes the implicit state vector output by the state variable construction module, the future observation vector output by the time series prediction module, the comprehensive anomaly score and the anomaly type into the assessment rule table to obtain the service status of the tunnel structure. The service status of the tunnel structure includes four levels: normal, attention, early warning, and danger.

[0021] A method for predicting the service status of tunnel structures based on a state-space model, the method comprising: Step S1: Integrate observation data from multiple sources, perform time alignment, anomaly removal, missing data completion, and normalization processing to generate observation vectors arranged according to a uniform time step. Step S2: Using the monitoring section as the basic unit, the observation vector is mapped into an implicit state vector consisting of deformation state, stress state, crack damage state, leakage state and environmental disturbance state. Step S3: Establish a state transition model to describe the time evolution of the hidden state vector:

[0022] in: This represents the hidden state vector at the current moment. Let be the hidden state vector of the previous time step. To standardize the time step, This is the state transition matrix, used to characterize the continuity of each state component and the influence relationships between them. For external driving vectors, For external driving input matrix, For maintenance and treatment vectors, In order to handle the impact matrix, For state disturbance terms; Step S4: Establish the relationship between the hidden state vector and the observation vector through the observation mapping model: in, For the actual observed vector, For the observation mapping matrix, For the observed noise term; When the data of the hidden state vector is missing, the corresponding component is masked by the missing data mask matrix. Step S5: Starting from the current state estimate, and combining the future external driving forces and maintenance treatment sequence, generate a rolling implicit state prediction sequence for the next preset number of days, i.e., the future observation vector; Step S6: Calculate the residual vector between the future observation vector and the actual observation vector, as well as the comprehensive anomaly score and anomaly type. When the comprehensive anomaly score exceeds the limit continuously, it is judged as a trend anomaly. Anomaly types include: structural deformation anomaly, crack propagation anomaly, and hydrologically driven leakage anomaly. Step S7: Determine the service status of the tunnel structure by using the implicit state vector, future observation vector, comprehensive anomaly score, and anomaly type service status assessment.

[0023] The beneficial effects of the technical solution provided in this application are: 1. This invention converts multi-source observation indicators into five implicit state components: deformation state, stress state, crack damage state, leakage state, and environmental disturbance state. This avoids judging the structural safety state based on only a single observation indicator or fixed threshold, enabling the system to be further elevated from the observation data layer to the structural service state layer, thereby improving the completeness and interpretability of the expression of the tunnel's true service state.

[0024] 2. This invention uses a state transition matrix, an external driving input matrix, a maintenance and treatment influence matrix, and a state disturbance covariance matrix to uniformly express the dynamic evolution relationship of the tunnel structure state with time, load, hydrology, temperature and humidity, surrounding disturbances, and maintenance and treatment. This can overcome the problem that traditional curve extrapolation methods are difficult to reflect the structural deterioration mechanism and external driving influence.

[0025] 3. This invention uses a sliding window method to update the state transition matrix and sets stability constraints on the main diagonal elements such as deformation state, cracking state, and leakage state. This allows the model to adapt to recent changes in operational status while avoiding drastic fluctuations in model parameters caused by short-term noise, thereby improving the stability and engineering applicability of long-term service trend prediction.

[0026] 4. This invention introduces sensor reliability parameters, observation noise covariance, and engineering sensitivity coefficients into the observation mapping process. It incorporates the historical correlation between observation indicators and implicit states, equipment online rate, missing measurement rate, status codes, and historical noise variance into the calculation, which can reduce the impact of sensor drift, sporadic noise, and single-point anomalies on the state estimation results.

[0027] 5. This invention sets up a missing measurement mask matrix. When a certain observation index is missing, only the corresponding observation component and the mapping matrix row are masked. The remaining valid observation indexes can still participate in the state correction, avoiding the interruption of the entire prediction, identification and evaluation process due to the short-term offline status of a single sensor, and improving the system's continuous operation capability under incomplete monitoring data conditions.

[0028] 6. Starting with the current state estimation results, this invention combines the future load sequence, temperature and humidity sequence, groundwater pressure sequence, and maintenance sequence to generate a future state sequence in rolling prediction windows of 7 days, 30 days, or 90 days, and simultaneously outputs the prediction confidence interval, enabling maintenance personnel to grasp the risk development trends such as accelerated deformation, crack propagation, and enhanced leakage in advance.

[0029] 7. This invention identifies anomalies based on the residual vector, residual covariance matrix, and comprehensive anomaly score between predicted and actual observations. It also distinguishes between instantaneous anomalies and trend anomalies by combining a continuous anomaly counter. This allows for the identification of signs of anomaly development before the observed indicators reach the traditional alarm threshold, thereby improving the early warning capability during tunnel operation.

[0030] 8. Based on the coordinated changes of indicators such as convergence deformation, arch settlement, lining stress, crack width, leakage, humidity and groundwater pressure, this invention distinguishes between structural deformation anomalies, crack expansion anomalies and hydrologically driven leakage anomalies, avoiding the simple judgment of single-point fluctuations as structural risks and improving the pertinence and accuracy of anomaly type identification.

