Service period tunnel performance evaluation method based on intra-domain self-calibration

Abstract

The invention relates to a service period tunnel performance evaluation method based on intra-domain self-calibration, and the method comprises the steps: obtaining the basic attribute information of a target tunnel, and building an evaluation domain according to the basic attribute information; obtaining monitoring information in the evaluation domain, constructing an index data matrix by using the monitoring information, and endowing each piece of index data with a weight; setting a mechanism association rule, calculating a mechanism association matrix and a statistical association matrix through the mechanism association rule, and obtaining a disease association matrix based on the mechanism association matrix and the statistical association matrix; and obtaining a positive ideal solution and a negative ideal solution of each index in the evaluation domain, combining the weight with the disease incidence matrix to obtain a comprehensive covariance matrix, calculating a standard mahalanobis distance between the target tunnel and the positive and negative ideal solutions by using the comprehensive covariance matrix, and obtaining a comprehensive health score. According to the invention, adaptive and quantifiable intelligent evaluation and early warning of tunnel performance are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel engineering infrastructure intelligent operation and maintenance, in particular to a service period tunnel performance evaluation method based on in-domain self-calibration. BACKGROUND

[0002] There are 18997 operating railway tunnels in China, with a total length of 24246km, of which 4917 are high-speed railway tunnels with a total length of 8279km; the total length of urban rail transit operating lines is 10923.8km, with a tunnel section ratio of over 68%; there are 28724 highway tunnels with a total length of 32596.6km. With the increase of service life, affected by factors such as complex environment, construction quality and operation load, diseases such as water leakage, lining cracking and structural deformation occur frequently, and the phenomenon of "sick service" of tunnels is common. In recent years, there have been shutdown incidents due to disease deterioration in many places, highlighting the urgency of safety evaluation of service period tunnels.

[0003] Accurate evaluation of the performance of service tunnels is the core of ensuring their long-term operation safety and scientific operation and maintenance decisions. Currently, there are two main technical paths in this field:

[0004] One is the empirical evaluation method based on standards, which uses existing industry standards as a basis to complete performance evaluation through manual inspection, expert scoring and standard comparison. Although this method has a clear process, it uses static and uniform evaluation thresholds, which are difficult to adapt to individual differences in structure, geology and operation load of different tunnels, and treats cracks, leakage and other diseases as independent factors, ignoring their internal correlation and chain effect, resulting in evaluation results lacking in pertinence and unable to predict systemic risks.

[0005] With the development of intelligentization, methods based on mathematical models and machine learning have been gradually applied, such as the analytic hierarchy process, fuzzy comprehensive evaluation method and deep learning model. Although these methods have improved objectivity to some extent, they still have inherent limitations: weight setting cannot completely avoid subjectivity and still relies on expert scoring, deep learning models have poor interpretability and form a "black box" decision, and most models are fixed after establishment and cannot dynamically evolve with the degradation of tunnel performance and data accumulation. In addition, these models face the problem of model applicability in data sparse scenarios when they are applied to new tunnels or special tunnel types with insufficient data.

[0006] Existing technical solutions disclose a fuzzy assessment method and system for the risk situation of tunnel engineering. The method includes: determining risk evaluation indicators and constructing an indicator layer based on the actual construction environment of the tunnel engineering; constructing a hierarchical model for fuzzy risk situation assessment based on the indicator layer; determining indicator weights and a fuzzy evaluation matrix based on expert evaluation opinions; and calculating the tunnel engineering risk situation index. By introducing fuzzy mathematics and the fuzzy hierarchical method, the problem of data uncertainty and its impact on the assessment results in the risk situation assessment of tunnel engineering is solved.

[0007] Existing technical solutions also disclose a method for evaluating the service performance of highway tunnel facilities based on big data. This method includes: constructing a two-level fuzzy comprehensive evaluation model covering evaluation indicators for engineering survey, engineering design, and engineering construction; establishing a data sample library based on the evaluation indicator values ​​of the object to be evaluated; determining calculation parameters through sample data; and using fuzzy calculation to obtain the evaluation values ​​of each level of indicators. This evaluation method can obtain reliable evaluation results based on the analysis of a large amount of indicator sample data, enabling theoretical prediction of the service performance of highway tunnel facilities.

[0008] Existing technical solutions also disclose a method for rapid diagnosis of the service performance and prediction of the development trend of shield tunnels. The method includes: acquiring monitoring data on shield tunnel settlement, cross-sectional convergence deformation, cracks, and spalling; establishing longitudinal bolt stress grading evaluation standards and diameter deformation ratio grading evaluation standards; judging the longitudinal and transverse service performance of the tunnel based on monitoring data; and predicting the development trend of cracks from an energy perspective based on crack monitoring data. This method is simple to operate, provides reasonable diagnosis and comprehensive evaluation, and does not require the establishment of a complex evaluation index system.

