Shield embankment settlement prediction and closed-loop control method based on multi-source parameter coupling modeling

Abstract

This invention discloses a method for predicting and controlling the settlement of shield tunnel embankments using multi-source parameter coupling modeling, specifically relating to the field of tunnels and underground engineering. The invention designs a closed-loop adaptive decision control method involving data acquisition, screening, prediction, interpretation, optimization, and feedback. Through a real-time feedback mechanism, the system continuously optimizes the prediction model and construction parameter adjustment strategies based on the actual effects during construction, thereby achieving dynamic adjustment and self-optimization capabilities during construction. By introducing interpretability processing, the settlement prediction model is transformed into an interpretable model, quantifying and clarifying the specific contribution and influence direction of each construction parameter on settlement changes. The system uses real-time feedback settlement monitoring data and construction status responses for incremental learning or online fine-tuning, enabling the settlement prediction model and multi-objective optimization model to continuously adapt to changes in strata and construction conditions, improving prediction accuracy and optimization effects, and ensuring the accuracy and robustness of long-term risk management during construction.

Description

Technical Field

[0001] This invention relates to the field of tunnels and underground engineering, specifically to a method for predicting and controlling settlement of shield tunnels using multi-source parameter coupling modeling. Background Technology

[0002] The shield tunneling method is one of the most commonly used tunnel construction methods in urban rail transit and cross-river tunnel projects. During the construction of shield tunnels under embankments, due to the poor stability of the embankment itself, the complex geological structure, and the significant changes in water and soil pressure, it is very easy to cause excessive surface settlement, which may lead to embankment deformation, leakage, or even safety accidents in severe cases.

[0003] A search revealed Chinese invention patent application CN113239439A, which proposes a surface settlement prediction system and method for shield tunneling construction. This method uses a meta-attribute extraction module, a settlement data generator training module, a settlement data generation module, and a real-time settlement prediction module to predict surface settlement during construction, improving the applicability and accuracy of the prediction, and eliminating the need for extensive data accumulation in the early stages.

[0004] Chinese invention patent application with publication number CN107092990A proposes a ground settlement prediction system and method for shield tunneling construction based on big data analysis. It utilizes a big data platform to collect, preprocess, extract features, and establish prediction models for shield tunneling construction data, and performs ground settlement prediction during the construction process, thereby improving the efficiency and real-time performance of settlement prediction.

[0005] In addition, Chinese invention patent application with publication number CN118013634A proposes a method and system for predicting ground settlement caused by shield tunneling construction. It discloses the construction of a spatial curve shield tunnel calculation model and an overall sinking model of the shield cross section using the shield machine's tunneling posture, and derives the analytical solution for settlement, thus theoretically realizing the prediction of ground settlement.

[0006] However, in practical engineering applications, the aforementioned publicly available settlement prediction methods and existing technologies generally have the following shortcomings:

[0007] On the one hand, existing methods fail to form a closed loop between prediction, attribution and regulation, resulting in the system as a whole being in an "open loop" state, unable to make dynamic adjustments and self-optimization based on actual construction feedback, and lacking the ability to continuously learn and adapt to engineering changes;

[0008] On the other hand, existing prediction models are mostly based on black-box algorithms, which cannot clearly explain the specific contribution of each construction parameter to settlement. This makes it difficult for construction personnel to obtain control basis from the prediction results, and they often rely on experience to adjust a single objective, rather than achieving quantitative collaborative optimization under multi-objective constraints. Summary of the Invention

[0009] To address the problems of difficulty in interpreting settlement prediction results and the disconnect between prediction results and construction control in existing technologies for shield tunneling, this invention provides a shield tunnel settlement prediction and closed-loop control method based on multi-source parameter coupling modeling. It proposes a closed-loop adaptive decision-making method applied to shield construction, consisting of "acquisition-screening-prediction-interpretation-optimization-feedback." This method predicts and attributes surface settlement and optimizes construction parameters to predict and protect against embankment deformation risks during shield construction, thus solving the problems mentioned in the background technology. The deformation includes at least the vertical displacement deformation corresponding to surface settlement.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting and controlling the settlement of a shield tunnel embankment using multi-source parameter coupling modeling, comprising the following steps:

[0011] S1: Collect multi-source construction parameter data during the shield tunneling process. The construction parameter data includes tunnel geometric parameter data, geological parameter data, and shield tunneling parameter data. Simultaneously record the corresponding surface settlement monitoring data to form an initial parameter set. Perform data preprocessing on the initial parameter set, including removing outliers, time alignment, using interpolation to fill in missing values, and normalizing the data to obtain a standardized multi-source parameter dataset.

