Intelligent evaluation method for tunnel service performance

Abstract

The invention discloses an intelligent evaluation method for tunnel service performance, which relates to the technical field of tunnel engineering, and comprises the following steps: selecting key tunnel disease types, designing service performance evaluation indexes, and collecting sample data required by tunnel service performance evaluation to form an evaluation index data set; and organizing tunnel engineering field experts, scoring partial tunnel samples according to the tunnel service performance evaluation indexes, and obtaining tunnel service performance scores. The multiple base machine learning models are integrated through the Bayesian model averaging method to form the tunnel service performance evaluation super-model, the advantages of all the base machine learning models are fully utilized, meanwhile, the insufficiency is compensated, and therefore the accuracy and stability of scoring are remarkably improved, and compared with a single model, the efficiency is improved. The tunnel service performance evaluation hyper-model can more comprehensively capture a complex relationship between tunnel service performance evaluation indexes and scores, reduces scoring errors caused by model deviation, and provides a more reliable result for tunnel health state evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel engineering, and in particular to an intelligent evaluation method for tunnel service performance. Background Art

[0002] Tunnel service performance evaluation is a key technology to ensure the performance of tunnels throughout their life cycle. At present, the commonly used tunnel service performance evaluation methods mainly include specification-based methods, mechanical model-based methods, mathematical model-based methods, and expert evaluation methods. Among them, the specification-based method classifies the tunnel service status according to the thresholds in the relevant specifications, and the evaluation results are not precise enough; the mechanical model-based method requires first establishing a tunnel mechanical model, and then evaluating the tunnel service performance based on the calculation results of the mechanical model. The implementation process is relatively cumbersome and difficult; the mathematical model-based method often uses fuzzy mathematical methods such as the hierarchical analysis method to conduct fuzzy comprehensive evaluation of tunnel service performance, which has high requirements for data collection and is often difficult to meet in actual tunnel projects; the expert scoring method invites experts in the field of tunnel engineering to score the tunnel service performance based on the tunnel service performance evaluation index values. It has the advantages of simplicity and practicality, and has therefore been widely used. However, the expert scoring method is less efficient when applied to large-scale tunnel service performance evaluation.

[0003] To improve the efficiency of the expert scoring method for tunnel service performance, some scholars have combined machine learning models with expert scoring methods and proposed a machine learning-based expert evaluation method for tunnel service performance. This method first invites experts to score a small number of tunnel samples to form a tunnel service performance evaluation dataset. Then, a machine learning model is used to learn the mapping between tunnel service performance evaluation index values ​​and tunnel service performance scores on the tunnel service performance evaluation dataset. Finally, the machine learning model is used instead of experts to score other similar tunnel samples, thereby improving the efficiency of the expert evaluation method for tunnel service performance when used for large-scale tunnel service performance evaluation.

[0004] However, the existing machine learning-based tunnel service performance expert scoring method generally uses a single machine learning model to learn the mapping between tunnel service performance evaluation index values ​​and tunnel service performance scores. The model scoring accuracy and stability are low. Therefore, there is an urgent need to improve the scoring accuracy and stability of the machine learning-based tunnel service performance expert scoring method. Summary of the Invention

[0005] The present invention aims to provide an intelligent evaluation method for tunnel service performance to solve the problems raised in the above background technology.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: An intelligent evaluation method for tunnel service performance includes the following steps: Step S1: Select key tunnel disease types, design service performance evaluation indicators, and collect sample data required for tunnel service performance evaluation to form an evaluation indicator data set; Step S2: Organize experts in the field of tunnel engineering to score some tunnel samples based on the tunnel service performance evaluation indicators, obtain tunnel service performance scores, and construct a tunnel service performance evaluation dataset; Step S3: training multiple base machine learning models to learn the mapping between tunnel service performance evaluation index values ​​and tunnel service performance scores; Step S4: using the Bayesian model averaging method to integrate multiple base machine learning models to construct a tunnel service performance evaluation super model; Step S5: Using the root mean square error, mean absolute error, and mean absolute percentage error as performance indicators, the tunnel service performance evaluation super model is tested to test the scoring stability of the model; Step S6: quantify the uncertainty of the tunnel service performance score. If the tunnel service performance evaluation super model passes all tests, it is put into large-scale tunnel service performance evaluation.

