Shield tunnel settlement control reinforcement method based on multi-source data fusion driving

Abstract

The invention relates to a shield tunnel settlement control reinforcement method based on multi-source data fusion driving, which comprises the following steps: S1, data acquisition: acquiring multi-source heterogeneous original data through a sensor network and database interface deployed in a shield tunneling machine and a construction area; s2, data preprocessing; s3: big data analysis modeling: comprising the following sub-steps: S31: normalizing physical significance; s32, carrying out model training; s4, real-time settlement risk assessment: predicting future settlement volume and dividing risk levels to obtain a risk assessment report; s5, intelligent reinforcement decision and execution: parameters of grouting equipment or a shield tunneling machine are controlled for real-time reinforcement; and the reinforced effect data is fed back to the step S1 to form closed-loop control. Compared with the prior art, the method has the advantages that the accuracy, robustness and physical interpretability of the settlement prediction model are improved, high precision and self-optimization of the prediction model are realized, and active and accurate intervention, high integration, self-optimization and the like are realized.

Description

Technical Field

[0001] This invention relates to the field of ground settlement control and reinforcement technology, and in particular to a method for controlling and reinforcing the settlement of shield tunnels based on multi-source data fusion. Background Technology

[0002] With the continuous development of urban underground space, shield tunneling has become the mainstream technology for tunnel construction due to its safety and efficiency. However, when shield machines traverse complex geological environments and existing buildings and structures, how to accurately control ground settlement and ensure construction safety and the stability of the surrounding environment remains a core challenge for the industry. Traditional settlement control methods rely heavily on engineers' experience and judgment, and on delayed responses based on limited monitoring data, resulting in insufficient proactive prediction and control capabilities. In recent years, with the rise of the Internet of Things and big data technologies, applying big data analytics to intelligent control of shield tunneling construction has become a clear development trend in the industry. By deploying sensor networks to collect massive amounts of construction data and using machine learning models for settlement prediction to guide reinforcement decisions, the foresight and accuracy of settlement control can theoretically be significantly improved.

[0003] In existing technologies, a typical method for controlling the settlement of tunnel boring machines (TBMs) by combining big data analytics has emerged. This method typically includes the following steps: collecting multi-source data such as geological data, TBM parameters, and historical settlement data through a sensor system; preprocessing the data, including cleaning and normalization, to form a structured dataset; using this dataset to train a machine learning model to establish a settlement prediction model; and finally, based on the model's prediction results, performing risk classification and triggering corresponding reinforcement measures. This type of method combines data-driven prediction with engineering control, representing an advanced level in the field.

[0004] However, existing technologies have a technical problem: the inherent scale differences in multi-source heterogeneous data lead to distortions in the feature importance of the constructed settlement prediction models. Specifically, existing technologies typically use methods such as min-max normalization in the data preprocessing stage to simply scale features with different physical dimensions, such as the extremely wide range of shield tunneling thrust and the extremely narrow range of soil compressibility coefficient, to a uniform numerical range ([0,1]). This purely mathematical normalization process does not consider the vastly different physical meanings and data distribution characteristics behind the features. When splitting features, machine learning models naturally tend to select features with wider numerical ranges and more split points. This may lead to the model over-relying on certain "easily split" features (thrust) that are not most strongly correlated with settlement physics, while underestimating the true contribution of some features with narrower numerical ranges but crucial physical meaning (specific soil parameters).

[0005] This technical problem can have serious consequences in practical engineering. While the trained model may exhibit high accuracy in fitting historical datasets, its learned patterns have poor generalization ability and weak physical interpretability. When construction conditions change slightly—for example, a change in the thrust range due to a new tunnel boring machine model, or the project entering a new geological unit with subtle differences from the training data distribution—the model's predictive performance will plummet, resulting in severe inaccuracies. These inaccuracies will directly lead to misjudgments of risk—either triggering false alarms, causing unnecessary work stoppages and resource waste; or, more dangerously, underreporting true settlement risks, resulting in delayed or insufficient reinforcement measures, ultimately leading to excessive surface subsidence, or even major safety accidents such as tunnel damage or damage to surrounding buildings.

[0006] Therefore, there is an urgent need in this field for a new method for controlling the settlement of shield tunnels that can fundamentally solve the problem of feature importance distortion caused by the scale difference of multi-source data, so that big data analysis models can more realistically reflect the physical contribution of each feature parameter, thereby improving the accuracy, robustness and engineering applicability of the prediction model. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a shield tunnel settlement control and reinforcement method based on multi-source data fusion, which improves the accuracy, robustness and physical interpretability of the settlement prediction model, realizes high precision and self-optimization of the prediction model, and achieves active and precise intervention, high integration and self-optimization.

[0008] The objective of this invention can be achieved through the following technical solutions: This invention provides a method for settlement control and reinforcement of shield tunnels based on multi-source data fusion, comprising the following steps: S1: Data Acquisition: Collect multi-source heterogeneous raw data through sensor networks and database interfaces deployed in the tunnel boring machine and construction area. The data includes geological exploration data, real-time construction parameter data of the tunnel boring machine, historical settlement monitoring data, and environmental factor data. S2: Data Preprocessing: The multi-source heterogeneous raw dataset collected in S1 is subjected to data cleaning, outlier removal, format normalization, and multi-source data fusion to form a structured dataset. The format normalization mentioned above is to uniformly convert the numerical values ​​of physical quantities with different dimensions to the numerical range of [0, 1]. S3: Big Data Analysis and Modeling: Receives the structured dataset generated by S2. This includes the following sub-steps: S31: Physical Meaning Normalization: Based on domain knowledge, physical influence factor transformation functions are defined for various feature variables in the structured dataset to uniformly convert feature values ​​with different physical dimensions into dimensionless settlement risk contribution scores, forming a set of physical meaning normalized feature vectors. The core processing objective is to uniformly convert feature variables with different physical dimensions into dimensionless settlement risk contribution scores with clear physical meaning, so as to ensure the comparability of the magnitude of each feature and the consistency of physical interpretation in the subsequent modeling process. S32: Model Training: Train the settlement prediction model using the physically normalized feature vector set formed in S31; S4: Real-time settlement risk assessment: After processing the real-time data of the current shield tunneling construction through steps S2 and S31, the data is input into the settlement prediction model trained in S32 for inference, predicting the future settlement amount and classifying the risk level, and obtaining a risk assessment report. S5: Intelligent reinforcement decision-making and execution: Based on the risk assessment report obtained in S4, it automatically makes decisions and generates reinforcement instructions based on preset control logic, controls the grouting equipment or shield machine parameters for real-time reinforcement, and feeds back the reinforcement effect data to step S1 to form a closed-loop control.