[0009] While the above evaluation systems and methods have achieved some success in assessing the service performance of tunnels, they still generally suffer from common problems:

[0010] First, the evaluation standards are static and fixed, making it difficult to adapt to the individual differences of different tunnels in terms of structural form, geological environment and operating conditions;

[0011] Second, it isolates various diseases and ignores their inherent coupling relationships and evolutionary connections, making it difficult to identify systemic risks. Third, the model lacks the ability to continuously learn and dynamically update, and is prone to insufficient adaptability or "model aging" problems when facing the evolution of service status or new types of tunnels. Summary of the Invention

[0012] To address the problems existing in the prior art, the present invention aims to provide a service-life tunnel performance evaluation method based on intra-domain self-calibration. By establishing a dynamic domain division mechanism and self-calibration process, a fundamental shift in evaluation standards from static fixation to dynamic adaptation is achieved. By introducing the TOPSIS comprehensive evaluation method based on weighted Mahalanobis distance, the statistical correlation between defect indicators is essentially incorporated at the mathematical level, solving the evaluation bias caused by isolated defect treatment. By designing a sliding window update mechanism and a parameter inheritance strategy based on similarity, the system is endowed with continuous evolution capabilities, overcoming the applicability challenges in model aging and data sparsity scenarios, and realizing adaptive, quantifiable, intelligent evaluation and early warning of tunnel performance.

[0013] To achieve the above objectives, the present invention provides the following solution:

[0014] A service life tunnel performance evaluation method based on intra-domain self-calibration includes:

[0015] Step 1: Obtain the basic attribute information of the target tunnel, and create an evaluation domain based on the basic attribute information;

[0016] Step 2: Obtain monitoring information within the evaluation domain, construct an indicator data matrix using the monitoring information, and assign weights to each indicator data.

[0017] Step 3: Set the mechanism association rules, calculate the mechanism association matrix and the statistical association matrix through the mechanism association rules, and obtain the disease association matrix based on the mechanism association matrix and the statistical association matrix;

[0018] Step 4: Obtain the positive and negative ideal solutions for each indicator within the evaluation domain. Combine the weights with the disease correlation matrix to obtain the comprehensive covariance matrix. Use the comprehensive covariance matrix to calculate the standard Mahalanobis distance between the target tunnel and the positive and negative ideal solutions to obtain the comprehensive health score.

[0019] Optionally, creating the evaluation domain includes:

[0020] Based on the basic attribute information, tunnel groups with high similarity are selected to form the evaluation domain, and a unique identifier is assigned to each evaluation domain for unified management by storing the evaluation domain in the domain configuration library.

[0021] Optionally, the weights assigned to each metric data include:

[0022] The indicators in the indicator data matrix are standardized, and the standard deviation of the standardized indicators and the correlation coefficient matrix between the indicators are calculated.

[0023] Based on the standard deviation and the correlation coefficient matrix, the information content is obtained, and the information content is normalized to obtain the objective weight of each indicator.

[0024] Optionally, calculating the standard deviation of the standardized indicators and the correlation coefficient matrix between the indicators includes:

[0025] Calculate the standard deviation of the standardized index:

[0026] ;

[0027] in, This represents the average value of the j-th standardized index. Let m represent the standard deviation of the j-th indicator, and m represent the total number of tunnels within the evaluation domain. This represents the defect index of the j-th standardized treatment of the i-th tunnel;

[0028] Calculate the correlation coefficient matrix between the indicators:

[0029] ;

[0030] in, Let be the correlation coefficient between the j-th indicator and the k-th indicator. This represents the defect index of the k-th standardized treatment of the i-th tunnel. This represents the average value of the k-th standardized indicator.

[0031] Optionally, the amount of information obtained is:

[0032] ;

[0033] in, Let j be the amount of information contained in the j-th indicator. This represents the standard deviation of the j-th indicator. Let be the correlation coefficient between the j-th indicator and the k-th indicator. This represents the total number of disease indicators.

[0034] Optionally, calculating the statistical correlation matrix includes:

[0035] ;

[0036] ;

[0037] in, Let be the statistical correlation coefficient between the i-th indicator and the j-th indicator. To evaluate the number of tunnels within the region, Let i be the standardized value of the i-th index for the k-th tunnel. Let be the average value of the i-th indicator within the domain.

[0038] Optionally, obtaining the disease correlation matrix includes:

[0039] ;

[0040] in, This is a disease correlation matrix. The fusion coefficient is... Let represent the influence intensity of the i-th disease index on the j-th disease index based on mechanistic knowledge. Let be the statistical correlation coefficient between the i-th indicator and the j-th indicator.

[0041] Optionally, obtaining the comprehensive covariance matrix includes:

[0042] Based on the aforementioned weights, calculate the weighted standard deviation vector for each indicator:

[0043] ;

[0044] in, Let be the weighted standard deviation vector of the j-th indicator;

[0045] Using the weighted standard deviation vector, a weighted standard deviation diagonal matrix is ​​constructed. Then, by combining this weighted standard deviation diagonal matrix with the disease correlation matrix, the comprehensive covariance matrix is ​​obtained.

[0046] ;

[0047] in, To synthesize the covariance matrix, It is a diagonal matrix of weighted standard deviations. This is the disease correlation matrix.

[0048] Optionally, calculating the standard Mahalanobis distance between the target tunnel and the positive and negative ideal solutions includes:

[0049] , ;

[0050] in, Let be the standard Mahalanobis distance between the target tunnel and the ideal solution. Let be the standard Mahalanobis distance between the target tunnel and the negative ideal solution. It is the inverse of the covariance matrix. Let be the standardized index vector of the target tunnel. This represents the matrix transpose operation. Let be the positive ideal solution vector. It is the negative ideal solution vector.