[0007] A further improvement of the technical solution of the present invention is that in step S1, the process of forming the evaluation index data set is: Based on the structural characteristics of tunnels, the causes and impacts of common diseases, a comprehensive analysis is conducted to determine the degree of harm that different diseases pose to tunnel safety, durability, and usability. This analysis identifies the types of tunnel diseases that have a significant impact on tunnel performance, including settlement, lateral movement, leakage, cracking, and concrete spalling. This analysis comprehensively reflects the health of tunnels during their service life. Based on the selected key disease types, corresponding tunnel service performance evaluation indicators are designed. These indicators include average relative settlement, average differential settlement, average convergence rate, total leakage area, total crack length, and total concrete spalling area, which can reflect the tunnel service performance from different perspectives. According to the designed tunnel service performance evaluation indicators, tunnel sample data with evaluation indicator values ​​are collected. The data sources include tunnel monitoring systems, engineering inspection reports and historical maintenance records to ensure the integrity and accuracy of the data and form an evaluation indicator data set.

[0008] A further improvement of the technical solution of the present invention is that in step S2, the process of constructing the tunnel service performance evaluation data set is: Organize experts from various fields of tunnel engineering, covering structural design, construction, monitoring, maintenance, etc., and formulate quantitative scoring rules based on the designed tunnel service performance evaluation indicators, clarify the scoring level division and corresponding value range, and ensure that the scoring standards are scientific, unified and operational. The scoring levels include excellent, good, qualified and unqualified; Provide some sample data to the experts, who will then perform a comprehensive score on each sample based on the scoring criteria and record the scoring basis. After the initial evaluation, organize an expert panel to cross-validate the scoring results. For samples with significant discrepancies, adjust the scores based on the actual service environment of the tunnel to ensure that the scoring results objectively reflect the health status of the tunnel. The final expert scoring results were summarized, and the average of all expert scores was defined as the Tunnel Serviceability Rating (TSR) of the sample data. The scores were then associated with the original indicator data to construct a structured tunnel serviceability evaluation dataset, which included the tunnel ID, indicator values, and comprehensive scores, and was annotated with the data source and scoring time.

[0009] A further improvement of the technical solution of the present invention is that the step S3 specifically includes: The tunnel service performance evaluation dataset was cleaned, and samples with more than 30% missing values ​​were removed. The cleaned tunnel service performance evaluation dataset was then divided into a training set and a test set. The training set was used for model training, and the test set was used for model performance verification. The data was divided proportionally, with 70% used as the training set and 30% as the test set, to ensure a reasonable data distribution. A variety of machine learning architectures, including partial least squares regression, support vector machine, and nearest neighbor regression, were selected. A base machine learning model was trained on the training set to learn the mapping between tunnel service performance evaluation indicators and tunnel service performance ratings (TSRs). The input of the base machine learning model was the service performance evaluation indicator, and the output was the tunnel service performance rating (TSR). Five-fold cross-validation was used to evaluate the model's generalization ability, and grid search was used to optimize hyperparameters and improve model performance. It should be noted that the training of the base machine learning model was implemented using the Python scikit-learn package. The trained base machine learning model is verified on the test set to evaluate the scoring accuracy and stability of the model. The model is adjusted and optimized based on the verification results to ensure that the model has good generalization ability in practical applications.

[0010] A further improvement of the technical solution of the present invention is that in step S4, the process of constructing the tunnel service performance evaluation super model is as follows: The weights of each base machine learning model are inferred using the Bayesian model averaging method. The posterior distribution of the weights is calculated using Bayesian evidence and likelihood functions, and the Metropolis-Hastings algorithm is used for sampling, so that the weights reflect the performance and uncertainty of each base machine learning model on the data. Based on the posterior distribution of the weights of the base machine learning models obtained by sampling, the advantages of multiple base models are integrated into a unified model to construct a tunnel service performance evaluation super model. The output of the super model is the weighted average of the outputs of each base machine learning model. The weight is determined by the Bayesian model averaging method, which improves the overall scoring accuracy and stability.

