A physical-data double-driven shield tunnel segment floating prediction and control system and method
By constructing a physical-data dual-driven shield tunnel segment uplift prediction and control system, and combining physical models with data-driven methods, the problems of accuracy and interpretability in segment uplift prediction during shield tunnel construction were solved, achieving high-precision prediction and closed-loop control.
Patent Information
- Application Number
- CN202610797262.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-04
- Publication Date
- 2026-08-25
AI Technical Summary
In existing shield tunnel construction, the methods for predicting and controlling segment uplift have problems such as difficulty in accurately determining parameters, oversimplification of load and boundary conditions, and inability to handle nonlinear changes in foundation stiffness. This results in a large deviation between theoretical predictions and actual monitoring values. Furthermore, data-driven methods lack physical constraints and have poor interpretability, insufficient generalization ability, and control decisions are mostly post-hoc remedial measures, lacking closed-loop management capabilities.
A physics-data dual-driven approach is adopted. By constructing a physical model based on Timoshenko beams and Winkler foundations, and combining spatiotemporal alignment and fusion of multi-source data, physical information neural networks and XGBoost models are used for prediction. The SHAP method is introduced for interpretability analysis to generate construction parameter optimization or shield attitude adjustment strategies.
It significantly improved prediction accuracy, enhanced the model's credibility and generalization ability, and realized closed-loop intelligent control from post-event remediation to pre-event intervention, effectively managing the floating of shield tunnel segments.
Smart Images
Figure CN122632697A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent control for shield tunnel construction, and particularly relates to a physical-data dual-driven shield tunnel segment floating prediction and control system and method. Background Technology
[0002] During shield tunnel construction, after the tunnel lining segments emerge from the shield tail, they are prone to uplift under the combined effects of grouting buoyancy, groundwater pressure, and ground reaction forces. Excessive uplift or drastic fluctuations can lead to quality defects such as segment misalignment, cracking, leakage, and joint opening. In severe cases, it can cause the tunnel axis to exceed limits, threatening structural safety and service life. The uplift problem is particularly prominent in complex strata (such as soft upper and hard lower strata, or water-rich strata) and large-diameter tunnels. Currently, the methods for predicting and controlling segment uplift are mainly divided into two categories: one is the physical model-driven method, such as simplifying the tunnel lining as a Timoshenko beam and using the Winkler model for the foundation, and solving the uplift displacement analytically or numerically. This method has a clear physical meaning. The other is the data-driven method, such as XGBoost and neural networks, which use historical construction data to build a black-box mapping model, resulting in higher prediction accuracy.
[0003] However, existing physical model-driven methods suffer from problems such as difficulty in accurately determining parameters, oversimplification of loads and boundary conditions, and inability to handle nonlinear variations in foundation stiffness, leading to significant discrepancies between theoretical predictions and actual monitoring values. While existing data-driven methods offer higher accuracy, they lack physical constraints, have poor interpretability, are difficult for field engineers to trust, and rely heavily on large amounts of high-quality data, resulting in insufficient generalization ability when operating conditions change. Furthermore, existing methods generally neglect the multi-source heterogeneity and spatiotemporal misalignment of construction parameters, and control decisions are mostly reactive, lacking a closed-loop management capability encompassing prediction, analysis, and control. Summary of the Invention
[0004] To address the aforementioned technical problems, this invention provides a physical-data dual-drive shield tunnel segment uplift prediction and control system, comprising: The physical model building module is used to construct a physical model of segment floating based on Timoshenko beams and Winkler foundation according to tunnel structural parameters and geological parameters, and to determine the governing equations, boundary conditions and load expressions of the physical model. The data acquisition and fusion module is used to collect multi-source data during the construction process, and to perform spatiotemporal alignment and fusion of the multi-source data according to the time and space correspondence to obtain a structured dataset; The dual-drive prediction module is used to solve the physical model using a physical information neural network based on the physical model and the spatial coordinates in the structured dataset to obtain a theoretical baseline prediction value; determine the prediction residual based on the theoretical baseline prediction value and the historical monitoring uplift amount in the structured dataset; learn the prediction residual using an XGBoost model based on the construction parameters in the structured dataset and the prediction residual to obtain a residual prediction value; and obtain the final uplift amount prediction value based on the theoretical baseline prediction value and the residual prediction value. The interpretability analysis module is used to quantify the contribution of each feature to the prediction result using the SHAP method based on the final predicted buoyancy value and the input features in the XGBoost model, and obtain the contribution value of each feature. The control decision module is used to generate construction parameter optimization strategies or shield attitude adjustment strategies based on the predicted final uplift value and the contribution values of each feature.
