A shield tunnel parameter prediction method and system based on deep learning

Abstract

The application discloses a shield tunnel parameter prediction method and system based on deep learning, relates to the technical field of geological engineering data analysis, and comprises the following steps: collecting dynamic working parameters of a shield machine and generating a point cloud matrix; performing position coding and high-frequency feature mapping on three-dimensional coordinates and multi-dimensional attribute information in the point cloud matrix to generate a standardized geological feature vector set; constructing a physical simulation deviation model, inputting a ground surface settlement prediction value into the physical simulation deviation model, outputting shield tunnel parameters, and generating a deviation analysis report based on the shield tunnel parameters; and applying adjusted model parameters to the physical simulation deviation model, generating updated shield tunnel parameters through iterative processing. Through the fusion of deep learning and physical simulation modeling, the application realizes high-precision prediction and dynamic optimization of shield tunnel parameters.

Description

Technical Field

[0001] This invention relates to the field of geological engineering data analysis technology, and in particular to a method and system for predicting shield tunnel parameters based on deep learning. Background Technology

[0002] Shield tunnel parameter prediction typically employs a method combining historical data statistical analysis with physical mechanism models to guide the optimization of tunneling parameters and the prediction of surface settlement. This approach relies on multi-source sensor data acquisition and spatiotemporal alignment technology, using traditional machine learning or numerical simulation methods to extract geological features and extrapolate settlement trends, providing a basis for decision-making in intelligent tunnel construction.

[0003] However, existing shield tunnel parameter prediction methods have limitations in capturing high-frequency details of various complex geological features, and their spatiotemporal correlation analysis of multi-source heterogeneous data is insufficient, making it difficult to further improve prediction accuracy and generalization ability. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a deep learning-based method for predicting shield tunnel parameters to address the problems of insufficient high-frequency geological feature representation and inadequate fusion of multi-source data.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for predicting shield tunnel parameters based on deep learning, comprising: collecting dynamic operating parameters of a shield machine and generating a point cloud matrix; performing position encoding and mapping processing on the three-dimensional coordinates and multi-dimensional attribute information in the point cloud matrix to generate a standardized geological feature vector set; fusing the standardized geological feature vector set with the dynamic operating parameters of the shield machine to generate a multi-dimensional spatiotemporal feature tensor, and performing a correlation analysis between geological settlement trend and equipment parameters based on the multi-dimensional spatiotemporal feature tensor to generate a predicted surface settlement value; constructing a physical simulation deviation model, inputting the predicted surface settlement value into the physical simulation deviation model, outputting shield tunnel parameters, and generating a deviation analysis report based on the shield tunnel parameters; adjusting the model parameters of the physical simulation deviation model with the deviation value of the shield tunnel parameters as the optimization target to generate adjusted model parameters; applying the adjusted model parameters to the physical simulation deviation model, and generating updated shield tunnel parameters through iterative processing.

[0007] As a preferred embodiment of the deep learning-based shield tunnel parameter prediction method of the present invention, the specific steps for generating the point cloud matrix are as follows: Time alignment is performed on the tunneling parameters in the dynamic working parameters of the shield machine to obtain a tunneling parameter sequence; the tunneling parameter sequence is aligned with the geological correlation parameters, position and attitude parameters, operating status parameters, and environmental and auxiliary parameters in the dynamic working parameters of the shield machine using coordinate system one to generate tunneling status data; spatial gridding is performed on the tunneling status data to generate the point cloud matrix. As a preferred embodiment of the deep learning-based shield tunnel parameter prediction method of the present invention, the specific steps for performing position encoding and mapping processing on the three-dimensional coordinates and multi-dimensional attribute information in the point cloud matrix are as follows: Based on the spatial distribution of the point cloud matrix, a mapping relationship framework from coordinates to geological features is established, and a continuous mapping function from spatial points to geological attribute values ​​is established; the high-frequency response range of the position encoding is expanded according to the continuous mapping function to construct a parameterized model that continuously represents geological features; position encoding is performed on each three-dimensional point in the point cloud matrix based on the parameterized model to obtain the encoded point cloud; high-frequency mapping processing is performed on the three-dimensional coordinates in the encoded point cloud to generate high-dimensional position features.