[0009] Furthermore, in S1, the geological exploration data includes soil type, groundwater level and geotechnical parameters, and the real-time construction parameter data of the tunnel boring machine includes thrust, torque, tunneling speed and soil pressure.

[0010] Furthermore, in S31, the feature variables in the structured dataset include the tunnel boring machine thrust feature value F and the settlement risk contribution score of the tunnel boring machine thrust feature value F. Through function S = - |F - Foptimal| / Foptimal is calculated. , in, This is the optimal thrust value determined based on the current geological conditions.

[0011] Furthermore, in S31, the feature variables in the structured dataset include soil compressibility coefficient feature values. Characteristic value of soil compression coefficient Score of contribution to settlement risk Through function Scompression = (av - amin) / (amax - amin) is calculated. , in and These are the minimum and maximum values ​​of the soil compression coefficient in the training dataset, respectively.

[0012] Furthermore, in S31, the feature variables in the structured dataset include the tunnel boring machine torque feature value T, and the settlement risk contribution score of the tunnel boring machine torque feature value T is: , Where T represents the real-time torque value of the tunnel boring machine. This represents the optimal torque value, which is derived from the power balance equation. Combined with optimal thrust value The torque and thrust are derived from the cutterhead angular velocity ω, ensuring that the torque and thrust are matched between energy consumption and formation disturbance.

[0013] Furthermore, in S31, the feature variables in the structured dataset include the tunneling speed feature value V, and the settlement risk contribution score of the tunneling speed feature value V is: , Where V represents the current tunneling speed of the tunnel boring machine. This represents the optimal tunneling speed, which is determined by the tunneling efficiency model. The model input is the soil compressibility coefficient. The parameters of groundwater level and cutterhead cutting resistance are used to output the speed value that maximizes construction efficiency while minimizing the risk of settlement.

[0014] Through the above processing, all key feature variables related to tunnel boring machine (TBM) construction are uniformly converted into scores reflecting the degree of deviation from the optimal control state. The closer the value is to zero, the lower the risk contribution; the greater the deviation, the higher the risk contribution. The final output feature vector set... , It includes all settlement risk contribution scores after physical normalization, which not only maintains the correspondence with the original physical parameters, but also eliminates the problem of feature importance distortion caused by differences in different dimensions and scales.

[0015] Furthermore, in S32, the physical meaning normalized feature vector set is used to train a machine learning model using a random forest regression algorithm. Historical settlement data is used as training labels, and the model hyperparameters are optimized through cross-validation to generate a settlement prediction model that can accurately reflect the physical importance of each feature.

[0016] Furthermore, the hyperparameter optimization process in S32 employs Bayesian optimization to find the optimal combination of hyperparameters that minimizes the root mean square error within the hyperparameter search space. The search expression for the optimal hyperparameters is: , in This represents a set of hyperparameter configurations. This indicates the model trained under this configuration. This represents the optimal combination of hyperparameters that minimizes RMSE.

[0017] The hyperparameter optimization process employs the Bayesian optimization method, with the hyperparameter search space as its input. This space contains the number of decision trees N and the maximum depth. Minimum number of samples for leaf nodes These parameters directly affect the model's complexity and generalization performance. The optimization objective is to minimize the root mean square error calculated based on k-fold cross-validation. , Where m is the number of samples in the validation set. Let j be the actual settlement value of the j-th sample. This is the model's predicted value; the smaller the value, the closer the model's prediction is to the actual settlement.

[0018] Bayesian optimization approximates the relationship between hyperparameters and model error by constructing a probabilistic surrogate model of the objective function on the hyperparameter search space. This surrogate model typically employs a Gaussian process, characterized by its ability to estimate the function value distribution in unexplored regions under a finite number of evaluations. During iteration, the acquisition function selects the next hyperparameter point to be evaluated based on the information from the current surrogate model. A commonly used acquisition function is the expected improvement (EI). Its core idea is to balance exploring unknown regions with utilizing known optimal regions to maximize possible error improvement. Through this mechanism, Bayesian optimization avoids the exhaustive approach of traditional grid search, finding the global optimum or near-optimal solution with fewer evaluations, thus significantly improving the efficiency of hyperparameter optimization and enhancing the training efficiency of the settlement prediction model. It achieves higher prediction accuracy and stability while ensuring physical interpretability.

[0019] Specifically, S32 uses the physical meaning-normalized feature vector set from S31. As input, each element in this vector set is a dimensionless score representing the contribution of settlement risk, ensuring that different physical features participate in the modeling on the same order of magnitude. The training process employs a random forest regression algorithm, and its model is defined as follows: Where N represents the number of decision trees, and its value ranges from 1 to 2. The increase in quantity improves the stability of the model, but at the same time increases the computational cost. This represents the i-th basis learner, whose input is the normalized feature vector. The output is a predicted value for settlement. The maximum depth of each decision tree during its growth is limited to [value missing]. Its value range is This constraint prevents overfitting and maintains the model's generalization ability by controlling the complexity of the tree. The training process is executed on the distributed computing framework Spark, improving the efficiency of model building by dividing large-scale sample data across multiple computing nodes for parallel processing. Model parameters are selected using a grid search method in the parameter space. Exhaustive combination attempts are performed, and k-fold cross-validation is used for evaluation under each set of parameters. The performance evaluation metric is the root mean square error. , where m represents the number of samples in the validation set. This represents the true settlement value of the j-th sample. This represents the predicted settlement value from the corresponding model. A smaller RMSE value indicates lower prediction error and better model performance. By selecting the parameter combination that minimizes RMSE from the parameter search space, a trained and optimized settlement prediction model is finally obtained. This model achieves high predictive accuracy while maintaining interpretability of physical meaning, providing a reliable reasoning tool for the real-time settlement risk assessment in the subsequent S4 step.

[0020] Furthermore, in S4, the risk level is based on the predicted settlement value. Safety threshold Divide by comparison; Safety threshold As specified in the tunnel structure safety code, this represents the maximum allowable settlement during construction and operation. When the predicted value falls below a certain percentage range, the risk level is divided into different grades, with the following logic: , in The subsidence risk levels are categorized as follows: Low indicates low risk, Medium indicates medium risk, and High indicates high risk.