[0011] A further improvement of the technical solution of the present invention is that the expression of the posterior distribution of the weights of the base machine learning model is as follows: ; Where: is the base machine learning model weight vector; is the prior distribution of the weight vector of the base machine learning model; is the likelihood function, reflecting the given The probability of observing the tunnel service performance evaluation training set when ; is the Bayesian evidence, which is constant given the training set; It is the posterior distribution of the weight vector of the fused weight prior distribution and the observed data; The Metropolis-Hastings algorithm is used to solve the posterior distribution of the weight vector of the fusion weight prior distribution and the observation data posterior basis machine learning model, and the input is determined to be the tunnel service performance evaluation training set , initial model weight vector , probability transfer function , likelihood function and number of samples , the output is as follows Sample from the posterior distribution ; The initial model weight vector is taken as uniform distribution, and the probability transfer function is taken as Gaussian probability density function. The expression of the probability transfer function is as follows: ; Where: yes The dimensionality of , i.e. the number of base machine learning models; is the covariance function of the Gaussian probability density function, which is taken as the identity matrix; is the current state; The expression of the likelihood function is as follows: ; Where: is the tunnel sample given by each basic machine learning model of value, is the number of training samples, The base machine learning model weight vector; The expression of the tunnel service performance evaluation super model is as follows: ; Where: is the tunnel service performance evaluation index value of a certain tunnel sample; It is The weights of the base machine learning models; It is The scoring of tunnel samples by a basic machine learning model value.

[0012] A further improvement to the technical solution of the present invention is that, in step S5, the process of model testing the tunnel service performance evaluation super model is as follows: the test set is input into the constructed tunnel service performance evaluation super model, replacing the expert scoring performance, generating a continuous prediction result, and calculating performance indicators including the root mean square error (RMSE, reflecting the degree of deviation between the predicted value and the true value), the mean absolute error (MAE, measuring the average level of absolute error), and the mean absolute percentage error (MAPE, evaluating the relative error ratio); Draw a distribution histogram of the prediction error to observe the central tendency, dispersion, and outlier distribution of the error, and determine whether the model has systematic deviations. Further calculate the standard deviation of the performance indicators and quantify the volatility of the model on different test samples. If the performance indicators remain stable under multiple random test set partitions, that is, the standard deviation is less than 10% of the mean, then the tunnel service performance evaluation super model is verified to have reliable scoring stability. Otherwise, the model structure or data preprocessing strategy needs to be adjusted.

[0013] A further improvement of the technical solution of the present invention is that the calculation expression of the root mean square error is as follows: ; The calculation expression of the mean absolute error is as follows: ; The calculation expression of the mean absolute percentage error is as follows: ; Where: is the root mean square error; is the mean absolute error; is the mean absolute percentage error; is the number of test samples; It is the first TSR values ​​of tunnel samples; The model gives the TSR values ​​of tunnel samples.

[0014] A further improvement of the technical solution of the present invention is that the step S6 specifically includes: The uncertainty of tunnel service performance scores is quantified by using the probabilistic characteristics of the tunnel service performance evaluation supermodel. The confidence interval of the scores is obtained by calculating the probability distribution of the scores, thereby evaluating the reliability of the scores. A comprehensive test of the tunnel service performance evaluation supermodel was conducted, including the scoring accuracy, stability, and rationality of the uncertainty quantification results, to ensure that the model performs stably on different test sets and that the uncertainty quantification results meet actual engineering needs. The verified tunnel service performance evaluation supermodel will then be applied to large-scale tunnel service performance evaluation tasks.

[0015] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art: 1. The present invention provides an intelligent evaluation method for tunnel service performance. It integrates multiple base machine learning models through the Bayesian model averaging method to form a tunnel service performance evaluation super model. It fully utilizes the advantages of each base machine learning model while compensating for its deficiencies, thereby significantly improving the accuracy and stability of the score. Compared with a single model, the tunnel service performance evaluation super model can more comprehensively capture the complex relationship between tunnel service performance evaluation indicators and scores, reduce scoring errors caused by model bias, and provide more reliable results for tunnel health status assessment.