[0005] Optionally, in the physical model construction module, the governing equations are expressed as: ; Where x represents the distance coordinate along the longitudinal direction of the tunnel, w(x) is the upward displacement of the tunnel segment, (EI)eq is the equivalent longitudinal bending stiffness of the tunnel, (kfGA)eq is the equivalent longitudinal shear stiffness of the tunnel, k is the foundation coefficient, R is the outer radius of the tunnel segment, and P(x) is the longitudinal load.
[0006] Optionally, in the physical model construction module, the longitudinal load in the load expression is represented as: ; Where P(x) is the longitudinal load, Fs is the static buoyancy of grouting, Fd is the dynamic buoyancy of grouting, and Fg is the self-weight of the tunnel segment.
[0007] Optionally, in the physical model construction module, the boundary conditions are set according to the following method: At the tail of the shield, the upward displacement of the tunnel segment is equal to the vertical deviation of the tail of the shield, and the rotation angle of the tunnel segment is equal to the pitch angle of the tail of the shield. At infinite distance from the shield tail, the upward displacement and rotation angle of the tunnel segments are both zero; At the junction of the hardened and unhardened sections of the slurry, the floating displacement, rotation angle, bending moment and shear force of the segments are continuous.
[0008] Optionally, the data acquisition and fusion module includes: The time alignment unit is used to align the construction parameters with the floating monitoring time based on the difference between the grouting completion time and the floating monitoring time. The spatial alignment unit is used to correct the spatial offset of earth pressure, total thrust, and cutterhead torque based on the difference in the number of rings between the shield tail and the cutterhead.
[0009] Optionally, in the dual-drive prediction module, when using a physical information neural network to solve the physical model, a loss function of the physical information neural network is constructed based on the residuals of the control equations and the residuals of the boundary conditions, and the physical information neural network is trained by minimizing the loss function.
[0010] Optionally, the control decision module includes: The controllability judgment unit is used to determine whether intervention is needed based on the comparison result between the predicted final buoyancy value and the preset safety threshold. The parameter optimization unit is used to optimize the controllable construction parameters by using a grid search method when the predicted final uplift value exceeds the safe range and the main control factor is a controllable construction parameter, based on the optimization objective of minimizing the deviation between the predicted uplift and the target uplift. The attitude compensation unit is used to generate an attitude compensation strategy, including the vertical deviation of the shield tail and the pitch angle, based on the adjustment of the shield machine attitude control target when the predicted value of the final uplift exceeds the safe range and the main control factor is an uncontrollable geological factor.
[0011] To address the aforementioned technical problems, this invention also provides a physical-data dual-driven method for predicting and controlling the uplift of tunnel segments, comprising: Develop a floating physical model and determine its governing equations, boundary conditions, and load expressions. Collect multi-source data during the construction process, and perform spatiotemporal alignment and fusion of the multi-source data according to the time and space correspondence to obtain a structured dataset; Based on the physical model and the spatial coordinates in the structured dataset, a physical information neural network is used to solve the physical model to obtain the theoretical baseline prediction value; The prediction residual is determined based on the theoretical baseline prediction value and the historical monitoring rise in the structured dataset; Based on the construction parameters in the structured dataset and the predicted residuals, the XGBoost model is used to learn the predicted residuals to obtain the predicted residual values. Based on the theoretical benchmark prediction value and the residual prediction value, the final upward float prediction value is obtained; Based on the predicted final float and the input features in the XGBoost model, the contribution of each feature to the prediction result is quantified using the SHAP method to obtain the contribution value of each feature. Based on the predicted final ascent and the contribution values of each feature, a construction parameter optimization strategy or a shield attitude adjustment strategy is generated.