[0008] As a preferred embodiment of the deep learning-based shield tunnel parameter prediction method of the present invention, the specific steps for fusing the standardized geological feature vector set with the shield machine dynamic operating parameters to generate a multidimensional spatiotemporal feature tensor are as follows: performing temporal alignment on the standardized geological feature vector set and the shield machine dynamic operating parameters to obtain a joint feature sequence; performing spatial recombination processing on the joint feature sequence to generate the multidimensional spatiotemporal feature tensor.

[0009] As a preferred embodiment of the deep learning-based shield tunnel parameter prediction method of the present invention, the step of performing correlation analysis between geological subsidence trend and equipment parameters based on the multidimensional spatiotemporal feature tensor to generate surface subsidence prediction values ​​includes the following steps: performing feature decomposition on the multidimensional spatiotemporal feature tensor to obtain the long-term trend tensor of geological subsidence and the equipment parameter tensor; and calculating the correlation between the long-term trend tensor and the equipment parameter tensor to obtain the surface subsidence prediction values.

[0010] As a preferred embodiment of the deep learning-based shield tunnel parameter prediction method of the present invention, the construction of the physical simulation deviation model includes the following steps: integrating the tunnel design axis coordinates, segment assembly parameters and soil mechanical parameters based on the interaction between the shield tunnel axis geometric parameters and the strata to obtain the physical simulation basic framework; obtaining historical construction deviation data; and performing correction on the segment stress and deformation parameters of the physical simulation basic framework according to the historical construction deviation data and the multi-condition mechanical response relationship to obtain the physical simulation deviation model.

[0011] As a preferred embodiment of the deep learning-based shield tunnel parameter prediction method of the present invention, the method involves: inputting the predicted surface settlement value into the physical simulation deviation model, outputting shield tunnel parameters, and generating a deviation analysis report based on the shield tunnel parameters. The specific steps are as follows: inputting the predicted surface settlement value into the physical simulation deviation model; performing local deformation correction on the spatial coordinates of the shield tunnel based on the physical simulation deviation model to obtain the shield tunnel parameters; performing a full-domain scan on the shield tunnel parameters to generate an original deviation dataset; performing multidimensional correlation analysis on the original deviation dataset; and generating the deviation analysis report based on the analysis results of the multidimensional correlation analysis.

[0012] As a preferred embodiment of the deep learning-based shield tunnel parameter prediction method of the present invention, the analysis results include main deviation characteristics, spatial distribution law of deviation, correlation evaluation between parameters, and deviation cause analysis.

[0013] As a preferred embodiment of the deep learning-based shield tunnel parameter prediction method of the present invention, the steps of applying the adjusted model parameters to the physical simulation deviation model and generating updated shield tunnel parameters through iterative processing are as follows: The adjusted model parameters are then updated into the physical simulation deviation model to obtain the updated physical simulation deviation model. Based on the updated physical simulation deviation model, deviation calculations are performed on the shield tunnel parameters to obtain deviation values; Based on the deviation value, gradient adjustment is performed on the shield tunnel parameters to generate the objective function of the shield tunnel parameters, and the physical simulation deviation is calculated by the finite difference method to obtain the gradient direction of the shield tunnel parameters. The shield tunnel parameters are iteratively updated based on the gradient direction of the shield tunnel parameters to generate the updated shield tunnel parameters.

