A method and system for predicting surface settlement of tunnel boring machines based on three-dimensional random fields

Abstract

This invention belongs to the technical field of geotechnical engineering and discloses a method and system for predicting shield tunneling surface settlement based on a three-dimensional random field. The method includes: determining empirical values ​​of statistical characteristic values ​​of spatial variability of soil parameters based on geological survey data, and constructing an empirical three-dimensional random field of soil parameters; performing numerical simulation analysis based on the empirical three-dimensional random field to obtain corresponding shield tunneling surface settlement data and establishing a surface settlement prediction dataset; training and acquiring a surrogate model; determining actual values ​​of statistical characteristic values ​​of spatial variability of soil parameters based on geological survey data of the target project, constructing an actual three-dimensional random field of soil parameters, and using the surrogate model to predict the shield tunneling surface settlement of the target project. This invention considers the surface settlement of shield tunneling under spatial variability, which is more accurate than deterministic analysis of soil parameters; the surrogate model established based on numerical simulation results for shield tunneling settlement prediction can meet the needs of rapid prediction and real-time analysis.

Description

Technical Field

[0001] This invention belongs to the technical field of geotechnical engineering, and more specifically, relates to a method and system for predicting surface settlement of tunnel boring machines based on a three-dimensional random field. Background Technology

[0002] Shield tunneling is commonly used in the construction of rail transit tunnels. However, the surface settlement caused during shield tunneling can pose safety risks to the tunnel itself and surrounding buildings. Therefore, this paper proposes a method for predicting surface settlement during shield tunneling, which can predict the deformation of the shield tunnel in advance and is of great significance for ensuring the safety of shield tunneling. Regarding the surface settlement problem during shield tunneling, existing technologies typically use numerical simulation methods to predict the surface settlement that occurs during shield tunneling. Numerical simulation methods use computers to simulate the surface settlement process caused by actual shield tunneling, and can visually demonstrate the surface settlement situation.

[0003] However, surface settlement is related to the physical and mechanical parameters of soil and rock masses, which exhibit spatial variability due to factors such as stratigraphic origin. In numerical simulations, treating soil parameters such as elastic modulus and internal friction angle as spatially correlated random variables can more realistically reflect the spatial variability of underground soil and rock masses, but it also significantly increases the complexity of the numerical simulation process. The method of analyzing shield tunnel surface settlement through numerical simulation incurs enormous computational costs under high-dimensional random input, making it difficult to meet the needs of rapid prediction and real-time analysis.

[0004] In summary, those skilled in the art need a rapid and reliable method to assess settlement risk and ensure the safety of tunnel boring machine (TBM) construction. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a method and system for predicting shield tunnel surface settlement based on three-dimensional random fields, which is used to solve the problem that the existing methods for shield tunnel surface settlement analysis through numerical simulation have high computational overhead and are difficult to meet the needs of rapid prediction and real-time analysis.

[0006] To achieve the above objectives, according to one aspect of the present invention, a method for predicting surface settlement of tunnel boring machines based on a three-dimensional random field is provided, comprising:

[0007] Based on the geological survey data of existing projects, the statistical characteristic values ​​and empirical values ​​of the spatial variability of soil layer parameters are determined, and an empirical three-dimensional random field of soil layer parameters is constructed based on the statistical characteristic values ​​and empirical values.

[0008] Based on the aforementioned empirical three-dimensional random field, numerical simulation analysis is performed using a pre-constructed three-dimensional refined finite difference model for shield tunneling construction to obtain corresponding shield tunneling surface settlement data, thereby establishing a surface settlement prediction dataset.

[0009] Based on the surface settlement prediction dataset, a proxy model is trained to obtain the relationship between the three-dimensional random field of soil parameters and the surface settlement of the shield tunnel.

[0010] Based on the geological survey data of the target project, the actual values ​​of the statistical characteristic values ​​of the spatial variability of soil layer parameters are determined. Based on the actual values ​​of the statistical characteristic values, an actual three-dimensional random field of soil layer parameters is constructed, and the surrogate model is used to predict the surface settlement of the shield tunnel of the target project.

