Integrated intelligent monitoring decision system and method for overlapping shield tunnel system

Abstract

The application discloses a kind of comprehensive intelligent monitoring decision system and method for overlapping shield tunnel system, system includes information acquisition system, iterative optimization state real-time evaluation system and intelligent maintenance decision system;First, preset model is trained by numerical calculation and other data;Operational measured data trains feedback model, and the output result of the weighted two models obtains tunnel deformation prediction value;Again, experience overlapping tunnel safety evaluation system and iterative optimization evaluation system are established, and the depth correlation of tunnel state, damage, maintenance decision is carried out by correlation degree intelligent real-time analysis, realize tunnel state real-time monitoring-tunnel damage real-time capture-tunnel maintenance decision real-time update, and the intelligent collocation of maintenance measure is carried out, then according to the operation effect feedback iterative optimization prediction model and safety evaluation system.The application realizes the real-time perception, intelligent diagnosis and maintenance of whole life cycle overall service performance of overlapping shield system, and improves the long-term service quality of overlapping shield tunnel structure.

Description

Technical Field

[0001] This invention belongs to the field of monitoring and maintenance technology for overlapping shield tunnels, specifically relating to a comprehensive intelligent monitoring and decision-making system and method for overlapping shield tunnel systems. Background Technology

[0002] With the continuous development of urban underground transportation construction, the number of subway shield tunnels is increasing, resulting in a large number of overlapping subway shield tunnels. Due to the mutual influence between overlapping subway shield tunnels in service, the structural defects of these tunnels are more prominent than those of single-track tunnels. Among these defects, water leakage, segment cracking, misalignment, uneven longitudinal settlement, and excessive lateral deformation are more serious, and in some sections, they even threaten the safe operation of the subway.

[0003] Traditional tunnel structure monitoring systems obtain information about tunnel structural defects through defect surveys. However, current systems primarily rely on human experience to assess and maintain tunnels, especially single-track tunnels. Considering the complex dynamic and static interactions between overlapping tunnels, current systems cannot simultaneously consider the operational status of each individual single-track tunnel within an overlapping tunnel system, as well as their mutual influences during operation, particularly for the specific structural forms and defect characteristics of overlapping shield tunnels. Furthermore, maintenance decisions made often fail to account for the complex interactions between overlapping tunnels, resulting in simplistic and unbalanced decision-making.

[0004] Therefore, in order to determine the structural safety status of overlapping shield tunnels and ensure the structural and operational safety of shield tunnels, it is necessary to propose a comprehensive intelligent monitoring and decision-making system for overlapping shield tunnel structures. This system should deeply correlate tunnel status, damage, and maintenance decisions through intelligent real-time correlation analysis, enabling real-time monitoring of tunnel status, real-time capture of tunnel damage, and real-time updating of tunnel maintenance decisions. This will allow for quick, accurate, and intelligent assurance of the structural and operational safety of the tunnels. Summary of the Invention

[0005] To address the problems existing in the above-mentioned background technology, the purpose of this invention is to provide a comprehensive intelligent monitoring and decision-making system and method for overlapping shield tunnel systems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems includes the following steps:

[0008] S1. Based on previous engineering monitoring data, establish a numerical model to obtain numerical simulation data, analyze the model to obtain calculation data, and use the calculation data as training data; this solves the problem of providing data support for the initial service of the tunnel due to data scarcity by using other data.

[0009] S2. Construct a preset model based on a Long Short-Term Memory (LSTM) network model. Use the training data as the input variable of the LSTM model in the preset model to train the preset model to obtain a preset algorithm. Preliminary prediction of tunnel deformation trends is made, and initial decisions for intelligent tunnel maintenance are provided. The preset algorithm will not be further trained after training is completed and put into use. To avoid 'data pollution', the preset model is encapsulated and trained separately, isolated from the later feedback algorithm data and model. It plays a transitional support role in the early stage of operation and an auxiliary decision-making role in the later stage of operation.

[0010] S3. Collect data such as internal forces of tunnel structures in overlapping areas, defects of new and old tunnels, and vibration of tunnel structures to obtain real-time monitoring data. The real-time monitoring data is divided into status information and functional damage information. The real-time monitoring data is then normalized.

[0011] S4. Construct a feedback model based on a Long Short-Term Memory (LSTM) network model using a feedback algorithm. The feedback algorithm iteratively updates the training results of the feedback model using changes in normalized real-time monitoring data. A real-time feedback mechanism is employed, using normalized real-time monitoring data as the input variable to the LSTM model within the feedback model. The feedback model is trained, and its input features are continuously adjusted based on the latest real-time monitoring data. The correlation between tunnel deformation and each monitoring variable is calculated within a fixed-length time window. Within each time window, the Spearman correlation coefficient is used to calculate the correlation between the monitoring variables and tunnel deformation, determining the intrinsic relationships between the data. Features with high correlation are selected as the input features for the LSTM model in the feedback model. The input features of the model are defined; the prediction results output by the preset model and the prediction results output by the feedback model are combined to output the final predicted value of tunnel deformation, and intelligent tunnel maintenance decisions are given based on the correlation between maintenance measures and tunnel status; the feedback algorithm monitors the real-time stress and displacement data of the tunnel and identifies the tunnel's operational status through image recognition during the operation and maintenance process, and continuously iterates the feedback, adjusting the weights of the influencing factors on tunnel deformation through iterative feedback to update the intelligent tunnel maintenance decisions; all training data of the feedback model are measured data during the operation period of this project, without injecting artificially calculated or simulated data, aiming to eliminate the interference caused by data sources with human errors and limitations to the algorithm; the feedback algorithm is a growth algorithm, which continues to train and learn during the tunnel's operation period, and continuously iterates and optimizes.

[0012] S5. Based on a large amount of actual engineering data, establish an experience-overlapping tunnel safety assessment system. Combine the experience-overlapping tunnel safety assessment system with the real-time monitoring data, the tunnel deformation trend predicted by the preset model and the feedback model, and establish an iterative optimization assessment system based on the correlation in the time dimension. The iterative optimization assessment system uses the inherent relationship between the experience-overlapping tunnel safety assessment system and each single indicator in the monitoring system, and uses correlation analysis to conduct a deep correlation in time between the actual tunnel operation status information and functional damage information. It continuously corrects and updates the experience-overlapping tunnel safety assessment system during the operation cycle, thereby continuously iterating and updating the system to optimize it and form a comprehensive health assessment system for overlapping tunnel structures.