[0014] Secondly, this invention provides a deep learning-based shield tunnel parameter prediction system, comprising: a parameter processing module for acquiring dynamic operating parameters of the shield machine and generating a point cloud matrix; a feature encoding module for performing position encoding and mapping processing on the three-dimensional coordinates and multi-dimensional attribute information in the point cloud matrix to generate a standardized geological feature vector set; a prediction analysis module for fusing the standardized geological feature vector set with the dynamic operating parameters of the shield machine to generate a multi-dimensional spatiotemporal feature tensor, and performing a correlation analysis between geological settlement trend and equipment parameters based on the multi-dimensional spatiotemporal feature tensor to generate a predicted surface settlement value; a model analysis module for constructing a physical simulation deviation model, inputting the predicted surface settlement value into the physical simulation deviation model to generate shield tunnel parameters, and generating a deviation analysis report based on the shield tunnel parameters; a parameter calibration module for adjusting the model parameters of the physical simulation deviation model with the deviation value of the shield tunnel parameters as the optimization target to generate adjusted model parameters; and a parameter update module for applying the adjusted model parameters to the physical simulation deviation model and generating updated shield tunnel parameters through iterative processing.

[0015] The beneficial effects of this invention are as follows: By fusing a standardized set of geological feature vectors with the dynamic operating parameters of the tunnel boring machine (TBM), a multidimensional spatiotemporal feature tensor is generated to achieve a high-dimensional spatiotemporal correlation representation of geological attributes and equipment status, thereby improving the accuracy and reliability of settlement trend prediction. Furthermore, this invention constructs and iteratively optimizes a physical simulation deviation model to correct and control the deviations of the TBM parameters, thereby enhancing the system's adaptability and decision support capabilities under complex geological conditions. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a deep learning-based method for predicting shield tunnel parameters.

[0018] Figure 2 A schematic diagram of point cloud matrix generation.

[0019] Figure 3 This is a flowchart for feature encoding and the generation of multidimensional spatiotemporal feature tensors.

[0020] Figure 4 A flowchart for building, updating, and optimizing a physical simulation deviation model. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0024] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for predicting shield tunnel parameters based on deep learning, including the following steps: S1: Collect dynamic operating parameters of the tunnel boring machine and generate a point cloud matrix.

[0025] In this invention, generating the point cloud matrix in step S1 includes the following steps A1 to A3: Step A1: Perform time alignment on the tunneling parameters in the dynamic working parameters of the tunnel boring machine to obtain the tunneling parameter sequence.

[0026] Specifically, through time alignment processing, the data points corresponding to the tunneling parameters in the dynamic working parameters of the tunnel boring machine are arranged and adjusted in chronological order to ensure that all data points have a consistent time reference, thereby obtaining the tunneling parameter sequence.

[0027] It should be noted that time alignment is achieved by adjusting timestamps to synchronize tunneling parameters with the dynamic operating parameters of the tunnel boring machine in the time dimension, thereby giving data points a consistent time reference.

[0028] The dynamic operating parameters of a tunnel boring machine (TBM) refer to the various types of parameters collected in real time by multiple sensors mounted on the TBM during tunnel construction. These parameters include: tunneling parameter sequences, such as time-series data on propulsion force, cutterhead torque, and propulsion speed; geological parameters, such as earth pressure, slurry pressure, and formation density; position and attitude parameters, such as GPS coordinates, mileage, pitch angle, and roll angle; operating status parameters, such as cutterhead speed, screw conveyor speed, and equipment temperature; and environmental and auxiliary parameters, such as grouting pressure, grouting volume, and ambient temperature.

[0029] It should be noted that a data point refers to each independent record in the tunneling status data, which contains the values ​​of tunneling parameters, geological correlation parameters, position and attitude parameters, system operating status parameters, and environmental and auxiliary parameters of the tunnel boring machine (TBM) dynamic operation. Step A2: Combine the tunneling parameter sequence in the TBM dynamic operation parameters with the geological correlation parameters, position and attitude parameters, operating status parameters, and environmental and auxiliary parameters in the TBM dynamic operation parameters, and execute coordinate system one to generate tunneling status data.

[0030] Specifically, the tunneling parameter sequence is unified into the same coordinate system with the geological correlation parameters, position and attitude parameters, operating status parameters, and environmental and auxiliary parameters in the dynamic working parameters of the tunnel boring machine through time synchronization and spatial coordinate transformation, so as to ensure that all data points are aligned in time and space, thereby generating tunneling status data.