[0011] According to the shield tunneling surface settlement prediction method based on three-dimensional random fields provided by the present invention, the pre-construction of a three-dimensional refined finite difference model for shield tunneling construction specifically includes:

[0012] Obtain the tunnel geometry parameters and geological survey data for shield tunneling construction; wherein, the tunnel geometry parameters include the tunnel outer diameter, segment ring width, lining thickness, tunnel burial depth, and lining material; and determine the soil layer thickness, soil elastic modulus, soil internal friction angle, soil density, soil cohesion, and soil Poisson's ratio based on the geological survey data.

[0013] Simulation software was used to establish a soil model based on geological survey data and tunnel geometry parameters during shield tunneling. The Mohr-Coulomb constitutive model was used for the soil, and the initial geostress field was generated using an elastoplastic solution method with varying parameters.

[0014] By setting an empty model to represent the excavated part of the soil model, and assigning a new constitutive model and material parameters to represent the tunnel segments and grouting layer, the shield tunneling excavation process is simulated in a loop to obtain surface settlement data of shield tunneling.

[0015] The simulation results of the three-dimensional refined finite difference model are compared with the monitoring data of the actual project to verify the rationality of the simulation results.

[0016] The shield tunneling surface settlement prediction method based on a three-dimensional random field provided by the present invention specifically includes the following: Constructing an empirical three-dimensional random field of soil layer parameters based on empirical values ​​of statistical eigenvalues.

[0017] Based on the mesh model in the pre-constructed three-dimensional refined finite difference model of shield tunneling construction, the positional relationship of each mesh is determined;

[0018] Based on statistical characteristic values ​​and empirical values, as well as the positional relationships of each grid, a Gaussian covariance model or a normal distribution model is used to construct a three-dimensional random field of soil parameters, thereby obtaining the empirical three-dimensional random field.

[0019] According to the shield tunneling surface settlement prediction method based on three-dimensional random field provided by the present invention, the soil layer parameters in the empirical three-dimensional random field of soil layer parameters are constructed based on statistical characteristic values ​​and empirical values. The soil layer parameters include the soil elastic modulus, soil internal friction angle and soil cohesion. Random field is generated for each soil layer parameter, and then the random fields of all soil layer parameters are combined to form the final three-dimensional random field.

[0020] The statistical characteristics of the spatial variability of soil parameters include the mean, variance, and multiple of the correlation lengths in three orthogonal directions.

[0021] According to the shield tunneling surface settlement prediction method based on a three-dimensional random field provided by the present invention, the surface settlement prediction dataset includes multiple sets of samples. Each set of samples includes input samples and output samples. The input samples are a three-dimensional random field of soil parameters. Specifically, the three-dimensional random field includes the grid information of the grid model in the pre-constructed three-dimensional refined finite difference model of shield tunneling construction and the soil parameter information of each grid. The output samples are shield tunneling surface settlement data, which are multiple surface settlement values ​​taken from the cross-section perpendicular to the tunnel axis.

[0022] According to the shield tunneling surface settlement prediction method based on three-dimensional random fields provided by the present invention, a proxy model between three-dimensional random fields and shield tunneling surface settlement is constructed using graph convolutional neural networks. Specifically, each input sample is constructed as a three-dimensional graph structure, which includes nodes and connecting edges. Each node corresponds to a grid in the grid model, and each node has node features, including the soil elastic modulus, soil internal friction angle, and soil cohesion of the corresponding grid. Edges between nodes are established based on the spatial proximity relationship of the grid to form graph data.

[0023] According to the shield tunneling surface settlement prediction method based on three-dimensional random fields provided by the present invention, when the surrogate model is trained, the output samples in the surface settlement prediction dataset are subjected to intrinsic orthogonal decomposition to obtain the modal basis and modal coefficient supervision values. The output of the surrogate model is the modal coefficient prediction value, and the shield tunneling surface settlement prediction data is obtained based on the modal basis reconstruction.

[0024] The shield tunneling surface settlement prediction method based on a three-dimensional random field provided by the present invention further includes:

[0025] Collect shield tunnel surface settlement monitoring data during the construction process of the target project. Based on the shield tunnel surface settlement monitoring data and the shield tunnel surface settlement prediction data output by the proxy model, optimize and correct the statistical characteristic values ​​of soil elastic modulus, soil internal friction angle and soil cohesion, and obtain the corrected statistical characteristic values.

