Submarine tunnel immersed tube health prediction method based on learning model

Abstract

The invention discloses a subsea tunnel immersed tube health prediction method based on a learning model, and the method comprises the steps: collecting various types of monitoring data, designing an index weight system of a tunnel health state, and determining the weight of each type of monitoring data according to expert experience scores; then data weighted fusion is carried out, monitoring data after weighted fusion are input into a deep learning model to serve as input and predictive variables, the deep learning model is trained, and the deep learning model is used for predicting future trend data according to historical monitoring data; and taking the trend data as new observation data, and inputting the new observation data into a Bayesian model to calculate the probability of future damage risks. According to the method, the Bayesian reasoning and the deep learning model are combined, the short-term health state change of the tunnel structure can be accurately predicted, the probability of damage occurrence is quantified through the Bayesian model, graded early warning prompt is carried out in combination with the health state risk index, and potential risks can be prevented in advance.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent construction technology and engineering structure health monitoring technology, and specifically relates to a method for predicting the health of an immersed tube in an undersea tunnel based on a learning model. Background Art

[0002] In modern infrastructure construction, tunnels, as crucial structures for transportation and water conservancy projects, are crucial for their long-term safety and stability. Due to the complexity of tunnel structures and the variability of their operating environments, effectively monitoring their health and proactively identifying potential risks has become a major challenge for the engineering community. Traditional tunnel health monitoring methods rely on manual inspections and periodic testing. This approach is not only time-consuming and labor-intensive, but also lacks sufficient real-time and sensitivity to sudden structural damage, making it difficult to meet the high safety requirements of modern tunnel projects. In recent years, with the rapid development of Internet of Things (IoT), sensor technologies, and big data processing technologies, intelligent structural health monitoring systems have emerged. These systems utilize a diverse network of sensors deployed within the tunnel to collect real-time monitoring data on various environmental and dynamic characteristics of the tunnel structure. These sensors provide 24 / 7 continuous monitoring of key areas of the tunnel structure, providing continuous data support for structural health. However, with so much sensor data, effectively processing and analyzing it to predict the health of immersed tubes in subsea tunnels remains a technical challenge. Summary of the Invention

[0003] The present invention mainly aims to overcome the deficiencies of the prior art and provides a method for predicting the health of immersed tubes in submarine tunnels based on a learning model.

[0004] The present invention is achieved through the following technical solutions: A method for predicting the health of an immersed tube in a submarine tunnel based on a learning model includes the following steps: Step 1: Acquire monitoring data of the target submarine immersed tube tunnel; the monitoring data includes: differential settlement data between tube segments, differential displacement data between tube segments, tension data between tube segments, differential displacement data between segments, tension data between segments, steel corrosion rate data, concrete resistivity data, and structural vibration data; Step 2: preprocess and normalize the monitoring data obtained in step 1 to form a time series data set; Step 3: Using various monitoring data as indicators, a weighting system for tunnel health status is constructed. The weights of various monitoring data are determined based on expert experience and scoring. Step 4: Based on the weights of various types of monitoring data obtained in Step 3, perform data weighted fusion, and input the weighted fusion monitoring data into the deep learning model as input and prediction variables to train the deep learning model, which is used to predict future trend data based on historical monitoring data; Step 5: Predict future trend data through the deep learning model trained in Step 4, and input the trend data as new observed data into the Bayesian model to calculate the probability of future damage risk.

[0005] In the above technical solution, the differential settlement data between pipe segments: obtained by displacement sensors, arranged at the joints between the pipe segments of the immersed tunnel; the differential displacement data between pipe segments: obtained by displacement sensors, arranged at the joints between the pipe segments of the immersed tunnel; the opening and closing amount data between pipe segments: obtained by displacement sensors, arranged at the joints between the pipe segments of the immersed tunnel; the differential displacement data between segments: obtained by displacement sensors, arranged at the joints between the segments of the immersed tunnel; the opening and closing amount data between segments: obtained by displacement sensors, arranged at the joints between the segments of the immersed tunnel; the steel corrosion rate data: obtained by durability sensors, arranged on the steel bars; the concrete resistivity data: obtained by concrete resistivity detection sensors, arranged on the side walls of the pipe segments of the immersed tunnel; the structural vibration data: obtained by acceleration sensors, arranged on the side walls of the pipe segments of the immersed tunnel.