[0031] 9. This invention integrates state estimation, future trends, anomaly scoring, and evaluation rule tables to output four service statuses: normal, watch out, early warning, and hazardous. It also provides section number, main contribution indicators, expected risk arrival time, and maintenance and treatment recommendations. This ensures that the early warning results not only reflect the current risk level but also provide a time-series basis for operation and maintenance decisions such as increased inspection frequency, drainage inspection, crack verification, grouting reinforcement, and lining repair. Attached Figure Description

[0032] The present application will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 The overall operation flowchart of the tunnel structure service status timing prediction system provided in the embodiments of the present invention; Figure 2 This is a flowchart of multi-source monitoring data access and unified observation vector generation provided in an embodiment of the present invention; Figure 3 This is a flowchart of implicit state construction and state space model closed-loop update provided in an embodiment of the present invention; Figure 4 A flowchart for timing prediction, anomaly identification, and service status assessment provided in this embodiment of the invention; Figure 5 This is a typical sensor deployment example diagram for an operational tunnel monitoring section provided in an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the mapping relationship between the implicit state vector and the observation index provided in an embodiment of the present invention. Figure 7 This invention provides a knowledge graph of observation indicators, implicit states, anomaly types, and maintenance treatments. Figure 8 This is a diagram illustrating the effect of multi-index time alignment and missing test mask processing provided in an embodiment of the present invention. Figure 9 Five types of hidden state component radar evaluation diagrams provided for embodiments of the present invention; Figure 10 The state transition matrix parameter heatmap and stability constraint distribution diagram provided in the embodiments of the present invention; Figure 11 This is a confidence interval diagram of the predicted crack width and leakage volume for the next 30 days provided in an embodiment of the present invention; Figure 12 This is a linkage identification diagram between observation residuals and comprehensive anomaly scores provided in an embodiment of the present invention; Figure 13 Multi-ring diagram of service status index composition and main contributing indicators provided in embodiments of the present invention; Figure 14 This is a visualization interface diagram of the comprehensive early warning of tunnel operation and service status provided in an embodiment of the present invention; Figure 15 This is a screenshot of the tunnel cross-section trend prediction and maintenance details interface provided in an embodiment of the present invention. Detailed Implementation

[0033] To provide a clearer understanding of the technical features, objectives, and effects of this application, the specific embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0034] The embodiments of this application provide a method for predicting the service status of tunnel structures based on a state-space model.

[0035] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating the steps of a method for predicting the service status of a tunnel structure based on a state-space model, as described in an embodiment of this application, including: Example 1: Please see Figures 1 to 15 A tunnel structure service status prediction system based on a state-space model, deployed in the cloud server and edge gateway of a tunnel operation monitoring platform, includes: Observation Data Preprocessing Module: This module, serving as the foundational support module of the system, is primarily deployed between the edge gateway and the cloud server. It is used to organize multi-source monitoring data collected on-site into a unified observation sequence that can be directly accessed by the state-space model. During tunnel operation, convergence gauges, arch settlement gauges, lining stress gauges, crack gauges, leakage gauges, temperature and humidity sensors, and groundwater pressure gauges are deployed at typical cross-sections. Data is accessed to the edge gateway via RS485, LoRa, 4G / 5G, or industrial Ethernet. The edge gateway records the cross-section number, sensor number, acquisition time, monitoring value, unit, and equipment status code according to a unified data format and uploads it to the time-series database.

[0036] In this embodiment, the system uses a uniform time step of 10 minutes to align data from different sampling frequencies to the same time point, forming an observation vector: ;in, To converge the deformation, For the settlement of the vault, For lining stress, The width of the crack. This refers to the leakage rate. For temperature, For humidity, The value represents groundwater pressure. For temperature and humidity data with sampling intervals less than 10 minutes, the current time step data is generated using window averaging; for data with sampling intervals greater than 10 minutes and missing data for no more than two cycles, linear interpolation is used for completion. If the duration of missing data exceeds a set length, the system does not perform interpolation but instead generates a missing data label for subsequent state estimation module identification.

[0037] To eliminate the differences in the dimensions of different observation indicators, the system uses historical stable period observation data. Standardization process: ;in, and Calculated from data of at least 30 days of stabilization period. Desirable ; This represents standardized historical stable-period observation data. For example, if the mean convergence deformation of a cross-section over 30 days is 4.20 mm with a standard deviation of 0.50 mm, and the current value is 5.10 mm, then the standardized result is approximately 1.80. The data, after time alignment, missing measurement processing, anomaly labeling, and standardization, serves as the unified input for both the state variable construction module and the observation mapping module.

[0038] State Variable Construction Module: This module, deployed in the state-space modeling layer of the cloud server, converts the standardized observation vectors output by Module 1 into implicit state vectors that characterize the actual service state of the tunnel structure. Instead of directly using monitoring values ​​such as convergence deformation, crown settlement, lining stress, crack width, and leakage as prediction objects, this module maps multi-source monitoring data (observation data from monitoring sections) into deformation states, stress states, crack damage states, leakage states, and environmental disturbance states based on the tunnel structure's service mechanism. This provides a unified state representation for subsequent state transition modeling.

[0039] In this embodiment, the state variable construction module uses a single monitoring section or several adjacent monitoring sections as the basic calculation unit, and calculates the time... The hidden state vector is defined as follows: Where T represents transpose, Indicates the state of deformation. Indicates the state of force. Indicates the state of cracking or damage. Indicates a leakage condition. This indicates the state of environmental disturbance. The above state variables are not equivalent to a single sensor value, but are calculated jointly from the current value, rate of change, cumulative change, and related indicators.