[0026] A modified three-dimensional random field for soil parameters is constructed based on the statistical eigenvalue correction value, and the modified three-dimensional random field will be used for future shield tunneling surface settlement prediction.

[0027] According to the shield tunneling surface settlement prediction method based on a three-dimensional random field provided by the present invention, obtaining the statistical characteristic value correction value specifically includes:

[0028] Based on the geological survey data of the target project, determine the actual values ​​of the statistical characteristic values ​​of the spatial variability of soil elastic modulus, soil internal friction angle, and soil cohesion, and then obtain the prior distribution.

[0029] By integrating the surface settlement monitoring data of the tunnel boring machine (TBM) and the surface settlement prediction data of the TBM output by the surrogate model, the posterior distribution of the soil elastic modulus, soil internal friction angle, and soil cohesion is updated using Bayesian theory combining likelihood and prior, and the statistical characteristic value correction value is obtained.

[0030] According to another aspect of the present invention, a shield tunneling surface settlement prediction system based on a three-dimensional random field is provided. The system includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the shield tunneling surface settlement prediction method based on a three-dimensional random field as described above.

[0031] In summary, compared with the prior art, the shield tunneling surface settlement prediction method and system based on three-dimensional random fields provided by this invention offer the following advantages:

[0032] 1. Considering the surface settlement of shield tunneling under spatial variability, a three-dimensional random field is generated based on statistical characteristic values ​​that can reflect the spatial variability distribution of soil parameters, which is more accurate than deterministic analysis of soil parameters; a surrogate model is established based on numerical simulation results to predict shield settlement, which significantly improves the prediction speed of shield surface settlement compared with finite difference calculation, and avoids large computational costs, thus meeting the needs of rapid prediction and real-time analysis.

[0033] 2. Using graph convolutional neural networks for shield tunneling surface settlement prediction can fully exploit the spatial correlation and coupling effect between grid points compared with traditional prediction methods, thereby improving prediction accuracy and model generalization ability. In addition, the graph structure is adaptable to irregular data distribution, while the grid established by finite element method is often irregular, making it more valuable for practical engineering applications.

[0034] 3. By integrating monitoring data to update and correct the statistical characteristic values ​​of spatial variability of soil parameters, and using the corrected statistical characteristic values ​​to predict future shield tunnel settlement, the real-time updates and optimizations based on monitoring data are incorporated into the prediction, which helps to improve the credibility of the proxy model. Attached Figure Description

[0035] Figure 1 This is a flowchart of the shield tunneling surface settlement prediction method based on three-dimensional random fields provided by the present invention.

[0036] Figure 2 This is a diagram of a shield tunneling surface settlement prediction method based on a three-dimensional random field, provided by the present invention.

[0037] Figure 3 This is a schematic diagram of the three-dimensional refined finite difference model under three-dimensional random fields according to the present invention.

[0038] Figure 4 This is a schematic diagram of the results of the three-dimensional refined finite difference model under three-dimensional random fields in this invention.

[0039] Figure 5 This is a schematic diagram of the geotechnical parameter update after integrating monitoring data according to the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0041] Please see Figure 1 This embodiment provides a method for predicting surface settlement of tunnel boring machines (TBMs) based on a three-dimensional random field. The method includes:

[0042] Based on the geological survey data of existing projects, the statistical characteristic values ​​and empirical values ​​of the spatial variability of soil layer parameters are determined, and an empirical three-dimensional random field of soil layer parameters is constructed based on the statistical characteristic values ​​and empirical values.

[0043] Based on the aforementioned empirical three-dimensional random field, numerical simulation analysis is performed using a pre-constructed three-dimensional refined finite difference model for shield tunneling construction to obtain corresponding shield tunneling surface settlement data, thereby establishing a surface settlement prediction dataset.

[0044] Based on the surface settlement prediction dataset, a proxy model is trained to obtain the relationship between the three-dimensional random field of soil parameters and the surface settlement of the shield tunnel.

[0045] Based on the geological survey data of the target project, the actual values ​​of the statistical characteristic values ​​of the spatial variability of soil layer parameters are determined. Based on the actual values ​​of the statistical characteristic values, an actual three-dimensional random field of soil layer parameters is constructed, and the surrogate model is used to predict the surface settlement of the shield tunnel of the target project.