[0021] The input to S4 includes the trained settlement prediction model. and the current real-time feature vector from step S1 and processed by steps S2 and S31. This feature vector maintains consistency with the historical training data in terms of dimensions and physical meaning, thereby ensuring the effectiveness of inference computation. The processing first involves... Input to model The future predicted time points are obtained through forward reasoning calculations. Predicted settlement value , in This represents the predicted settlement value. This indicates a future point in time.

[0022] This tiered approach not only ensures consistency with safety standards but also enables rapid response to different levels of risk. This step ultimately generates a risk assessment report. , The report includes predicted subsidence. Corresponding risk level This information is passed as input to step S5 for subsequent intelligent reinforcement decisions and execution, thereby enabling real-time risk warning and dynamic control during the tunnel boring machine (TBM) construction process.

[0023] Settlement prediction model The structure is built based on the gradient boosting decision tree algorithm, and its input is a set of physically normalized feature vectors. Each element in this vector set represents the contribution of different construction and geological features to settlement risk. The model is formed by the additive ensemble of multiple weak predictors, and the overall expression is as follows: , Where K represents the total number of weak learners. The larger the value, the stronger the expressive power of the model, but the computational complexity also increases. η represents the learning rate, which controls the contribution of each weak learner to the overall model. Too large a value may lead to unstable convergence, while too small a value will increase the number of iterations. Let represent the k-th regression tree, whose function is to learn a set of mapping relationships by partitioning the feature space, so that the input features can more closely approximate the true settlement value. The model training process uses a forward stepwise algorithm. At the k-th iteration, the current model's predicted output is... , Then, the negative gradient of the loss function L with respect to the actual settlement at the predicted value is calculated, and this negative gradient is used as the new target residual. The formula for calculating the negative gradient is: , in This represents the true settlement value of the j-th training sample. This represents the predicted settlement value from the previous model. Let represent the residual of the j-th sample in the k-th iteration. A new regression tree is generated by fitting this residual sequence. This information is then incorporated into the model to gradually reduce prediction errors. This process continues in each iteration until the set number of iterations KK or the residual convergence condition is met. Through ensemble learning, the model continuously corrects its shortcomings in settlement prediction, resulting in a final settlement prediction model. It significantly improves both generalization ability and prediction accuracy, and maintains logical consistency with physical normalization characteristics, thus providing reliable support for real-time settlement risk assessment.

[0024] Furthermore, in S5, deterministic decision-making logic is executed based on the risk level; when At this time, the system needs to perform synchronous grouting to slow down the settlement trend. The formula for calculating the required grouting volume is as follows: , Where Q represents the grouting volume in cubic meters; k represents the permeability coefficient, which is obtained by querying data in the geological parameter database that matches the current soil type, and is used to characterize the permeability and diffusion capacity of the grout in the soil. This represents the settlement compensation amount; the value here is the predicted settlement amount. This refers to the settlement that needs to be compensated for through grouting. Based on the calculated Q, the system further sets the target grouting pressure. It generates synchronous grouting commands that include grouting volume and target pressure; like If the predicted settlement exceeds the safety threshold, the decision system generates a composite reinforcement instruction set, which includes shield parameter adjustment commands, synchronous grouting commands, and secondary reinforcement grouting commands. The shield parameter adjustment commands make the tunneling process more stable by appropriately reducing thrust and torque. Synchronous grouting is used to directly suppress the current settlement expansion trend. Secondary reinforcement grouting is carried out at key locations after construction to enhance the strength of the soil structure.

[0025] The input for S5 is the risk assessment report from step S4. The report includes predicted subsidence. and the corresponding risk level .

[0026] The final reinforcement control command output by this step is: , in This indicates a command to adjust the parameters of the tunnel boring machine. This indicates a secondary reinforcement grouting instruction. The result is sent to the executing agency in real time to ensure that reinforcement measures under different risk levels can be implemented efficiently, thereby forming a closed-loop safety control throughout the construction process.

[0027] The reinforcement control command The execution process is specifically manifested as a closed-loop pressure control and flow tracking process for the grouting pump. The input is the instruction sequence output from step S5, including the grouting volume Q and the target pressure. Where Q corresponds to the required grouting volume of the reinforced area. The settlement control threshold is calculated from the model output in step S3, ensuring that the formation reinforcement strength is met while avoiding formation disturbance caused by excessive pressure. The actuator controller uses a proportional-integral-derivative algorithm, with the core output being the pump speed control signal. The formula is , The error term is defined as follows: , This is a real-time grouting pressure curve. Parameters The proportional gain amplifies the real-time error to quickly correct the pump speed. To achieve integral gain, eliminate accumulated errors and ensure final pressure stability. , The differential gain is used to suppress pressure fluctuations and improve the system's dynamic response. This is achieved through... Real-time calculations allow the controller to continuously adjust. This allows for the adjustment of the pump's speed and instantaneous flow rate, ensuring the grouting process is close to... The process proceeds smoothly under stable conditions. The grouting volume Q is controlled by real-time accumulation from an integral flow sensor. When the accumulated grouting volume reaches the target value, a shutdown logic is triggered, thus completing the reinforcement task. This control process ensures that the grout diffuses uniformly under a stable pressure field, forming a continuous reinforced body, maintaining consistency with the stratum response parameters obtained from the normalized modeling in step S3, and achieving a dynamic closed loop for tunnel crossing settlement control.

[0028] Furthermore, the thrust signal filtering sub-step in S1 uses the raw thrust data collected by the sensors of the tunnel boring machine's hydraulic system. As input, this signal is accompanied by high-frequency noise generated by hydraulic pump valve switching and mechanical vibration during transmission and sampling. To obtain a stable and reliable thrust characteristic sequence, a first-order low-pass digital filter is used for real-time processing, and its discrete form is as follows: , in Indicates the current time The filtered output value, This is the filtering result from the previous time step. The smoothing coefficient is determined based on the main frequency response of the tunnel boring machine's hydraulic system, ensuring that high-frequency interference is suppressed without weakening the dynamic changes of the effective signal. The closer the value of is to 1, the closer the filtering result is to the original input. It has a faster response speed but weaker noise suppression capability; when A smaller value results in a smoother filter output and enhanced noise suppression, but a slower system response. Through the aforementioned iterative relationship, the filter weighted and fused the new input signal with the result from the previous moment in each sampling period, gradually attenuating high-frequency components outside the noise band and retaining only the low-frequency components relevant to the hydraulic system characteristics, thereby generating a smooth and reliable thrust signal. The signal is then transmitted to the data preprocessing stage in step S2 to ensure the accuracy and consistency of the subsequent normalization modeling and settlement prediction inputs, and to maintain logical connection with the physical normalization processing in step S3, so that the response characteristics of the control system to the tunnel boring machine thrust can be accurately mapped to the reinforcement control commands. The generation process.