[0040] For deformation state The system comprehensively considers the standardized values ​​and rates of change of convergence deformation and crown settlement to calculate the deformation state components: ;in, and These are the standardized convergence deformation and the crown settlement, respectively. To standardize the time step, to This is a weighting coefficient, which can be calibrated based on historical monitoring data, structural safety assessment experience, or engineering specification limits. For example, if the uniform time step of an operating tunnel is 10 minutes, and if the convergence deformation of a certain section continues to increase while the crown settlement rises synchronously, then... It will be affected by both the current deformation amplitude and the rate of change, avoiding the use of a single point deformation value to judge the state.

[0041] For the state of cracking The system is constructed by comprehensively considering crack width, crack growth rate, and crack duration. ;in, The standardized crack width, This is the normalized value corresponding to the duration of continuous crack growth. , and This is used to assign weights to the crack damage status. For example, if the width of a crack increases from 0.16mm to 0.20mm, even though it has not yet reached the alarm limit, if it shows an increasing trend for 7 consecutive days, the crack damage status component will increase earlier, providing a trend basis for subsequent anomaly identification.

[0042] For leakage conditions The system couples and expresses leakage, humidity, and groundwater pressure: ;in, , , These are the standardized leakage rate, humidity, and groundwater pressure, respectively. , and The weights are used to differentiate between structural leakage development driven by simple increases in ambient humidity and those driven by groundwater pressure. For example, when humidity rises while groundwater pressure and leakage also increase simultaneously, the system identifies the leakage state of that section as a higher level, rather than simply viewing it as an environmental fluctuation.

[0043] Force state Constructed from the lining stress and its rate of change, environmental disturbance state The weighting coefficients are constructed from temperature, humidity, and water pressure fluctuations. During the weighting calibration of each state component, the system uses historical stable-period monitoring data, engineering limits, and operational data from similar tunnels to jointly determine the weighting coefficients. Engineering limits refer to the allowable or warning values ​​for indicators such as convergence deformation, crown settlement, lining stress, crack width, leakage, and groundwater pressure, determined according to design documents, current specifications, operational management requirements, or project alarm rules. Operational data from similar tunnels refers to existing tunnel monitoring data, inspection records, and maintenance data that are similar to the current tunnel in terms of structural form, surrounding rock conditions, operational loads, monitoring indicators, or types of defects. Using these data to jointly calibrate the weighting coefficients avoids the bias in state component weights caused by relying solely on short-term monitoring samples from a single tunnel. After each state component is calculated, the system combines them into a unified implicit state vector and writes it into the state space model cache. Thus, the subsequent state transition modeling module no longer processes discrete sensor raw values, but rather state variables that reflect the evolution characteristics of the tunnel structure's service state. The core innovation of this module lies in elevating multi-source observation data from the "indicator layer" to the "structural state layer," which not only preserves the engineering meaning of the monitoring data but also establishes a unified computational foundation for state transition, observation mapping, trend prediction, and service assessment.

[0044] As one example, each state component consists of the current observed index value, the change in adjacent time steps, the duration of continuous growth, and related index coupling terms. The state variable construction results are written into the state space model cache.

[0045] State Transition Modeling Module: Deployed in the state-space modeling layer of a cloud server, this module describes the evolution of the implicit service state of the tunnel structure over time. It receives the implicit state vector output from the state variable construction module and, combined with vehicle or train loads, groundwater pressure changes, temperature and humidity fluctuations, seasonal factors, surrounding construction disturbances, and maintenance records, establishes a dynamic model of the tunnel structure's state changing from the previous moment to the current moment. Unlike methods that only extrapolate monitoring curves, this module incorporates both "structural degradation inertia" and "external environmental driving effects" into the calculation, making the prediction results more consistent with the long-term service characteristics of tunnel structures during the operational phase.

[0046] In this embodiment, the state transition model is expressed in discrete-time state-space: ;in, This represents the hidden state vector at the current moment. This is the hidden state vector from the previous time step. To standardize the time step; It is a state transition matrix used to characterize the self-continuity and mutual influence between deformation, stress, cracking, leakage and environmental disturbance states; External driving vector; For external driving input matrix; For maintenance and treatment vectors; To address the impact matrix; This is the state disturbance term.

[0047] The external driving vector can be set according to the actual engineering requirements: ;in, This indicates the load intensity of vehicles or trains within the current time window. This indicates the change in groundwater pressure. Indicates the amount of temperature change. Indicates the amount of change in humidity. This indicates the intensity of surrounding construction or vibration disturbance. Taking an operating railway tunnel as an example, the system counts the number of trains passing and axle load levels in 10-minute time steps. When the train load intensity increases and the groundwater pressure rises simultaneously during a certain period, the model will enhance the coupling effect between the stress state, crack state, and leakage state, rather than treating a sudden increase in a single observation indicator as an independent anomaly.