[0046] refer to Figure 2 This embodiment illustrates a method for predicting surface settlement of tunnel boring machines based on a three-dimensional random field, including:

[0047] The tunnel geometric parameters and geological survey data for shield tunneling are obtained, and a three-dimensional refined finite difference model is established based on the tunnel geometric parameters and geological survey data. The tunnel geometric parameters include the tunnel outer diameter, segment ring width, lining thickness, tunnel burial depth, and lining material. The geological survey data includes a detailed survey report of the shield tunneling section, an engineering geological profile, and a general plan of the shield tunneling section.

[0048] Based on geological exploration data, the statistical characteristic values ​​of the spatial variability of the elastic modulus, internal friction angle, and cohesion of the strata were determined. A three-dimensional random field model based on the elastic modulus, internal friction angle, and cohesion was constructed and imported into a three-dimensional refined finite difference model for numerical analysis.

[0049] The surface settlement values ​​of the shield tunnel obtained from the three-dimensional random field finite difference model were extracted to construct a surface settlement prediction dataset;

[0050] Based on the obtained surface subsidence prediction dataset, a proxy model between a three-dimensional random field and shield tunneling surface subsidence is constructed using intrinsic orthogonal decomposition and graph convolutional neural network.

[0051] Collect monitoring data during the tunnel boring machine (TBM) construction process, and use Bayesian theory to update the statistical characteristics of elastic modulus, internal friction angle, and cohesion. Use the updated geotechnical parameters for future surface settlement prediction.

[0052] Specifically, refer to Figure 3 The pre-construction of a refined three-dimensional finite difference model for shield tunneling specifically includes:

[0053] Obtain the tunnel geometry parameters and geological survey data for shield tunneling construction; wherein, the tunnel geometry parameters include the tunnel outer diameter, segment ring width, lining thickness, tunnel burial depth, and lining material; determine the soil layer thickness, soil elastic modulus, soil internal friction angle, soil density, soil cohesion, and soil Poisson's ratio based on the geological survey data; for example, manually obtain the cutterhead excavation diameter of 6.47m, tunnel radius of 3.1m, segment thickness of 35cm, and segment ring width of 1.5m from actual shield tunneling projects;

[0054] Simulation software (such as FLAC3d) is used to establish a soil model based on the soil layer thickness and tunnel geometry parameters in the geological survey data of shield tunneling construction. For example, the soil model is 60m high, 51m wide, and 33m long. Boundary conditions are set to constrain the horizontal, front-back, and bottom displacements of the model. The Mohr-Coulomb constitutive model is used for the soil, and the initial geostress field is generated by the elastoplastic solution method with changing parameters.

[0055] By setting an empty model (null) to represent the excavated part of the soil model, and assigning a new constitutive model and material parameters to represent the tunnel segments and grouting layer, the shield tunneling excavation process is simulated in a loop to obtain surface settlement data of shield tunneling.

[0056] The simulation results of the three-dimensional refined finite difference model are compared with the monitoring data of the actual project to verify the rationality of the simulation results.

[0057] In some embodiments, constructing an empirical three-dimensional random field of soil parameters based on empirical values ​​of statistical eigenvalues ​​specifically includes:

[0058] Based on the mesh model in the pre-built 3D refined finite difference model of shield tunneling construction, the positional relationship of each mesh is determined; for example, the mesh model of the established 3D refined finite difference model of shield tunneling construction can be exported using the Python language embedded in FLAC3d.

[0059] Based on statistical characteristic values ​​and empirical values, as well as the positional relationships of each grid, a Gaussian covariance model or a normal distribution model is used to construct a three-dimensional random field of soil parameters, thereby obtaining the empirical three-dimensional random field.

[0060] In an empirical three-dimensional random field constructed based on statistical characteristic values ​​of soil layer parameters, the soil layer parameters include the soil elastic modulus, the internal friction angle, and the soil cohesion. A random field is generated for each soil layer parameter separately, and then the random fields of all soil layer parameters are combined to form the final three-dimensional random field. For example, based on a mesh model, a three-dimensional random field for the soil elastic modulus is first generated, then a three-dimensional random field for the internal friction angle, and then a three-dimensional random field for the soil cohesion. Finally, the three parameter values ​​are simultaneously assigned on the same mesh to obtain the final three-dimensional random field. The statistical characteristic values ​​of the spatial variability of soil layer parameters include the mean, variance, and multiple correlation lengths in three orthogonal directions.