[0006] In the above technical solution, in Step 2, the preprocessing includes missing value processing and outlier detection and correction to ensure data quality and integrity.

[0007] In the above technical solution, in Step 2, the missing value processing includes: if the missing rate of the monitoring data at a certain time point exceeds the set threshold, the data at this time point will be deleted; if there are continuous missing values within the sampling period, choose to fill the missing data with the first valid value before or after.

[0008] In the above technical solution, in Step 2, for outlier detection and correction: use the Z-score-based method for outlier detection and correct the identified abnormal data.

[0009] In the above technical solution, in Step 2, the normalization process is to scale all data into the range of [0, 1] to ensure that all data have the same scale.

[0010] In the above technical solution, in step 3, the index weight system for the tunnel health status includes two levels: the target layer and the index layer. Among them, the target layer represents four main dimensions in tunnel health assessment, including: deformation between structural pipe joints, deformation between structural node segments, durability of reinforced concrete, and load response, which are four target layers. Under each target layer, specific indexes are included. Among them, the target layer of deformation between structural pipe joints includes three indexes: differential settlement between pipe segments, differential displacement between pipe segments, and opening and closing amount between pipe segments; the target layer of deformation between structural node segments includes two indexes: differential displacement between segments and opening and closing amount between segments; the target layer of durability of reinforced concrete includes two indexes: steel corrosion rate and concrete resistivity; the target layer of load response includes one index: structural vibration. First, conduct expert experience scoring on the four target layers to obtain the weights of each target layer. The total weight of the four target layers is 1. Then, conduct expert experience scoring on the indexes under each target layer to obtain the weights of each index under each target layer. The sum of the weights of all indexes under each target layer is 1. Finally, multiply the weight of the target layer by the weight of the index under it to obtain the comprehensive weight of the index. This comprehensive weight is used as the final weight of each index, that is, the weights of various monitoring data are obtained.

[0011] In the above technical solution, in step 5, the Bayesian health status prediction model is: ; Among them, is the posterior probability of the tunnel structure suffering damage B under the condition of new observed data . This value is used to measure the probability of future damage risk under the current data; is the probability of the observed data appearing when damage B occurs, that is, the likelihood function of the model; is the prior probability of structural damage, which is obtained through historical data and expert experience evaluation; is the total probability of the observed data, which is statistically calculated based on actual monitoring data.

[0012] In the above technical solution, in step 5, calculate the health status risk index DRI, and its formula is: ; According to the calculated health status risk index DRI, divide the warning level and generate a health warning prompt.

[0013] In the above technical solution, the generated health warning prompt information is fed back to the operator through a visual interface, which is convenient for timely taking maintenance and repair measures.

[0014] The advantages and beneficial effects of the present invention are: The present invention designs a method for predicting the health of immersed tubes of a subsea tunnel based on a learning model. By collecting various monitoring data and designing an index weight system for the tunnel health status, the weights of various monitoring data are determined according to expert experience scoring. Then, data weighted fusion is performed, and the weighted fusion monitoring data is input into a deep learning model as input and prediction variables to train the deep learning model. This deep learning model is used to predict future trend data based on historical monitoring data. The trend data is used as new observation data and input into a Bayesian model to calculate the probability of future damage risk. The present invention combines Bayesian inference and a deep learning model, can accurately predict the short-term health status change of the tunnel structure, quantify the probability of damage occurrence through the Bayesian model, and perform hierarchical early warning prompts in combination with the health status risk index (DRI), and can prevent potential risks in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flowchart of the method for predicting the health of immersed tubes of a subsea tunnel based on a learning model according to the present invention.

[0016] For those of ordinary skill in the art, other related drawings can be obtained based on the above drawings without creative efforts. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The following will describe in detail the embodiments of the technical solution of the present invention in conjunction with the drawings. The following embodiments are only used to illustrate the technical solution of the present invention more clearly, so they are only examples and cannot be used to limit the protection scope of the present invention.