[0012] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.
[0013] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.
[0014] Compared with the prior art, the present invention has the following advantages and technical effects: First, a Physical Information Neural Network (PINN) is used to solve the Timoshenko beam-Winkler foundation physical model, obtaining theoretical benchmark values that conform to mechanical laws. Then, XGBoost is used to learn and predict residuals. Through a dual-drive mechanism of "physical benchmark + error correction," the prediction accuracy is significantly improved, solving the problem of inaccurate predictions by a single model. Second, the SHAP method is introduced to perform interpretability analysis on the prediction results, quantifying the contribution of each feature and distinguishing between controllable and uncontrollable factors. At the same time, a spatiotemporal alignment mechanism is designed to eliminate misalignment of multi-source data, enhancing the model's credibility and generalization ability. Finally, based on the predicted values and error tracing results, construction parameter optimization or shield attitude adjustment strategies are proactively generated, realizing closed-loop intelligent control from post-event remediation to pre-event intervention. Attached Figure Description
[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the system structure according to an embodiment of the present invention. Detailed Implementation
[0016] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0017] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0018] Example 1 like Figure 1 As shown, this embodiment provides a physical-data dual-driven shield tunnel segment uplift prediction and control system, including: The physical model building module is used to construct a physical model of segment floating based on Timoshenko beams and Winkler foundation according to tunnel structural parameters and geological parameters, and to determine the governing equations, boundary conditions and load expressions of the physical model. The data acquisition and fusion module is used to collect multi-source data during the construction process, and to perform spatiotemporal alignment and fusion of the multi-source data according to the time and space correspondence to obtain a structured dataset; The dual-drive prediction module is used to solve the physical model using a physical information neural network based on the physical model and the spatial coordinates in the structured dataset to obtain a theoretical baseline prediction value; determine the prediction residual based on the theoretical baseline prediction value and the historical monitoring uplift amount in the structured dataset; learn the prediction residual using an XGBoost model based on the construction parameters in the structured dataset and the prediction residual to obtain a residual prediction value; and obtain the final uplift amount prediction value based on the theoretical baseline prediction value and the residual prediction value. The interpretability analysis module is used to quantify the contribution of each feature to the prediction result using the SHAP method based on the final predicted buoyancy value and the input features in the XGBoost model, and obtain the contribution value of each feature. The control decision module is used to generate construction parameter optimization strategies or shield attitude adjustment strategies based on the predicted final uplift value and the contribution values of each feature.
[0019] Optionally, in the physical model construction module, the governing equations are: ; in x Let w(x) represent the distance coordinates along the longitudinal direction of the tunnel, and w(x) represent the upward displacement of the tunnel segment. The equivalent longitudinal bending stiffness of the tunnel. denoted as the longitudinal equivalent shear stiffness of the tunnel, k is the subgrade coefficient, R is the outer radius of the tunnel segment, and P(x) is the longitudinal load.
[0020] Furthermore, in the physical model construction module, the longitudinal load P(x) includes the grouting static buoyancy force. grouting dynamic buoyancy and the weight of the tunnel segments The load expression is: ; in This refers to the length of the unhardened section of the grout, which is determined by both the initial setting time of the grout and the tunnel boring machine's advance speed.
[0021] Furthermore, in the physical model construction module, the boundary conditions are set as follows: shield tail Displacement equals the vertical deviation of the shield tail, and rotation angle equals the pitch angle of the shield tail. Far from the tail of the shield Displacement and rotation are both zero; The boundary between the hardened and unhardened sections Displacement, rotation, bending moment, and shear force are continuous.