[0048] The state transition matrix is ​​not a fixed constant, but is calibrated using a sliding window based on historical state sequences and external driving sequences. The system can select the most recent 30 or 90 days of data as the training window and update the parameters using a least squares method with regularization constraints. ;in, The length of the sliding window. To smooth the constraint coefficients and prevent drastic fluctuations in the transition matrix due to short-term noise, for conventionally operating tunnels, the following can be used: The time steps were set to 4320, corresponding to approximately 30 days of data sampled over 10 minutes. The value can be between 0.01 and 0.10. The system can also set engineering stability constraints on the state transition matrix. For example, the main diagonal coefficients for deformation and cracking states are generally limited to between 0.90 and 1.05 to avoid the model generating divergent predictions that do not conform to long-term service patterns.

[0049] Regarding the impact of maintenance and treatment, the system encodes operations such as grouting, drainage, crack sealing, lining reinforcement, and speed-limited operation into treatment vectors. For example, after completing a drainage diversion, The drainage component is set from 0 to 1 and decays over several consecutive time steps. After the crack is sealed, the components corresponding to the crack damage state and leakage state are corrected according to the treatment influence matrix. In this way, the model can distinguish between "natural deterioration trend" and "state decline after maintenance intervention", avoiding incorrect extrapolation based on the original deterioration trend after maintenance measures are implemented.

[0050] The state disturbance term describes the uncertainty caused by unmodeled factors, and its covariance can be dynamically estimated based on historical prediction biases. ;in, Let be the state perturbation covariance matrix. The state value is predicted based on the previous time step; Indicates the first The actual observed state value at a given moment. When a section is in the rainy season or during a period of drastic fluctuations in groundwater pressure, the historical prediction deviation increases, and the system will simultaneously increase the disturbance covariance, widening the confidence interval of subsequent predictions, thus reminding maintenance personnel that the uncertainty of the state evolution is high during this period.

[0051] The core innovation of this module lies in treating the service state of tunnel structures as a continuously evolving latent state process. It expresses the transmission relationships between various state components through a state transition matrix, the influence of loads, hydrology, and environmental factors through an external driving matrix, and the changes in the state evolution path caused by maintenance measures through a treatment impact matrix. This module is closely integrated with the previous module: the state variable construction module provides a unified latent state representation, and the state transition modeling module further describes the changes in latent states over time, providing a computable, updatable, and interpretable dynamic model foundation for subsequent observation mapping, time-series prediction, anomaly identification, and service state assessment.

[0052] Observation Mapping Module: This module, deployed in the state-space modeling layer of a cloud server, is used to establish the correspondence between implicit service state vectors and actual field monitoring data. The state variable construction module obtains... These are structural state-level variables and cannot be directly and completely read by sensors. Convergence deformation, crown settlement, lining stress, crack width, leakage, temperature and humidity, and groundwater pressure, collected by field sensors, belong to the observation-level variables. The role of the observation mapping module is to establish a computable relationship between the two, enabling the system to use actual monitoring data to reverse-correct the implicit state and maintain stable state estimation even under sensor noise, localized missing measurements, or single-point anomalies.

[0053] In this embodiment, the observation mapping model is expressed as follows: ;in, For the observation vector, The hidden state vector, For the observation mapping matrix, This represents the observation noise term. The observation mapping matrix is ​​used to express the sensitivity of different observation indicators to different implicit states. For example, convergence deformation and crown settlement mainly correspond to the deformation state, lining stress mainly corresponds to the stress state, crack width mainly corresponds to the crack damage state, and leakage, humidity, and groundwater pressure together correspond to the leakage state and the environmental disturbance state.

[0054] To ensure the mapping relationship aligns with the actual tunnel structure engineering, the system determines matrix elements based on the sensitivity of observation indicators, cross-sectional location, sensor reliability, and historical correlation. For the first... The first observation indicator and the first The mapping coefficients between the state components can be initially calculated using the following formula: ;in, The correlation coefficient between observed indicators and state components (deformation state component, stress state component, crack damage state component, leakage state component, and environmental disturbance state component) in historical data. For the first Reliability coefficient of sensor type For the first The engineering sensitivity coefficient of state components. The historical correlation coefficient refers to the correlation parameter between the monitoring index and the state component, calculated based on historical monitoring data. It can be determined using Pearson correlation coefficient, Spearman correlation coefficient, or normalized covariance. The engineering sensitivity coefficient refers to the response strength coefficient of the monitoring index to changes in service condition, determined based on the monitoring point location, structural stress characteristics, type of defects, and engineering experience. For example, if the crack gauge has a long-term online rate of 98% and stable data fluctuations, then... A value of 0.98 can be used; if the communication of a humidity sensor is unstable, the reliability coefficient can be reduced to 0.70, thereby reducing its impact on leakage status estimation.

[0055] For monitoring noise, the system establishes a noise covariance matrix based on the observation residuals during stable operation:

[0056] in, These correspond to the observation noise variances for convergence, settlement, stress, cracks, leakage, temperature, humidity, and groundwater pressure, respectively. The observation noise term refers to the deviation in observed values ​​caused by sensor measurement errors, environmental interference, communication fluctuations, and short-term random disturbances. The observation residual is the difference between the actual observed vector and the model-predicted observed vector at the same moment, used to estimate the observation noise and determine the degree of deviation in the monitoring data. Taking an operating tunnel as an example, if the standard deviation of the noise during the stabilization period of the convergence meter is 0.05 mm, the corresponding variance can be taken as 0.0025. If the leakage monitoring is significantly affected by fluctuations in on-site drainage, the noise variance can be appropriately increased to prevent the model from over-correcting the hidden state when leakage observation values ​​fluctuate in the short term.