[0061] For example, in one specific embodiment, the geological survey data involves 6 soil layers. The statistical characteristic values ​​of the spatial variability of soil layer parameters corresponding to each soil layer can be obtained based on the geological survey data, and then the three-dimensional random field model corresponding to different soil layers can be obtained.

[0062] In this embodiment, the Gaussian covariance model is represented as follows:

[0063] ;

[0064] ;

[0065] in, The distance between any two points is Covariance between physical quantities over time; Represents structural variance (the variance of the structurable spatial variation). This represents the principal correlation length (a single value for isotropic random fields); the larger the value, the smoother the random field. , , Indicates the length related to the x, y, and z directions; This represents the sub-scale variance, reflecting measurement errors or small abrupt changes smaller than the grid scale; when When, the total variance is ; In this embodiment, the hysteresis distance is represented. This indicates a length-weighted summation in the x, y, and z directions. , , This represents the hysteresis distance components in the x, y, and z directions. In this embodiment, the generalized Gaussian covariance is used to characterize only the zero-mean fluctuations of the random field; the mean field of the parameter space is specified separately. Given, the final three-dimensional random field is... Generate, where With a mean of 0 and a covariance of A stationary random field whose covariance satisfies the above covariance formula, where x represents the position coordinate.

[0066] After generating the three-dimensional random field, the three-dimensional random field model based on elastic modulus, internal friction angle, and cohesion is imported one-to-one into the three-dimensional refined finite difference model for numerical analysis using the Python language embedded in FLAC3d. The shield tunnel surface settlement value is the result of numerical analysis of the three-dimensional refined finite difference model under the corresponding stratum's three-dimensional random field. In this embodiment, 50 surface settlement values ​​are taken from the cross-section perpendicular to the tunnel axis, such as... Figure 4 As shown.

[0067] The surface settlement prediction dataset comprises multiple sets of samples, such as 500 sets of samples corresponding to three-dimensional random fields under different strata. Each set of samples includes input samples and output samples. The input samples are three-dimensional random fields of soil parameters, specifically including the grid information of the grid model in the pre-constructed three-dimensional refined finite difference model of shield tunneling construction and the soil parameter information of each grid. The output samples are shield tunnel surface settlement data, which are multiple surface settlement values ​​on the cross-section perpendicular to the tunnel axis. To facilitate alignment with monitoring data, the output is defined as the displacement results of a preset set of monitoring points, including: the coordinates (x, y) of each monitoring point; and the settlement amount of that point (usually the vertical displacement component, denoted as z, positive upwards). The grid information includes the coordinates of the grid in the x, y, and z directions in the model and the grid number.

[0068] In some embodiments, a proxy model (e.g., a GCN model) between a 3D random field and shield tunneling surface settlement is constructed using a graph convolutional neural network. The graph structure of the 3D mesh is constructed using the graph convolutional neural network, specifically by building a 3D graph structure for each input sample. This 3D graph structure includes nodes and connecting edges. Each node corresponds to a grid cell in the mesh model; that is, each grid cell is considered a graph node. Each node has node features, including the soil elastic modulus, soil internal friction angle, and soil cohesion corresponding to the grid. Edges between nodes are established based on the spatial proximity of the grid to form graph data. Standard normalization processing can also be performed on the node features of all input samples according to their dimensions (using the same mean and standard deviation for training and inference) to obtain a numerically stable and dimensionally consistent input representation. Finally, all samples are converted into feature tensors of shape (500, N, 3) (N is the number of nodes, and 3 is the feature dimension of each node), and their adjacency relationships or edge indices are stored for graph convolution calculations.

[0069] During training, the surrogate model performs intrinsic orthogonal decomposition (POD) on the output samples in the surface settlement prediction dataset to obtain the modal basis and modal coefficient supervision values. The output of the surrogate model is the modal coefficient prediction value, and the shield tunnel surface settlement prediction data is obtained based on the modal basis reconstruction.