[0029] The closed-loop control system of the method obtains actual settlement measurements from settlement monitoring points. The feedback is implemented by collecting this value in real time after the hardening process is completed and sending it back to the system database as a new input sample. This feedback data not only includes... It also relates to the tunnel boring machine's propulsion conditions and grouting execution commands at the time. and actual grouting pressure The process data are paired and stored to form structured time-series sample pairs, enabling the data acquisition step S1 to continuously expand into a dynamically updated historical working condition database. Subsequently, in the model training step S3, the settlement prediction model... The update process employs an incremental learning mechanism, the core of which lies in using new samples to fine-tune the original model parameters rather than completely retraining, thus maintaining the model's historical learning ability while enhancing its adaptability to new working conditions. Let the current model parameter vector be... The prediction residual corresponding to the new sample is The updated parameters are: , in L represents the learning rate, which controls the magnitude of parameter updates, and L is the value based on the residual. loss function, Indicates in The gradient at the location. Through this update, the prediction error of the model for new samples is partially offset, and the prediction capability gradually self-corrects in subsequent construction. As feedback data continues to enter the system, the prediction model can adapt to different geological conditions and construction disturbances, maintaining dynamic perception and prediction accuracy of settlement risk, thus forming a positive feedback loop with the reinforcement execution in step S5, enabling the settlement prediction and control capability to be continuously optimized and remain effective in the long term.

[0030] Compared with the prior art, the present invention has the following advantages: (1) Improved the accuracy, robustness, and physical interpretability of the settlement prediction model. Eliminated the distortion of feature importance caused by scale differences in multi-source heterogeneous data. Traditional methods use pure mathematical methods such as min-max normalization to process data without considering the physical meaning behind the features. This may cause machine learning models to overemphasize features with large numerical ranges but weak physical correlations (such as thrust), while underestimating features with small numerical ranges but key features (such as specific soil parameters), resulting in poor model generalization ability and inaccurate predictions when construction conditions change. This invention proposes a core step of "physical meaning normalization." Instead of simply scaling the data, it uses domain knowledge to define a physical influence factor transformation function for each feature, converting it into a dimensionless settlement risk contribution score. For tunnel boring machine thrust, the score reflects the degree to which it deviates from the optimal value; for soil compressibility coefficient, the score reflects its relative risk level under those geological conditions. After this transformation, all features are of the same order of magnitude and have clear physical meaning before entering the model. This allows the model to truly learn the physical contribution of each parameter to settlement, rather than being misled by the data scale. Therefore, the model's prediction results are not only more accurate but also have a clear physical mechanism, maintaining stable predictive performance even when encountering new geological conditions or tunnel boring machine parameters.

[0031] (2) High accuracy and self-optimization of the prediction model were achieved. By integrating learning and automated hyperparameter optimization, the potential of the data was fully explored, and the continuous evolution of the model was achieved by using closed-loop feedback, ensuring the long-term reliability of the system. Existing technology models may be relatively simple, and hyperparameters rely on manual experience for adjustment, making it difficult to capture complex nonlinear relationships; moreover, once the model is trained, it becomes fixed and cannot adapt to the dynamically changing environment during construction.

[0032] This invention employs ensemble learning algorithms such as Gradient Boosting Decision Tree (GBDT) to effectively learn the complex interactions between high-dimensional features, thereby enhancing predictive capabilities. Automated optimization: Advanced algorithms such as Bayesian optimization are used to automatically search for the optimal hyperparameter combination of the model, avoiding the blindness and inefficiency of manual parameter tuning and ensuring the model is always in an optimal performance state. Incremental learning mechanism: The system uses actual settlement monitoring data after reinforcement as feedback to perform online incremental updates to the prediction model. This system not only has high prediction accuracy under current working conditions but also adapts to changes in the construction process, forming an organic whole with learning capabilities, maintaining a highly reliable settlement early warning capability over the long term.

[0033] (3) Proactive and precise intervention was achieved. Intelligent prediction and decision execution were closely coupled to achieve adaptive and precise reinforcement control based on quantitative risk, effectively suppressing settlement. The prediction results based on the model were accurately compared with the safety threshold, and the risk was automatically divided into low, medium and high levels. Differentiated precise control: For different risk levels, the system automatically generates and executes differentiated reinforcement instructions. For example, the required precise grouting volume is automatically calculated for medium risk; for high risk, a composite reinforcement instruction set (adjusting shield parameters + synchronous grouting + secondary reinforcement grouting) is triggered. Closed-loop execution control: The PID control algorithm is used at the execution layer to perform closed-loop precise control of grouting pressure, etc., to ensure that the reinforcement measures are implemented stably and accurately according to the instructions.

[0034] This has enabled a shift in settlement control from "experience-driven" to "data model-driven," and from "extensive" to "precise." It can proactively prevent settlement risks and take the most appropriate intervention measures before or in the early stages of a risk, greatly improving control efficiency and safety assurance levels.