[0013] S6. The tunnel condition is assessed based on the comprehensive health assessment system for overlapping tunnel structures, and tunnel maintenance decisions are intelligently matched based on the correlation between tunnel condition and maintenance decisions. As the tunnel is in operation, the correlation between tunnel condition and tunnel maintenance decisions is iterated in real time, and the comprehensive health assessment system for overlapping tunnel structures and the final intelligent tunnel maintenance decisions are continuously iterated and optimized.

[0014] Furthermore, the status information includes: soil stress around overlapping tunnels, pore water pressure, arch settlement, overhang heave, uneven settlement, tunnel seepage, wheel-rail periodic wear, ground vibration, track slab vibration, vehicle body vibration, and tunnel wall vibration. The functional damage information includes segment misalignment, segment cracking, and cross-sectional deformation.

[0015] Furthermore, the aforementioned experience-overlapping tunnel safety assessment system includes the assessment of the safety level of the transverse deformation of the shield tunnel, the assessment of the structural safety level within the 3D range of the longitudinal settlement trough, and the assessment of the tunnel's health status.

[0016] Furthermore, the LSTM model includes an input layer, an LSTM layer, a fully connected layer, and an output layer. The LSTM layer learns the time-series features in the input data and controls the information flow through forget gates, input gates, and output gates. The fully connected layer maps the time-series features learned by the LSTM layer onto the tunnel deformation prediction and outputs the prediction result.

[0017] ;

[0018] in, To predict the amount of tunnel deformation, This refers to the hidden state output by the LSTM layer.

[0019] Furthermore, the prediction results output by the preset model and the prediction results output by the feedback model are combined to output the final predicted value of tunnel deformation. for:

[0020] ;

[0021] in, Output the predicted value of tunnel deformation for the preset model. The feedback model outputs a predicted value for tunnel deformation. The fusion weights represent the degree of confidence in the results of the preset model. In the initial stages of model operation, if sufficient real-time monitoring data is not yet available, the preset model can be used as the primary model with fixed weights. As the system runs and real-time data increases, the feedback model gradually stabilizes. At this point, error feedback can be used for dynamic weighting. The fusion weights are calculated by incorporating the model's recent prediction errors.

[0022] ;

[0023] in, The prediction error of the preset model, This refers to the prediction error of the feedback model. The smaller the error, the greater the prediction weight, achieving the purpose of dynamic trust adjustment. The weight of the feedback algorithm increases continuously as the operation cycle progresses; after the operation cycle reaches a certain stage, the feedback algorithm gradually matures and becomes dominant.

[0024] Furthermore, the calculation formula for the forget gate is as follows:

[0025] ;

[0026] in, This refers to the part of the memory state from the previous moment that needs to be retained. and These are the weight matrix and bias term of the forget gate, respectively; It is the hidden state of the previous moment, containing historical deformation information; for Input data at any given time;

[0027] The calculation formula for the input gate is:

[0028] ;

[0029] in, The input gate vector controls the data that should be written into the current memory cell from the current input data. and These are the weight matrix and bias term of the input gate, respectively;

[0030] In addition, LSTM also computes candidate memory cells. , representing the potential information currently being input:

[0031] ;

[0032] in, and The weight matrix and bias term of each candidate memory unit;

[0033] The formula for updating the state of a memory cell is:

[0034] ;

[0035] in, for The state of the memory unit at any given moment. for The state of the memory cell at time -1;

[0036] The formula for calculating the output gate is:

[0037] ;

[0038] in, The output gate vector controls the output of the hidden state at the current time step, determining the predicted value and state propagation. For the sigmoid function, and These are the weight matrix and bias term of the output gate, respectively;

[0039] The hidden state Generate using the following formula:

[0040] .

[0041] Furthermore, during the training of the preset model, the weight parameters are adjusted by minimizing the mean squared error (MSE) loss function to optimize the LSTM model parameters; the MSE loss function is calculated as follows:

[0042] ;

[0043] in, It predicts the amount of tunnel deformation. Predicting the actual deformation of the tunnel. It is the number of samples;

[0044] The loss function in the feedback model simultaneously considers the consistency between real-time data errors and the predictions of the preset model. The loss function in the feedback model is as follows:

[0045] ;

[0046] in, , The preset model and the feedback model are used to predict the deformation of the tunnel during the training process, respectively.

[0047] Furthermore, the method for determining the inherent relationships between the various data points is as follows:

[0048] A fixed time window of 30 days was used, and the correlation between tunnel deformation and each monitored variable was recalculated every 30 days. Within each time window, the Spearman correlation coefficient was used. The calculation formula is as follows:

[0049] ;

[0050] In the formula, For each pair of observations, the rank difference The number of data points;

[0051] When performing correlation analysis, in addition to the current state quantity information monitoring data... In addition, lagged variables are introduced. , Calculate the effect of the capture variable on the functional impairment information variable respectively. The delayed impact; among them, This is status information monitoring data from 15 days ago. This is monitoring data on status information from 30 days ago;

[0052] Within each time window, based on the results of correlation analysis, the correlation and lag effects are considered together, and the monitoring variable data with high correlation to functional impairment information variables are selected as the input data of the LSTM model.

[0053] The present invention further provides a system for implementing the above-described integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems, comprising:

[0054] Information Acquisition System: This system collects data on internal forces, defects in new and old tunnels, and tunnel vibration in the overlapping tunnel area, enabling real-time monitoring of the safety and operational status of the overlapping shield tunnels. The system transmits real-time monitoring data wirelessly. It includes a sensor monitoring module and an image monitoring module. The sensor monitoring module monitors soil stress, pore water pressure, arch settlement, overhang, uneven settlement, tunnel seepage, segment misalignment, segment cracking, and cross-sectional deformation around the overlapping tunnels. The control module is used to monitor wheel-rail periodic wear, tunnel seepage, segment misalignment, segment cracking, cross-sectional deformation, ground vibration, track slab vibration, vehicle vibration, and tunnel wall vibration. The real-time monitoring data is divided into status information and functional damage information. The soil stress, pore water pressure, wheel-rail periodic wear, ground vibration, track slab vibration, vehicle vibration, and tunnel wall vibration around overlapping tunnels are status information. The arch settlement, invert heave, uneven settlement, tunnel seepage, segment misalignment, segment cracking, and cross-sectional deformation are functional damage information.