[0051] Optionally, the method further includes:

[0052] A sliding window update mechanism is set up. When the preset time window in the sliding window update mechanism reaches the update cycle, new monitoring data within the window period is collected, steps 1-4 are re-executed, and a standard update is determined by comparing the rate of change between the old and new threshold vectors. The rate of change is calculated as follows:

[0053] ;

[0054] in, For the rate of change, , These represent the old and new threshold vectors, respectively. For the i-th component of the new threshold vector, For the i-th component of the old threshold vector, The total number of preset performance levels;

[0055] If the rate of change exceeds the preset stability criterion, the new parameter will be stored as a new version in the domain configuration library.

[0056] A parameter inheritance initialization mechanism is set up. Using the parameter inheritance initialization mechanism, the comprehensive similarity between the target tunnel and each existing evaluation domain in terms of key attribute features is calculated. The evaluation domain with the highest similarity is selected as the parameter source. That is, when the comprehensive similarity exceeds a preset threshold, the entire set of evaluation parameters generated by data driving is inherited from that evaluation domain, and the evaluation parameters are used as the initial evaluation standard for the new tunnel.

[0057] The beneficial effects of this invention are as follows:

[0058] This invention generates adaptive evaluation thresholds for different tunnels through a dynamic domain division mechanism, determines index weights entirely based on monitoring data using the CRITIC objective weighting method, incorporates statistical correlations between defects at the mathematical level into the TOPSIS model based on weighted Mahalanobis distance, and designs a sliding window update mechanism and parameter inheritance strategy to achieve continuous model evolution and rapid adaptation to sparse data. Finally, it constructs a closed-loop intelligent evaluation system with a complete process of "evaluation-calibration-early warning-update".

[0059] This invention establishes individualized evaluation standards for different tunnels through a dynamic domain division mechanism, and uses the CRITIC objective weighting method to determine the index weights entirely based on data-driven methods, eliminating the influence of expert subjectivity and solving the problems of static and rigid evaluation standards and strong subjectivity of weights.

[0060] This invention introduces the TOPSIS comprehensive evaluation method based on weighted Mahalanobis distance, which inherently incorporates the statistical correlation between disease indicators at the mathematical level. At the same time, it combines a disease correlation network based on authoritative literature and intra-domain statistical calibration to achieve quantitative analysis of the coupling relationship of various diseases and systematic risk warning.

[0061] This invention designs a parameter update mechanism and stability criterion based on a sliding window, enabling the evaluation system to continuously optimize as the tunnel state evolves and data accumulates, thus overcoming the model aging problem. Through a parameter inheritance strategy based on similarity, it effectively solves the model initialization problem in data-sparse scenarios, improving the applicability of the system in newly built and special tunnels. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 This is a flowchart of a service life tunnel performance evaluation method based on intra-domain self-calibration according to an embodiment of the present invention;

[0064] Figure 2 This is a flowchart of the TOPSIS scoring process according to an embodiment of the present invention. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0067] Explanation of technical terms:

[0068] In this invention, "domain" specifically refers to the "evaluation domain." It is a set of tunnels selected from all tunnels through a dynamic domain segmentation mechanism that share highly similar key attributes (such as structural type and geological conditions). The core of this concept lies in establishing a fair and reasonable "reference system" or "control group" for performance evaluation, ensuring that the evaluation is conducted among "similar items," thereby achieving cross-domain comparability and individual adaptability of the evaluation results.

[0069] CRITIC Algorithm: CRITIC (Criteria Importance Through IntercriteriaCorrelation) is an objective weighting method. This method calculates weights based on the comparative strength and conflict between evaluation indicators. Comparative strength is measured by the standard deviation of the indicators; a larger standard deviation indicates greater data fluctuation and more information contained. Conflict is measured by the correlation coefficient between indicators; the weaker the correlation, the stronger the independence and the more information contained. In this invention, this method is used to determine the objective weights of each disease indicator entirely based on the monitoring data itself, completely eliminating the influence of expert subjectivity.

[0070] Weighted Mahalanobis Distance: Mahalanobis distance is a distance metric that considers the covariance structure between data features. It measures the distance between a point and a data distribution. Unlike Euclidean distance, it is not affected by dimensions and considers the correlation between features. Weighted Mahalanobis distance further introduces weights for each feature. In this invention, by introducing weighted Mahalanobis distance into the TOPSIS model, the statistical correlation between disease indicators is essentially incorporated, enabling the system to provide a more sensitive and practically accurate risk assessment when multiple highly correlated diseases worsen simultaneously.

[0071] TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) is a commonly used multi-objective decision analysis method. Its core idea is to construct "positive ideal solutions" (optimal solutions) and "negative ideal solutions" (worst solutions) in a multi-dimensional space, and then rank each evaluated object by calculating the distance to these two ideal solutions. In this invention, this method is used to calculate the comprehensive health score of tunnels, and its combination with weighted Mahalanobis distance constitutes the core algorithm for comprehensive evaluation.

[0072] Dynamic domain segmentation: Dynamic domain segmentation refers to the process of dynamically selecting a subset of tunnels with high similarity from a large group of tunnels based on a set of preset homogeneity keys (such as structural type, geological conditions, etc.) to form an evaluation "domain". In this invention, this mechanism establishes a fair and reasonable "reference system" for evaluation, ensuring the comparability of evaluation results between different tunnels, and is the basis for realizing individualized and adaptive evaluation standards.