[0057] During real-time operation, the observation mapping module maps the predicted state output by the state transition modeling module to the predicted observation value. : The system then compares the predicted and actual observations and passes the differences to the subsequent time-series prediction and anomaly identification modules. If the measured crack width is 0.22 mm at a certain moment, while the model predicts 0.18 mm based on the current crack damage state, and this deviation exists for multiple consecutive time steps, the system will not immediately treat it as single-point noise. Instead, it will combine the crack damage state, leakage state, and groundwater pressure changes to determine whether the implicit state needs to be corrected.

[0058] When some sensors are missing data, the observation mapping module masks the corresponding observation components using a missing data mask matrix: ;in, For a diagonal mask matrix, the elements corresponding to valid observations are set to 1, and the elements corresponding to missing observations are set to 0. This represents the effective observation vector after processing with the missing measurement mask; This represents the observation matrix after the missing measurement masking process. In this way, the system can continue to complete state estimation using the remaining valid monitoring data. For example, if a groundwater pressure gauge at a certain cross-section goes offline for a short period, the system can still maintain its leakage status judgment based on leakage volume, humidity, and crack width, avoiding the interruption of the entire prediction chain due to the failure of a single sensor.

[0059] The core innovation of this module lies in establishing an explicit mapping relationship between sensor observations and the implicit service state of the tunnel structure. This allows the system to both correct the model using field monitoring data and maintain continuous and stable state estimation under conditions of data noise, missing measurements, and single-point anomalies. This module receives the predicted state output from the state transition modeling module and provides an observational basis for subsequent time-series prediction modules. It also provides the anomaly identification module with information on the deviation between predicted and measured observations.

[0060] Time-series prediction module: This module is deployed in the prediction calculation layer of the cloud server. After the state variable construction, state transition modeling, and observation mapping are completed, it outputs the trend of tunnel structure service status changes within a certain time window in the future. This module receives the state transition parameters generated by module three, the observation mapping relationship generated by module four, and the state estimation results at the current moment. It performs rolling calculations according to the set prediction step size to form a prediction sequence of indicators such as future deformation, stress, cracks, and leakage.

[0061] In this embodiment, the system uses the current state estimation result, corrected by actual monitoring data, as the prediction starting point. The prediction window can be set to the next 7 days, 30 days, or 90 days. If the field data uses a uniform time step of 10 minutes, then the 7-day prediction window corresponds to 1008 prediction steps. Based on the availability of future load, hydrological, and environmental inputs, the system employs two methods: for train or vehicle loads, it can call upon the operation plan to form a future load sequence; for temperature, humidity, and groundwater pressure, it can use recent averages, weather forecasts, or historical data from the same period to generate an external driving sequence.

[0062] The multi-step state prediction process is represented as follows: ;in, For the prediction step number, Indicates based on the current time Information predicts the future The hidden state vector of each step Indicates the future number The state transition matrix of the step, Indicates the future number The external driving vector of the step, Indicates the future number The external driving input matrix of the step, Indicates the future number Step-by-step maintenance and treatment vector, Indicates the future number The impact matrix of the treatment steps. If no maintenance treatment is scheduled during the forecast period, the corresponding treatment variable is set to zero; if drainage, grouting or crack sealing is planned, it is written into the forecast sequence according to the treatment time and duration of effect.

[0063] After the state prediction is completed, the system generates predicted values ​​for future observation indicators based on the observation mapping relationship: ;in, Indicates that by the future... The predicted vector of future monitoring indicators is obtained by mapping the hidden state vector. Indicates the future number The system uses an observation mapping matrix for each step. Predicted results include future curves for indicators such as convergence deformation, crown settlement, lining stress, crack width, and leakage. For example, if the current crack width of an operating tunnel section is 0.18 mm, and the system predicts it will increase to 0.27 mm within the next 30 days, while groundwater pressure continues to rise, this result will be transmitted to the anomaly identification module to determine whether there is a coupled development trend between crack damage and leakage.

[0064] To avoid misjudgment caused by a single predicted value, the system simultaneously outputs the prediction confidence interval: ;in, For the first The prediction interval for each observed indicator To correspond to the prediction variance, This represents the confidence level coefficient. In engineering applications, a 95% confidence level can be used, corresponding to... =1.96. When groundwater pressure fluctuates significantly or historical prediction errors increase, the prediction interval automatically widens, indicating higher prediction uncertainty during this period.

[0065] This module outputs the future trend curve, prediction range, indicator growth rate, and the expected time to reach the warning threshold, and transmits the results to the anomaly identification module and the service status assessment module. In this way, the system can not only determine whether the current state is abnormal, but also identify potential risks of accelerated deformation, crack propagation, and increased leakage in the future, providing a time basis for operating units to formulate inspection plans and maintenance arrangements.

[0066] Anomaly Detection Module: Deployed in the prediction and early warning layer of the cloud server, this module identifies whether the tunnel structure exhibits trend deviations, abrupt disturbances, or continuous deterioration based on the discrepancy between predicted and actual observations. Unlike single fixed threshold alarms, this module utilizes normal evolution prediction results obtained from a state-space model, comparing measured data with predicted data to detect abnormal changes before reaching alarm limits.