[0031] Step A3: Perform spatial grid reconstruction on the tunneling status data to generate a point cloud matrix.

[0032] Specifically, the three-dimensional coordinates (x, y, z) of each data point are calculated based on the position and attitude parameters in the tunneling status data. The three-dimensional coordinates and the corresponding original shield machine dynamic working parameters (including the tunneling parameters, geological correlation parameters, position and attitude parameters, operating status parameters, and environmental and auxiliary parameters collected in step S1) are arranged in a grid according to spatial position to form a two-dimensional matrix. Each row of the two-dimensional matrix corresponds to a spatial point, and the columns correspond to the three-dimensional coordinate values ​​and multi-dimensional attribute values, thereby generating a point cloud matrix.

[0033] It should be noted that in this invention, a spatial point refers to the position representation of a data point in three-dimensional space after coordinate transformation. Each spatial point not only has three-dimensional coordinates (x, y, z) but also carries all multi-dimensional attribute information of the corresponding data point. Therefore, a spatial point can be understood as a mapping representation of a data point in a unified spatial coordinate system. The three-dimensional coordinates of each spatial point are calculated by performing coordinate transformation on the position and attitude parameters in the tunneling status data, representing the position values ​​of the spatial point in the X, Y, and Z axes of the unified spatial coordinate system. The multi-dimensional attribute information represents all dynamic working parameter values ​​of the tunnel boring machine associated with each spatial point, specifically including the original values ​​of tunneling parameters, geological correlation parameters, position and attitude parameters, operating status parameters, and specific values ​​of environmental and auxiliary parameters. S2: Perform position encoding and mapping processing on the three-dimensional coordinates and multi-dimensional attribute information in the point cloud matrix to generate a standardized set of geological feature vectors.

[0034] In this invention, the specific operations for generating the standardized geological feature vector set in step S2 include the following steps B1 to B4: Step B1: Based on the spatial distribution of the point cloud matrix, establish a framework for mapping relationships from coordinates to geological features, and establish a continuous mapping function from spatial points to geological attribute values.

[0035] Furthermore, the point cloud matrix has a spatial range, which refers to the three-dimensional spatial region defined by the maximum and minimum values ​​of the three-dimensional coordinates of all data points in the point cloud matrix in the X, Y, and Z directions. Specifically, it is the three-dimensional spatial region defined by the maximum and minimum values ​​of the three-dimensional coordinates of all data points in the point cloud matrix in the X direction [X...]. min ,X max ], Y direction [Y min ,Y max ], Z direction [Z min Z max A rectangular space.

[0036] Furthermore, the three-dimensional coordinates are mapped to the geological attribute values ​​of the corresponding grid cells, establishing a framework for the mapping relationship between coordinates and geological features. The specific operations are as follows: Establish a regular grid based on the spatial extent of the point cloud matrix; A bilinear interpolation method is used to calculate a weighted average of the geological attribute values ​​of data points in four neighboring grid cells around the target location for any target location. The weights are determined based on the Euclidean distance from the target location to each neighboring data point. The Euclidean distance is the straight-line distance between the three-dimensional coordinates (x, y, z) of the target location and the three-dimensional coordinates of the neighboring data points. The weights are inversely proportional to the Euclidean distance, that is, the closer the data point is, the greater the weight, and the farther the data point is, the smaller the weight. This yields continuous geological attribute values ​​for the target location, thereby establishing a continuous mapping function from spatial points to geological attribute values.

[0037] It should be noted that this invention establishes a correspondence between three-dimensional spatial coordinates and geological attribute values ​​through a continuous mapping function. The input to the continuous mapping function is the three-dimensional coordinates (x, y, z) at any location in a unified spatial coordinate system, and the output is the geological attribute value corresponding to that arbitrary location. By constructing the continuous mapping function using bilinear interpolation, the geological attribute values, which originally only existed at discrete spatial points, are extended to any location within the spatial range of the point cloud matrix, thereby achieving the transformation from discrete sampling to a continuous field.