[0073] Sliding window update mechanism: A sliding window is a data processing technique that defines a fixed-length time window. As time progresses, the window slides forward, discarding old data and incorporating new data. In this invention, this mechanism is used to periodically (e.g., every 12 months) trigger the self-update of the evaluation system, recalculating weights and thresholds using the latest monitoring data. This allows the system to continuously evolve along with the tunnel state, effectively overcoming the model's "aging" problem.

[0074] Tunnel Defect Association Network / Matrix: A network model used to quantify the interrelationships between different tunnel defects (such as cracks, leaks, and deformations), typically represented in matrix form. Nodes in this network represent defects, edges represent the relationships between defects, and edge weights represent the strength of these relationships. In this invention, this network is constructed by integrating mechanistic associations defined in authoritative literature with statistical correlations from within-domain data. It is crucial for transitioning from "isolated evaluation" to "systematic diagnosis," and is used to provide early warning of systemic risks triggered by the chain reaction of defects.

[0075] Conformal prediction: Conformal prediction is a mathematical framework that provides a reliable confidence interval for the output of any prediction model. Its core idea is to quantify the uncertainty of the prediction by measuring the "compatibility" or "similarity" of a new sample (such as a tunnel to be evaluated) with a set of known, clearly labeled historical samples (such as tunnels with assigned ratings). If the new sample is very similar to historical samples, the prediction certainty is high and the confidence interval is narrow; conversely, if the new sample is very special and differs greatly from historical samples, the prediction uncertainty is high and the confidence interval is wide. In this invention, this method is used to generate a confidence interval for a comprehensive health score. This interval essentially answers the question, "How reliable is this score?" When this confidence interval intersects with the boundary threshold of the performance rating, it means that there is uncertainty in the true performance rating of the tunnel, and the system will trigger a "manual review suggestion" prompt.

[0076] like Figure 1 As shown in the figure, this embodiment discloses a service-life tunnel performance evaluation method based on domain self-calibration, including: Step 1, obtaining basic attribute information of the target tunnel, and creating an evaluation domain based on the basic attribute information; Step 2, obtaining monitoring information within the evaluation domain, constructing an index data matrix using the monitoring information, and assigning weights to each index data; Step 3, setting mechanistic association rules, calculating the mechanistic association matrix and statistical association matrix through the mechanistic association rules, and obtaining a disease association matrix based on the mechanistic association matrix and statistical association matrix; Step 4, obtaining the positive ideal solution and negative ideal solution of each index within the evaluation domain, combining the weights with the disease association matrix to obtain a comprehensive covariance matrix, and using the comprehensive covariance matrix to calculate the standard Mahalanobis distance between the target tunnel and the positive and negative ideal solutions to obtain a comprehensive health score.

[0077] Furthermore, creating an evaluation domain includes: selecting tunnel groups with high similarity based on basic attribute information to form an evaluation domain, and assigning a unique identifier to store the evaluation domain in a domain configuration library for unified management.

[0078] Specifically, this is implemented based on the tunnel operation and maintenance database that has been widely established in the field. This database contains static attribute data and dynamic status data of the tunnel. The static attribute data mainly includes tunnel name, route information, usage classification, structural type, surrounding rock grade, hydrological conditions, start and completion time, design data, and service life, etc. The dynamic status data mainly includes data on apparent defects, deep defects, and material deterioration obtained from regular inspections and real-time monitoring. These are data resources that tunnel operating units naturally accumulate in their daily operation and maintenance work.

[0079] The overall technical process begins with dynamic domain-based database construction. The system first retrieves basic attribute information of the target tunnel from the tunnel operation and maintenance database, including its purpose subclass, structural type, surrounding rock grade, hydrological conditions, and service life. Based on these attribute characteristics, the system performs intelligent retrieval and matching in the database, filtering out a group of tunnels with high similarity to form a logical evaluation domain. This process lays the foundation for subsequent individualized evaluations by establishing a reasonable "similar reference system." The system assigns a unique identifier to each newly created evaluation domain and stores the domain's basic definition information in a dedicated domain configuration library. The domain configuration library uses version control, recording historical parameter changes for each evaluation domain, providing data support for continuous system optimization.

[0080] Furthermore, assigning weights to each indicator data includes: standardizing the indicators in the indicator data matrix, calculating the standard deviation of the standardized indicators and the correlation coefficient matrix between the indicators; obtaining the information content based on the standard deviation and the correlation coefficient matrix, normalizing the information content, and obtaining the objective weights of each indicator.

[0081] Specifically, after completing the domain segmentation, the system enters the in-depth domain analysis phase. This phase executes two key computational tasks in parallel: objective weighting analysis and disease correlation analysis. In the objective weighting analysis, the system extracts the latest monitoring data of all tunnels within the evaluation domain from the tunnel operation and maintenance database, constructing an indicator data matrix. Where m represents the total number of tunnels in the evaluation domain, n represents the total number of selected defect indicators, and xij represents the original monitoring value of the j-th defect indicator of the i-th tunnel. The CRITIC algorithm is used for weight calculation. The core steps of this algorithm are:

[0082] First, the raw data is standardized to eliminate the influence of dimensions: ,in This represents the minimum value of the j-th index among all tunnels. This represents the maximum value of the j-th index across all tunnels. This represents the standardized index value, with a range of [0,1].

[0083] Next, the standard deviation of each indicator is calculated to measure the strength of the comparison between the indicators: ,in This represents the average value of the j-th standardized index. This represents the standard deviation of the j-th indicator. The larger the value, the more significant the difference in this indicator between different tunnels.