[0067] In this embodiment, the system first calculates the observation residuals: ;in, This represents the current actual observation vector. The current observation vector is the one predicted at the previous time step. This is the residual vector. If the measured crack width is 0.24 mm and the predicted value is 0.19 mm, then the crack residual is 0.05 mm. If this deviation occurs continuously, it indicates that the crack development rate may be higher than the normal evolution level.

[0068] To eliminate differences in the dimensions of different indicators and noise, the system calculates a comprehensive anomaly score: ;in, For abnormal scoring, Let be the residual covariance matrix. When a threshold is exceeded, the system marks it as an anomaly candidate; if multiple consecutive time steps exceed the threshold, it is determined to be an abnormal trend. For example, with a uniform time step of 10 minutes, if the threshold is exceeded more than 4 times within 1 hour, it can be determined to be a persistent anomaly.

[0069] The system further distinguishes anomaly types based on the synergistic relationship of indicators: if convergence deformation, crown settlement, and lining stress residual increase synchronously, it is judged as structural deformation anomaly; if crack width and lining stress residual increase synchronously, it is judged as crack expansion anomaly; if leakage, humidity, and groundwater pressure residual increase together, it is judged as hydrologically driven leakage anomaly; if only a single sensor spikes, and related indicators do not change synchronously, it is marked as a candidate for equipment anomaly. This module outputs the anomaly location, anomaly type, anomaly score, and main contributing indicators, and transmits them to the service status assessment module.

[0070] Service Status Assessment Module: This module, deployed in the cloud server's operation and maintenance application layer, integrates the implicit states output by the state variable construction module, the future trends output by the time series prediction module, and the anomaly scores output by the anomaly identification module to generate the tunnel structure's service status level and maintenance recommendations. This module belongs to the results fusion and engineering application module, primarily transforming the aforementioned model calculation results into assessment results that operation and management personnel can directly understand and implement.

[0071] In this embodiment, the system performs weighted fusion of deformation state, stress state, crack damage state, leakage state, and environmental disturbance state to form a service status index: ;in, This is the service status index. , , , These represent the weights for each state component. The weights can be set according to the tunnel type, structural location, and operational management requirements. For example, in operating railway tunnels, cracks and deformations have a significant impact on structural safety, and therefore, weights can be set accordingly. =0.30、 =0.20、 =0.25、 =0.15、 =0.10.

[0072] The system classifies status levels based on both the service status index and the anomaly score. For example, when... A score <0.35 and an abnormal score not exceeding the limit is considered normal; when 0.35 ≤ When <0.60, it is considered as a watchlist; when 0.60≤ A warning is issued when the score is <0.80 or when abnormal scores exceed the limit consecutively; when A value ≥0.80 with persistent anomalies is considered dangerous. If a cross-section is calculated to have... =0.68. If the crack width and groundwater pressure residual exceed the threshold for 1 hour, the system will output an "early warning" level and mark it as a crack damage-leakage coupling risk.

[0073] To facilitate the interpretation of the evaluation results, the system calculates the contribution rate of each state component: ;in, For the first The contribution rate of the state-type components to the service condition index. If the contribution rate of the crack damage state reaches 42% and the contribution rate of the leakage state reaches 28%, the system will classify crack propagation and leakage development as the main sources of risk.

[0074] The module ultimately outputs the section number, status level, key contribution indicators, future trends, expected risk arrival time, and maintenance recommendations. For the "Attention" level, more frequent inspections are recommended; for the "Warning" level, specialized testing, drainage checks, or crack verification are recommended; for the "Hazard" level, measures such as speed limits, closure inspections, grouting reinforcement, or lining repair are recommended. The assessment results are simultaneously written to the operation and maintenance database and displayed in the visualization interface as section risk color bars, trend curves, and a list of remedial measures.

[0075] As one example, the normal state corresponds to each state component of the implicit state vector and the comprehensive anomaly score meeting the preset threshold; the state of concern corresponds to one state component in the implicit state vector continuously increasing; the state of warning corresponds to multiple state components increasing simultaneously or the anomaly score continuously exceeding the limit; the state of danger corresponds to the high-risk state component continuously increasing and the anomaly score continuously exceeding the limit; the assessment results include the section number, state level, main contribution indicators, expected risk arrival time and maintenance and treatment recommendations.

[0076] Figure 1 The overall operation flowchart of the tunnel structure service status time-series prediction system; Figure 2 Flowchart for multi-source monitoring data access and unified observation vector generation; Figure 3 Flowchart for implicit state construction and closed-loop update of the state space model; Figure 4 Flowchart for time series prediction, anomaly identification, and service status assessment; Figure 5 Example diagram of sensor deployment at a typical monitoring section of an operational tunnel; Figure 6 This is a schematic diagram illustrating the mapping relationship between the hidden state vector and the observed index. Figure 7 A knowledge graph of observation indicators, implicit states, anomaly types, and maintenance treatments; Figure 8 The diagram shows the effect of multi-indicator time alignment and missing test mask processing; Figure 9 Radar evaluation diagrams for five types of implicit state components; Figure 10 Heatmap of state transition matrix parameters and distribution of stability constraints; Figure 11 Confidence interval plot for predicting crack width and leakage volume over the next 30 days; Figure 12 A diagram showing the linkage between observed residuals and comprehensive anomaly scores; Figure 13 Multi-ring diagram of the components and main contributing indicators of the service status index; Figure 14A comprehensive early warning visualization interface for the tunnel's operational status; Figure 15 This is a screenshot showing the interface details of tunnel cross-section trend prediction and maintenance treatment.