[0035] (4) High integration and self-optimization. By connecting each link into a closed loop through data flow, a highly integrated and self-optimizing intelligent construction solution is formed. This invention is a complete systems engineering project, encompassing the entire process from data acquisition (S1), preprocessing (S2), intelligent modeling (S3), risk assessment (S4), to intelligent decision-making and execution (S5), and feeding back the final results to the starting point, forming a complete "perception-analysis-decision-execution-learning" closed loop. This holistic design ensures smooth information flow and highly coordinated decision-making and action. It is no longer a mere accumulation of individual technologies, but an intelligent control system capable of self-perception, self-decision-execution, and self-optimization. Attached Figure Description

[0036] Figure 1 The flowchart shows a shield tunnel settlement control and reinforcement method driven by multi-source data fusion. Figure 2 A flowchart for preprocessing S2 data; Figure 3 A flowchart for normalizing the physical meaning of S31; Figure 4 A flowchart for real-time settlement risk assessment of S4 and intelligent reinforcement decision-making and execution of S5; Figure 5 This is a flowchart of the closed-loop feedback and model update for settlement control of shield tunnels. Detailed Implementation

[0037] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Component models, material names, connection structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0038] Example 1 This embodiment provides a method for controlling and reinforcing the settlement of shield tunnels based on multi-source data fusion, such as... Figure 1-5 As shown, it includes the following steps: S1: Data Acquisition: Collect multi-source heterogeneous raw data through sensor networks and database interfaces deployed in the tunnel boring machine and construction area. The data includes geological exploration data, real-time construction parameter data of the tunnel boring machine, historical settlement monitoring data, and environmental factor data. S2: Data Preprocessing: The multi-source heterogeneous raw dataset collected in S1 is subjected to data cleaning, outlier removal, format normalization, and multi-source data fusion to form a structured dataset. The format normalization mentioned above is to uniformly convert the numerical values ​​of physical quantities with different dimensions to the numerical range of [0, 1]. S3: Big Data Analysis and Modeling: Receives the structured dataset generated by S2. This includes the following sub-steps: S31: Physical Meaning Normalization: Based on domain knowledge, physical influence factor transformation functions are defined for various feature variables in the structured dataset to uniformly convert feature values ​​with different physical dimensions into dimensionless settlement risk contribution scores, forming a set of physical meaning normalized feature vectors. The core processing objective is to uniformly convert feature variables with different physical dimensions into dimensionless settlement risk contribution scores with clear physical meaning, so as to ensure the comparability of the magnitude of each feature and the consistency of physical interpretation in the subsequent modeling process. S32: Model Training: Train the settlement prediction model using the physically normalized feature vector set formed in S31; S4: Real-time settlement risk assessment: After processing the real-time data of the current shield tunneling construction through steps S2 and S31, the data is input into the settlement prediction model trained in S32 for inference, predicting the future settlement amount and classifying the risk level, and obtaining a risk assessment report. S5: Intelligent reinforcement decision-making and execution: Based on the risk assessment report obtained in S4, it automatically makes decisions and generates reinforcement instructions based on preset control logic, controls the grouting equipment or shield machine parameters for real-time reinforcement, and feeds back the reinforcement effect data to step S1 to form a closed-loop control.

[0039] Example 2 This embodiment provides a method for controlling and reinforcing the settlement of shield tunnels based on multi-source data fusion, such as... Figure 1-5 As shown, it includes the following steps: S1: Data Acquisition: Collect multi-source heterogeneous raw data through sensor networks and database interfaces deployed in the tunnel boring machine and construction area. The data includes geological exploration data, real-time construction parameter data of the tunnel boring machine, historical settlement monitoring data, and environmental factor data. S2: Data Preprocessing: The multi-source heterogeneous raw dataset collected in S1 is subjected to data cleaning, outlier removal, format normalization, and multi-source data fusion to form a structured dataset. The format normalization mentioned above is to uniformly convert the numerical values ​​of physical quantities with different dimensions to the numerical range of [0, 1]. S3: Big Data Analysis and Modeling: Receives the structured dataset generated by S2. This includes the following sub-steps: S31: Physical Meaning Normalization: Based on domain knowledge, physical influence factor transformation functions are defined for various feature variables in the structured dataset to uniformly convert feature values ​​with different physical dimensions into dimensionless settlement risk contribution scores, forming a set of physical meaning normalized feature vectors. The core processing objective is to uniformly convert feature variables with different physical dimensions into dimensionless settlement risk contribution scores with clear physical meaning, so as to ensure the comparability of the magnitude of each feature and the consistency of physical interpretation in the subsequent modeling process. S32: Model Training: Train the settlement prediction model using the physically normalized feature vector set formed in S31; S4: Real-time settlement risk assessment: After processing the real-time data of the current shield tunneling construction through steps S2 and S31, the data is input into the settlement prediction model trained in S32 for inference, predicting the future settlement amount and classifying the risk level, and obtaining a risk assessment report. S5: Intelligent reinforcement decision-making and execution: Based on the risk assessment report obtained in S4, it automatically makes decisions and generates reinforcement instructions based on preset control logic, controls the grouting equipment or shield machine parameters for real-time reinforcement, and feeds back the reinforcement effect data to step S1 to form a closed-loop control.

[0040] In a specific implementation, in S1, the geological exploration data includes soil type, groundwater level and geotechnical parameters, and the real-time construction parameter data of the tunnel boring machine includes thrust, torque, tunneling speed and soil pressure.

[0041] In a specific implementation, S31, the feature variables in the structured dataset include the shield machine thrust feature value F and the settlement risk contribution score of the shield machine thrust feature value F. Through function S = - |F - Foptimal| / Foptimal is calculated. , in, This is the optimal thrust value determined based on the current geological conditions.

[0042] In a specific implementation, S31, the feature variables in the structured dataset include soil compressibility coefficient feature values. Characteristic value of soil compression coefficient Score of contribution to settlement risk Through function Scompression = (av - amin) / (amax - amin) is calculated. , in and These are the minimum and maximum values ​​of the soil compression coefficient in the training dataset, respectively.

[0043] In a specific implementation, in S31, the feature variables in the structured dataset include the shield machine torque feature value T, and the settlement risk contribution score of the shield machine torque feature value T is: , Where T represents the real-time torque value of the tunnel boring machine. This represents the optimal torque value, which is derived from the power balance equation. Combined with optimal thrust value The torque and thrust are derived from the cutterhead angular velocity ω, ensuring that the torque and thrust are matched between energy consumption and formation disturbance.

[0044] In a specific implementation, S31, the feature variables in the structured dataset include the tunneling speed feature value V, and the settlement risk contribution score of the tunneling speed feature value V is: , Where V represents the current tunneling speed of the tunnel boring machine. This represents the optimal tunneling speed, which is determined by the tunneling efficiency model. The model input is the soil compressibility coefficient. The parameters of groundwater level and cutterhead cutting resistance are used to output the speed value that maximizes construction efficiency while minimizing the risk of settlement.

