Method for dynamically optimizing model weights for immersed tunnel structural health assessment

Abstract

The application discloses a kind of immersed tube tunnel structure health assessment model weight dynamic optimization method, obtains the measured data of immersed tube tunnel monitoring parameter, compares and calculates each corresponding threshold with measured data, and the measured health score of each monitoring parameter is obtained, and the comprehensive measured health score A is obtained;Based on deep learning model, historical monitoring data modeling is carried out, and the prediction data of each monitoring parameter in future period is predicted, and the prediction data is compared and calculated with each corresponding threshold, and the prediction health score of each monitoring parameter is obtained, and the comprehensive prediction health score B is obtained;Establish immersed tube tunnel structure health assessment model C, C= x A+ y B, weight x And y Bayesian dynamic update is carried out.The application can effectively improve the precision and adaptability of health assessment model by collecting future sensor data to calculate corresponding measured score, and realizing fusion score weight dynamic update combined with Bayesian estimation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of safety monitoring of immersed tunnel engineering, and particularly relates to a method for dynamically optimizing the weight of an immersed tunnel structure health assessment model. BACKGROUND

[0002] As an important form of underwater tunnel engineering, the structural health monitoring of an immersed tunnel is crucial to the long-term safety and operational stability of the tunnel. In the prior art, the structural health assessment of an immersed tunnel is mainly based on a variety of sensor measured parameters (such as inter-segment displacement, segment displacement, structural vibration response, etc.) to construct a structural health assessment model. However, the existing assessment model has the following shortcomings:

[0003] 1. Prediction and measurement are disconnected: future prediction data and real-time measurement data are not effectively integrated, making it difficult to capture the dynamic evolution of the structure state;

[0004] 2. Fixed weight of measurement and prediction fusion: the fusion model does not effectively combine future sensor prediction values and measured data for dynamic optimization;

[0005] 3. Lack of uncertainty quantification: the assessment results lack confidence analysis and cannot provide risk quantification support for maintenance decisions. SUMMARY

[0006] The purpose of the present application is to overcome the shortcomings of the prior art and provide a method for dynamically optimizing the weight of an immersed tunnel structure health assessment model, which further improves the accuracy of the health assessment model by updating the weight proportion of prediction and measurement data in the immersed tunnel structure health assessment model in real time.

[0007] The present application is achieved by the following technical solutions:

[0008] A method for dynamically optimizing the weight of an immersed tunnel structure health assessment model, comprising the following steps:

[0009] Step 1: Based on the sensors arranged in the immersed tunnel, real-time measurement data of the monitoring parameters of the immersed tunnel are obtained, including inter-segment deformation, segment deformation, reinforced concrete durability, and load response. After comparing and calculating the measured data of these monitoring parameters with their respective threshold values, the measured health scores of each monitoring parameter are obtained;

[0010] Step 2: The measured health scores of each monitoring parameter are fused with corresponding weight coefficients to calculate the comprehensive measured health score A;

[0011] Step 3: modeling the historical monitoring data based on a deep learning model to predict the prediction data of each monitoring parameter in the future period t; then comparing the prediction data of each monitoring parameter with the respective threshold value to obtain the prediction health score of each monitoring parameter;

[0012] Step 4: fusing the prediction health scores of each monitoring parameter by giving corresponding weight coefficients to calculate the comprehensive prediction health score B;

[0013] Step 5: fusing the comprehensive measured health score A and the comprehensive prediction health score B by giving corresponding weight coefficients to establish a health assessment model C of the immersed tunnel structure, wherein x is the weight coefficient of the comprehensive measured health score A, y is the weight coefficient of the comprehensive prediction health score B, and x+y = 1;

[0014] Step 6: dynamically updating the weights x and y by Bayes;

[0015] Step 7: outputting the evaluation value of the health assessment model C based on the updated weights x and y .

[0016] In the above technical solution, the measured health score of the monitoring parameter is valued between 0-100, and the prediction health score of the monitoring parameter is valued between 0-100.