[0022] Furthermore, the data acquisition and fusion module includes: Time alignment unit: Converts the grouting completion time into the difference between the grouting completion time and the floating monitoring time. The calculation formula is as follows: ; in This refers to the time required to complete the grouting process. To ensure that the construction parameters correspond to the uplift monitoring time, the construction parameters are aligned with the uplift monitoring time. Spatial alignment unit: This unit corrects for spatial offset in earth pressure, total thrust, and cutterhead torque. The calculation formula is as follows: ; in The difference in the number of rings between the shield tail and the cutterhead ensures that the construction parameters correspond spatially to the segment positions.
[0023] Furthermore, in the dual-drive prediction module, PINN takes spatial coordinate x as input and floating displacement w(x) as output. It is trained by minimizing a loss function that includes the residuals of the governing equations and boundary conditions, thus obtaining the theoretical baseline prediction value wPINN(x) without relying on measured data. The loss function of PINN is: ; in: ; ; Furthermore, in the dual-drive prediction module, the XGBoost model is used to learn the nonlinear relationship between the PINN prediction residuals and construction characteristics; the residuals are defined as... ; in The actual floating amount in historical monitoring data; the XGBoost model takes 15 key construction parameters obtained after feature selection as input, and the residuals as input. Training is performed on the output; the final predicted float value is: ; The 15 key construction parameters include: earth pressure, water content, internal friction angle, tunneling speed, total grouting volume, grouting pressure, grouting volume of each hole, grout mix ratio, advancing distance, overburden load, tunnel depth, total thrust, cutterhead torque, cutterhead rotation speed, and grouting completion time.
[0024] Furthermore, the interpretability analysis module employs the SHAP method; for the j-th feature of the i-th sample, its SHAP value is... satisfy: ; in This represents the XGBoost model's prediction for this sample. Predict the mean for the entire sample; when A value greater than 0 indicates that the feature has a positive contribution to the prediction result (promoting upward movement). A value less than 0 indicates a negative contribution (suppressing upward movement); by sorting the SHAP values, we can identify the main controlling factors affecting upward movement and distinguish between controllable and uncontrollable factors.
[0025] Furthermore, the control decision module includes: Controllability assessment unit: Based on the comparison between the predicted rise and the preset safety threshold, it determines whether intervention is required; Parameter optimization unit: When the predicted uplift exceeds the safe range and the main controlling factor is a controllable construction parameter, a grid search method is used to optimize the controllable parameters. The optimization objective function is: ; Where u is a vector of controllable construction parameters. To determine the target upward displacement, the optimal parameter combination is obtained through iterative optimization and then output to the on-site construction personnel for execution. Attitude compensation unit: When the predicted uplift exceeds the safe range and the main controlling factor is an uncontrollable geological factor, the attitude control target of the tunnel boring machine is adjusted, including the vertical deviation of the shield tail and the pitch angle, and the uplift effect is offset by attitude compensation.
[0026] Example 2 This embodiment provides a physical-data dual-driven method for predicting and controlling the uplift of tunnel segments, including: Based on the tunnel structural parameters and geological parameters, a physical model for segment floating based on Timoshenko beams and Winkler foundations was constructed, and the governing equations, boundary conditions, and load expressions of the physical model were determined. Collect multi-source data during the construction process, and perform spatiotemporal alignment and fusion of the multi-source data according to the time and space correspondence to obtain a structured dataset; Based on the physical model and the spatial coordinates in the structured dataset, a physical information neural network is used to solve the physical model to obtain the theoretical baseline prediction value; The prediction residual is determined based on the theoretical baseline prediction value and the historical monitoring rise in the structured dataset; Based on the construction parameters in the structured dataset and the predicted residuals, the XGBoost model is used to learn the predicted residuals to obtain the predicted residual values. Based on the theoretical benchmark prediction value and the residual prediction value, the final upward float prediction value is obtained; Based on the predicted final float and the input features in the XGBoost model, the contribution of each feature to the prediction result is quantified using the SHAP method to obtain the contribution value of each feature. Based on the predicted final ascent and the contribution values of each feature, a construction parameter optimization strategy or a shield attitude adjustment strategy is generated.