[0045] Through the above processing, all key feature variables related to tunnel boring machine (TBM) construction are uniformly converted into scores reflecting the degree of deviation from the optimal control state. The closer the value is to zero, the lower the risk contribution; the greater the deviation, the higher the risk contribution. The final output feature vector set... , It includes all settlement risk contribution scores after physical normalization, which not only maintains the correspondence with the original physical parameters, but also eliminates the problem of feature importance distortion caused by differences in different dimensions and scales.

[0046] In a specific implementation, in S32, the physical meaning normalized feature vector set is used to train a machine learning model using a random forest regression algorithm. Historical settlement data is used as training labels, and the model hyperparameters are optimized through cross-validation to generate a settlement prediction model that can accurately reflect the physical importance of each feature.

[0047] In a specific implementation, the hyperparameter optimization process in S32 employs the Bayesian optimization method to find the optimal hyperparameter combination that minimizes the root mean square error within the hyperparameter search space. The search expression for the optimal hyperparameters is: , in This represents a set of hyperparameter configurations. This indicates the model trained under this configuration. This represents the optimal combination of hyperparameters that minimizes RMSE.

[0048] The hyperparameter optimization process employs the Bayesian optimization method, with the hyperparameter search space as its input. This space contains the number of decision trees N and the maximum depth. Minimum number of samples for leaf nodes These parameters directly affect the model's complexity and generalization performance. The optimization objective is to minimize the root mean square error calculated based on k-fold cross-validation. , Where m is the number of samples in the validation set. Let j be the actual settlement value of the j-th sample. This is the model's predicted value; the smaller the value, the closer the model's prediction is to the actual settlement.

[0049] Bayesian optimization approximates the relationship between hyperparameters and model error by constructing a probabilistic surrogate model of the objective function on the hyperparameter search space. This surrogate model typically employs a Gaussian process, characterized by its ability to estimate the function value distribution in unexplored regions under a finite number of evaluations. During iteration, the acquisition function selects the next hyperparameter point to be evaluated based on the information from the current surrogate model. A commonly used acquisition function is the expected improvement (EI). Its core idea is to balance exploring unknown regions with utilizing known optimal regions to maximize possible error improvement. Through this mechanism, Bayesian optimization avoids the exhaustive approach of traditional grid search, finding the global optimum or near-optimal solution with fewer evaluations, thus significantly improving the efficiency of hyperparameter optimization and enhancing the training efficiency of the settlement prediction model. It achieves higher prediction accuracy and stability while ensuring physical interpretability.

[0050] Specifically, S32 uses the physical meaning-normalized feature vector set from S31. As input, each element in this vector set is a dimensionless score representing the contribution of settlement risk, ensuring that different physical features participate in the modeling on the same order of magnitude. The training process employs a random forest regression algorithm, and its model is defined as follows: Where N represents the number of decision trees, and its value ranges from 1 to 2. The increase in quantity improves the stability of the model, but at the same time increases the computational cost. This represents the i-th basis learner, whose input is the normalized feature vector. The output is a predicted value for settlement. The maximum depth of each decision tree during its growth is limited to [value missing]. Its value range is This constraint prevents overfitting and maintains the model's generalization ability by controlling the complexity of the tree. The training process is executed on the distributed computing framework Spark, improving the efficiency of model building by dividing large-scale sample data across multiple computing nodes for parallel processing. Model parameters are selected using a grid search method in the parameter space. Exhaustive combination attempts are performed, and k-fold cross-validation is used for evaluation under each set of parameters. The performance evaluation metric is the root mean square error. , where m represents the number of samples in the validation set. This represents the true settlement value of the j-th sample. This represents the predicted settlement value from the corresponding model. A smaller RMSE value indicates lower prediction error and better model performance. By selecting the parameter combination that minimizes RMSE from the parameter search space, a trained and optimized settlement prediction model is finally obtained. This model achieves high predictive accuracy while maintaining interpretability of physical meaning, providing a reliable reasoning tool for the real-time settlement risk assessment in the subsequent S4 step.

[0051] In a specific implementation, in S4, the risk level is based on the predicted settlement value. Safety threshold Divide by comparison; Safety threshold As specified in the tunnel structure safety code, this represents the maximum allowable settlement during construction and operation. When the predicted value falls below a certain percentage range, the risk level is divided into different grades, with the following logic: , in The subsidence risk levels are categorized as follows: Low indicates low risk, Medium indicates medium risk, and High indicates high risk.

[0052] The input to S4 includes the trained settlement prediction model. and the current real-time feature vector from step S1 and processed by steps S2 and S31. This feature vector maintains consistency with the historical training data in terms of dimensions and physical meaning, thereby ensuring the effectiveness of inference computation. The processing first involves... Input to model The future predicted time points are obtained through forward reasoning calculations. Predicted settlement value , in This represents the predicted settlement value. This indicates a future point in time.

[0053] This tiered approach not only ensures consistency with safety standards but also enables rapid response to different levels of risk. This step ultimately generates a risk assessment report. , The report includes predicted subsidence. Corresponding risk level This information is passed as input to step S5 for subsequent intelligent reinforcement decisions and execution, thereby enabling real-time risk warning and dynamic control during the tunnel boring machine (TBM) construction process.

[0054] Settlement prediction model The structure is built based on the gradient boosting decision tree algorithm, and its input is a set of physically normalized feature vectors. Each element in this vector set represents the contribution of different construction and geological features to settlement risk. The model is formed by the additive ensemble of multiple weak predictors, and the overall expression is as follows: , Where K represents the total number of weak learners. The larger the value, the stronger the expressive power of the model, but the computational complexity also increases. η represents the learning rate, which controls the contribution of each weak learner to the overall model. Too large a value may lead to unstable convergence, while too small a value will increase the number of iterations. Let represent the k-th regression tree, whose function is to learn a set of mapping relationships by partitioning the feature space, so that the input features can more closely approximate the true settlement value. The model training process uses a forward stepwise algorithm. At the k-th iteration, the current model's predicted output is... , Then, the negative gradient of the loss function L with respect to the actual settlement at the predicted value is calculated, and this negative gradient is used as the new target residual. The formula for calculating the negative gradient is: , in This represents the true settlement value of the j-th training sample. This represents the predicted settlement value from the previous model. Let represent the residual of the j-th sample in the k-th iteration. A new regression tree is generated by fitting this residual sequence. This information is then incorporated into the model to gradually reduce prediction errors. This process continues in each iteration until the set number of iterations KK or the residual convergence condition is met. Through ensemble learning, the model continuously corrects its shortcomings in settlement prediction, resulting in a final settlement prediction model. It significantly improves both generalization ability and prediction accuracy, and maintains logical consistency with physical normalization characteristics, thus providing reliable support for real-time settlement risk assessment.