[0084] Then, the correlation coefficient matrix between the indicators is calculated to measure the conflict between them. , This represents the correlation coefficient between the j-th indicator and the k-th indicator, with a value range of [-1, 1]. The closer a value is to 0, the stronger the independence between the two indicators.

[0085] Based on the above calculation results, the information content of each indicator is calculated as follows: ,in, The standard deviation represents the amount of information contained in the j-th indicator. The larger the value and the weaker its correlation with other indicators, the more information the indicator contains.

[0086] Finally, the information content is normalized to obtain the objective weights of each indicator: , This represents the objective weight of the j-th indicator. This represents the sum of information content from all indicators and is used for weight normalization. It also constructs the overall objective weight vector. .

[0087] This calculation process fully considers the comparative strength and conflict of the indicators, ensuring that the weight allocation is entirely based on data-driven principles and eliminating the influence of subjective human factors.

[0088] In disease correlation analysis, the system first retrieves predefined mechanistic correlation rules from authoritative literature and expert knowledge bases to form an initial qualitative correlation network. Then, it calculates statistical correlation degrees using actual monitoring data within the evaluation domain to quantitatively calibrate the mechanistic rules. Specifically, it calculates the correlation coefficient matrix between disease indicators and weights it with the mechanistic rule weights to ultimately generate a quantitative disease correlation matrix. The specific steps of the disease association analysis module are as follows:

[0089] The mechanism association matrix is ​​initialized by retrieving predefined mechanism association rules from the domain knowledge base and constructing the initial mechanism association matrix. ,in, This represents the influence intensity of the i-th disease index on the j-th disease index based on mechanistic knowledge, with a value range of [0,1], where 0 indicates no direct influence and 1 indicates complete influence.

[0090] Calculate the statistical correlation matrix based on the actual monitoring data within the evaluation domain. ,in The statistical correlation coefficient between the i-th and j-th indicators reflects the strength of the association at the data level, and is calculated using the following formula: m represents the number of tunnels within the evaluation domain. The standardized value of the i-th indicator for the k-th tunnel. This represents the average value of the i-th indicator within the domain.

[0091] The correlation coefficient after weighted fusion of mechanistic and statistical correlations is obtained: And generate the final disease correlation matrix: For the fusion coefficient The value ranges from [0,1] and needs to be dynamically adjusted based on the reliability of domain knowledge. For associations with clear and generally accepted mechanisms, a larger value is taken, indicating greater confidence in the mechanism; for associations with unclear or controversial mechanisms, a smaller value is taken, indicating greater reliance on statistical data.

[0092] Furthermore, obtaining the comprehensive covariance matrix includes: calculating the weighted standard deviation vector of each indicator based on the weights; constructing a weighted standard deviation diagonal matrix using the weighted standard deviation vector; and using the weighted standard deviation diagonal matrix in conjunction with the disease correlation matrix to obtain the comprehensive covariance matrix.

[0093] Specifically, based on the aforementioned analysis results, the system enters the core comprehensive evaluation stage. This stage employs a comprehensive evaluation method based on an improved TOPSIS, which improves upon the introduction of weighted Mahalanobis distance to replace traditional Euclidean distance. The covariance structure of the Mahalanobis distance is constructed by fusing objective weights with the disease correlation matrix, thus retaining the core ideas of TOPSIS while achieving an essential integration of disease correlation and a sensitive perception of systemic risks. Figure 2 As shown. The specific implementation steps are as follows:

[0094] First, determine the ideal solution, and then calculate the optimal value (positive ideal solution) and worst value (negative ideal solution) of each index within the evaluation domain: , ,in and These are the minimum and maximum values ​​of the j-th standardized index within the domain, respectively.

[0095] To construct the integrated covariance matrix of the fusion weights, first calculate the weighted standard deviation vector of each indicator. ,in This is equivalent to scaling the original standard deviation using CRITIC weights, thus amplifying the dispersion of important indicators in the covariance matrix. Next, a weighted standard deviation diagonal matrix is ​​constructed. Calculate the comprehensive covariance matrix Simultaneously, the inverse matrix of the covariance matrix is ​​calculated. .

[0096] Calculate the standard Mahalanobis distance using the combined covariance inverse matrix with fused weights constructed in step 2. Calculate the standard Mahalanobis distance between the target tunnel and the positive and negative ideal solutions: , .

[0097] Calculate the overall health score The score S ranges from [0,1], with a value closer to 1 indicating better health.

[0098] This improved evaluation method can fundamentally incorporate the statistical correlation between disease indicators. When multiple strongly correlated diseases deteriorate simultaneously, the system will give a significantly reduced score, achieving a keen early warning of systemic risks.

[0099] After obtaining the comprehensive score, the system initiates a threshold self-calibration process. This process aggregates the score data of all tunnels within the evaluation domain. As input, an unsupervised clustering algorithm (such as K-means) is used to automatically classify performance levels. The clustering process finds the optimal cluster centers based on the natural distribution characteristics of the scoring data, forming a hierarchical threshold vector. Where k represents the number of preset performance levels (e.g., four levels: healthy, sub-healthy, attention, and deterioration). This represents the threshold for the i-th level. This data-driven threshold determination method ensures that the evaluation criteria match the actual situation of the tunnel group within the current domain.