[0070] Specifically, the process involves using the vertical displacement vector of each sample at a monitoring point on the ground surface as the original output. First, the settlement data of all samples is mean-reduced according to the monitoring point dimension. Then, singular value decomposition (SVD) is performed on the mean-reduced sample matrix. The top r principal modes whose cumulative energy reaches a preset threshold are selected as the modal basis (i.e., the POD basis). The settlement vector of each sample is projected as r-dimensional modal coefficients (i.e., POD coefficients). During the training phase, these low-dimensional modal coefficients are used as supervision signals. During the inference phase, the predicted modal coefficients are multiplied by the modal basis and added back to the mean to reconstruct the complete settlement field. The intrinsic orthogonal decomposition (POD) basis is obtained by decomposing the settlement data of a batch of numerical simulation samples once during the training phase and is fixed for reconstruction after training, without being recalculated during the application phase. That is, the modal basis and mean field are determined and fixed during the training phase. During the inference / application phase, only the modal coefficients are predicted by the surrogate model, multiplied by the fixed modal basis, and added back to the mean to reconstruct the settlement field.

[0071] In some embodiments, the constructed graph convolutional neural network (e.g., GCN model) includes multiple graph convolutional layers, activation layers (using ReLU function), pooling layers, and fully connected layers. The specific structure may include three graph convolutional layers, each followed by a ReLU activation function, and finally outputting the predicted value of shield tunnel surface settlement through a fully connected layer.

[0072] A hyperparameter grid search is performed on the graph convolutional neural network model to obtain the optimal hyperparameter combination, thereby improving the accuracy of shield tunneling surface settlement prediction. In this embodiment, 500 samples are divided into training and test sets in a 7:3 ratio. Initial learning rate, dropout rate, etc., are selected as hyperparameters to be optimized. The average loss of the test set is used as the objective, and the hyperparameter combination corresponding to the minimum average loss is selected as the final parameters. The selected hyperparameters are used to train the constructed graph convolutional neural network model. In this embodiment, the mean squared error loss function is used, and the optimizer is an adaptive moment estimator (such as Adam). The trained graph convolutional neural network model is tested to evaluate its ability to predict shield tunneling surface settlement.

[0073] Furthermore, the shield tunneling surface settlement prediction method based on three-dimensional random fields in this embodiment also includes:

[0074] Collect shield tunnel surface settlement monitoring data during the construction process of the target project. Based on the shield tunnel surface settlement monitoring data and the shield tunnel surface settlement prediction data output by the proxy model, optimize and correct the statistical characteristic values ​​of soil elastic modulus, soil internal friction angle and soil cohesion, and obtain the corrected statistical characteristic values.

[0075] A modified three-dimensional random field for soil parameters is constructed based on the statistical eigenvalue correction value, and the modified three-dimensional random field will be used for future shield tunneling surface settlement prediction.

[0076] First, the three-dimensional surface subsidence field is reconstructed using the modal coefficients predicted by the graph convolutional neural network model:

[0077] The surrogate model has been trained before processing to obtain modal bases. Full sample mean And the surrogate model's prediction of the modal coefficients of the current parameter field. The processing operation is based on the reconstruction formula. The low-dimensional coefficients are back-projected into vertical displacement vectors (or vertical displacements of surface grid nodes) on the monitoring point set; after processing, a predicted settlement field consistent with the monitoring point set is obtained. This serves as the output of the model for subsequent likelihood construction.

[0078] In some embodiments, obtaining the statistical feature value correction value specifically includes:

[0079] Based on the geological survey data of the target project, determine the actual values ​​of the statistical characteristic values ​​of the spatial variability of soil elastic modulus, soil internal friction angle, and soil cohesion, and then obtain the prior distribution.

[0080] The elastic modulus can be obtained based on the geological survey report and historical data before processing. internal friction angle Cohesion The statistical information (such as mean, variance / standard deviation, and multiple spatial correlation lengths; for example, the spatial correlation length can be determined based on empirical values, and optimization can be performed only on the mean and variance, or simultaneously on the mean, variance, and spatial correlation length); the processing operation is to select and parameterize a suitable distribution family (e.g., , Take the log-normal or Gaussian distribution. (Use Gaussian coefficients, and if necessary, set up a joint covariance and spatial correlation model) to form a parameter (or statistical hyperparameter) vector. prior distribution After processing, a prior model that can be used for Bayesian updates is obtained.