[0012] S2: Based on the multi-source parameter dataset after data preprocessing, a correlation analysis is performed between the multi-source construction parameters and the surface settlement monitoring data to evaluate the degree of correlation between each construction parameter and the change in surface settlement. Based on the correlation analysis results, key construction parameter combinations that are highly correlated with the change in surface settlement are selected, and weakly correlated parameters are eliminated. The key construction parameter combinations are used as input variables for the subsequent settlement prediction model.

[0013] S3: Key construction parameters are used as inputs to the subsequent settlement prediction model, and surface settlement monitoring data are used as outputs to construct an XGBoost surface settlement prediction model to characterize the nonlinear mapping relationship between key construction parameters and surface settlement. A Bayesian optimization algorithm is introduced to adaptively optimize the core parameters of the XGBoost surface settlement prediction model. The core parameters include tree structure parameters (maximum depth, minimum child node weight) and training parameters (learning rate, number of subtrees). The optimal parameter combination is determined with the goal of minimizing the prediction error.

[0014] S4: The SHAP method is used to conduct interpretability analysis on the Bayesian-optimized XGBoost surface settlement prediction model. By calculating the SHAP values ​​of each key construction parameter, the contribution of each key construction parameter to the surface settlement prediction results is quantified. The positive and negative influence directions of each key construction parameter on surface settlement prediction are analyzed, and the construction parameters that play a dominant role in surface settlement prediction are identified, providing a quantitative basis for the optimization of construction parameters.

[0015] S5: Based on the contribution of key construction parameters and their positive and negative influence directions obtained in S4, and combined with construction safety constraints, equipment operation constraints and construction efficiency indicators, a multi-objective optimization model is constructed. A multi-objective optimization algorithm is used to collaboratively optimize the key construction parameters and generate a Pareto optimal solution set that satisfies the constraints, so as to minimize surface subsidence, maximize construction efficiency and minimize energy consumption.

[0016] For the Pareto optimal solution set, a multi-index decision screening process is performed on the Pareto optimal solution set, including:

[0017] (1) Dimensionless processing is performed on each objective function, and the objective weights are set according to the actual needs of construction;

[0018] (2) Based on the dimensionless results and target weights, a comprehensive evaluation calculation is performed on each Pareto solution to obtain the corresponding comprehensive goodness value, and the Pareto solutions are ranked accordingly.

[0019] (3) Based on the SHAP analysis results, impose constraints or penalty mechanisms on key parameters with high contribution to limit their drastic fluctuations in order to improve the stability and safety of the scheme. Specifically, identify key parameters with high contribution based on the average absolute SHAP value of each construction parameter to the predicted target, determine the stability constraint space of each key construction parameter according to the SHAP dependency, construct a penalty function associated with its SHAP contribution for parameters outside the stability constraint range, and introduce the penalty function into the comprehensive evaluation calculation of the Pareto solution to reduce the ranking priority of the solution corresponding to the high-sensitivity parameter. When any key parameter with high contribution exceeds the safety control boundary, the corresponding Pareto solution is removed.

[0020] (4) Select the combination of construction parameters with the best overall quality from the sorted Pareto solution set as the optimal solution;

[0021] The final determined optimal combination of construction parameters is used as a reference for adjusting specific construction processes, thereby achieving dynamic optimization and control of the tunnel boring process.