[0055] The iterative optimization real-time assessment system includes an experience-based overlapping tunnel safety assessment system established based on a large amount of actual engineering data, and an iterative optimization assessment system adjusted in real time based on actual tunnel operation status feedback. The iterative optimization assessment system uses correlation analysis to deeply correlate actual tunnel operation status information and functional damage information over time, thereby continuously updating the overlapping tunnel safety assessment system. Combining real-time monitoring data, preset models, and feedback model predictions of tunnel deformation trends, an iterative optimization assessment system is established based on correlation over time. This system deeply correlates status information and functional damage information through the existing experience-based overlapping tunnel safety assessment system and the inherent connections between individual indicators in the monitoring system, continuously correcting them within the operation cycle, thereby iteratively optimizing the comprehensive health assessment system of the overlapping tunnel structure.

[0056] The intelligent maintenance decision-making system comprises an initial preset algorithm decision-making system and a feedback algorithm decision-making system. The initial preset algorithm decision-making system includes a preset model built using an LSTM model based on a preset algorithm. A numerical model is established using past engineering monitoring data as numerical simulation data, and computational data is obtained through model analysis to form the preset algorithm. The feedback algorithm decision-making system includes a feedback model built using an LSTM model based on a feedback algorithm. The model's training results are iteratively updated based on changes in normalized real-time monitoring data to form the feedback algorithm. The feedback algorithm system, leveraging the initial preset algorithm decision-making system, continuously updates and improves itself through iterative feedback during actual monitoring and maintenance. It provides intelligent maintenance decisions by collecting real-time stress and displacement data of the tunnel and identifying the tunnel's operational status through image recognition. It continuously optimizes maintenance decisions by analyzing the correlation between the adopted maintenance and control measures and the tunnel's status. Simultaneously, the feedback algorithm decision-making system is continuously iteratively optimized based on the adopted intelligent maintenance decisions and feedback on the tunnel's operational status.

[0057] Furthermore, the maintenance decision-making is divided into direct optimization measures for a single disease and related optimization measures. Direct optimization measures directly select whether to maintain the current measure or try other measures based on the effect of the measures taken, and finally select the optimal measure after trying all measures. Related optimization measures obtain multi-objective optimization measures by comprehensively regulating multiple disease factors and after correlation analysis. Finally, the control measures of the maintenance decision are selected by comprehensively comparing the correlation obtained by the decision system analysis and combining direct optimization measures and related optimization measures to achieve the best macro-multi-control operation and maintenance effect.

[0058] Compared with the shortcomings and deficiencies of existing technologies, the present invention has the following beneficial effects:

[0059] 1. This invention integrates three major modules: an information acquisition system, an iterative optimization real-time assessment system, and an intelligent maintenance decision-making system. It establishes an intelligent monitoring and decision-making system for the health status of overlapping tunnel structures. The system obtains computational data through preliminary numerical calculations and analytical model calculations to train a preset model, thus solving the problem of insufficient data in the early stages. At the same time, the preset algorithm is encapsulated separately to avoid affecting the algorithm trained with data obtained from actual engineering projects.

[0060] 2. By automatically collecting data on the internal forces of overlapping tunnel structures, defects in new and old tunnels, and vibration test data of tunnel structures, the real-time monitoring data is further divided into status information and functional damage information. With the help of a dual-algorithm decision-making system consisting of a preset algorithm and a feedback algorithm, the correlation between the operating status and damage of overlapping tunnels is analyzed in real time; thereby iteratively revising the comprehensive health assessment system of overlapping tunnel structures.

[0061] 3. Based on the comprehensive health assessment system for overlapping tunnel structures, maintenance and control measures are implemented while correlation analysis is performed with tunnel status data. This allows for iterative optimization, providing optimal safety warnings or intelligent decision-making for maintenance plans based on real-time assessments of the overall tunnel performance.

[0062] 4. By intelligently combining direct optimization measures with related optimization measures, real-time perception, diagnosis, and intelligent maintenance of the overall service performance of the overlapping shield tunnel system throughout its entire life cycle are achieved, thereby improving the long-term service quality of the overlapping shield tunnel structure. Attached Figure Description

[0063] Figure 1 This is an operation diagram of the integrated intelligent monitoring and decision-making system for overlapping shield tunnel systems provided in this embodiment of the invention;

[0064] Figure 2 This is a schematic diagram of the organization of the integrated intelligent monitoring and decision-making system for overlapping shield tunnel systems provided in an embodiment of the present invention;

[0065] Figure 3 These are the tunnel model and observation point plan view (a) and the stratum model and observation point profile view (b) used in the model training of the preset algorithm provided in this embodiment of the invention.

[0066] Figure 4 This is an iterative diagram of the intelligent maintenance decision-making system provided in the embodiments of the present invention;

[0067] Figure 5 This is a flowchart of the integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems provided in an embodiment of the present invention;

[0068] Figure 6 This is an intelligent maintenance decision map based on correlation analysis provided in an embodiment of the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0070] The integrated intelligent monitoring and decision-making system for overlapping shield tunnel systems has the following operation diagram and organizational diagram: Figure 1 and 2 As shown, the system includes:

[0071] Information Acquisition System: This system collects data on the internal forces of the tunnel structure in the overlapping area, defects in both new and old tunnels, and vibration test data of the tunnel structure. It monitors the safety status and operational status of the overlapping shield tunnels in real time. The system transmits real-time monitoring data via a wireless transmission module. The system includes a sensor monitoring module and an image monitoring module. The sensor monitoring module monitors soil stress, pore water pressure, arch settlement, overhang, uneven settlement, tunnel seepage, segment misalignment, segment cracking, and cross-sectional deformation around the overlapping tunnels. The image monitoring module monitors wheel-rail periodic wear, tunnel seepage, segment misalignment, segment cracking, cross-sectional deformation, surface vibration, track slab vibration, vehicle body vibration, and tunnel wall vibration.