[0081] By integrating the surface settlement monitoring data of the tunnel boring machine (TBM) and the surface settlement prediction data of the TBM output by the surrogate model, the posterior distribution of the soil elastic modulus, soil internal friction angle, and soil cohesion is updated using Bayesian theory combining likelihood and prior, and the statistical characteristic value correction value is obtained.

[0082] Obtain measured settlement data (and (Corresponding to the same set of monitoring points), prediction of the surrogate model under parameter θ. and observation error covariance (Given by monitoring accuracy or experience); the processing operation is to establish an observation model. Assuming Thus, the likelihood is obtained. And according to Bayes' theorem To update, numerically, MCMC, SMC, or variational inference can be used to obtain a posterior approximation; after processing, the following is obtained: , , The posterior distribution and its statistics (posterior mean, variance, confidence interval, and may include updates to spatially relevant parameters).

[0083] Then, the posterior distributions of elastic modulus, internal friction angle, and cohesion were used to predict surface settlement:

[0084] Before processing, prepare the construction control parameters and boundary conditions for the predicted working condition, as well as the posterior distribution. The processing operation involves drawing samples from the posterior distribution. Alternatively, a three-dimensional random field can be generated using posterior statistical eigenvalues ​​and input into the surrogate model to obtain the predicted modal coefficients. , and according to The settlement field is reconstructed to form a predicted distribution. If necessary, a high-fidelity numerical model is used to verify the results on a small number of samples. After processing, point predictions and uncertainty quantification results of surface settlement are obtained (such as posterior mean, confidence interval, and quantile curves). Statistical feature value corrections are repeated when new monitoring data arrives to achieve online calibration and continuous optimization of predictions. Figure 5 As shown.

[0085] In other embodiments, a shield tunneling surface settlement prediction system based on a three-dimensional random field is also provided. The system includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the shield tunneling surface settlement prediction method based on a three-dimensional random field as described above.

[0086] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting surface settlement of tunnel boring machines based on three-dimensional random fields, characterized in that, include: Based on the geological survey data of existing projects, the statistical characteristic values ​​and empirical values ​​of the spatial variability of soil layer parameters are determined, and an empirical three-dimensional random field of soil layer parameters is constructed based on the statistical characteristic values ​​and empirical values. Based on the aforementioned empirical three-dimensional random field, numerical simulation analysis is performed using a pre-constructed three-dimensional refined finite difference model for shield tunneling construction to obtain corresponding shield tunneling surface settlement data, thereby establishing a surface settlement prediction dataset. Based on the surface settlement prediction dataset, a proxy model is trained to obtain the relationship between the three-dimensional random field of soil parameters and the surface settlement of the shield tunnel. Based on the geological survey data of the target project, the actual values ​​of the statistical characteristic values ​​of the spatial variability of soil layer parameters are determined. Based on the actual values ​​of the statistical characteristic values, an actual three-dimensional random field of soil layer parameters is constructed, and the surrogate model is used to predict the surface settlement of the shield tunnel of the target project. The surface settlement prediction dataset includes multiple sets of samples. Each set of samples contains input samples and output samples. The input samples are three-dimensional random fields of soil parameters. Specifically, the three-dimensional random fields include the grid information of the grid model in the pre-constructed three-dimensional refined finite difference model of shield tunneling and the soil parameter information of each grid. The output samples are shield tunnel surface settlement data, which are multiple surface settlement values ​​taken from the cross-section perpendicular to the tunnel axis. A proxy model between a three-dimensional random field and the surface settlement of a shield tunnel is constructed using a graph convolutional neural network. Specifically, each input sample is constructed as a three-dimensional graph structure, which includes nodes and connecting edges. Each node corresponds to a grid in the grid model, and each node has node features, including the soil elastic modulus, soil internal friction angle, and soil cohesion of the corresponding grid. Edges between nodes are established based on the spatial proximity of the grid to form graph data.