[0100] Before outputting the evaluation results, the system performs a confidence assessment. A conformal prediction method is used to calculate the confidence interval for the comprehensive score of the target tunnel. Specifically, a conformity metric is constructed by measuring the compatibility between the new sample and the historical sample set. This is applied to the comprehensive score of the target tunnel. Its non-consistency score is calculated as follows: , This represents the predicted score of the target tunnel based on similar tunnel data. This represents an estimate of local variability, reflecting the uncertainty of the prediction. Based on the calculated inconsistent score distribution, a confidence interval [L, U] is determined such that... When the interval intersects with any grading threshold, the system automatically triggers a "manual review suggestion" prompt, effectively preventing misjudgment of the critical state and improving the reliability of the evaluation results.

[0101] Furthermore, the method also includes: setting a sliding window update mechanism, when the preset time window in the sliding window update mechanism reaches the update cycle, collecting new monitoring data within the window period, re-executing steps 1-4, and determining whether a standard update is needed by comparing the change rate of the new and old threshold vectors. If the change rate exceeds the preset stability criterion, the new parameters are stored as a new version in the domain configuration library; setting a parameter inheritance initialization mechanism, using the parameter inheritance initialization mechanism to calculate the comprehensive similarity between the target tunnel and each existing evaluation domain in terms of key attribute features, selecting the evaluation domain with the highest similarity as the parameter source, that is, when the comprehensive similarity exceeds the preset threshold, inheriting the complete set of data-driven generated evaluation parameters from that evaluation domain, and using the evaluation parameters as the initial evaluation standard for the new tunnel.

[0102] Specifically, to maintain the system's timeliness, this invention designs a sliding window update mechanism. The system sets a fixed time window (e.g., 12 months). When the update cycle is reached, it automatically collects new monitoring data from within the window period and re-executes the entire calculation process from objective weighting to threshold self-calibration. The system determines whether a standard update is needed by comparing the rate of change between the old and new threshold vectors. The rate of change is calculated as follows: ,in , These represent the old and new threshold vectors, respectively. If the rate of change exceeds the preset stability criterion (e.g., 5%), the new parameter will be stored as the new version in the domain configuration library.

[0103] To address the technical challenge of establishing effective evaluation models for newly constructed tunnels or tunnels with scarce monitoring data through dynamic domain segmentation, this invention designs a parameter inheritance initialization mechanism. This mechanism calculates the comprehensive similarity between the target tunnel and existing evaluation domains in the system across key attribute characteristics (including usage subclass, structural type, surrounding rock grade, hydrological conditions, and service life), selecting the evaluation domain with the highest similarity as the parameter source. When the similarity exceeds a preset threshold, the system automatically inherits a complete set of data-driven evaluation parameters from that source domain—including objective weight vectors, disease correlation matrices, and grading threshold vectors—and uses these parameters as the initial evaluation criteria for the new tunnel. This ensures that the tunnel obtains a scientifically sound performance evaluation capability from the initial construction phase, achieving seamless integration of operation and maintenance management throughout its entire lifecycle.

[0104] Corresponding to the above method, this invention provides a complete system implementation plan. The system consists of a data acquisition module, a dynamic domain segmentation module, an intra-domain analysis module, a comprehensive scoring module, a threshold self-calibration module, a confidence assessment module, a rolling update module, a parameter inheritance module, a result output and early warning module, and a domain configuration library. All modules are connected through standardized data interfaces, forming a complete "evaluation-calibration-early warning-update" closed-loop system.

[0105] Based on the core concept of this invention, those skilled in the art can conceive of various alternative solutions. Grouping based on features can be extended to attributes such as design speed and burial depth; unsupervised clustering can employ Gaussian mixture models or the DBSCAN algorithm; confidence assessment can use the Bootstrap method; and objective weighting can employ the entropy weight method, etc. These alternative solutions all achieve the inventive objective of this invention and fall within the scope of protection of this invention.

[0106] This invention overcomes the limitations of traditional static evaluation standards by introducing dynamic domain division and threshold self-calibration mechanisms, achieving a fundamental shift in evaluation standards from "uniform and fixed" to "individually adaptive." Employing the TOPSIS model based on CRITIC objective weighting and weighted Mahalanobis distance, it not only completely eliminates the influence of subjective human factors on weight setting but also fundamentally incorporates the statistical correlation between disease indicators at the mathematical level, enabling the system to keenly identify and warn of cascading risks.

[0107] This invention possesses continuous evolution and sparse data adaptability. Through a sliding window update mechanism, model parameters can be dynamically optimized as the tunnel state evolves and data accumulates, effectively overcoming the "aging" problem of traditional models. For newly built or data-scarce tunnels, a similarity-based parameter inheritance strategy enables rapid initialization, maintaining high reliability even in scenarios with insufficient data.

[0108] The technical system of this invention forms a complete closed loop of "evaluation-calibration-early warning-update", which can significantly reduce operation and maintenance costs while improving the timeliness of early warning, and provides scientific and reliable decision support for intelligent evaluation of tunnel performance during service.

[0109] This embodiment discloses a service life tunnel performance evaluation method based on intra-domain self-calibration, including:

[0110] 1. System initialization and data preprocessing:

[0111] The system data acquisition module extracts static attributes and dynamic monitoring data of the target tunnel from the operation and maintenance database. Static attributes: Purpose subclass = "Highway double-track tunnel", Structural type = "Composite lining", Surrounding rock grade = "Class IV", Hydrological conditions = "Medium water abundance", Service life = 8 years. Dynamic data preprocessing: The system extracts features from the original defect records (such as multiple cracks, seepage points, deformation indicators) and synthesizes comprehensive evaluation indicators. Finally, a standardized indicator vector containing 10 feature indicators is formed for subsequent calculations.