[0022] S6: Continuously collect multi-source construction parameter data and corresponding surface settlement monitoring data during subsequent construction, and compare and analyze the actual monitoring results with the model prediction results to calculate the prediction error and construction parameter response deviation; input new data into the surface settlement prediction model and multi-objective optimization model in real time, and optimize the prediction model parameters through incremental learning or online fine-tuning; adjust the weight allocation, constraint boundary and optimization objective priority of key construction parameters according to the model update results, realize the collaborative iterative optimization of the prediction model and the optimization decision model, and form a closed-loop control system of data acquisition—settlement prediction—parameter interpretation—multi-objective optimization—construction control—feedback correction, thereby continuously improving the accuracy of surface settlement prediction and the optimization effect of construction parameters.

[0023] Preferably, the correlation analysis adopts the grey relational analysis method, and the specific operation of the analysis method is as follows:

[0024] First, the multi-source construction parameter data and corresponding surface settlement monitoring data were subjected to min-max standardization to eliminate the influence of dimensional differences. Then, the correlation coefficient between each construction parameter and the vertical displacement of the ground surface was calculated, and the corresponding grey relational degree was also calculated. The construction parameters were then sorted according to the grey relational degree, and those with a grey relational degree higher than 70% were selected as input variables for the settlement prediction model. The calculation formula is as follows:

[0025] Initial parameter set:

[0026] ;

[0027] Indicates a reference feature, , representing the i-th reference feature; the reference features are the factors that affect settlement; Indicate target features, , representing the i-th target feature, which is a feature related to settlement index. , indicating the sample size;

[0028] Min-max normalization is performed on the reference and target features:

[0029] ;

[0030] In the formula, and These represent the standardized reference features and target features, respectively. , representing the i-th reference feature; , representing the j-th target feature;

[0031] Correlation coefficient:

[0032] ;

[0033] In the formula, Represents the correlation coefficient; The resolution coefficient, The smaller the value, the more significant the difference in the correlation coefficient; it is generally taken as 0.5.

[0034] Grey relational degree:

[0035] ;

[0036] In the formula, for and The grey relational degree is such that the closer the grey relational degree value is to 1, the stronger the correlation. n represents the number of features.

[0037] Based on the calculation results of the gray relational degree, construction parameters with a gray relational degree higher than 70% are selected to form a set of key construction parameters. The set of key construction parameters is sorted according to the time series of the construction process to construct a data sample set with key construction parameters as feature variables and surface settlement monitoring data at the corresponding time as target variables.

[0038] Preferably, the above data sample set is divided into training and testing sets. The first 70% of the data sample set arranged in chronological order of construction time is used as the training set, and the last 30% is used as the testing set. The set of key construction parameters is used as input feature variables and input into the XGBoost surface settlement prediction model for model training and prediction, thereby establishing the mapping relationship between key construction parameters and surface settlement. The XGBoost surface settlement prediction model is constructed using an XGBoost model based on a gradient boosting framework. This model uses classification and regression trees as base learners. By constructing multiple regression trees as base learners and continuously fitting the prediction residuals of previous models during iterative training, it gradually approximates the nonlinear mapping relationship between construction parameters and surface settlement. The expected boost acquisition function of the improved learner is:

[0039] ;

[0040] In the formula, Hyperparameters; threshold For observations A quantile, determined by the hyperparameter Decide; This indicates the construction parameters under the condition that the observed value is less than the threshold. The probability density of occurrence; This indicates that the parameter is set when the observed value is greater than or equal to the threshold. The probability density of occurrence It represents the probability density function of the observed value y occurring within the interval.

[0041] Preferably, an adaptive parameter optimization strategy based on historical sample prediction error feedback is adopted to adjust the structural and training parameters of the XGBoost surface subsidence prediction model. During the training process of the XGBoost surface subsidence prediction model, the error between the predicted and actual values ​​is compared, and this error is used to adjust the parameters of the XGBoost surface subsidence prediction model, such as the learning rate and regularization coefficient. This also includes adjustments to the model's structural parameters, including model complexity. Through this adaptive optimization strategy, the model can continuously learn and gradually improve its performance, especially when facing new data or environments. The specific adjustments are achieved through the following steps:

[0042] Prediction error feedback:

[0043] ;

[0044] In the formula, This represents the prediction error for the i-th sample; This represents the model's predicted values, providing computational data for the desired improvement of the acquisition function; Indicates the actual monitored value;

[0045] Learning rate adjustment formula:

[0046] ;

[0047] In the formula, This represents a control factor for adjusting the learning rate, controlling the magnitude of the adjustment. This indicates the prediction error. If the error is small, the learning rate can be increased appropriately to accelerate the learning process.