[0017] In the above technical solution, wherein Q is the pipe joint deformation measured health score obtained by comparing the actual monitoring value of the pipe joint deformation with the threshold value, W is the segment deformation measured health score obtained by comparing the actual monitoring value of the segment deformation with the threshold value, E is the reinforced concrete durability measured health score obtained by comparing the actual monitoring value of the reinforced concrete durability with the threshold value, and R is the load response measured health score obtained by comparing the actual monitoring value of the load response with the threshold value.

[0018] In the above technical solution, the framework of the Bayesian dynamic weight includes establishing a prior distribution, a likelihood function and a posterior distribution;

[0019] In the above technical solution, wherein is the pipe joint deformation prediction health score obtained by comparing the prediction value of the pipe joint deformation with the threshold value, is the segment deformation prediction health score obtained by comparing the prediction value of the segment deformation with the threshold value, is the reinforced concrete durability prediction health score obtained by comparing the prediction value of the reinforced concrete durability with the threshold value. The load response prediction health score is obtained after comparison calculation of the load response prediction value and the threshold value thereof.

[0020] In the above technical solution, the Bayesian dynamic weight framework includes: establishing a prior distribution, a likelihood function and a posterior distribution;

[0021] Prior distribution: let the weight be x obeys Beta distribution , initial parameters ;

[0022] Likelihood function: let the future measured value be obeys Gaussian distribution with the health assessment model C output error; wherein, for , the future measured value of the health assessment model C is calculated by collecting the sensor measured data in the future time period;

[0023] Posterior distribution: the weight is updated through the Bayesian theorem.

[0024] In the above technical solution, when the future measured value is received, the posterior distribution is calculated, then the update value of the weight x corresponding to the maximum posterior probability is solved, and the prior distribution parameters are updated.

[0025] In the above technical solution, in step 7, when the evaluation value deviates from the safety threshold range, a warning signal is triggered, and a maintenance instruction is generated.

[0026] The advantages and beneficial effects of the present application are:

[0027] The present application introduces a Bayesian estimation mechanism, adjusts the contribution degree of measured and predicted data in real time through Bayesian weight updating, and improves the response speed of the model to the structural state change;

[0028] The present application uses Bayesian estimation, combines real-time sensor data and model prediction, and can gradually reduce error accumulation. The acquisition of each observation data will trigger error feedback correction, thereby improving the accuracy of the model and avoiding the error expansion problem in the traditional method;

[0029] Many traditional health assessment methods rely on expert experience or are based on artificially set rules. Such methods not only have limitations in real-time performance, but also are easily disturbed by human factors, leading to inconsistency of the evaluation results. The weight updating process of the Bayesian estimation in the present application is automated, which can adaptively adjust according to the measured data and historical prediction data without relying too much on expert experience. The automated updating process improves the real-time performance and accuracy of the model, reduces human interference and bias;

[0030] Through the Bayesian estimation weight dynamic optimization method, the immersed tunnel structure health assessment model can more accurately reflect the structure health condition, adapt to external environment changes and data updates in real time, and avoid the limitations of traditional methods. This method not only improves the accuracy and robustness of the evaluation results, but also reduces the need for manual intervention and improves the scalability and long-term stability of the model, providing a more intelligent and efficient solution for tunnel structure health monitoring and evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 The flowchart of the immersed tunnel structure health assessment model weight dynamic optimization method.

[0032] Figure 2 The flowchart of the Bayesian dynamic weight updating in the present application. DETAILED DESCRIPTION

[0033] In order to better understand the technical scheme of the present application by those skilled in the art, the technical scheme of the present application will be further described in combination with specific embodiments.

[0034] The present application designs an immersed tunnel structure health assessment model weight dynamic optimization method, referring to the accompanying drawings Figure 1 , comprising the following steps:

[0035] Step 1: Based on the multiple types of sensors arranged in the immersed tunnel, real-time acquisition of the measured data of the key monitoring parameters of the immersed tunnel is carried out, the key monitoring parameters include: pipe joint deformation, segment deformation, reinforced concrete durability and load response parameters, etc. After comparing and calculating the measured data of these monitoring parameters with their respective threshold values, the measured health scores of each monitoring parameter are obtained.