[0027] The governing equation is expressed as follows: The longitudinal equivalent bending stiffness of the tunnel multiplied by the fourth derivative of the segment uplift displacement, minus the longitudinal equivalent shear stiffness of the tunnel multiplied by the second derivative of the segment uplift displacement, plus the subgrade coefficient multiplied by the segment uplift displacement equals the longitudinal load. The upward displacement of the tunnel segment is a function of the longitudinal distance coordinate along the tunnel.
[0028] The longitudinal load in the load expression is represented as follows: The static buoyancy of grouting plus the dynamic buoyancy of grouting minus the weight of the tunnel lining segments; Among them, the static buoyancy and dynamic buoyancy of grouting only act within the length of the unhardened section of the grout, and the length of the unhardened section of the grout is determined according to the initial setting time of the grouting and the tunnel boring machine's advancing speed.
[0029] The boundary conditions are set in the following manner: At the tail of the shield, the upward displacement of the tunnel segment is equal to the vertical deviation of the tail of the shield, and the rotation angle of the tunnel segment is equal to the pitch angle of the tail of the shield. At infinite distance from the shield tail, the upward displacement and rotation angle of the tunnel segments are both zero; At the junction of the hardened and unhardened sections of the slurry, the floating displacement, rotation angle, bending moment and shear force of the segments are continuous.
[0030] Spatiotemporal alignment and fusion of the multi-source data includes: Based on the difference between the grouting completion time and the uplift monitoring time, the construction parameters are aligned with the uplift monitoring time. Spatial offset corrections are performed on earth pressure, total thrust, and cutterhead torque based on the difference in the number of rings between the shield tail and the cutterhead.
[0031] When solving the physical model using a physical information neural network, a loss function for the physical information neural network is constructed based on the residuals of the governing equations and the residuals of the boundary conditions, and the physical information neural network is trained by minimizing this loss function.
[0032] When learning the predicted residual using the XGBoost model, the XGBoost model is trained using 15 key construction parameters obtained after feature selection as input and the predicted residual as output. The 15 key construction parameters include: earth pressure, water content, internal friction angle, tunneling speed, total grouting volume, grouting pressure, grouting volume of each hole, grout mix ratio, advancing distance, overburden load, tunnel depth, total thrust, cutterhead torque, cutterhead rotation speed, and grouting completion time.
[0033] When using the SHAP method to quantify the contribution of each feature to the prediction result, the SHAP value corresponding to each feature is calculated based on the predicted value of the current sample by the XGBoost model and the mean of the predictions of all samples. When the SHAP value is greater than zero, it indicates that the feature makes a positive contribution to buoyancy; when the SHAP value is less than zero, it indicates that the feature makes a negative contribution to buoyancy.
[0034] Strategies for generating construction parameter optimization or shield attitude adjustment include: Based on the comparison between the predicted final buoyancy value and the preset safety threshold, it is determined whether intervention is necessary. When the predicted final uplift amount exceeds the safe range and the main control factor is a controllable construction parameter, the controllable construction parameter is optimized by a grid search method based on the optimization objective of minimizing the deviation between the predicted uplift amount and the target uplift amount. When the predicted final ascent exceeds the safe range and the main controlling factor is an uncontrollable geological factor, an attitude compensation strategy including the vertical deviation of the shield tail and the pitch angle is generated according to the adjustment target of the shield machine attitude control.
[0035] The final predicted buoyancy value is obtained in the following manner: The final predicted value of the upward float is obtained by adding the theoretical baseline prediction value output by the physical information neural network and the residual prediction value output by the XGBoost model. The physical information neural network optimizes network parameters by automatically differentiating the residuals of the control equations and boundary conditions and backpropagating. The XGBoost model is trained with 15 key construction parameters as input and the difference between theoretical benchmark predictions and historical monitoring uplift values as the learning objective.
[0036] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.
[0037] On the other hand, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.