[0048] Regularization coefficient adjustment formula;

[0049] ;

[0050] In the formula, This represents the adjusted regularization coefficient; Indicates the current regularization coefficient; This represents the hyperparameter that controls the growth of the regularization coefficient; This indicates the prediction error.

[0051] Preferably, incremental learning is used to monitor the data collected during construction in real time and adjust it according to the new data to ensure that the settlement prediction is continuously optimized as construction progresses. Incremental learning is a machine learning method that allows the model to be updated and adjusted in real time when it receives new data without having to completely retrain the model. For settlement prediction models and multi-objective optimization models, incremental learning helps the model to be gradually optimized as the data changes during the construction process, thereby improving the model's prediction accuracy and decision-making ability.

[0052] Incremental learning formula:

[0053] ;

[0054] The parameters of the model can be adjusted using gradient descent or other optimization algorithms. Assuming the model's loss function is... ,in The parameters of the model are represented by the error of historical samples, and are updated through the steps described above; where... Indicates the current parameters of the model; This indicates the updated parameters. This represents the learning rate, which controls the step size for parameter updates; This represents the gradient of the loss function with respect to the model parameters;

[0055] Error-based adaptive optimization:

[0056] ;

[0057] In the formula, This represents the exponentially weighted average of the gradient, reflecting changes in the gradient. This represents a constant and is used to prevent division by zero errors.

[0058] The present invention has the following advantages:

[0059] This invention constructs a closed-loop adaptive decision-making system of data collection, screening, prediction, interpretation, optimization, and feedback. Through a real-time feedback mechanism, the system can continuously optimize the prediction model and construction parameter adjustment strategy based on the actual effects during construction, thereby achieving the ability to dynamically adjust and self-optimize during construction. Unlike existing black-box prediction models, this invention introduces interpretability processing to transform the settlement prediction model into an interpretable model, quantifying and clarifying the specific contribution and direction of influence of each construction parameter on settlement changes.

[0060] The system uses real-time feedback of settlement monitoring data and construction status response to perform incremental learning or online fine-tuning, enabling the settlement prediction model and multi-objective optimization model to continuously adapt to changes in strata and construction conditions, improve prediction accuracy and optimization effect, and ensure the accuracy and robustness of long-term risk management during construction. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the process for predicting and protecting the deformation of a large-diameter shield tunneling embankment according to the present invention.

[0062] Figure 2 This is a ranking diagram of feature contribution in an embodiment of the present invention;

[0063] Figure 3 This is a diagram of the base learner ensemble model in an embodiment of the present invention. Detailed Implementation

[0064] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] like Figure 1 As shown, this embodiment uses a tunnel boring machine (TBM) project on a subway line as an application example. During the TBM tunneling process, construction parameters and ground surface vertical displacement monitoring data are collected simultaneously to construct a settlement prediction model. Specifically, this includes:

[0066] Step 1: During the tunnel boring machine (TBM) excavation, extract multi-source construction parameter data related to the TBM's construction status from the TBM construction monitoring log, and collect ground surface vertical displacement monitoring data at the corresponding locations. These construction parameters include tunneling speed, cutterhead rotation speed, rolling angle, grout flow rate, grout discharge flow rate, deviation flow rate, grout specific gravity, discharge specific gravity, thrust, and torque. Combine the construction parameters corresponding to each tunneling cycle with the ground surface vertical displacement monitoring values ​​at the same time to form a sample data set. By continuously collecting construction data during the TBM excavation process, construct no fewer than 1000 sets of synchronous samples to form the original dataset, providing a data foundation for the subsequent construction of the settlement prediction model.