[0112] 2. Dynamic domain partitioning:

[0113] The dynamic domain segmentation module performs similarity matching in the global tunnel database based on the attribute characteristics of the target tunnel, filtering out a group of tunnels with high similarity. Matching conditions include the same structural type, surrounding rock grade, hydrological conditions, and service life within ±3 years. Ultimately, 42 tunnels were successfully matched, forming an evaluation domain. The system assigns a unique identifier to each domain and stores its basic definition information in the domain configuration database.

[0114] 3. In-depth domain analysis:

[0115] After entering the in-depth analysis phase within the domain, the system executes two core computational tasks in parallel. In the objective weighting analysis, the system extracts indicator data from all tunnels within the evaluation domain to construct a standardized data matrix. Where m=42 represents the number of tunnels and n=10 represents the number of indicators. The CRITIC algorithm is used for calculation: first, the standard deviation of each indicator is calculated. As a measure of comparative strength, such as the standard deviation of the maximum crack width. Then calculate the correlation coefficient matrix between the indicators. For example, the correlation coefficient between the maximum crack width and the total leakage area. Next, the information content of each indicator is calculated. ,For example Finally, the objective weight vector is obtained through normalization. ,in This process is entirely data-driven, completely eliminating the influence of subjective human factors.

[0116] The concurrent disease association analysis first retrieves a predefined mechanism association matrix from the domain knowledge base. ,in, The intensity of the mechanism-related influence of cracks on leakage is represented; then, a statistical correlation matrix is ​​calculated based on actual monitoring data within the evaluation domain. ,in This represents the statistical impact of cracks on leakage; finally, a weighted fusion formula is used. Generate the final disease correlation matrix R, and take... hour, This process effectively integrates mechanistic knowledge with statistical data.

[0117] 4. Overall Evaluation (Improved TOPSIS and Weighted Mahalanobis Distance):

[0118] The comprehensive scoring module uses an improved TOPSIS method for calculation. First, the ideal solution is determined. and negative ideal solution , representing the optimal and worst values ​​of each indicator, respectively. Then, a comprehensive covariance matrix integrating weights and correlations is constructed, first by calculating the weighted standard deviation vector of each indicator. ,in Next, a weighted standard deviation diagonal matrix is ​​constructed. Calculate the comprehensive covariance matrix Simultaneously, the inverse matrix of the covariance matrix is ​​calculated. Next, the weighted Mahalanobis distance between the target tunnel and the positive and negative ideal solutions is calculated: the distance to the positive ideal solution. Distance to the negative ideal solution Finally, a comprehensive health score is calculated. This improved method inherently incorporates the statistical correlation between disease indicators, enabling more sensitive risk warnings when multiple strongly correlated diseases worsen simultaneously.

[0119] 5. Threshold self-calibration:

[0120] After obtaining the comprehensive score, the threshold self-calibration module inputs all 42 tunnel scores within the evaluation domain into the K-means clustering algorithm, automatically classifying performance levels and generating a grading threshold vector T=[0.90,0.75,0.50,0.20]. Based on this threshold, the score of 0.74 for this highway tunnel is classified as "sub-healthy" (0.50≤S<0.75). Subsequently, the confidence assessment module uses a conformal prediction method to calculate the confidence interval of this score as [0.70,0.78]. Since this interval intersects with the grading threshold of 0.75, the system automatically triggers a "manual review suggestion" prompt, effectively preventing misjudgment of the critical state.

[0121] 6. System Updates and Parameter Inheritance:

[0122] To maintain system timeliness, a sliding window update mechanism was designed. Every 12 months, the system automatically collects new monitoring data and re-executes the entire calculation process from objective weighting to threshold self-calibration. The decision to update parameters is made by comparing the rate of change between the old and new threshold vectors; the rate of change is calculated as follows: ,in , These represent the old and new threshold vectors, respectively. If the rate of change exceeds the preset stability criterion (e.g., 5%), the new parameters are stored as the new version in the domain configuration library. For newly built or data-sparse tunnels, the parameter inheritance module calculates the attribute similarity between the tunnel and the existing evaluation domain. When the similarity exceeds the 80% threshold, it automatically inherits the complete set of evaluation parameters (W, R, T) from the source domain, achieving rapid initialization and seamless integration into the evaluation system.

[0123] 7. Output and Early Warning Module:

[0124] The results output and early warning module generates a complete evaluation report including tunnel name, comprehensive score, performance level, confidence interval, core risk indicators, and related risk warnings. This report is then pushed to relevant personnel through the operation and maintenance platform, forming a complete closed loop from evaluation to early warning. The entire implementation process demonstrates the core advantages of this invention, which achieves accurate and intelligent evaluation of the performance of in-service tunnels through innovative mechanisms such as dynamic domain segmentation, objective weighting, correlation analysis, and adaptive calibration.