[0018] The present invention designs a method for predicting the health of immersed tubes of a subsea tunnel based on a learning model. Refer to the attached Figure 1 , including the following steps: Step 1, obtain the monitoring data of the target subsea immersed tube tunnel.

[0019] The monitoring data includes: differential settlement data between pipe sections, differential displacement data between pipe sections, opening and closing amount data between pipe sections, differential displacement data between segments, opening and closing amount data between segments, steel bar corrosion rate data, concrete resistivity data, and structural vibration data.

[0020] Differential settlement data between pipe sections: obtained by displacement sensors, with a numerical range of 0 - 20 mm; arranged at the joints between pipe sections of the immersed tube tunnel.

[0021] Differential displacement data between pipe sections: obtained by displacement sensors, with a numerical range of 0 - 20 mm; arranged at the joints between pipe sections of the immersed tube tunnel.

[0022] Inter-segment tension and contraction data: obtained through displacement sensors with a value range of 0-20mm; arranged at the joints between the segments of the immersed tube tunnel.

[0023] Inter-segment differential displacement data: obtained through displacement sensors with a value range of 0-15mm; deployed at the joints between segments of the immersed tube tunnel; a segment refers to a segment composed of multiple continuous pipe sections, and inter-segment displacement is a key focus of geological changes.

[0024] Inter-segment tension and contraction data: obtained through displacement sensors with a value range of 0-15mm; deployed at the joints between segments of the immersed tube tunnel.

[0025] Rebar corrosion rate data: obtained through durability sensors placed on the rebar.

[0026] Concrete resistivity data: obtained through concrete resistivity detection sensors, which are installed on the side walls of the pipe sections of the immersed tube tunnel.

[0027] Structural vibration data: obtained through acceleration sensors with a value of 0.001-1.5g. The acceleration sensors are arranged on the side walls of the immersed tube tunnel segments.

[0028] Step 2: Preprocess and normalize the monitoring data obtained in step 1 to form a time series data set.

[0029] Preprocessing, including missing value handling and outlier detection and correction, to ensure data quality and integrity: 1. Missing value processing: If the missing rate of monitoring data at a certain time point exceeds 30%, the data at that time point will be deleted to avoid significant impact on the overall data set. For a small number of consecutive missing data (such as missing 3-5 points), forward or backward filling methods are used to fill in the missing data. Specifically, if If the data at a certain moment is missing, the previous valid data can be used. or filling: ; 2. Outlier detection and correction: The Z-score-based method is used for outlier detection. The outlier judgment formula is: ; in is the sample mean, is the sample standard deviation; if , then the point is considered as an outlier and is corrected.

[0030] Example description: Assume that the original data is: , the processed data is: Through these processes, the raw data is converted into high-quality datasets suitable for model training and analysis, ensuring the accuracy and reliability of subsequent model predictions.

[0031] Normalization is to scale all data to the interval [0, 1] to ensure that all data have the same scale. The formula for data normalization is: ; Step 3: Using various monitoring data as indicators, a weight system for tunnel health status is constructed. The weights of various monitoring data are determined based on expert experience scores (subjective evaluation).

[0032] The indicator weight system for constructing tunnel health status includes two levels: target layer and indicator layer. Among them, the target layer represents the four main dimensions in tunnel health assessment, including "deformation between structural pipe nodes", "deformation between structural node segments", "reinforced concrete durability" and "load response". Under each target layer, there are specific indicators, namely indicator layers. Specifically, the target layer for deformation between structural pipe nodes includes three indicators: differential settlement between pipe segments, differential displacement between pipe segments and tension between pipe segments; the target layer for deformation between structural node segments includes two indicators: differential displacement between segments and tension between segments; the target layer for reinforced concrete durability includes two indicators: steel corrosion rate and concrete resistivity; and the load response target layer includes one indicator: structural vibration.

[0033] First, the four target layers are scored based on expert experience to obtain the weight of each target layer, and the total weight of the four target layers is 1.

[0034] Then, the indicators under each target layer are scored based on expert experience to obtain the weight of each indicator under each target layer. The sum of the weights of all indicators under each target layer is 1.