[0038] Example 3 This embodiment provides a physical-data dual-driven method for predicting and controlling the uplift of tunnel segments, including: This embodiment uses a shield tunnel project in a certain section of a rail transit line as the application object to fully demonstrate the implementation process and effects of the present invention. The section is 2875.295m long, consisting of approximately 1600 rings (ring width 1.8m), with a tunnel depth ranging from 7.5m to 63m. The strata it traverses are mainly sandstone, sandy mudstone, silty clay, and plain fill, with surrounding rock grades IV to VI. The shield tunneling machine is an earth pressure balance shield tunneling machine with a cutterhead diameter of 8.83m. The tunnel lining uses precast reinforced concrete segments with an outer diameter of 8.5m, an inner diameter of 7.7m, and a thickness of 0.4m. The concrete is C50, with a permeability grade of P12. The segments are divided into 7 sections circumferentially (1K+2B+4A), with 14 M30 bolts per ring in the longitudinal joints and 19 M30 bolts per ring in the circumferential joints. The segments are classified into four types: A, B, C, and D, each suitable for different surrounding rock grades and environmental conditions. The section passes under the Liangtan River multiple times (minimum clearance between the riverbed and the tunnel is 8.5m), is close to underground passages (clear distance 2m), and has pile foundations (minimum horizontal clearance 5.8m), requiring high safety standards.
[0039] Construction data for the first 500 rings (mileage YK31+733~YK32+633) were collected, including grouting parameters (pressure, flow rate, grout density, and grouting completion time for each grouting hole), propulsion parameters (earth pressure, total thrust, cutterhead torque, and propulsion speed), geological parameters (water content, internal friction angle, overburden load, burial depth, and surrounding rock grade), and segment uplift monitoring values (stable uplift amount 24 hours after grouting).
[0040] To address the spatiotemporal misalignment issue of multi-source data, the following alignment process is performed: Time alignment: Convert the grouting completion time to the difference between the grouting completion time and the floating monitoring time. For example, if the grouting is completed at 8:00 and the monitoring is carried out at 8:00 the next day, the time difference is 24 hours.
[0041] Spatial alignment: Parameters such as earth pressure, total thrust, and cutterhead torque reflect the strata in front of the cutterhead, while the tunnel segment is located about 8 rings behind the shield tail. Therefore, these parameters are shifted forward by 8 rings, that is, the earth pressure of the Rth ring is taken as the measured value of the R-8th ring, to ensure that the features correspond to the actual position of the tunnel segment.
[0042] After the above processing, a structured dataset with rings as units and features aligned with the target is formed, with a total of 500 valid samples.
[0043] 1. Physical Information Neural Network (PINN) Benchmark Prediction: The tunnel lining is considered a Timoshenko beam, and the foundation is modeled using the Winkler model. Based on design data, the equivalent bending stiffness, equivalent shear stiffness, foundation coefficient (by stratum), segment outer radius, grouting buoyancy, and self-weight are determined. The load acts only on the unhardened grout section (14.4m in length, corresponding to 8 rings) and decreases linearly along the longitudinal direction. Boundary conditions: the displacement and rotation at the shield tail are equal to the measured shield tail attitude; the displacement and rotation at a distance are zero; and the displacement, rotation, bending moment, and shear force are continuous at the boundary between the hardened and unhardened sections.
[0044] The PINN network employs six fully connected layers, each with 256 neurons. It takes spatial coordinate x (distance from the shield tail) as input and outputs the upward displacement at that location. The loss function is a weighted sum of the residuals from the governing equations and the residuals from the boundary conditions. After 10,000 training epochs, the loss converges, yielding the theoretical baseline prediction value wPINN.
[0045] 2. XGBoost residual correction: The measured upward displacement and The difference The learning objective is to use the XGBoost model as input, with 15 key features (selected after being ranked by importance, including earth pressure, water content, internal friction angle, grouting volume in each hole, total thrust, grouting pressure, etc.) as input, and the output being the residual prediction value ΔwXGBoost. Model parameters: maximum depth 3, learning rate 0.1, number of trees 100.