[0072] The information collection coordinates are not only focused on the individual tunnel itself and the surrounding soil, but also on the soil around overlapping tunnels and the entire system, and the data is collected and processed in a unified manner.

[0073] The information acquisition system utilizes multiple sensors and wireless transmission technology to achieve real-time monitoring of tunnel-related data, effectively solving the problem of data lag and improving the accuracy and timeliness of monitoring; at the same time, it provides a basis for efficient and accurate decision-making.

[0074] The iterative optimization real-time assessment system comprises an experience-based overlapping tunnel safety assessment system established based on extensive actual engineering data, and an iterative optimization assessment system. The iterative optimization assessment system further refines and corrects the experience-based overlapping tunnel safety assessment system by incorporating specific data from the data acquisition system and using algorithmic predictions and decision feedback. Simultaneously, the iterative optimization assessment system uses correlation analysis to deeply correlate actual tunnel operation status information with functional damage information over time, thereby continuously updating the overlapping tunnel safety assessment system. By identifying the inherent relationships between various individual indicators in the artificially defined experience-based overlapping tunnel safety assessment system, the iterative optimization assessment system establishes a comprehensive assessment system that considers all evaluation indicators, more accurately assessing and predicting tunnel operation status, and facilitating maintenance decision-making.

[0075] Traditional decision-making systems, which rely on image acquisition and sensor data processing followed by a human-defined evaluation framework, heavily rely on human experience. Essentially, they merely reduce the workload of manual data collection and are prone to human error. Furthermore, traditional human-based evaluation systems can only assess individual indicators, failing to comprehensively evaluate the interrelationships between them and neglecting the inherent connections and patterns among monitoring indicators. This invention, however, involves image acquisition and recognition alongside human-defined standards in the decision-making process. Through iterative optimization of the algorithm and decision-making based on monitoring feedback after decision maintenance, the evaluation system and decision-making are continuously improved through learning, building upon the human-defined standards. The evaluation system is continuously optimized and refined based on actual engineering conditions.

[0076] Intelligent maintenance decision-making system: including initial preset algorithm decision-making system and feedback algorithm decision-making system.

[0077] The initial preset algorithm decision system was formed by training a set of pre-established 1:1 numerical simulation and analytical models, empirical judgments, and engineering examples. The preset model training used tunnel models and observation point plan maps, as well as stratigraphic models and observation point profile maps, as shown in the following figures. Figure 3 As shown in the diagram. The feedback algorithm system, leveraging an initially pre-set algorithm decision-making system, continuously iterates and updates its system based on feedback from the tunnel's operational status during actual monitoring and maintenance. By collecting real-time stress and displacement data and using image recognition to assess the tunnel's operational status, it provides intelligent maintenance decisions. Through correlation analysis between the implemented maintenance and control measures and the tunnel's status, it continuously optimizes maintenance decisions. Simultaneously, based on the intelligent maintenance decisions and feedback from the tunnel's operational status, the feedback algorithm decision-making system is continuously iterated and optimized. The iteration diagram of the intelligent maintenance decision-making system is shown in the diagram. Figure 4 As shown.

[0078] The initial preset algorithm decision system includes a preset model built based on an LSTM model of the preset algorithm. The computational data analyzed by the numerical model serves as the input variable of the LSTM model in the preset model, training the preset model to obtain the preset algorithm. The feedback algorithm decision system includes a feedback model built based on an LSTM model of the feedback algorithm. The training results of the feedback model are continuously updated iteratively based on the changes in normalized real-time monitoring data to form the feedback algorithm. The LSTM model includes an input layer, an LSTM layer, a fully connected layer, and an output layer. The input variable of the preset model is numerical simulation, which is based on previous engineering monitoring data. The input variable of the feedback model is normalized real-time monitoring data, including: soil stress around overlapping tunnels, pore water pressure, arch settlement, invert heave, uneven settlement, tunnel seepage, segment misalignment, segment cracking, and cross-sectional deformation, etc. The hidden states output by the LSTM layer are mapped to the final predicted tunnel deformation through the fully connected layer, outputting the prediction result.

[0079] The feedback algorithm system, with the help of the initially preset algorithm decision system, continuously updates and improves itself through iterative feedback during the actual monitoring and maintenance process. It provides intelligent maintenance decisions by collecting real-time stress and displacement data of the tunnel and identifying the tunnel's operational status through image recognition. At the same time, the feedback algorithm decision system is continuously iterated and optimized based on the intelligent maintenance decisions taken and the feedback on the tunnel's operational status.

[0080] To achieve accurate structural prediction of tunnel deformation systems, an initial LSTM model is trained using numerical simulation data. Numerical simulation data provides a theoretically sound preliminary model, helping to validate the model architecture and predictive capabilities. Training with numerically simulated data before acquiring real monitoring data provides a stable starting point for subsequent model fine-tuning and optimization, thus accelerating the model's learning process in practical applications. Deformation prediction is a critical issue in the health monitoring of overlapping shield tunnels. Due to the influence of external loads, temperature changes, and other factors, the deformation of the tunnel may change over time. This invention employs an LSTM model, using both numerical simulation data and real-time data as input, to accurately predict tunnel deformation.

[0081] like Figure 5 As shown, the monitoring and decision-making method of the integrated intelligent monitoring and decision-making system for overlapping shield tunnel systems is as follows:

[0082] S1. Based on previous engineering monitoring data, establish a numerical model to obtain numerical simulation data, analyze the model to obtain calculation data, and use the calculation data as training data; solve the problem of providing data support for the tunnel at the beginning of its service due to data scarcity by using other data.

[0083] S2. Construct a preset model based on a preset algorithm for the LSTM model. Use training data as the input variable for the LSTM model in the preset model to train the preset model, initially predict the deformation trend of the tunnel, and provide initial decisions for intelligent tunnel maintenance. The preset algorithm will not be further trained after training is completed and put into use. To avoid 'data pollution', the preset model is encapsulated and trained separately, isolated from the later feedback algorithm data and model. It plays a transitional support role in the early stage of operation and an auxiliary decision-making role in the later stage of operation.