[0125] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for evaluating the performance of a tunnel in service based on in-domain self-calibration, characterized in that, The method comprises the following steps: Step 1, obtaining basic attribute information of a target tunnel, and creating an evaluation domain according to the basic attribute information; Step 2, obtaining monitoring information in the evaluation domain, constructing an index data matrix by using the monitoring information, and assigning a weight to each index data; Step 3, setting a mechanism correlation rule, calculating a mechanism correlation matrix and a statistical correlation matrix by using the mechanism correlation rule, and obtaining a disease correlation matrix based on the mechanism correlation matrix and the statistical correlation matrix; Step 4, obtaining a positive ideal solution and a negative ideal solution of each index in the evaluation domain, obtaining a comprehensive covariance matrix by combining the weight and the disease correlation matrix, and calculating a standard Mahalanobis distance between the target tunnel and the positive ideal solution and the negative ideal solution by using the comprehensive covariance matrix to obtain a comprehensive health score.

2. The in-domain self-calibration based service life tunnel performance evaluation method according to claim 1, characterized in that, The creation of the evaluation domain comprises the following steps: According to the basic attribute information, a tunnel group with high similarity is screened out to form the evaluation domain, and a unique identifier is assigned to store the evaluation domain in a domain configuration library for unified management.

3. The in-domain self-calibration based service life tunnel performance evaluation method according to claim 1, characterized in that, The weight assignment to each index data comprises the following steps: Standardizing the indexes in the index data matrix, calculating the standard deviation of the standardized indexes and the correlation coefficient matrix between the indexes; Based on the standard deviation and the correlation coefficient matrix, an information amount is obtained, and the information amount is normalized to obtain the objective weight of each index.

4. The in-domain self-calibration based service life tunnel performance evaluation method according to claim 3, characterized in that, The calculation of the standard deviation of the standardized indexes and the correlation coefficient matrix between the indexes comprises the following steps: The standard deviation of the standardized indexes is calculated: ； wherein, represents the average value of the jth standardized index value, represents the standard deviation of the jth index, and m represents the total number of tunnels within the evaluation field, represents the jth standardized disease index of the ith tunnel; The correlation coefficient matrix between the indexes is calculated: ； wherein, is the correlation coefficient of the jth index and the kth index, represents the kth normalized disease index of the ith tunnel, represents the average value of the kth normalized index.

5. The in-domain self-calibration based service life tunnel performance evaluation method according to claim 3, characterized in that, The information amount is obtained: ； wherein, the amount of information contained in the jth index, denotes the standard deviation of the jth index, is the correlation coefficient of the jth index and the kth index, is the total number of disease indices.

6. The in-domain self-calibration based service life tunnel performance evaluation method according to claim 1, characterized in that, The calculation of the statistical correlation matrix comprises the following steps: ； ； wherein, is the statistical correlation coefficient of the i-th indicator and the j-th indicator, is the number of tunnels in the evaluation domain, is the normalized value of the i-th indicator of the k-th tunnel, is the average value of the i-th indicator in the domain.

7. The in-domain self-calibration based service life tunnel performance evaluation method according to claim 1, characterized in that, The disease correlation matrix is obtained: ； wherein, is a disease correlation matrix, is a fusion coefficient, is the influence strength of the i-th disease indicator on the j-th disease indicator based on mechanistic knowledge, is the statistical correlation coefficient of the i-th indicator and the j-th indicator.

8. The in-domain self-calibration based service life tunnel performance evaluation method according to claim 1, characterized in that, The comprehensive covariance matrix is obtained, which comprises the following steps: Based on the weight, a weighted standard deviation vector of each index is calculated: ； wherein, is the weighted standard deviation vector for the jth index; A weighted standard deviation diagonal matrix is constructed by using the weighted standard deviation vector, and the comprehensive covariance matrix is obtained by using the weighted standard deviation diagonal matrix in combination with the disease correlation matrix: ； wherein, is a combined covariance matrix, is a weighted standard deviation diagonal matrix, is a disease association matrix.

9. The in-domain self-calibration based service life tunnel performance evaluation method according to claim 1, characterized in that, The calculation of the standard Mahalanobis distance between the target tunnel and the positive ideal solution and the negative ideal solution comprises the following steps: ， ； wherein, is the standard Mahalanobis distance between the target tunnel and the positive ideal solution, is the standard Mahalanobis distance between the target tunnel and the negative ideal solution, is the inverse of the covariance matrix, is the normalized indicator vector of the target tunnel, denotes the transpose operation of a matrix, is the positive ideal solution vector, is the negative ideal solution vector.

10. The in-domain self-calibration based service life tunnel performance evaluation method according to claim 1, characterized in that, The method further comprises the following steps: A sliding window update mechanism is set, when a preset time window in the sliding window update mechanism reaches an update period, new monitoring data in the window period is collected, steps 1-4 are re-executed, and whether standard updating is needed is determined by comparing the change rate of new and old threshold vectors: ； wherein, is a rate of change, , respectively represent a new and an old threshold vector, is an i-th component of the new threshold vector, is an i-th component of the old threshold vector, is a total number of preset performance levels; If the change rate exceeds a preset stability criterion, the new parameters are stored in the domain configuration library as a new version; A parameter inheritance initialization mechanism is set, the comprehensive similarity of a target tunnel and each existing evaluation domain in key attribute characteristics is calculated by using the parameter inheritance initialization mechanism, and the evaluation domain with the highest similarity is selected as the parameter source, that is, when the comprehensive similarity exceeds a preset threshold, a complete set of evaluation parameters generated by data driving is inherited from the evaluation domain, and the evaluation parameters are used as the initialization evaluation standard of a newly built tunnel.

Images