[0046] Final predicted value On the validation set (rings 401–500), the model has a coefficient of determination R² = 0.95 and a mean absolute error of 3.2 mm, which is better than a single physics model (R² = 0.88) or a single XGBoost model (R² = 0.92).
[0047] Taking rings 401-450 (underpass of the Liangtan River section) as an example, SHAP analysis was performed on the XGBoost residual model. The results show that water content has the largest positive contribution to the residuals (mean SHAP +2.8 mm), indicating that high water content significantly promotes uplift, which is an uncontrollable geological factor. The grouting ratio in the upper and middle sections has a significant negative contribution (mean SHAP -1.5 mm), and reducing this ratio can suppress uplift, which is a controllable construction parameter. Total thrust and grouting pressure difference also have positive contributions and are controllable. The overburden load has a negative contribution and is uncontrollable. Furthermore, C-type segments (used in Class V underpass sections) are more prone to uplift than A-type segments, requiring strengthened control.
[0048] In the 451-500 ring (from the section under the Liangtan River to the receiving section of the North Logistics Park, with surrounding rock grade V-VI and using D-type segments), the predicted average uplift is 56mm, exceeding the safety threshold of 50mm (calculated based on a net distance of 2m between underground passages). SHAP analysis shows that the ratio of upper and middle grouting is a controllable main controlling factor, and the system judges it as "controllable".
[0049] A grid search was used to optimize the grouting ratio between the upper and middle sections. The objective function was to minimize the deviation between the predicted upward displacement and the target value of 45 mm. The search range was 1.5–2.5, with an optimal ratio of 1.7. Simultaneously, combined with SHAP analysis, it was recommended to reduce the total thrust from 38000 kN to 35000 kN and control the grouting pressure difference within 0.05 MPa.
[0050] On-site construction was carried out as recommended: the grouting ratio was gradually adjusted from 2.0 to 1.7, the total thrust was adjusted to 35,000 kN, and the grouting pressure differential was kept below 0.05 MPa. Post-implementation monitoring data showed that the uplift was 55-58 mm before implementation (rings 451-460); during implementation (rings 461-480), it decreased to 48-52 mm; and after implementation (rings 481-500), it stabilized at 42-46 mm. Segment misalignment decreased from 5 mm to below 2 mm, with no cracking or leakage, and the tunnel axis deviation met design requirements. The uplift in the sections passing under the Liangtan River and near the underground passage was controlled below 45 mm, without affecting existing buildings or structures.
[0051] This embodiment verifies the effectiveness of the present invention in achieving high-precision prediction, interpretable analysis, and active control through a dual physical-data approach under complex geological and environmental conditions.
[0052] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A physical-data dual-driven shield tunnel segment uplift prediction and control system, characterized in that, include: The physical model building module is used to construct a physical model of segment floating based on Timoshenko beams and Winkler foundation according to tunnel structural parameters and geological parameters, and to determine the governing equations, boundary conditions and load expressions of the physical model. The data acquisition and fusion module is used to collect multi-source data during the construction process, and to perform spatiotemporal alignment and fusion of the multi-source data according to the time and space correspondence to obtain a structured dataset; The dual-drive prediction module is used to solve the physical model using a physical information neural network based on the physical model and the spatial coordinates in the structured dataset to obtain a theoretical baseline prediction value; determine the prediction residual based on the theoretical baseline prediction value and the historical monitoring uplift amount in the structured dataset; learn the prediction residual using an XGBoost model based on the construction parameters in the structured dataset and the prediction residual to obtain a residual prediction value; and obtain the final uplift amount prediction value based on the theoretical baseline prediction value and the residual prediction value. The interpretability analysis module is used to quantify the contribution of each feature to the prediction result using the SHAP method based on the final predicted buoyancy value and the input features in the XGBoost model, and obtain the contribution value of each feature. The control decision module is used to generate construction parameter optimization strategies or shield attitude adjustment strategies based on the predicted final buoyancy value and the contribution values of each feature.