[0084] S3. Collect data on tunnel structural internal forces, new and old tunnel defects, and tunnel structural vibration in the overlapping area to obtain real-time monitoring data. The real-time monitoring data is divided into status information and functional damage information, and the real-time monitoring data is normalized.

[0085] S4. A feedback model based on an LSTM model is constructed. The feedback algorithm is formed by iteratively updating the training results of the feedback model using changes in normalized real-time monitoring data. A real-time feedback mechanism is adopted, using normalized real-time monitoring data as the input variable of the LSTM model in the feedback model. The feedback model is trained, and the input features of the LSTM model in the feedback model are continuously adjusted according to the latest real-time monitoring data. The correlation between tunnel deformation and each monitoring variable is calculated in a fixed-length time window. Within each time window, the Spearman correlation coefficient is used to calculate the correlation between the monitoring variable and tunnel deformation. Features with high correlation are selected as the input features of the LSTM model in the feedback model. The prediction results output by the preset model and the prediction results output by the feedback model are combined to output the final predicted value of tunnel deformation. Intelligent tunnel maintenance decisions are given based on the correlation between maintenance measures and tunnel status. During actual monitoring and operation, the feedback algorithm continuously iterates and adjusts the weights of factors affecting tunnel deformation through iterative feedback, updating the intelligent tunnel maintenance decisions. All training data for the feedback model are actual measured data from the operation period of this project. No artificially calculated or simulated data is injected, aiming to eliminate the interference of data sources with human error and limitations on the algorithm. The feedback algorithm is a growth algorithm, which continues to train and learn during the tunnel's operation period and continuously iterates and optimizes.

[0086] S5. Based on a large amount of actual engineering data, establish an experience-based safety assessment system for overlapping tunnels, as shown in Table 1-4:

[0087] Table 1. Safety Level Assessment of Lateral Deformation of Shield Tunnels under the Horizontal Egg-Shaped Deformation Mode

[0088]

[0089] Table 2 Safety Level Assessment for Lateral Deformation of Shield Tunnels

[0090]

[0091] Table 3. Structural Safety Level Assessment Table within the 3D Range of the Longitudinal Settlement Trench

[0092]

[0093] The Tunnel Service Performance Index (TSI) is a comprehensive evaluation of the health status of tunnel sections based on tunnel defects (including uneven settlement, convergence deformation, leakage area, crack length, and segment damage area) as proposed in the "Technical Specification for Structural Safety Protection of Urban Rail Transit" (CJJ / T 202-2013). Its prediction formula is as follows:

[0094] ;

[0095] In the formula, This represents the average relative settlement. For average differential settlement; This represents the average convergence deformation change. The leakage area per 100 rings; The length of the crack per 100 rings; The area of ​​the stripping zone per 100 rings; , , , , , , All are undetermined regression parameters;

[0096] The TSI evaluation formula for soft soil areas is as follows:

[0097] ;

[0098] The TSI rating is based on the "Specification for Service Performance Appraisal of Shield Tunnel Structures" (DGTJ08-2123-2013), as shown in Table 4, and is divided into five safety levels:

[0099] Table 4. Tunnel Health Status Assessment Standards

[0100]

[0101] A comprehensive correlation analysis was established. In addition to the influencing factors of the traditional evaluation system, each factor was classified using sensors and image recognition. The collected data was further divided into state-related information and functional impairment information based on the collection category. State-related information included wheel-rail periodic wear, surface vibration, track slab vibration, vehicle body vibration, tunnel wall vibration, soil stress around overlapping tunnels, and pore water pressure. Functional impairment information included arch settlement, overhang heave, uneven settlement, tunnel seepage, segment misalignment, segment cracking, and cross-sectional deformation. In the correlation analysis, state-related information was linked to functional impairment information to further optimize the influencing factors and evaluation criteria.

[0102] Identify the inherent relationships between the various data points:

[0103] A fixed time window of 30 days was used, and the correlation between tunnel deformation and various monitored variables (such as soil stress and vibration) was recalculated every 30 days. Within each time window, the Spearman correlation coefficient was used to calculate the correlation between the monitored variables and tunnel deformation, and features with high correlation were selected as inputs. (Spearman correlation coefficient) The calculation formula is as follows:

[0104] ;

[0105] in, For each pair of observations, the rank difference This represents the number of data points.

[0106] like Figure 6 As shown, tunnel deformation typically exhibits a strong hysteresis effect. Therefore, when conducting correlation analysis, in addition to monitoring data of the current state variables, other factors must also be considered. In addition, lagged variables are introduced. , Calculate the effect of the capture variable on the functional impairment information variable respectively. The delayed effect, i.e. , and ,in, This is status information monitoring data from 15 days ago. This is monitoring data on status information from 30 days ago;

[0107] Within each time window, based on the results of correlation analysis, the correlation and lag effect are considered together, and the monitoring variable data with high correlation to functional impairment information variables are selected as the input data of the LSTM model.

[0108] Meanwhile, the weight of each influencing factor changes continuously with the operation cycle. By iterating, the weight of the influencing factors is continuously adjusted, thereby enabling real-time and accurate assessment of the tunnel's health status.

[0109] By combining real-time monitoring data, preset models, and feedback models to predict tunnel deformation trends, and establishing an iterative optimization evaluation system based on the correlation in the time dimension, the iterative optimization evaluation system deeply correlates state-related information and functional damage-related information through existing experience-overlapping tunnel safety evaluation systems and the inherent connections between various individual indicators in the monitoring system. It continuously corrects these issues within the operation cycle, thereby continuously iterating and updating the experience-overlapping tunnel safety evaluation system to optimize it and form a comprehensive health evaluation system for overlapping tunnel structures.

[0110] S6. Evaluate the intelligent maintenance decision of the tunnel based on the comprehensive health assessment system of overlapping tunnel structures, and determine the final intelligent maintenance decision of the tunnel based on the assessment results. As the tunnel is in operation, the real-time monitoring data and the intelligent maintenance decision of the tunnel are iterated accordingly, and the comprehensive health assessment system of overlapping tunnel structures and the final intelligent maintenance decision of the tunnel are also continuously iterated and optimized.