[0067] Step 2: To reduce the interference of redundant construction parameters on the settlement prediction model training and improve model training efficiency and prediction stability, this embodiment uses grey relational analysis to analyze the construction parameters. First, the construction parameter data and settlement monitoring data are standardized to eliminate the influence of dimensional differences. Then, the correlation coefficient between each construction parameter and the vertical displacement of the ground surface is calculated, and the corresponding grey relational degree is further calculated. The construction parameters are sorted according to the grey relational degree, and those with a grey relational degree higher than 70% are selected as input variables for the settlement prediction model for subsequent model construction. The calculation formula is as follows:

[0068] Initial parameter set:

[0069] ;

[0070] Indicates a reference feature, , representing the i-th reference feature; the reference features are the factors that affect settlement; Indicate target features, , representing the i-th target feature, which is a feature related to settlement index. , indicating the sample size;

[0071] Min-max normalization is performed on the reference and target features:

[0072] ;

[0073] In the formula, and These represent the standardized reference features and target features, respectively. , representing the i-th reference feature; , where j represents the j-th target feature.

[0074] Correlation coefficient:

[0075] ;

[0076] In the formula, Represents the correlation coefficient; The resolution coefficient, The smaller the value, the more significant the difference in the correlation coefficient; generally, 0.5 is taken.

[0077] Grey relational degree:

[0078] ;

[0079] In the formula, for and The grey relational degree is such that the closer the grey relational degree value is to 1, the stronger the correlation. n represents the number of features.

[0080] The results of the grey relational analysis are shown in Table 1 and Figure 2 As shown:

[0081] Table 1 Results of Grey Relational Analysis

[0082] Parameter name Slurry flow rate Slurry flow rate Specific gravity of pulp Specific gravity of feed tunneling speed Cutter head speed thrust Roll angle Torque Deviation flow Grey relational degree 0.725 0.710 0.709 0.696 0.691 0.667 0.662 0.647 0.601 0.551

[0083] Step 3: The surface subsidence prediction model is constructed using the XGBoost model based on the gradient boosting framework. This model uses classification and regression trees as base learners, and constructs multiple regression trees as base learners (e.g., ...). Figure 3 As shown in the figure, during iterative training, the prediction residuals of the preceding model are continuously fitted to establish a surface settlement prediction model to characterize the nonlinear mapping relationship between construction parameters and surface settlement. The model not only outputs the predicted settlement value but also the confidence interval, providing quantitative uncertainty information for subsequent risk assessment and multi-objective decision-making. The improved expected acquisition function in the prediction model is:

[0084] ;

[0085] In the formula, Hyperparameters; threshold Typically set as observation value One of the quantiles is determined by the hyperparameter; This is expressed as the parameter when the observed value is less than the threshold. The probability density of occurrence This is expressed as the parameter being equal to or greater than a threshold when the observed value is greater than or equal to the threshold. The probability density of occurrence This represents the probability density function of the parameter y occurring within the interval.

[0086] During tunnel boring machine (TBM) construction, an adaptive parameter optimization strategy based on historical sample prediction error feedback is employed to automatically adjust the structural and training parameters of the prediction model. During model training, the parameters are adjusted by comparing the error between predicted and actual values. When the prediction error is large, the learning rate is adjusted from 0.01 to 0.009 to ensure training stability; conversely, when the error is small, the learning rate can be increased to accelerate the learning process. Furthermore, the regularization coefficient is dynamically adjusted based on the model's error feedback to avoid overfitting. If the error is large, the regularization coefficient is adjusted from 0.1 to 0.115 to enhance the model's generalization ability. Specific adjustments can be achieved through the following steps:

[0087] Prediction error feedback:

[0088] ;

[0089] In the formula, This represents the prediction error for the i-th sample; This represents the model's predicted value; This represents the actual monitored value.

[0090] Learning rate adjustment formula:

[0091] ;

[0092] In the formula, This represents a control factor for adjusting the learning rate, controlling the magnitude of the adjustment. This represents the prediction error. If the error is small, the learning rate can be increased appropriately to accelerate the learning process.

[0093] Regularization coefficient adjustment formula;

[0094] ;

[0095] In the formula, This represents the adjusted regularization coefficient; Indicates the current regularization coefficient; This represents the hyperparameter that controls the growth of the regularization coefficient; This indicates the prediction error.