[0036] For example, Q is the pipe joint deformation measured health score obtained by comparing and calculating the pipe joint deformation actual monitoring value with its threshold value, W is the segment deformation measured health score obtained by comparing and calculating the segment deformation actual monitoring value with its threshold value, E is the reinforced concrete durability measured health score obtained by comparing and calculating the reinforced concrete durability actual monitoring value with its threshold value, and R is the load response measured health score obtained by comparing and calculating the load response actual monitoring value with its threshold value.

[0037] The measured health scores of these monitoring parameters are valued between 0 and 100, and the score size represents the degree of abnormality of the monitoring parameter, 100 points represent that the parameter is within the threshold range, and the smaller the score, the greater the deviation from the normal threshold range.

[0038] Step 2: The measured health scores of each monitoring parameter are respectively given corresponding weight coefficients for fusion, and then the comprehensive measured health score A is calculated. For example: wherein the values before Q, W, E and R are weight coefficients, which are calculated by expert questionnaire method.

[0039] Step 3: Based on a deep learning model (such as LSTM or Transformer), the historical monitoring data is modeled to predict the prediction data of each monitoring parameter in the future period t; then, after comparing and calculating the prediction data of each monitoring parameter with the respective threshold value, the prediction health score of each monitoring parameter is obtained.

[0040] Step 4: The prediction health scores of each monitoring parameter are respectively given corresponding weight coefficients for fusion, and then the comprehensive prediction health score B is calculated.

[0041] For example: wherein is the pipe segment deformation prediction health score obtained by comparing and calculating the pipe segment deformation prediction value and the threshold value thereof, is the segment deformation prediction health score obtained by comparing and calculating the segment deformation prediction value and the threshold value thereof, is the reinforced concrete durability prediction health score obtained by comparing and calculating the reinforced concrete durability prediction value and the threshold value thereof, is the load response prediction health score obtained by comparing and calculating the load response prediction value and the threshold value thereof. 、 、 and the values before them are weight coefficients, which are calculated by expert questionnaire method.

[0042] Step 5: Considering the current measured health state of the immersed tunnel and the predicted health state in the future period t, the comprehensive measured health score A and the comprehensive prediction health score B are given corresponding weight coefficients for fusion, so as to establish an immersed tunnel structure health assessment model C, wherein x and y are the weights corresponding to the comprehensive measured health score A and the comprehensive prediction health score B respectively, and satisfy x+y=1.

[0043] Step 6: The weights x and y are dynamically updated by Bayesian method.

[0044] Since the calculation of the above health assessment model C is based on the measured value of the immersed tunnel structure on the day and the deep learning prediction result of the immersed tunnel in the future period t, the corresponding weights x and yNot fixed, but need to be dynamically updated and adjusted with the actual monitoring state of the sensor and the prediction accuracy of deep learning. For this purpose, a Bayesian dynamic weight update strategy is adopted to realize dynamic update of the weights x and y .

[0045] The framework of the Bayesian dynamic weight update strategy includes establishing a prior distribution, a likelihood function and a posterior distribution.

[0046] Prior distribution: let the weight x obeys the Beta distribution Beta α , β ), the initial parameters .

[0047] Likelihood function: let the future measured value obeys the Gaussian distribution of the output error of the health assessment model C. Among them, is the variance (noise error); for , the future measured value of the health assessment model C can be calculated by collecting the sensor measured data in the future time period t, for example: , wherein respectively correspond to the inter-pipe deformation, inter-segment deformation, reinforced concrete durability and load response measured health scores calculated by the sensor measured data in the future time period t.

[0048] Posterior distribution: update the weight

[0049] ;

[0050] Weight dynamic update includes:

[0051] Maximum a posteriori estimation (MAP): solve the update value x of the weight corresponding to the maximum posterior probability:

[0052] ;

[0053] Online parameter update: update the Beta parameters of the prior distribution according to the posterior distribution:

[0054] ;

[0055] Among them, is the learning rate.

[0056] Specifically, when the weight is dynamically updated, refer to the attached Figure 2 , input A, B and , initialize the parameters ; each time a future measurement value is received , the posterior distribution is calculated , then the weight that maximizes the posterior probability is solved x , the updated value of , and the prior distribution parameters and are updated and . Since the weight x + y =1, the updated value of the weight x is obtained, and the updated value of the weight y , y =1- x .

[0057] Step 7: Based on the updated weights x and y , health assessment and risk warning are carried out.