[0016] 2. The present invention provides an intelligent evaluation method for tunnel service performance. It utilizes the probabilistic characteristics of a tunnel service performance evaluation supermodel to quantify the uncertainty of tunnel service performance scores. By calculating the probability distribution and confidence interval of the scoring results, the reliability of the scores can be evaluated. Uncertainty quantification provides tunnel managers with richer information, helping them better understand the credibility of the scoring results, thereby enabling more scientific decisions to be made when formulating maintenance and repair plans, and avoiding inappropriate decisions caused by overconfidence or underestimation of risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0018] Figure 1 This is a schematic diagram of the workflow of an intelligent evaluation method for tunnel service performance according to the present invention; Figure 2 Schematic diagram of the super model and expert scoring values ​​on the tunnel service performance evaluation test set of the present invention; Figure 3 Schematic diagram of the mean performance indicators of the super model and the base machine learning model of the present invention on the tunnel service performance evaluation test set; Figure 4 Schematic diagram of the ratio of expert scores to model scores on the tunnel service performance evaluation test set of the present invention; Figure 5 Figure 3 is a schematic diagram of the uncertainty quantification of the service performance scoring results of the super-model tunnel of the present invention (the five horizontal lines in each box plot in the figure represent the minimum value, first quantile, median, third quantile, and maximum value of TSR from bottom to top, respectively). DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] Example 1, as Figures 1 to 5 As shown, the present invention provides an intelligent evaluation method for tunnel service performance, comprising the following steps: Step S1: Select key tunnel disease types, design service performance evaluation indicators, and collect sample data required for tunnel service performance evaluation to form an evaluation indicator data set. Based on the tunnel structure characteristics, common disease causes and impacts, comprehensively analyze the degree of harm of different diseases to tunnel safety, durability and use functions, and screen out tunnel disease types that have a significant impact on tunnel service performance, including settlement, lateral displacement, leakage, cracking and concrete spalling, which can fully reflect the health status of the tunnel during service. Among them, settlement refers to the vertical displacement of the tunnel structure, which is caused by uneven stratum settlement, improper construction or long-term load. Settlement can easily lead to deformation of the tunnel structure, affecting the geometry and structural stability of the tunnel. In severe cases, it may cause tunnel collapse. Settlement has a great impact on the safety and durability of tunnels, especially in areas with soft soil foundations or complex geological conditions; lateral displacement refers to the horizontal displacement of the tunnel structure, which is caused by lateral pressure of the stratum, disturbance during construction or long-term load. Lateral displacement can easily lead to deformation of the tunnel structure, affecting the geometric shape and structural stability of the tunnel. Lateral displacement has a great impact on the safety and durability of tunnels, especially in areas with complex geological conditions; water leakage refers to water seepage or leakage inside the tunnel, which is caused by damage to the waterproof layer, construction defects or long-term use. Water leakage will cause the internal structure of the tunnel to become damp, accelerate the corrosion of concrete, reduce the durability of the tunnel, and affect the durability and use of the tunnel. The function is greatly affected, which can easily lead to damage to the equipment inside the tunnel and structural corrosion; cracking refers to cracks on the surface of the tunnel structure, which is caused by construction defects, ground deformation or long-term load. Cracks can easily lead to reduced strength of the tunnel structure, affecting the safety and durability of the tunnel. Cracking has a greater impact on the safety and durability of the tunnel, especially in high-stress areas; concrete spalling refers to the spalling phenomenon on the surface of the tunnel concrete, which is caused by concrete aging, corrosion or construction defects. Spalling will lead to reduced strength of the tunnel structure, affecting the safety and durability of the tunnel. Concrete spalling has a greater impact on the safety and durability of the tunnel, especially in high humidity and corrosive environments. Based on the selected key disease types, the corresponding tunnel design Service performance evaluation indicators, including average relative settlement, average differential settlement, average convergence rate, total leakage area, total crack length and total concrete spalling area, which can reflect the service performance of the tunnel from different angles. The average relative settlement reflects the settlement difference in different parts of the tunnel, helping to evaluate the overall stability of the tunnel; the average differential settlement further quantifies the unevenness of settlement in different parts of the tunnel, helping to identify potential structural problems; the average convergence rate reflects the degree of lateral deformation of the tunnel, helping to evaluate the lateral stability of the tunnel; the total leakage area quantifies the severity of leakage, helping to evaluate the waterproof performance of the tunnel; the total crack length quantifies the distribution of cracks, helping to evaluate the structural integrity of the tunnel;The total concrete spalling area quantifies the severity of spalling and helps assess the structural integrity of the tunnel. Based on the designed tunnel service performance evaluation indicators, tunnel sample data with evaluation indicator values ​​is collected. Data sources include tunnel monitoring systems, engineering inspection reports, and historical maintenance records to ensure data integrity and accuracy, forming an evaluation indicator dataset. The evaluation index data set can be recorded as: ; Where: It is Tunnel samples; It is The first tunnel sample Evaluation index value; is the number of evaluation indicators; is the number of tunnel samples; Step S2: Organize experts in the field of tunnel engineering to score some tunnel samples based on the tunnel service performance evaluation indicators, obtain tunnel service performance scores, and construct a tunnel service performance evaluation data set. Organize experts in various fields of tunnel engineering, covering structural design, construction, monitoring, maintenance and other directions, and formulate quantitative scoring rules based on the designed tunnel service performance evaluation indicators, clarify the scoring level division and corresponding numerical range, and ensure that the scoring standards are scientific, unified and operational. Among them, the scoring levels include excellent, good, qualified and unqualified. Excellent means that the tunnel service performance is very good, all indicators are within the normal range, and there are no obvious defects; good means that the tunnel service performance is good, some indicators have slight deviations, but It does not affect the overall use function; qualified means that the tunnel service performance basically meets the requirements, some indicators have obvious deviations, but there are no major safety hazards; unqualified means that the tunnel service performance is poor, multiple indicators seriously exceed the standards, and there are major safety hazards. Some sample data are provided to experts, and each sample is comprehensively scored according to the scoring criteria, and the scoring basis is recorded. After completing the initial evaluation, an expert group is organized to cross-validate the scoring results. For samples with large differences, the scores are adjusted in combination with the actual service environment of the tunnel to ensure that the scoring results objectively reflect the health status of the tunnel. The final scoring results of the experts are summarized, and the average of all expert scores is defined as the tunnel serviceability rating (TSR) of the sample data. The score is then associated with the original indicator data to construct a structured tunnel service performance evaluation dataset, which includes the tunnel ID, each indicator value and the comprehensive score, and the data source and scoring time are marked; The tunnel service performance evaluation dataset can be written as: ; Where: It is TSR values ​​of tunnel samples rated by experts; for the tunnel samples rated by the experts; Step S3, train multiple base machine learning models to learn the mapping between tunnel service performance evaluation index values ​​and tunnel service performance scores, clean the tunnel service performance evaluation data set, remove samples with missing values ​​exceeding 30%, and divide the cleaned tunnel service performance evaluation data set into a training set and a test set, wherein the training set is used for model training and the test set is used for model performance verification. A proportional division is adopted, 70% as the training set and 30% as the test set to ensure a reasonable data distribution, select a variety of machine learning architectures including partial least squares regression, support vector machine and nearest neighbor regression, train the base machine learning model on the training set, and learn the tunnel service performance evaluation. The mapping between the value index value and the tunnel service performance rating (TSR) is performed. The input of the base machine learning model is the service performance evaluation index, and the output is the tunnel service performance rating (TSR). 5-fold cross-validation is used to evaluate the generalization ability of the model, and grid search is used to optimize the hyperparameters to improve the model performance. It should be noted that the training of the base machine learning model is implemented using the Python scikit-learn package. The trained base machine learning model is verified on the test set to evaluate the scoring accuracy and stability of the model. The model is then adjusted and optimized based on the verification results to ensure that the model has good generalization ability in practical applications. No. The learning process of a basic machine learning model is expressed as: ; Where: It is The mapping between the tunnel service performance evaluation index values ​​learned by the individual machine learning models and the tunnel service performance score values; Step S4: Using the Bayesian model averaging method to integrate multiple base machine learning models, construct a tunnel service performance evaluation super model, and use the Bayesian model averaging method to infer the weights of each base machine learning model, wherein the posterior distribution of the weights is calculated using Bayesian evidence and likelihood function, and the Metropolis-Hastings algorithm is used for sampling so that the weights reflect the performance and uncertainty of each base machine learning model on the data. Based on the posterior distribution of the base machine learning model weights obtained by sampling, the advantages of multiple base models are integrated into a unified model to construct a tunnel service performance evaluation super model. The output of the super model is the weighted average of the outputs of each base machine learning model, and the weights are determined by the Bayesian model averaging method, thereby improving the overall scoring accuracy and stability. The expression of the posterior distribution of the weights of the basic machine learning model is as follows: ; Where: is the base machine learning model weight vector; is the prior distribution of the weight vector of the base machine learning model; is the likelihood function, reflecting the given The probability of observing the tunnel service performance evaluation training set when ; is the Bayesian evidence, which is constant given the training set; It is the posterior distribution of the weight vector of the fused weight prior distribution and the observed data; The Metropolis-Hastings (MH) algorithm is used to solve the posterior distribution of the weight vector of the fusion weight prior distribution and the observation data posterior basis machine learning model, and the input is determined to be the tunnel service performance evaluation training set. , initial model weight vector , probability transfer function , likelihood function and number of samples , the output is as follows Sample from the posterior distribution , specifically: In each iteration t, the Metropolis-Hastings algorithm samples ( In each step, according to The candidate samples generated) and sampled from the probability density function of the uniform distribution ( is the probability density function of the uniform distribution, is a random number sampled from a uniformly distributed probability density function), calculate the acceptance probability , by comparison and random numbers sampled from a uniformly distributed probability density function ,if ,but ,otherwise, ;