[0035] Finally, the weight of the target layer is multiplied by the weight of the indicators below it to obtain the comprehensive weight of the indicators. This comprehensive weight is used as the final weight of each indicator, that is, the weight of each type of monitoring data is obtained.

[0036] This weighting system can more comprehensively and accurately characterize the key control factors of the tunnel health status. See Table 1 below, which is a data table of the weight distribution results of the tunnel health status indicator weighting system in this embodiment; Table 1: Weight distribution results of the tunnel health status indicator weight system

[0037] Step 4: Based on the weights of various types of monitoring data obtained in Step 3, perform data weighted fusion, and input the weighted-fused monitoring data into the deep learning model as input and prediction variables to train the deep learning model, which is used to predict future trend data based on historical monitoring data.

[0038] Specifically, a deep learning model using the PyTorch framework is adopted. The number of model layers is 2, the number of hidden layer units is 64, the number of training epochs of the model is set to 100, the batch size is 32, and the learning rate is 0.001. Before the start of model training, the model is initialized with parameters. The weights are initialized using the Xavier initialization method to ensure that the model starts learning from a better starting point. During training, the change in the training loss value is monitored in real time. The initial loss value is 0.19, which drops to 0.04 at the 20th training epoch and finally converges to 0.02 at the 50th epoch, indicating that the learning effect of the model is gradually improving.

[0039] After completing the model training, 10% of the data is used as the validation set to verify and evaluate the model. The mean squared error (MSE) on the validation set is 0.04, which is close to the error of the training set, indicating that the model performs well in preventing overfitting.

[0040] Step 5: Predict the future trend data through the deep learning model trained in Step 4, and use the trend data as new observed data to input into the Bayesian model to calculate the probability of future damage risk.

[0041] The Bayesian health state prediction model is: ; Among them, is the posterior probability of the tunnel structure suffering damage B under the condition of the new observed data , and this value is used to measure the probability of future damage risk under the current data; is the probability of the observed data appearing when damage B occurs, that is, the likelihood function of the model; is the prior probability of structural damage, which is obtained through historical data and expert experience evaluation; is the total probability of the observed data, which is statistically calculated based on actual monitoring data.

[0042] Furthermore, calculate the health state risk index DRI, and its formula is: ; According to the calculated health state risk index DRI, divide the warning levels and generate health warning prompts, specifically as follows: DRI < 0.2: It is in a normal state, the structural health state is excellent, and the damage risk is low.

[0043] 0.2 ≤ DRI<0.4: The structure is slightly exceeded, with slight displacement or stress fluctuation. However, the structure is in good condition and still within the safe range.

[0044] 0.4 ≤ DRI<0.6: This is an excessive state, the structure has certain risks, and monitoring and possible preventive maintenance measures are required.

[0045] DRI ≥ 0.6: Severely exceeded, structural health status has deteriorated significantly, there is a high risk of damage, and immediate maintenance or repair measures are required Furthermore, the generated health warning prompt information is fed back to the operator through a visual interface, facilitating timely maintenance and repair measures.

[0046] The embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned examples, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. A method for predicting the health of immersed tubes in submarine tunnels based on a learning model, characterized in that: The following steps are involved: Step 1: Acquire monitoring data of the target submarine immersed tube tunnel; the monitoring data includes: differential settlement data between tube segments, differential displacement data between tube segments, tension data between tube segments, differential displacement data between segments, tension data between segments, steel corrosion rate data, concrete resistivity data, and structural vibration data; Step 2: preprocess and normalize the monitoring data obtained in step 1 to form a time series data set; Step 3: Using various monitoring data as indicators, a weighting system for tunnel health status is constructed. The weights of various monitoring data are determined based on expert experience and scoring. Step 4: Perform data weighted fusion based on the weights of various monitoring data obtained in step 3. Input the weighted fusion monitoring data into the deep learning model as input and prediction variables to train the deep learning model. The deep learning model is used to predict future trend data based on historical monitoring data. Step 5: Use the deep learning model trained in step 4 to predict future trend data, use the trend data as new observation data, and input it into the Bayesian model to calculate the probability of future injury risk.