[0058] Health assessment: Based on the updated weights x and y , the evaluation value of the health assessment model C is output.

[0059] Risk warning: When the evaluation value deviates from the safe threshold range, a warning signal is triggered, and maintenance instructions are generated.

[0060] The above is an exemplary description of the present application. It should be noted that any simple modification, modification or other equivalent replacement that does not deviate from the core of the present application and can not require creative labor by those skilled in the art falls within the protection scope of the present application.

Claims

1. A method for dynamic optimization of weights in a health assessment model for immersed tunnel structures, characterized in that, Includes the following steps: Step 1: Based on the sensors deployed in the immersed tunnel, real-time measured data of the immersed tunnel monitoring parameters are obtained. The monitoring parameters include: inter-segment deformation, inter-segment deformation, reinforced concrete durability and load response. After comparing and calculating the measured data of these monitoring parameters with their respective thresholds, the measured health score of each monitoring parameter is obtained. Step 2: Assign corresponding weighting coefficients to the measured health scores of each monitoring parameter and integrate them to calculate the comprehensive measured health score A; Step 3: Model historical monitoring data based on deep learning model to predict the predicted data of each monitoring parameter in the future time period t; then compare the predicted data of these monitoring parameters with their respective thresholds to obtain the predicted health score of each monitoring parameter. Step 4: Assign corresponding weight coefficients to the predicted health scores of each monitoring parameter and fuse them to calculate the comprehensive predicted health score B; Step 5: The measured health score A and the predicted health score B are weighted and fused to establish a health assessment model C for the immersed tunnel structure. , where x is the weighting coefficient of the comprehensive measured health score A, y is the weighting coefficient of the comprehensive predicted health score B, and x+y=1; Step 6: Weighting x and y Perform Bayesian dynamic updates; Step 7: Based on the updated weights x and y Output the evaluation value of health assessment model C.

2. The dynamic optimization method for weights in the immersed tunnel structural health assessment model according to claim 1, characterized in that: The measured health scores of the monitored parameters range from 0 to 100, and the predicted health scores of the monitored parameters also range from 0 to 100.

3. The dynamic optimization method for weights in the immersed tunnel structural health assessment model according to claim 1, characterized in that: Where Q is the measured health score of inter-segment deformation calculated by comparing the actual monitored value of inter-segment deformation with its threshold, W is the measured health score of inter-segment deformation calculated by comparing the actual monitored value of inter-segment deformation with its threshold, E is the measured health score of reinforced concrete durability calculated by comparing the actual monitored value of reinforced concrete durability with its threshold, and R is the measured health score of load response calculated by comparing the actual monitored value of load response with its threshold.

4. The dynamic optimization method for weights in the immersed tunnel structural health assessment model according to claim 1, characterized in that: ,in The predicted health score for inter-segment deformation is obtained by comparing the predicted value of inter-segment deformation with its threshold. The inter-segment deformation prediction health score is obtained by comparing the predicted inter-segment deformation value with its threshold. The predicted health score for reinforced concrete durability is obtained by comparing the predicted durability value with its threshold. The load response prediction health score is obtained by comparing the predicted load response value with its threshold.

5. The dynamic optimization method for weights in the immersed tunnel structural health assessment model according to claim 1, characterized in that: The framework for Bayesian dynamic weights includes: establishing the prior distribution, the likelihood function, and the posterior distribution; Prior distribution: Let the weights be... x Follows a Beta distribution initial parameters ; Likelihood function: assuming future measured values The output error of the health assessment model C follows a Gaussian distribution; where, for It calculates the future measured values ​​of health assessment model C by collecting sensor data over a future time period. ; Posterior distribution: update weights using Bayes' theorem.

6. The dynamic optimization method for weights in the immersed tunnel structural health assessment model according to claim 5, characterized in that: When the weights are dynamically updated, each time a future measured value is received... Calculate the posterior distribution, and then solve for the weights that maximize the posterior probability. x The updated values ​​are then used to update the prior distribution parameters.

7. The dynamic optimization method for weights in the immersed tunnel structural health assessment model according to claim 1, characterized in that: In step 7, when the assessed value deviates from the safety threshold range, an early warning signal is triggered and a maintenance instruction is generated.

Images