[0055] In a specific implementation, in S5, deterministic decision-making logic is executed based on the risk level; when At this time, the system needs to perform synchronous grouting to slow down the settlement trend. The formula for calculating the required grouting volume is as follows: , Where Q represents the grouting volume in cubic meters; k represents the permeability coefficient, which is obtained by querying data in the geological parameter database that matches the current soil type, and is used to characterize the permeability and diffusion capacity of the grout in the soil. This represents the settlement compensation amount; the value here is the predicted settlement amount. This refers to the settlement that needs to be compensated for through grouting. Based on the calculated Q, the system further sets the target grouting pressure. It generates synchronous grouting commands that include grouting volume and target pressure; like If the predicted settlement exceeds the safety threshold, the decision system generates a composite reinforcement instruction set, which includes shield parameter adjustment commands, synchronous grouting commands, and secondary reinforcement grouting commands. The shield parameter adjustment commands make the tunneling process more stable by appropriately reducing thrust and torque. Synchronous grouting is used to directly suppress the current settlement expansion trend. Secondary reinforcement grouting is carried out at key locations after construction to enhance the strength of the soil structure.

[0056] The input for S5 is the risk assessment report from step S4. The report includes predicted subsidence. and the corresponding risk level .

[0057] The final reinforcement control command output by this step is: , in This indicates a command to adjust the parameters of the tunnel boring machine. This indicates a secondary reinforcement grouting instruction. The result is sent to the executing agency in real time to ensure that reinforcement measures under different risk levels can be implemented efficiently, thereby forming a closed-loop safety control throughout the construction process.

[0058] The reinforcement control command The execution process is specifically manifested as a closed-loop pressure control and flow tracking process for the grouting pump. The input is the instruction sequence output from step S5, including the grouting volume Q and the target pressure. Where Q corresponds to the required grouting volume of the reinforced area. The settlement control threshold is calculated from the model output in step S3, ensuring that the formation reinforcement strength is met while avoiding formation disturbance caused by excessive pressure. The actuator controller uses a proportional-integral-derivative algorithm, with the core output being the pump speed control signal. The formula is , The error term is defined as follows: , This is a real-time grouting pressure curve. Parameters The proportional gain amplifies the real-time error to quickly correct the pump speed. To achieve integral gain, eliminate accumulated errors and ensure final pressure stability. , The differential gain is used to suppress pressure fluctuations and improve the system's dynamic response. This is achieved through... Real-time calculations allow the controller to continuously adjust. This allows for the adjustment of the pump's speed and instantaneous flow rate, ensuring the grouting process is close to... The process proceeds smoothly under stable conditions. The grouting volume Q is controlled by real-time accumulation from an integral flow sensor. When the accumulated grouting volume reaches the target value, a shutdown logic is triggered, thus completing the reinforcement task. This control process ensures that the grout diffuses uniformly under a stable pressure field, forming a continuous reinforced body, maintaining consistency with the stratum response parameters obtained from the normalized modeling in step S3, and achieving a dynamic closed loop for tunnel crossing settlement control.

[0059] In a specific implementation, the thrust signal filtering sub-step in S1 uses the raw thrust data collected by the sensors of the tunnel boring machine's hydraulic system. As input, this signal is accompanied by high-frequency noise generated by hydraulic pump valve switching and mechanical vibration during transmission and sampling. To obtain a stable and reliable thrust characteristic sequence, a first-order low-pass digital filter is used for real-time processing, and its discrete form is as follows: , in Indicates the current time The filtered output value, This is the filtering result from the previous time step. The smoothing coefficient is determined based on the main frequency response of the tunnel boring machine's hydraulic system, ensuring that high-frequency interference is suppressed without weakening the dynamic changes of the effective signal. The closer the value of is to 1, the closer the filtering result is to the original input. It has a faster response speed but weaker noise suppression capability; when A smaller value results in a smoother filter output and enhanced noise suppression, but a slower system response. Through the aforementioned iterative relationship, the filter weighted and fused the new input signal with the result from the previous moment in each sampling period, gradually attenuating high-frequency components outside the noise band and retaining only the low-frequency components relevant to the hydraulic system characteristics, thereby generating a smooth and reliable thrust signal. The signal is then transmitted to the data preprocessing stage in step S2 to ensure the accuracy and consistency of the subsequent normalization modeling and settlement prediction inputs, and to maintain logical connection with the physical normalization processing in step S3, so that the response characteristics of the control system to the tunnel boring machine thrust can be accurately mapped to the reinforcement control commands. The generation process.

[0060] The closed-loop control system of the method obtains actual settlement measurements from settlement monitoring points. The feedback is implemented by collecting this value in real time after the hardening process is completed and sending it back to the system database as a new input sample. This feedback data not only includes... It also relates to the tunnel boring machine's propulsion conditions and grouting execution commands at the time. and actual grouting pressure The process data are paired and stored to form structured time-series sample pairs, enabling the data acquisition step S1 to continuously expand into a dynamically updated historical working condition database. Subsequently, in the model training step S3, the settlement prediction model... The update process employs an incremental learning mechanism, the core of which lies in using new samples to fine-tune the original model parameters rather than completely retraining, thus maintaining the model's historical learning ability while enhancing its adaptability to new working conditions. Let the current model parameter vector be... The prediction residual corresponding to the new sample is The updated parameters are: , in L represents the learning rate, which controls the magnitude of parameter updates, and L is the value based on the residual. loss function, Indicates in The gradient at the location. Through this update, the prediction error of the model for new samples is partially offset, and the prediction capability gradually self-corrects in subsequent construction. As feedback data continues to enter the system, the prediction model can adapt to different geological conditions and construction disturbances, maintaining dynamic perception and prediction accuracy of settlement risk, thus forming a positive feedback loop with the reinforcement execution in step S5, enabling the settlement prediction and control capability to be continuously optimized and remain effective in the long term.

[0061] Components not described in detail in this embodiment are all existing components that can be purchased through public channels.