[0021] The initial model weight vector is taken as uniform distribution, and the probability transfer function is taken as Gaussian probability density function. The expression of the probability transfer function is as follows: ; Where: yes The dimensionality of , i.e. the number of base machine learning models; is the covariance function of the Gaussian probability density function, which is taken as the identity matrix; is the current state; The expression of the likelihood function is as follows: ; Where: is the tunnel sample given by each basic machine learning model TSR value, is the number of training samples, The base machine learning model weight vector; The expression of the tunnel service performance evaluation super model is as follows: ; Where: is the tunnel service performance evaluation index value of a certain tunnel sample; It is The weights of the base machine learning models; It is TSR value of tunnel samples scored by a basic machine learning model; Step S5: Using root mean square error, mean absolute error, and mean absolute percentage error as performance indicators, the tunnel service performance evaluation supermodel is tested to test the model's scoring stability. The test set is input into the constructed tunnel service performance evaluation supermodel to replace the expert scoring performance, generate continuous prediction results, and calculate performance indicators including root mean square error (RMSE, reflecting the degree of deviation between the predicted value and the true value), mean absolute error (MAE, measuring the average level of absolute error), and mean absolute percentage error (MAPE, evaluating the relative error ratio). Through joint analysis of multiple performance indicators, the stability of the model under different error scales is comprehensively evaluated to avoid the limitations of a single performance indicator. A distribution histogram of the prediction error is plotted to observe the central tendency, dispersion, and outlier distribution of the error to determine whether the model has systematic bias. The standard deviation of the performance indicator is further calculated to quantify the volatility of the model on different test samples. If the performance indicator remains stable under multiple random test set partitions, that is, the standard deviation is less than 10% of the mean, then the tunnel service performance evaluation supermodel has been verified to have reliable scoring stability. Otherwise, the model structure or data preprocessing strategy needs to be adjusted. The calculation expression of the root mean square error is as follows: ; The calculation expression of mean absolute error is as follows: ; The calculation expression of mean absolute error is as follows: ; Where: is the root mean square error; is the mean absolute error; is the mean absolute percentage error; is the number of test samples; It is the first TSR values ​​of tunnel samples; The model gives the TSR values ​​of tunnel samples; Step S6: quantify the uncertainty of the tunnel service performance score. If the tunnel service performance evaluation super model passes all tests, it will be put into large-scale tunnel service performance evaluation. The probability characteristics of the tunnel service performance evaluation super model are used to quantify the uncertainty of the tunnel service performance score. By calculating the probability distribution of the scoring results, the confidence interval of the score is obtained to evaluate the reliability of the score. The tunnel service performance evaluation super model is comprehensively tested, including the scoring accuracy, stability and rationality of the uncertainty quantification results, to ensure that the model performs stably on different test sets and that the uncertainty quantification results meet actual engineering requirements. The verified tunnel service performance evaluation super model is then applied to large-scale tunnel service performance evaluation tasks. By utilizing its efficient, accurate and stable characteristics, the service performance of a large number of tunnels can be quickly evaluated, providing a scientific basis for tunnel maintenance, repair and management, and ensuring the safe operation of tunnels.