2. The system according to claim 1, characterized in that, In the physical model construction module, the governing equations are expressed as follows: ； Where x represents the distance coordinate along the longitudinal direction of the tunnel, w(x) is the upward displacement of the tunnel segment, (EI)eq is the equivalent longitudinal bending stiffness of the tunnel, (kfGA)eq is the equivalent longitudinal shear stiffness of the tunnel, k is the foundation coefficient, R is the outer radius of the tunnel segment, and P(x) is the longitudinal load.
3. The system according to claim 2, characterized in that, In the physical model construction module, the longitudinal load in the load expression is represented as: ； Where P(x) is the longitudinal load, Fs is the static buoyancy of grouting, Fd is the dynamic buoyancy of grouting, and Fg is the self-weight of the tunnel segment.
4. The system according to claim 2, characterized in that, In the physical model construction module, the boundary conditions are set in the following manner: At the tail of the shield, the upward displacement of the tunnel segment is equal to the vertical deviation of the tail of the shield, and the rotation angle of the tunnel segment is equal to the pitch angle of the tail of the shield. At infinite distance from the shield tail, the upward displacement and rotation angle of the tunnel segments are both zero; At the junction of the hardened and unhardened sections of the slurry, the floating displacement, rotation angle, bending moment and shear force of the segments are continuous.
5. The system according to claim 1, characterized in that, The data acquisition and fusion module includes: The time alignment unit is used to align the construction parameters with the floating monitoring time based on the difference between the grouting completion time and the floating monitoring time. The spatial alignment unit is used to correct the spatial offset of earth pressure, total thrust, and cutterhead torque based on the difference in the number of rings between the shield tail and the cutterhead.
6. The system according to claim 1, characterized in that, In the dual-drive prediction module, when the physical information neural network is used to solve the physical model, a loss function of the physical information neural network is constructed based on the residuals of the control equations and the residuals of the boundary conditions, and the physical information neural network is trained by minimizing the loss function.
7. The system according to claim 1, characterized in that, The control decision module includes: The controllability judgment unit is used to determine whether intervention is needed based on the comparison result between the predicted final buoyancy value and the preset safety threshold. The parameter optimization unit is used to optimize the controllable construction parameters by using a grid search method when the predicted final uplift value exceeds the safe range and the main control factor is a controllable construction parameter, based on the optimization objective of minimizing the deviation between the predicted uplift and the target uplift. The attitude compensation unit is used to generate an attitude compensation strategy, including the vertical deviation of the shield tail and the pitch angle, based on the adjustment of the shield machine attitude control target when the predicted value of the final uplift exceeds the safe range and the main control factor is an uncontrollable geological factor.
8. A method for predicting and controlling the uplift of shield tunnel segments driven by both physical and data principles, characterized in that... include: Based on the tunnel structural parameters and geological parameters, a physical model for segment floating based on Timoshenko beams and Winkler foundations was constructed, and the governing equations, boundary conditions and load expressions of the physical model were determined. Collect multi-source data during the construction process, and perform spatiotemporal alignment and fusion of the multi-source data according to the time and space correspondence to obtain a structured dataset; Based on the physical model and the spatial coordinates in the structured dataset, a physical information neural network is used to solve the physical model to obtain the theoretical baseline prediction value; The prediction residual is determined based on the theoretical baseline prediction value and the historical monitoring rise in the structured dataset; Based on the construction parameters in the structured dataset and the predicted residuals, the XGBoost model is used to learn the predicted residuals to obtain the predicted residual values. Based on the theoretical baseline prediction value and the residual prediction value, the final upward float prediction value is obtained; Based on the predicted final float and the input features in the XGBoost model, the contribution of each feature to the prediction result is quantified using the SHAP method to obtain the contribution value of each feature. Based on the predicted final ascent and the contribution values of each feature, a construction parameter optimization strategy or a shield attitude adjustment strategy is generated.
9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, The processor implements the method of claim 8 when executing the computing program.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of claim 8.