[0096] To evaluate the performance of the established shield tunnel settlement prediction model, the coefficient of determination (R²) was selected. 2 The root mean square error (RMSE) and absolute error (MAE) were used as evaluation indicators, and the calculation results are shown in Table 2. Experimental results show that the settlement prediction model after parameter adaptive optimization can accurately predict the trend of surface settlement during the shield tunneling under the embankment, and the prediction results have a high degree of consistency with the measured settlement values.

[0097] Table 2 Comparison of Indicators for Four Different Prediction Models

[0098] Predictive Model Coefficient of determination Root mean square error absolute error Support Vector Machine 0.8479 0.4883 0.3984 Random Forest 0.8874 0.4207 0.3416 XGBoost 0.9093 0.3735 0.3039 BOHB-XGBoost 0.9637 0.2463 0.1958

[0099] Step 4: By analyzing the feature contribution of the model output, the influence of each key construction parameter on the settlement prediction results is quantified, and the direction of influence of each construction parameter on surface settlement changes is clarified, thereby identifying the construction parameters that play a dominant role in settlement changes. The interpretability processing preferably employs a feature attribution method based on game theory to achieve an interpretable expression of the settlement prediction model's decision-making process.

[0100] Based on the interpretability analysis of the optimal prediction model, the contribution of features, in descending order, is as follows: slurry discharge specific gravity, cutterhead rotation speed, slurry feed specific gravity, slurry discharge flow rate, slurry feed flow rate, tunneling speed, and thrust. Slurry discharge specific gravity, cutterhead rotation speed, slurry feed specific gravity, and slurry discharge flow rate are the main construction parameters affecting surface settlement changes. Among them, slurry discharge specific gravity has the most significant impact on settlement, indicating that imbalance in the slurry system is an important cause of surface settlement. At the same time, appropriately increasing the slurry feed specific gravity can enhance the ground support capacity, thereby effectively reducing the risk of surface settlement.

[0101] Step 5: Based on the quantitative contribution analysis results obtained in Step 4, the slurry specific gravity, cutterhead rotation speed, slurry feed specific gravity, and slurry discharge flow rate were identified as key influencing parameters. A multi-objective optimization model was constructed with the core objectives of minimizing surface settlement, maintaining normal tunneling efficiency, and ensuring the stability of the shield's attitude, while incorporating multiple constraints such as equipment capacity, slurry mix ratio, and schedule requirements. Subsequently, a genetic algorithm was used to solve the model, generating a set of Pareto optimal solutions with different objective focuses. Scheme A focuses on settlement control (reducing the tunneling speed by 20% and simultaneously increasing the grouting pressure by 0.1 MPa), while Scheme B focuses on efficiency balance (reducing the tunneling speed by 10% and fine-tuning the cutterhead rotation speed). Finally, considering that the project is currently located in a high-risk section directly below the embankment, Scheme A, with the highest settlement control weight, was selected as the final decision and converted into an executable set of quantitative construction instructions, which was then transmitted to the shield control system for execution.

[0102] Step 6: Based on the latest feedback data obtained in Step 5, the settlement prediction model is fine-tuned online to dynamically adapt to changes in geological conditions and improve the accuracy of subsequent predictions. For the shield tunnel settlement prediction model and the multi-objective optimization model, the model will perform incremental learning based on the data from each tunneling cycle and real-time surface settlement monitoring data. For example, when geological conditions change during tunneling, the model will compare the error between the predicted surface settlement and the actual monitored value in real time, and make dynamic adjustments based on the error, such as increasing the tunneling speed from 0.5 m / min to 1.5 m / min, or the cutterhead speed from 0.1 rpm to 0.4 rpm, or adjusting the grout flow rate from 5 m³ / h to 30 m³ / h, to adapt to changes in geological conditions and thus improve the accuracy of subsequent predictions.

[0103] Incremental learning formula:

[0104] ;

[0105] The parameters of the model can be adjusted using gradient descent or other optimization algorithms. Assuming the model's loss function is... ,in The parameters of the model are represented by the error from historical samples, and are updated through the steps described above. In the formula... Indicates the current parameters of the model; This indicates the updated parameters. This represents the learning rate, which controls the step size for parameter updates; This represents the gradient of the loss function with respect to the model parameters.