2. The method for predicting the health of immersed tubes in a submarine tunnel based on a learning model according to claim 1, wherein: Differential settlement data between pipe segments: obtained through displacement sensors, arranged at the joints between pipe segments of the immersed tube tunnel; Differential displacement data between pipe segments: obtained through displacement sensors, arranged at the joints between pipe segments of the immersed tube tunnel; Inter-segment tension and contraction data: obtained through displacement sensors, arranged at the joints between pipe segments of the immersed tube tunnel; Differential displacement data between segments: obtained through displacement sensors, arranged at the joints between segments of the immersed tube tunnel; Inter-segment tension and contraction data: obtained through displacement sensors, arranged at the joints between segments of the immersed tube tunnel; Steel bar corrosion rate data: obtained through durability sensors, arranged on steel bars; Concrete resistivity data: obtained through concrete resistivity detection sensors, arranged on the side walls of the pipe segments of the immersed tube tunnel; Structural vibration data: obtained through acceleration sensors, arranged on the side walls of the pipe segments of the immersed tube tunnel.

3. The method for predicting the health of the immersed tube of a submarine tunnel based on a learning model according to claim 1, wherein: In step 2, the preprocessing includes missing value processing and outlier detection and correction.

4. The method for predicting the health of immersed tubes of a submarine tunnel based on a learning model according to claim 1, wherein: In step 2, missing value processing includes: if the missing rate of monitoring data at a certain time point exceeds the set threshold, the data at that time point will be deleted; if there are consecutive missing values within the sampling period, the missing data will be filled with the first valid value before or after.

5. The learning model-based method for predicting the health of immersed tubes in submarine tunnels according to claim 1 is characterized by: In step 2, outlier detection and correction: a Z-score-based method is used to detect outliers and correct the identified abnormal data.

6. The method for predicting the health of immersed tubes of a submarine tunnel based on a learning model according to claim 1, wherein: In step 2, the normalization process is to scale all data to the range [0, 1].

7. The learning model-based method for predicting the health of immersed tubes in submarine tunnels according to claim 1 is characterized by: In step 3, the indicator weight system for tunnel health status is constructed, which includes two levels: the target layer and the indicator layer. The target layer represents the four main dimensions in tunnel health assessment, including deformation between structural tube nodes, deformation between structural node segments, durability of reinforced concrete, and load response. Under each target layer, specific indicators are included. The target layer for deformation between structural tube nodes includes three indicators: differential settlement between tube segments, differential displacement between tube segments, and tension between tube segments; the target layer for deformation between structural node segments includes two indicators: differential displacement between segments and tension between segments; the target layer for durability of reinforced concrete includes two indicators: steel corrosion rate and concrete resistivity; and the target layer for load response includes one indicator: structural vibration. First, the four target layers are scored based on expert experience to obtain the weights of each target layer, and the total weight of the four target layers is 1; then, the indicators under each target layer are scored based on expert experience to obtain the weights of each indicator under each target layer, and the sum of the weights of all indicators under each target layer is 1; finally, the weight of the target layer is multiplied by the weights of the indicators under it to obtain the comprehensive weight of the indicator. This comprehensive weight is used as the final weight of each indicator to obtain the weight of each type of monitoring data.

8. The method for predicting the health of the immersed tube of a submarine tunnel based on a learning model according to claim 1, characterized in that: In step 5, the Bayesian health status prediction model is: ； in, For new observation data The posterior probability of damage B occurring in the tunnel structure under the condition of , which is used to measure the probability of future damage risk under the current data; is the observed data when damage B occurs Probability of occurrence; is the prior probability of structural damage, which is obtained through historical data and expert experience evaluation; The total probability of observing the data is calculated statistically based on the actual monitoring data.

9. The method for predicting the health of the immersed tube of a submarine tunnel based on a learning model according to claim 8, characterized in that: In step 5, the health risk index DRI is calculated using the following formula: ； Based on the calculated health risk index DRI, warning levels are divided and health warning prompts are generated.

10. The method for predicting the health of immersed tubes in submarine tunnels based on a learning model according to claim 9 is characterized in that: The generated health warning prompt information is fed back to the operator through a visual interface.

Images