[0062] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A method for settlement control and reinforcement of shield tunnels based on multi-source data fusion, characterized in that, Includes the following steps: S1: Data Acquisition: Collect multi-source heterogeneous raw data through sensor networks and database interfaces deployed in the tunnel boring machine and construction area. The data includes geological exploration data, real-time construction parameter data of the tunnel boring machine, historical settlement monitoring data, and environmental factor data. S2: Data Preprocessing: The multi-source heterogeneous raw dataset collected in S1 is subjected to data cleaning, outlier removal, format normalization, and multi-source data fusion to form a structured dataset. ; S3: Big Data Analysis and Modeling: Receives the structured dataset generated by S2. This includes the following sub-steps: S31: Physical meaning normalization: Define physical influence factor transformation functions for each type of feature variable in the structured dataset, and uniformly convert feature values ​​of different physical dimensions into dimensionless settlement risk contribution scores to form a set of physical meaning normalized feature vectors. S32: Model Training: Train the settlement prediction model using the physically normalized feature vector set formed in S31; S4: Real-time settlement risk assessment: After processing the real-time data of the current shield tunneling construction through steps S2 and S31, the data is input into the settlement prediction model trained in S32 for inference, predicting the future settlement amount and classifying the risk level, and obtaining a risk assessment report. S5: Intelligent reinforcement decision-making and execution: Based on the risk assessment report obtained in S4, it automatically makes decisions and generates reinforcement instructions based on preset control logic, controls the grouting equipment or shield machine parameters for real-time reinforcement, and feeds back the reinforcement effect data to step S1 to form a closed-loop control.

2. The method for settlement control and reinforcement of shield tunnels based on multi-source data fusion as described in claim 1, characterized in that, In S1, the geological exploration data includes soil type, groundwater level and geotechnical parameters, and the real-time construction parameter data of the tunnel boring machine includes thrust, torque, tunneling speed and soil pressure.

3. The method for settlement control and reinforcement of shield tunnels based on multi-source data fusion as described in claim 1, characterized in that, In S31, the feature variables in the structured dataset include the shield machine thrust feature value F and the settlement risk contribution score of the shield machine thrust feature value F. Through function S = - |F - Foptimal| / Foptimal is calculated. ， in, This is the optimal thrust value determined based on the current geological conditions.

4. The method for settlement control and reinforcement of shield tunnels based on multi-source data fusion as described in claim 1, characterized in that, In S31, the feature variables in the structured dataset include soil compression coefficient feature values. Characteristic value of soil compression coefficient Score of contribution to settlement risk Through function Scompression = (av - amin) / (amax - amin) is calculated. ， in and These are the minimum and maximum values ​​of the soil compression coefficient in the training dataset, respectively.

5. The method for settlement control and reinforcement of shield tunnels based on multi-source data fusion as described in claim 1, characterized in that, In S31, the feature variables in the structured dataset include the shield machine torque feature value T, and the settlement risk contribution score of the shield machine torque feature value T is: ， Where T represents the real-time torque value of the tunnel boring machine. This represents the optimal torque value, which is derived from the power balance equation. Combined with optimal thrust value The torque and thrust are derived from the cutterhead angular velocity ω, ensuring that the torque and thrust are matched between energy consumption and formation disturbance.

6. The method for settlement control and reinforcement of shield tunnels based on multi-source data fusion as described in claim 1, characterized in that, In S31, the feature variables in the structured dataset include the tunneling speed feature value V, and the settlement risk contribution score of the tunneling speed feature value V is: ， Where V represents the current tunneling speed of the tunnel boring machine. This represents the optimal tunneling speed, which is determined by a tunneling efficiency model. The model input is the soil compressibility coefficient. The parameters of groundwater level and cutterhead cutting resistance are used to output the speed value that maximizes construction efficiency while minimizing the risk of settlement.

7. The method for settlement control and reinforcement of shield tunnels based on multi-source data fusion as described in claim 1, characterized in that, In S32, the physical meaning normalized feature vector set is used, and a random forest regression algorithm is used to train the machine learning model. Historical settlement data is used as training labels, and the model hyperparameters are optimized through cross-validation to generate a settlement prediction model that can accurately reflect the physical importance of each feature.

8. The method for settlement control and reinforcement of shield tunnels based on multi-source data fusion as described in claim 7, characterized in that, The hyperparameter optimization process in S32 employs Bayesian optimization, searching for the optimal combination of hyperparameters that minimizes the root mean square error within the hyperparameter search space. The search expression for the optimal hyperparameters is: ， in This represents a set of hyperparameter configurations. This represents the model trained under a certain set of hyperparameters. This represents the optimal combination of hyperparameters that minimizes RMSE.

9. The method for settlement control and reinforcement of shield tunnels based on multi-source data fusion as described in claim 1, characterized in that, In S4, the risk level is based on the predicted settlement value. Safety threshold Divide by comparison; Safety threshold As specified in the tunnel structure safety code, this represents the maximum allowable settlement during construction and operation. When the predicted value falls below a certain percentage range, the risk level is divided into different grades, with the following logic: ， in The subsidence risk levels are categorized as follows: Low indicates low risk, Medium indicates medium risk, and High indicates high risk.

10. The method for settlement control and reinforcement of shield tunnels based on multi-source data fusion as described in claim 1, characterized in that, In S5, deterministic decision-making logic is executed based on the risk level; when At this time, the system needs to perform synchronous grouting to slow down the settlement trend. The formula for calculating the required grouting volume is as follows: ， Where Q represents the grouting volume in cubic meters; k represents the permeability coefficient, which is obtained by querying data in the geological parameter database that matches the current soil type, and is used to characterize the permeability and diffusion capacity of the grout in the soil. This represents the settlement compensation amount; the value here is the predicted settlement amount. This refers to the settlement that needs to be compensated for through grouting; based on the calculated Q, the system further sets the target grouting pressure. It generates synchronous grouting commands that include grouting volume and target pressure; like If the predicted settlement exceeds the safety threshold, the decision system generates a composite reinforcement instruction set, which includes shield parameter adjustment commands, synchronous grouting commands, and secondary reinforcement grouting commands. The shield parameter adjustment commands make the tunneling process more stable by appropriately reducing thrust and torque. Synchronous grouting is used to directly suppress the current settlement expansion trend. Secondary reinforcement grouting is carried out at key locations after construction to enhance the strength of the soil structure.

Images