2. The shield tunneling surface settlement prediction method based on three-dimensional random fields as described in claim 1, characterized in that, The pre-construction of a refined three-dimensional finite difference model for shield tunneling construction specifically includes: Obtain the tunnel geometry parameters and geological survey data for shield tunneling construction; wherein, the tunnel geometry parameters include the tunnel outer diameter, segment ring width, lining thickness, tunnel burial depth, and lining material; and determine the soil layer thickness, soil elastic modulus, soil internal friction angle, soil density, soil cohesion, and soil Poisson's ratio based on the geological survey data. Simulation software was used to establish a soil model based on geological survey data and tunnel geometry parameters during shield tunneling. The Mohr-Coulomb constitutive model was used for the soil, and the initial geostress field was generated using an elastoplastic solution method with varying parameters. By setting an empty model to represent the excavated part of the soil model, and assigning a new constitutive model and material parameters to represent the tunnel segments and grouting layer, the shield tunneling excavation process is simulated in a loop to obtain surface settlement data of shield tunneling. The simulation results of the three-dimensional refined finite difference model are compared with the monitoring data of the actual project to verify the rationality of the simulation results.

3. The shield tunneling surface settlement prediction method based on three-dimensional random fields as described in claim 1, characterized in that, The empirical three-dimensional random field for constructing soil layer parameters based on statistical eigenvalues ​​and empirical values ​​specifically includes: Based on the mesh model in the pre-constructed three-dimensional refined finite difference model of shield tunneling construction, the positional relationship of each mesh is determined; Based on statistical characteristic values ​​and empirical values, as well as the positional relationships of each grid, a Gaussian covariance model or a normal distribution model is used to construct a three-dimensional random field of soil parameters, thereby obtaining the empirical three-dimensional random field.

4. The shield tunneling surface settlement prediction method based on three-dimensional random fields as described in claim 3, characterized in that, The empirical three-dimensional random field for soil parameters, constructed based on statistical eigenvalues, includes soil elastic modulus, soil internal friction angle, and soil cohesion. Random fields are generated for each soil parameter separately, and then the random fields of all soil parameters are combined to form the final three-dimensional random field. The statistical characteristics of the spatial variability of soil parameters include the mean, variance, and multiple of the correlation lengths in three orthogonal directions.

5. The shield tunneling surface settlement prediction method based on three-dimensional random fields as described in claim 1, characterized in that, During training, the surrogate model performs intrinsic orthogonal decomposition on the output samples in the surface settlement prediction dataset to obtain the modal basis and modal coefficient supervision values. The output of the surrogate model is the modal coefficient prediction value, and the shield tunnel surface settlement prediction data is obtained based on the modal basis reconstruction.

6. The shield tunneling surface settlement prediction method based on three-dimensional random fields as described in claim 1, characterized in that, Also includes: Collect shield tunnel surface settlement monitoring data during the construction process of the target project. Based on the shield tunnel surface settlement monitoring data and the shield tunnel surface settlement prediction data output by the proxy model, optimize and correct the statistical characteristic values ​​of soil elastic modulus, soil internal friction angle and soil cohesion, and obtain the corrected statistical characteristic values. A modified three-dimensional random field for soil parameters is constructed based on the statistical eigenvalue correction value, and the modified three-dimensional random field will be used for future shield tunneling surface settlement prediction.

7. The shield tunneling surface settlement prediction method based on three-dimensional random fields as described in claim 6, characterized in that, Obtaining the statistical feature value correction value specifically includes: Based on the geological survey data of the target project, determine the actual values ​​of the statistical characteristic values ​​of the spatial variability of soil elastic modulus, soil internal friction angle, and soil cohesion, and then obtain the prior distribution. By integrating the surface settlement monitoring data of the tunnel boring machine (TBM) and the surface settlement prediction data of the TBM output by the surrogate model, the posterior distribution of the soil elastic modulus, soil internal friction angle, and soil cohesion is updated using Bayesian theory combining likelihood and prior, and the statistical characteristic value correction value is obtained.

8. A shield tunneling surface settlement prediction system based on a three-dimensional random field, characterized in that, The system includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it performs the shield tunneling surface settlement prediction method based on a three-dimensional random field as described in any one of claims 1-7.

Images