[0111] For the LSTM model used in both the preset model and the feedback model, tunnel deformation prediction is performed after the model training is completed and the evaluation is satisfactory. In the specific model, it is assumed that at time step... Sensor data is ,Right now for The input data at time t; the output predicted deformation of the LSTM model is .

[0112] The LSTM layer of the LSTM model learns the time-series features in the input data and controls the information flow through the forget gate, input gate, and output gate. The fully connected layer maps the time-series features learned by the LSTM layer onto the deformation prediction of the tunnel and outputs the prediction result.

[0113] The formula for calculating the forgetting gate is:

[0114] ;

[0115] in, This refers to the part of the memory state from the previous moment that needs to be retained. and These are the weight matrix and bias term of the forget gate, respectively; It is the hidden state from the previous moment, containing historical deformation information.

[0116] The formula for calculating the input gate is:

[0117] ;

[0118] in, The input gate vector controls the data that should be written into the current memory cell from the current input data. and These are the weight matrix and bias term of the input gate, respectively.

[0119] In addition, LSTM also computes candidate memory cells. , representing the potential information currently being input:

[0120] ;

[0121] in, and These are the weight matrix and bias term for the candidate memory units, respectively.

[0122] The formula for updating the state of a memory cell is:

[0123] ;

[0124] in, for The state of the memory unit at any given moment. for The state of the memory cell at time -1.

[0125] The formula for calculating the output gate is:

[0126] ;

[0127] in, The output gate vector controls the output of the hidden state at the current time step, determining the predicted value and state propagation. For the sigmoid function, and These are the weight matrix and bias term of the output gate, respectively.

[0128] Ultimately, the hidden state h t Generate using the following formula:

[0129] .

[0130] Hidden state output by LSTM layer The deformation will be mapped to the predicted tunnel through the fully connected layer. Output the prediction results:

[0131] .

[0132] During the pre-set model training process, the weight parameters are adjusted and the model parameters are optimized by minimizing the mean squared error (MSE) loss function; the MSE loss function is calculated as follows:

[0133] ;

[0134] in, It predicts the amount of tunnel deformation. Predicting the actual deformation of the tunnel. That is the number of samples.

[0135] The loss function in the feedback model considers both the real-time data error and the consistency of the pre-defined model prediction. The loss function in the feedback model is as follows:

[0136] ;

[0137] in, , The preset model and the feedback model are used to predict the deformation of the tunnel during the training process, respectively.

[0138] Once the model training is complete and the evaluation is satisfactory, it can be used to predict tunnel deformation. The dual-algorithm system has decision weights in recognition and decision-making. Initially, considering that the feedback algorithm is not trained or has been trained only slightly, the default algorithm decision has a larger weight, while the feedback algorithm has a smaller weight. The role of the weight weight is that when the two algorithms give decisions simultaneously, if the decisions of the two algorithms overlap, they are adopted; otherwise, the weighted values ​​are used to unify the deformation prediction.

[0139] To improve the stability and reliability of the prediction results, the prediction results output from the preset model and the feedback model are combined to output a final unified prediction value for tunnel deformation. :

[0140] ;

[0141] in, Output the predicted value of tunnel deformation for the preset model. The feedback model outputs a predicted value for tunnel deformation. The fusion weights represent the degree of confidence in the results of the preset model. In the initial stages of model operation, if sufficient real-time monitoring data is not yet available, the preset model can be used as the primary model with fixed weights. As the system runs and real-time data increases, the feedback model gradually stabilizes. At this point, error feedback can be used for dynamic weighting. The fusion weights are calculated by incorporating the model's recent prediction errors.

[0142] ;

[0143] in, The prediction error of the preset model, This refers to the prediction error of the feedback model. The smaller the error, the greater the prediction weight, achieving the purpose of dynamic trust adjustment. The weight of the feedback algorithm increases continuously as the operation cycle progresses; after the operation cycle reaches a certain stage, the feedback algorithm gradually matures and becomes dominant.

[0144] Tunnel maintenance decisions are made based on a combination of factors, including the tunnel's technical condition, operational requirements, and environmental factors. The main considerations are as follows:

[0145] The depth of the upper and lower tunnels is correlated, and the operational status information is deeply correlated with the functional impairment information, for example:

[0146] Tunnel structural defects (both upper and lower tunnels): lining cracks, segment misalignment, etc.; Track structural defects (both upper and lower tunnels): rail corrugation, increased track irregularities, etc.; Correlate with status information: soil surrounding the tunnel (both upper and lower tunnels): soil dynamic stress, pore water pressure, train ride smoothness, comfort (both upper and lower tunnels), surface vibration and vibration of surrounding buildings, soil displacement, soil dynamic acceleration, etc.; and then correlate maintenance measures with operational status information: adjust train speed, maintain tracks, add vibration reduction measures, replace tracks, adjust train dispatch frequency, and daily operation frequency based on issues such as pore water pressure, deviatoric stress, soil settlement, train ride smoothness, and comfort around the tunnel.

[0147] At the same time, disease treatment decisions are categorized:

[0148] The treatment decision is further divided into direct optimization measures for single diseases and related optimization measures, among which:

[0149] For a single disease, the direct optimization measures can be determined by judging the effect of the measures taken and choosing to maintain the measures or try other measures; after trying all measures, the optimal measures can be selected by comparison.

[0150] Correlation optimization measures can primarily regulate one or two factors while also providing some regulation and improvement for other disease factors, thus providing comprehensive regulation optimization measures; such measures yield multi-objective optimization measures after correlation analysis;

[0151] The final control measures can be selected by comprehensively comparing several correlation optimization measures and direct optimization measures through the correlation degree obtained by the decision system analysis, so as to achieve the overall optimal control and maintenance effect.

[0152] Based on the inspection results, the existing defects in the tunnel are assessed to determine their severity and development trend, and then corresponding treatment plans are formulated. For example:

[0153] Direct optimization measures: As the service life increases, mechanical and electrical equipment and fire-fighting facilities within the tunnel may age and become damaged. When the maintenance costs of the equipment are too high or it cannot meet the requirements for safe operation, replacement needs to be considered. For example, when the lighting fixtures in the tunnel are severely aged and their luminous efficiency is significantly reduced, they should be replaced with new energy-saving fixtures in a timely manner. Smaller cracks can be treated with surface sealing methods; while more serious cracks may require measures such as drilling and grouting reinforcement. When dealing with tunnel water leakage, materials such as polymer waterproof membranes and waterproof coatings can be selected; when reinforcing the tunnel lining, materials such as carbon fiber cloth and steel plates can be used. The optimal solution is determined through comparison and selection based on preliminary trials and feedback over a period of time.