[0022] Example 2, as Figures 1 to 5 As shown in Example 1, the Shanghai Metro Tunnel Service Performance Evaluation Training Set is taken. On the Shanghai Metro Tunnel Service Performance Evaluation Test Set, the average TSR value of the tunnel service performance evaluation super model is the same as the TSR value given by the expert. Figure 2 The performance index values ​​of the tunnel service performance evaluation super model on the five test sets are shown in Table 1. The RMSE, MAE, and MAPE are less than 0.4, 0.3, and 9%, respectively, indicating that it has a high scoring accuracy. The average index values ​​of the tunnel service performance evaluation super model and the three basic machine learning models used in this embodiment - Partial Least Squares Model (PLS), K-Nearest Neighbor Regression Model (KNN), and Support Vector Regression Model (SVR) on the five test sets are shown in Table 1. Figure 3 Specifically, the RMSE, MAE, and MAPE of the tunnel service performance evaluation super model decreased by 11.3%, 13.5%, and 17.6% respectively compared to the best base machine learning model, indicating that the tunnel service performance evaluation super model has significantly improved its tunnel service performance scoring accuracy. Table 1. Indicators of the super model on the Shanghai Metro Tunnel Service Performance Evaluation Test Set index Test Set 1 Test Set 2 Test set 3 Test set 4 Test set 5 RMSE 0.247 0.333 0.345 0.298 0.332 MAE 0.181 0.273 0.298 0.259 0.244 MAPE 0.050 0.076 0.084 0.073 0.085 like Figure 4As shown in the figure, the coefficient of variation of the scoring error of the tunnel service performance evaluation super model on the test samples is reduced by 28.6% compared with the optimal base machine learning model, which significantly improves the stability of the tunnel service performance scoring of the base machine learning model. When the tunnel service performance evaluation super model passes all tests (the coefficient of variation of the scoring error on the test samples is reduced by 28.6% compared with the optimal base machine learning model, and the RMSE, MAE, and MAPE of the tunnel service performance evaluation super model are all very small (especially the MAPE is less than 10%, meeting the requirements of engineering accuracy)), it can be put into use for large-scale tunnel service performance evaluation.