[0077] A method for predicting the service status of tunnel structures based on a state-space model, the method comprising: Step S1: Integrate observation data from multiple sources, perform time alignment, anomaly removal, missing data completion, and normalization processing to generate observation vectors arranged according to a uniform time step. Step S2: Using the monitoring section as the basic unit, the observation vector is mapped into an implicit state vector consisting of deformation state, stress state, crack damage state, leakage state and environmental disturbance state. Step S3: Establish a state transition model to describe the time evolution of the hidden state vector:

[0078] in: This represents the hidden state vector at the current moment. Let be the hidden state vector of the previous time step. To standardize the time step, This is the state transition matrix, used to characterize the continuity of each state component and the influence relationships between them. For external driving vectors, For external driving input matrix, For maintenance and treatment vectors, In order to handle the impact matrix, For state disturbance terms; Step S4: Establish the relationship between the hidden state vector and the observation vector through the observation mapping model: in, For the actual observed vector, For the observation mapping matrix, For the observed noise term; When the data of the hidden state vector is missing, the corresponding component is masked by the missing data mask matrix. Step S5: Starting from the current state estimate, and combining the future external driving forces and maintenance treatment sequence, generate a rolling implicit state prediction sequence for the next preset number of days, i.e., the future observation vector; Step S6: Calculate the residual vector between the future observation vector and the actual observation vector, as well as the comprehensive anomaly score and anomaly type. When the comprehensive anomaly score exceeds the limit continuously, it is judged as a trend anomaly. Anomaly types include: structural deformation anomaly, crack propagation anomaly, and hydrologically driven leakage anomaly. Step S7: Determine the service status of the tunnel structure by using the implicit state vector, future observation vector, comprehensive anomaly score, and anomaly type service status assessment.

[0079] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure.

[0080] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A system for predicting the service status of tunnel structures based on a state-space model, characterized in that, The system includes: The observation data preprocessing module is used to collect and preprocess the observation data of the monitoring section to obtain the observation vector. The observation data include: convergence deformation, crown settlement, lining stress, crack width, leakage, temperature and humidity, and groundwater pressure; The state variable construction module uses a single monitoring section or a group of adjacent monitoring sections as the basic computational unit to convert the observation vector into an implicit state vector: in, Represents the deformation state components. Indicates the force state components, Indicates the component representing the crack damage state. Indicates the leakage status component. Represents the environmental disturbance state component; T represents transpose; The deformation state components consist of standardized convergent deformation, lining stress, arch settlement and its rate of change. The crack damage state component consists of crack width, crack growth rate, and continuous growth duration. The leakage status component consists of leakage amount, temperature and humidity, and groundwater pressure. The weight coefficients of each state component in the implicit state vector are determined by historical stable period observation data, engineering preset limits and tunnel operation data of the same type, and are updated according to a set cycle. The state variable construction module is used to construct implicit state vectors based on observation vectors, including five types of states: deformation state, stress state, crack damage state, leakage state, and environmental disturbance state. The state transition modeling module is used to construct discrete-time state transition models; The state transition modeling module is used to establish the following discrete-time state transition model: in: This represents the hidden state vector at the current moment. Let be the hidden state vector of the previous time step. To standardize the time step, This is the state transition matrix, used to characterize the continuity of each state component and the influence relationships between them. For external driving vectors, For external driving input matrix, For maintenance and treatment vectors, In order to handle the impact matrix, This refers to the state disturbance term; The external driving vectors include: vehicle or train load intensity, groundwater pressure change, temperature change, humidity change, and surrounding construction disturbance intensity. The state transition matrix is ​​updated using a sliding window method, and the length of the sliding window is determined based on a uniform sampling period. The main diagonal elements of the state transition matrix are subject to stability constraints, with the values ​​of the main diagonal elements corresponding to the deformation state, cracking state, and leakage state limited to between 0.90 and 1.

05. The maintenance and treatment vectors are coded separately for drainage, grouting, crack sealing, lining reinforcement and speed-limited operation. The treatment impact matrix records the treatment start time, duration of effect, and state fallback coefficient; The observation mapping module is used to build observation mapping models. The time series prediction module is used to generate future observation vectors based on implicit state vectors, discrete-time state transition models, and observation mapping models. An anomaly identification module is used to obtain a comprehensive anomaly score and anomaly type based on future observation vectors and actual observation vectors; The service status assessment module is used to determine the service status of the tunnel structure based on the implicit state vector, future observation vector, comprehensive anomaly score, and anomaly type.