[0106] Error-based adaptive optimization:

[0107] ;

[0108] In the formula, This represents the exponentially weighted average of the gradient, reflecting changes in the gradient. This represents a constant and is used to prevent division by zero errors.

[0109] After model initialization, parameter optimization is not fixed but rather an adaptive adjustment based on real-time monitoring data, part of the overall closed-loop system. For example, if the model's prediction error is large at a certain moment (0.5 cm), the model will adjust its learning rate and regularization coefficient based on the real-time data fed back in step 6. When the prediction error is greater than 0.3 cm, the learning rate will automatically decrease from 0.01 to 0.008 to ensure gradual model convergence; conversely, if the error is small, the learning rate can be increased to accelerate the training process. Through this closed-loop feedback mechanism, the system achieves a fundamental shift from single static control to continuous dynamic optimization, forming an intelligent control system with autonomous perception, evaluation, learning, and improvement capabilities. This ensures high-precision and robust control over the deformation risk of the embankment during long-term and complex shield tunneling construction.

[0110] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for predicting and controlling the settlement of a shield tunnel embankment using multi-source parameter coupling modeling, characterized by: Includes the following steps: S1: Collect multi-source construction parameter data during the shield tunneling process, and simultaneously record the corresponding surface settlement monitoring data to form an initial parameter set; The initial parameter set is preprocessed to obtain a standardized multi-source parameter dataset; S2: Based on the multi-source parameter dataset after data preprocessing, a correlation analysis is performed between the multi-source construction parameters and the surface settlement monitoring data to evaluate the degree of correlation between each construction parameter and the change in surface settlement. Based on the correlation analysis results, key construction parameter combinations that are highly correlated with the change in surface settlement are selected, and weakly correlated parameters are eliminated. The key construction parameter combinations are used as input variables for the subsequent settlement prediction model. S3: Construct an XGBoost surface settlement prediction model to characterize the nonlinear mapping relationship between key construction parameters and surface settlement; introduce a Bayesian optimization algorithm to adaptively optimize the core parameters of the XGBoost surface settlement prediction model, and determine the optimal parameter combination with the goal of minimizing the prediction error; S4: The SHAP method is used to conduct interpretability analysis on the Bayesian-optimized XGBoost surface settlement prediction model. By calculating the SHAP values ​​of each key construction parameter, the contribution of each key construction parameter to the surface settlement prediction results and the direction of its positive and negative influence are quantified. S5: Based on the contribution and positive and negative influence directions of the key construction parameters obtained in S4, and combined with construction safety constraints, equipment operation constraints and construction efficiency indicators, a multi-objective optimization model is constructed to collaboratively optimize the key construction parameters and generate a Pareto optimal solution set that satisfies the constraints. For the Pareto optimal solution set, a multi-index decision screening process is performed on the Pareto optimal solution set, and the construction parameter combination with the best comprehensive quality is selected as the optimal solution from the sorted Pareto solution set. The final determined optimal combination of construction parameters is used as a reference for adjusting specific construction processes, thereby achieving dynamic optimization and control of the tunnel boring process. S6: Continuously collect multi-source construction parameter data and corresponding surface settlement monitoring data during subsequent construction, compare and analyze the actual monitoring results with the model prediction results, calculate the prediction error and construction parameter response deviation; input new data into the surface settlement prediction model and multi-objective optimization model in real time, and optimize the prediction model parameters through incremental learning or online fine-tuning; adjust the weight allocation, constraint boundary and optimization objective priority of key construction parameters according to the model update results, so as to achieve collaborative iterative optimization of the prediction model and the multi-objective optimization model; Through incremental learning, data collected during construction is monitored in real time, and adjustments are made based on new data to ensure that settlement predictions are continuously optimized as construction progresses, as follows: Incremental learning formula: ； Assume the model's loss function is ,in The parameters of the model are represented by the error of historical samples, and are updated through the steps described above; where... Indicates the current parameters of the model; This indicates the updated parameters. This represents the learning rate, which controls the step size for parameter updates; This represents the gradient of the loss function with respect to the model parameters; Error-based adaptive optimization: ； In the formula, This represents the exponentially weighted average of the gradient, reflecting changes in the gradient. This represents a constant to prevent division by zero errors.