[0154] Correlation optimization measures: Based on the specific conditions and types of defects in the tunnel, appropriate correlation optimization measures are selected. For example, measures such as adjusting vehicle speed and adjusting the number of daily operating trains are used to determine the correlation between each tunnel's operational status parameter and the maintenance adjustment measures, thereby selecting the optimal correlation optimization measure for each status.

[0155] Traditional detection systems only monitor and evaluate a single tunnel line, thus failing to consider the interactions and influences between overlapping tunnels. This invention's dual-algorithm decision-making system comprehensively evaluates and analyzes sensor and image acquisition information from both tunnel lines and their surroundings within overlapping tunnels. This not only satisfies the operational management needs of a single tunnel but also analyzes the mutual influences between the two tunnel lines within an overlapping system, enabling holistic decision-making. The analysis of tunnel operational defects is more comprehensive and thorough, and the corresponding measures of the decision-making system are more multifaceted. Through the intelligent combination of correlated optimization measures and direct optimization measures, the system achieves optimal overall operational management results. Furthermore, by utilizing the positive and negative feedback from tunnel structure and operation obtained after implementing maintenance decisions, the system further optimizes maintenance decisions, continuously iterating to find the best solutions to corresponding problems.

[0156] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A comprehensive intelligent monitoring and decision-making method for overlapping shield tunnel systems, characterized in that, Includes the following steps: S1. Based on previous engineering monitoring data, establish a numerical model to obtain numerical simulation data, analyze the model to obtain calculation data, and use the calculation data as training data. S2. Construct a preset model based on the LSTM model, use the training data as the input variable of the LSTM model in the preset model, train the preset model to obtain the preset algorithm, preliminarily predict the deformation trend of the tunnel, and give the initial decision for intelligent tunnel maintenance. S3. Collect data on the internal forces of the tunnel structure in the overlapping area, the defects of the new and old tunnels, and the vibration data of the tunnel structure to obtain real-time monitoring data. The real-time monitoring data is divided into status information and functional damage information. The real-time monitoring data is then normalized. S4. Construct a feedback model based on the LSTM model using the feedback algorithm. The feedback algorithm is formed by iteratively updating the training results of the feedback model based on the changes in the normalized real-time monitoring data. A real-time feedback mechanism is adopted, using normalized real-time monitoring data as the input variable of the LSTM model in the feedback model to train the feedback model. The input features of the LSTM model in the feedback model are continuously adjusted according to the latest real-time monitoring data. The correlation between tunnel deformation and each monitoring variable is calculated in a fixed-length time window. Within each time window, the Spearman correlation coefficient is used to calculate the correlation between the monitoring variable and tunnel deformation. Features with high correlation are selected as the input features of the LSTM model in the feedback model. The prediction results output by the preset model and the prediction results output by the feedback model are combined to output the final predicted value of tunnel deformation. Based on the correlation between maintenance measures and tunnel condition, intelligent tunnel maintenance decisions are given. The feedback algorithm monitors the real-time stress and displacement data of the tunnel and identifies the tunnel's operational status through image recognition during the actual operation and maintenance process. It continuously iterates and provides feedback, constantly adjusting the weights of factors affecting tunnel deformation through iterative feedback, and updating the intelligent maintenance decision-making for the tunnel. S5. Based on a large amount of actual engineering data, establish an experience-overlapping tunnel safety assessment system. Combine the experience-overlapping tunnel safety assessment system with the real-time monitoring data, the tunnel deformation trend predicted by the preset model and the feedback model, and establish an iterative optimization assessment system based on the correlation in the time dimension. The iterative optimization assessment system uses the inherent relationship between the experience-overlapping tunnel safety assessment system and each single indicator in the monitoring system, and uses correlation analysis to conduct a deep correlation in time between the actual tunnel operation status information and functional damage information. It continuously corrects and updates the experience-overlapping tunnel safety assessment system during the operation cycle, thereby continuously iterating and updating the system to optimize it and form a comprehensive health assessment system for overlapping tunnel structures. S6. Assess the tunnel condition based on the comprehensive health assessment system for overlapping tunnel structures, and intelligently match tunnel maintenance decisions based on the correlation between tunnel condition and maintenance decisions. As the tunnels operate, the correlation between tunnel status and tunnel maintenance decisions is iterated in real time, and the comprehensive health assessment system for overlapping tunnel structures and the final intelligent tunnel maintenance decisions are also continuously iterated and optimized.

2. The integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems as described in claim 1, characterized in that, The status information includes: soil stress around overlapping tunnels, pore water pressure, wheel-rail periodic wear, ground vibration, track slab vibration, vehicle vibration, and tunnel wall vibration; the functional damage information includes crown settlement, invert heave, uneven settlement, tunnel seepage, segment misalignment, segment cracking, and cross-sectional deformation.

3. The integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems as described in claim 1, characterized in that, The aforementioned experience-overlapping tunnel safety assessment system includes the assessment of the safety level of the transverse deformation of the shield tunnel, the assessment of the structural safety level within the 3D range of the longitudinal settlement trough, and the assessment of the tunnel's health status.

4. The integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems as described in claim 1, characterized in that, The LSTM model comprises an input layer, an LSTM layer, a fully connected layer, and an output layer. The LSTM layer learns the time-series features in the input data and controls the information flow through forget gates, input gates, and output gates. The fully connected layer maps the time-series features learned by the LSTM layer onto the tunnel deformation prediction and outputs the prediction result. ； in, To predict the amount of tunnel deformation, This refers to the hidden state output by the LSTM layer.

5. The integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems as described in claim 4, characterized in that, The prediction results output by the preset model and the prediction results output by the feedback model are combined to output the final predicted value of tunnel deformation. for: ； in, Output the predicted value of tunnel deformation for the preset model. The feedback model outputs a predicted value for tunnel deformation. The fusion weights, representing the degree of confidence in the results of the pre-defined model, are calculated by incorporating the model's recent prediction errors: ； in, The prediction error of the preset model, This represents the prediction error of the feedback model.