[0023] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An intelligent evaluation method for tunnel service performance, characterized in that: The following steps are involved: Step S1: Select key tunnel disease types, design service performance evaluation indicators, and collect sample data required for tunnel service performance evaluation to form an evaluation indicator data set; Step S2: Organize experts in the field of tunnel engineering to score some tunnel samples based on the tunnel service performance evaluation indicators, obtain tunnel service performance scores, and construct a tunnel service performance evaluation dataset; Step S3: training multiple base machine learning models to learn the mapping between tunnel service performance evaluation index values ​​and tunnel service performance scores; Step S4: using the Bayesian model averaging method to integrate multiple base machine learning models to construct a tunnel service performance evaluation super model; Step S5: Using the root mean square error, mean absolute error, and mean absolute percentage error as performance indicators, perform model testing on the tunnel service performance evaluation super model; Step S6: quantify the uncertainty of the tunnel service performance score. If the tunnel service performance evaluation super model passes all tests, it is put into large-scale tunnel service performance evaluation.

2. The intelligent evaluation method for tunnel service performance according to claim 1, characterized in that: In step S1, the process of forming the evaluation index data set is as follows: Based on the structural characteristics of tunnels, the causes and impacts of common defects, a comprehensive analysis was conducted to determine the degree of harm that different defects pose to tunnel safety, durability, and usability. The types of defects that have a significant impact on tunnel performance were identified, including settlement, lateral movement, water leakage, cracking, and concrete spalling. Based on the selected key disease types, corresponding tunnel service performance evaluation indicators are designed. The tunnel service performance evaluation indicators include average relative settlement, average differential settlement, average convergence rate, total leakage area, total crack length, and total concrete spalling area. According to the designed tunnel service performance evaluation indicators, tunnel sample data with evaluation indicator values ​​recorded are collected to form an evaluation indicator data set.

3. The intelligent evaluation method for tunnel service performance according to claim 1, characterized in that: In step S2, the process of constructing the tunnel service performance evaluation data set is as follows: Organize experts from various fields of tunnel engineering to develop quantitative scoring rules based on the designed tunnel service performance evaluation indicators, clarify the scoring levels and corresponding value ranges, among which the scoring levels include excellent, good, qualified and unqualified; Provide some sample data to the experts, and then give each sample a comprehensive score based on the scoring criteria and record the basis for the score. After completing the initial evaluation, organize an expert panel to cross-validate the scoring results. The final expert scoring results are summarized, and the average of all expert scores is defined as the tunnel service performance score (TSR) of the sample data. The scores are then associated with the original indicator data to construct a structured tunnel service performance evaluation dataset, which includes the tunnel ID, indicator values, and comprehensive scores, and is annotated with the data source and scoring time.