2. The tunnel structure service status prediction system based on a state-space model as described in claim 1, characterized in that, The observation mapping module establishes the following observation mapping model: in, This refers to the actual monitored observation vector. For the observation mapping matrix, The hidden state vector, For the observed noise term; The elements in the observation mapping matrix are jointly determined by the historical correlation coefficient between the observation index and the state components in the implicit state vector, the sensor reliability coefficient, and the engineering sensitivity coefficient. The engineering sensitivity coefficient is the coefficient of the strength of the response of the observed index to changes in service status; The observation indicators include: convergence deformation, crown settlement, lining stress, crack width, leakage, temperature and humidity, and groundwater pressure. The sensor reliability coefficient is determined by the sensor online rate, missing rate, equipment status code, and historical noise variance. The observation mapping module is equipped with a missing measurement mask matrix. When a certain observation index is missing, the observation vector corresponding to the observation index and the data corresponding to the row of the observation mapping matrix are set to 0 through the missing measurement mask matrix, and the remaining valid observation vectors continue to participate in the state estimation. The observed noise term is estimated using the observed residual vector during the stable operation phase of the tunnel, and the noise variance is recorded separately for convergence deformation, crown settlement, lining stress, crack width, leakage, temperature, humidity and groundwater pressure.

3. The tunnel structure service status prediction system based on a state-space model as described in claim 1, characterized in that, The time-series prediction module uses the current time... State estimation results Starting from the predicted step size, a sequence of future states is generated step by step to obtain the future observation vector: in: For the prediction step number, This is the hidden state vector for the k-th step in the future, predicted based on information from the current moment. Let be the state transition matrix for the k-th step in the future. Let be the external driving vector for the k-th future step. Let be the external driving input matrix for the k-th future step. Let this be the maintenance treatment vector for the k-th step in the future. This is the impact matrix for the action taken at the k-th step in the future; Indicates that by the future... The future observation vector is obtained by mapping the hidden state vector step by step. Indicates the future number The observation mapping matrix of the step; Current moment State estimation results The latent state vector after data correction is at time [time]. Utilize time Actual observation data The predicted values ​​were corrected. Then, the optimal estimate is obtained; The external driving vectors for the k-th future step include: the future load sequence, the future temperature and humidity sequence, and the groundwater pressure sequence; The future load sequence is generated by the vehicle or train operation plan; Future temperature and humidity sequences and groundwater pressure sequences are generated from recent averages, weather forecasts, or historical data from the same period.

4. The tunnel structure service status prediction system based on a state-space model as described in claim 1, characterized in that, The anomaly identification module calculates a comprehensive anomaly score based on the future observation vector and the actual observation vector of the tunnel: in: For comprehensive anomaly scoring, This is the observation residual vector between the actual observation vector and the future observation vector. The residual covariance matrix is... Indicates matrix transpose; When the comprehensive anomaly score When the threshold is exceeded, a transient anomaly candidate is generated; When the comprehensive anomaly score of multiple consecutive time steps An abnormal trend is generated when the first threshold is exceeded. The observed residual vectors include: convergence deformation residual vector, crown settlement residual vector, lining stress residual vector, crack width residual vector, leakage residual vector, temperature and humidity residual vector, and groundwater pressure residual vector. Anomaly types include: structural deformation anomalies, crack propagation anomalies, and hydrologically driven leakage anomalies; When the residual vectors of convergence deformation, crown settlement, and lining stress increase simultaneously, it is marked as structural deformation anomaly; When the residual vectors of crack width and lining stress increase simultaneously, it is marked as an abnormal crack propagation. When the residual vectors of leakage, temperature and humidity, and groundwater pressure increase synchronously, it is marked as a hydrologically driven leakage anomaly.

5. The tunnel structure service status prediction system based on a state-space model as described in claim 1, characterized in that, The service status assessment module writes the implicit state vector output by the state variable construction module, the future observation vector output by the time series prediction module, the comprehensive anomaly score and the anomaly type into the assessment rule table to obtain the service status of the tunnel structure. The service status of the tunnel structure includes four levels: normal, attention, early warning, and danger.

6. A method for predicting the service status of tunnel structures based on a state-space model, implemented based on a tunnel structure service status prediction system based on a state-space model as described in any one of claims 1-5, characterized in that, The method includes: Step S1: Integrate observation data from multiple sources, perform time alignment, anomaly removal, missing data completion, and normalization processing to generate observation vectors arranged according to a uniform time step. Step S2: Using the monitoring section as the basic unit, the observation vector is mapped into an implicit state vector consisting of deformation state, stress state, crack damage state, leakage state and environmental disturbance state. Step S3: Establish a state transition model to describe the time evolution of the hidden state vector: in: This represents the hidden state vector at the current moment. Let be the hidden state vector of the previous time step. To standardize the time step, This is the state transition matrix, used to characterize the continuity of each state component and the influence relationships between them. For external driving vectors, For external driving input matrix, For maintenance and treatment vectors, In order to handle the impact matrix, This refers to the state disturbance term; Step S4: Establish the relationship between the hidden state vector and the observation vector through the observation mapping model: in, For the actual observed vector, For the observation mapping matrix, For the observed noise term; When the data of the hidden state vector is missing, the corresponding component is masked by the missing data mask matrix. Step S5: Starting from the current state estimate, and combining the future external driving forces and maintenance treatment sequence, generate a rolling implicit state prediction sequence for the next preset number of days, i.e., the future observation vector; Step S6: Calculate the residual vector between the future observation vector and the actual observation vector, as well as the comprehensive anomaly score and anomaly type. When the comprehensive anomaly score exceeds the limit continuously, it is judged as a trend anomaly. Anomaly types include: structural deformation anomaly, crack propagation anomaly, and hydrologically driven leakage anomaly. Step S7: Determine the service status of the tunnel structure by using the implicit state vector, future observation vector, comprehensive anomaly score, and anomaly type service status assessment.

Images