6. The integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems as described in claim 4, characterized in that, The calculation formula for the forget gate is as follows: ； in, This refers to the part of the memory state from the previous moment that needs to be retained. and These are the weight matrix and bias term of the forget gate, respectively; It is the hidden state of the previous moment, containing historical deformation information; for Input data at any given time; The calculation formula for the input gate is: ； in, The input gate vector controls the data that should be written into the current memory cell from the current input data. and These are the weight matrix and bias term of the input gate, respectively; In addition, LSTM also computes candidate memory cells. , representing the potential information currently being input: ； in, and These are the weight matrix and bias term of the candidate memory units, respectively; The formula for updating the state of a memory cell is: ； in, for The state of the memory unit at any given moment. for The state of the memory cell at time -1; The formula for calculating the output gate is: ； in, The output gate vector controls the output of the hidden state at the current time step, determining the predicted value and state propagation. For the sigmoid function, and These are the weight matrix and bias term of the output gate, respectively; The hidden state Generate using the following formula: 。 7. The integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems as described in claim 4, characterized in that, During the training of the preset model, the weight parameters are adjusted and the LSTM model parameters are optimized using the MSE loss function; the MSE loss function is calculated as follows: ； in, It predicts the amount of tunnel deformation. Predicting the actual deformation of the tunnel. It is the number of samples; The loss function in the feedback model simultaneously considers the consistency between real-time data errors and the predictions of the preset model. The loss function in the feedback model is as follows: ； in, , The preset model and the feedback model are used to predict the deformation of the tunnel during the training process, respectively.

8. The integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems as described in claim 1, characterized in that, The method for determining the inherent relationships between various data points is as follows: A fixed time window of 30 days was used, and the correlation between tunnel deformation and each monitored variable was recalculated every 30 days. Within each time window, the Spearman correlation coefficient was used. The calculation formula is as follows: ； In the formula, For each pair of observations, the rank difference The number of data points; When performing correlation analysis, in addition to the current state quantity information monitoring data... In addition, lagged variables are introduced. , Calculate the effect of the capture variable on the functional impairment information variable respectively. The delayed impact; among them, This is status information monitoring data from 15 days ago. This is monitoring data on status information from 30 days ago; Within each time window, based on the results of correlation analysis, the correlation and lag effects are considered together, and the monitoring variable data with high correlation to functional impairment information variables are selected as the input data of the LSTM model.

9. A system for implementing the integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems as described in any one of claims 1-8, characterized in that, include: Information Acquisition System: Used to collect data on internal forces, defects in new and old tunnels, and vibration data of tunnel structures in overlapping areas, and to monitor the safety status and operation of overlapping shield tunnels in real time; the information acquisition system transmits real-time monitoring data through a wireless transmission module; the information acquisition system includes a sensor monitoring module and an image monitoring module. The sensor monitoring module is used to monitor soil stress, pore water pressure, crown settlement, invert heave, uneven settlement, tunnel seepage, segment misalignment, segment cracking, and cross-sectional deformation around the overlapping tunnels; the image monitoring module... The module is used to monitor wheel-rail periodic wear, tunnel seepage, segment misalignment, segment cracking, cross-sectional deformation, ground vibration, track slab vibration, vehicle vibration, and tunnel wall vibration. The real-time monitoring data is divided into status information and functional damage information. Status information includes soil stress around overlapping tunnels, pore water pressure, arch settlement, invert heave, uneven settlement, tunnel seepage, wheel-rail periodic wear, ground vibration, track slab vibration, vehicle vibration, and tunnel wall vibration. Functional damage information includes segment misalignment, segment cracking, and cross-sectional deformation. The iterative optimization real-time assessment system includes an experience-based overlapping tunnel safety assessment system established based on a large amount of actual engineering data, and an iterative optimization assessment system that is adjusted in real time based on the actual tunnel operation status. The iterative optimization assessment system uses correlation analysis to deeply correlate information on the actual tunnel operation status with information on functional damage over time, thereby continuously iterating and updating the overlapping tunnel safety assessment system. Combining real-time monitoring data, preset models, and feedback models to predict tunnel deformation trends, an iterative optimization evaluation system is established based on the correlation in the time dimension. The iterative optimization evaluation system deeply correlates state-related information and functional damage-related information through existing experience in overlapping tunnel safety evaluation systems and the inherent connections between various single indicators in the monitoring system, and continuously corrects them within the operation cycle, thereby iteratively optimizing the comprehensive health evaluation system of overlapping tunnel structures. The intelligent maintenance decision-making system comprises an initial preset algorithm decision-making system and a feedback algorithm decision-making system. The initial preset algorithm decision-making system includes a preset model built based on an LSTM model with a preset algorithm, and the feedback algorithm decision-making system includes a feedback model built based on an LSTM model with a feedback algorithm. The feedback algorithm system, with the help of the initial preset algorithm decision-making system, continuously monitors and iterates feedback during the operation and maintenance process, continuously updating and improving itself. It provides intelligent maintenance decisions by collecting real-time stress and displacement data of the tunnel and identifying the tunnel's operational status through image recognition. It continuously optimizes maintenance decisions by analyzing the correlation between the adopted maintenance and control measures and the tunnel's status. Simultaneously, it iteratively optimizes the feedback algorithm decision-making system based on the adopted intelligent maintenance decisions and feedback on the tunnel's operational status.

10. The system for the integrated intelligent monitoring and decision-making method for overlapping shield tunnel systems as described in claim 9, characterized in that, The maintenance decision-making process is divided into direct optimization measures for single diseases and related optimization measures. Direct optimization measures are determined by judging the effect of the measures taken and choosing to maintain the measures or try other measures. Finally, the optimal measures are selected after trying all measures. Related optimization measures are obtained by comprehensively regulating multiple disease factors and obtaining multi-objective optimization measures after correlation analysis. Finally, the control measures of the maintenance decision are selected by comprehensively comparing the correlation obtained by the decision system analysis and combining direct optimization measures and related optimization measures to achieve the best macro-level multi-control operation and maintenance effect.

Images