4. The intelligent evaluation method for tunnel service performance according to claim 1, characterized in that: The step S3 specifically includes: The tunnel service performance evaluation dataset was cleaned, samples with more than 30% missing values ​​were removed, and the cleaned tunnel service performance evaluation dataset was divided into a training set and a test set; A variety of machine learning architectures, including partial least squares regression, support vector machine, and nearest neighbor regression, were selected. A base machine learning model was trained on the training set to learn the mapping between tunnel service performance evaluation index values ​​and tunnel service performance scores. The input of the base machine learning model was the service performance evaluation index, and the output was the tunnel service performance score. Five-fold cross-validation was used to evaluate the model's generalization ability, and grid search was used to optimize hyperparameters. The trained base machine learning model is verified on the test set to evaluate the scoring accuracy and stability of the model, and the model is adjusted and optimized based on the verification results.

5. The intelligent evaluation method for tunnel service performance according to claim 4, characterized in that: In step S4, the process of constructing the tunnel service performance evaluation super model is as follows: The weights of each base machine learning model are inferred using the Bayesian model averaging method. The posterior distribution of the weights is calculated using Bayesian evidence and likelihood functions, and the Metropolis-Hastings algorithm is used for sampling, so that the weights reflect the performance and uncertainty of each base machine learning model on the data. Based on the posterior distribution of the weights of the base machine learning models obtained by sampling, the advantages of multiple base models are integrated into a unified model to construct a super model for tunnel service performance evaluation. The output of the super model is the weighted average of the outputs of each base machine learning model, and the weights are determined by the Bayesian model averaging method.

6. The intelligent evaluation method for tunnel service performance according to claim 5, characterized in that: The expression of the posterior distribution of the weights of the base machine learning model is as follows: ； Where: is the base machine learning model weight vector; is the prior distribution of the weight vector of the base machine learning model; is the likelihood function, reflecting the given The probability of observing the tunnel service performance evaluation training set when ; is the Bayesian evidence, which is constant given the training set; It is the posterior distribution of the weight vector of the fused weight prior distribution and the observed data; use The algorithm solves the posterior distribution of the weight vector of the fusion weight prior distribution and the observed data, and determines the input as the tunnel service performance evaluation training set , initial model weight vector , probability transfer function , likelihood function and number of samples , the output is as follows Sample from the posterior distribution ; The initial model weight vector is taken as uniform distribution, and the probability transfer function is taken as Gaussian probability density function. The expression of the probability transfer function is as follows: ； Where: yes The dimensionality of , i.e. the number of base machine learning models; is the covariance function of the Gaussian probability density function, which is taken as the identity matrix; is the current state; The expression of the likelihood function is as follows: ； Where: is the tunnel sample given by each basic machine learning model of value, is the number of training samples, The base machine learning model weight vector; The expression of the tunnel service performance evaluation super model is as follows: ； Where: is the tunnel service performance evaluation index value of a certain tunnel sample; It is The weights of the base machine learning models; It is The scoring of tunnel samples by a basic machine learning model value.

7. The intelligent evaluation method for tunnel service performance according to claim 6, characterized in that: In step S5, the process of performing model testing on the tunnel service performance evaluation super model is as follows: The test set is input into the constructed tunnel service performance evaluation super model to replace the expert scoring performance, generate continuous prediction results, and calculate performance indicators including root mean square error, mean absolute error and mean absolute percentage error respectively; Draw a distribution histogram of the prediction error to observe the central tendency, dispersion, and outlier distribution of the error, and determine whether the model has systematic deviations. Further calculate the standard deviation of the performance indicators and quantify the volatility of the model on different test samples. If the performance indicators remain stable under multiple random test set partitions, that is, the standard deviation is less than 10% of the mean, then the tunnel service performance evaluation super model is verified to have reliable scoring stability. Otherwise, the model structure or data preprocessing strategy needs to be adjusted.

8. The intelligent evaluation method for tunnel service performance according to claim 7, characterized in that: The calculation expression of the root mean square error is as follows: ； The calculation expression of the mean absolute error is as follows: ； The calculation expression of the mean absolute percentage error is as follows: ； Where: is the root mean square error; is the mean absolute error; is the mean absolute percentage error; is the number of test samples; It is the first Tunnel samples value; The model gives the Tunnel samples value.

9. The intelligent evaluation method for tunnel service performance according to claim 8, characterized in that: The step S6 specifically includes: The probability characteristics of the tunnel service performance evaluation super model are used to quantify the uncertainty of the tunnel service performance score. The confidence interval of the score is obtained by calculating the probability distribution of the score results. The tunnel service performance evaluation super model is comprehensively tested, including the scoring accuracy, stability and rationality of the uncertainty quantification results, and the verified tunnel service performance evaluation super model is then applied to large-scale tunnel service performance evaluation tasks.

Images