2. The method for predicting and controlling the settlement of a shield tunnel embankment using multi-source parameter coupling modeling as described in claim 1, characterized in that: The correlation analysis employed the grey relational analysis method, and the specific steps of the analysis are as follows: First, the multi-source construction parameter data and corresponding surface settlement monitoring data were subjected to min-max standardization. The correlation coefficients between each construction parameter and the vertical displacement of the ground surface were calculated, along with the corresponding grey relational degree. The construction parameters were then sorted according to their grey relational degree, and those with a grey relational degree higher than 70% were selected as input variables for the settlement prediction model. The calculation formula is as follows: Initial parameter set: ； Indicates a reference feature, , representing the i-th reference feature; The reference characteristics are the various factors that affect settlement; Indicate target features, , representing the i-th target feature, which is a feature related to settlement index. , indicating the sample size; Min-max normalization is performed on the reference and target features: ； In the formula, and These represent the standardized reference features and target features, respectively. , representing the i-th reference feature; , representing the j-th target feature; Correlation coefficient: ； In the formula, Represents the correlation coefficient; The resolution coefficient; Grey relational degree: ； In the formula, for and The grey relational degree is such that the closer the grey relational degree value is to 1, the stronger the correlation. n represents the number of features. Based on the calculation results of the gray relational degree, construction parameters with a gray relational degree higher than 70% are selected to form a set of key construction parameters. The set of key construction parameters is sorted according to the time series of the construction process to construct a data sample set with key construction parameters as feature variables and surface settlement monitoring data at the corresponding time as target variables.

3. The method for predicting and controlling the settlement of a shield tunnel embankment using multi-source parameter coupling modeling as described in claim 2, characterized in that: The data sample set is divided into training and testing sets. The first 70% of the data sample set arranged in chronological order of construction time is used as the training set, and the last 30% is used as the testing set. The set of key construction parameters is used as input feature variables and fed into the XGBoost surface settlement prediction model for model training and prediction, thereby establishing a mapping relationship between key construction parameters and surface settlement. The XGBoost surface settlement prediction model is constructed using an XGBoost model based on a gradient boosting framework. This model uses classification and regression trees as base learners. By constructing multiple regression trees as base learners and continuously fitting the prediction residuals of previous models during iterative training, it gradually approximates the nonlinear mapping relationship between construction parameters and surface settlement. The improved expected boost acquisition function of the base learner is: ； In the formula, Hyperparameters; threshold For observations A quantile, determined by the hyperparameter Decide; This indicates the construction parameters under the condition that the observed value is less than the threshold. The probability density of occurrence; This indicates that the parameter is set when the observed value is greater than or equal to the threshold. The probability density of occurrence It represents the probability density function of the observed value y occurring within the interval.

4. The method for predicting and controlling the settlement of a shield tunnel embankment using multi-source parameter coupling modeling as described in claim 1, characterized in that: An adaptive parameter optimization strategy based on historical sample prediction error feedback is adopted to adjust the structural and training parameters of the XGBoost surface subsidence prediction model. During the training process of the XGBoost surface subsidence prediction model, the error between the predicted and actual values ​​is compared, and this error is used to adjust the parameters of the XGBoost surface subsidence prediction model. The specific adjustment is achieved through the following steps: Prediction error feedback: ； In the formula, This represents the prediction error for the i-th sample; This represents the model's predicted values, providing computational data for the desired improvement of the acquisition function; Indicates the actual monitored value; Learning rate adjustment formula: ； In the formula, This represents a control factor for adjusting the learning rate, controlling the magnitude of the adjustment. Indicates prediction error; Regularization coefficient adjustment formula: ； In the formula, This represents the adjusted regularization coefficient; Indicates the current regularization coefficient; This represents the hyperparameter that controls the growth of the regularization coefficient; This indicates the prediction error.

Images