[0038] Therefore, the working principle of the continuous mapping function is as follows: For any target location in a unified spatial coordinate system, the continuous mapping function identifies several spatial points closest to the target location. Based on the geological attribute values ​​of these spatial points and the distance relationship between the spatial points and the target location, the geological attribute value of the target location is calculated. The closer the spatial points are, the greater their influence on the target location; the farther the spatial points are, the smaller their influence on the target location. Therefore, this invention can use distance weighting to enable the continuous mapping function to reasonably estimate the corresponding geological attribute value for any target location's three-dimensional coordinates (x, y, z) within the spatial range of the point cloud matrix, even if the target location is not the original spatial point location.

[0039] Preferably, the conversion from discrete spatial points to a continuous spatial field in step B1 allows the position encoding processing in step B2 and the high-frequency mapping processing in step B4 to be performed at any target location within the spatial range of the point cloud matrix, rather than being limited to the location of the original spatial points, thereby improving the accuracy and continuity of geological feature representation.

[0040] It should be noted that the bilinear interpolation method in this invention obtains spatial data of geological attribute values ​​by performing linear interpolation calculations in two directions respectively. This invention uses the bilinear interpolation method to process the spatial gridding and reorganization stage of the point cloud matrix, interpolating discrete geological attribute values ​​to generate a continuous mapping function. Specifically, this is achieved by weighted averaging of the geological attribute values ​​of adjacent grid points in the horizontal and vertical directions to obtain the geological attribute value of any point. The geological attribute values ​​in this invention refer to continuous variables used to characterize the physical and mechanical properties of strata during shield tunneling. These values ​​are collected in real time through field exploration data, ground-penetrating radar detection, or sensors, reflecting the stability, excavability, and impact on the surrounding environment of strata at different spatial locations. They serve as a fundamental component of multidimensional attribute information in the point cloud matrix, together with the three-dimensional coordinates.

[0041] Step B2: Construct a parameterized model that continuously represents geological features by expanding the high-frequency response range of the location encoding based on the continuous mapping function.

[0042] Specifically, based on the mapping relationship from spatial points to geological attribute values ​​defined by the continuous mapping function, the high-frequency response range is expanded by enhancing the density and range of frequency components in the location encoding, thereby constructing a parametric model that continuously represents geological features based on learnable parameters.

[0043] It should be noted that the high-frequency response range refers to the range of capabilities of the location encoding function to characterize rapidly changing or high-frequency details in spatial location. The parameterized model for continuously characterizing geological features in this invention is a mathematical model that converts three-dimensional coordinates into high-dimensional encoded vectors. The parameterized model uses multiple sine and cosine basis functions of different frequencies to map each three-dimensional coordinate (x, y, z) to a high-dimensional feature space. The construction process of the parameterized model includes: setting basis functions at multiple frequency levels, with the frequency range determined based on the scale of geological feature changes revealed by the continuous mapping function; encoding the three-dimensional coordinates using basis functions of different frequencies; and combining the multi-frequency encoding results to form a high-dimensional location encoded vector. Therefore, the parameterized model in this invention can guarantee the ability to capture changes in geological features at different scales, from macroscopic to microscopic.

[0044] Step B3: Perform position encoding on each 3D point in the point cloud matrix based on the parameterized model to obtain the encoded point cloud.

[0045] Specifically, the three-dimensional coordinates (x, y, z) of each spatial point in the point cloud matrix are input into a parameterized model that continuously represents geological features. The parameterized model calculates the high-dimensional position encoding vector corresponding to each three-dimensional coordinate through multi-frequency basis functions.

[0046] The calculation process of the multi-frequency basis function is as follows: The parameterized model pre-sets N different frequency values, which increase exponentially from low to high, covering spatial scales from meters to millimeters. For each frequency value, the parameterized model multiplies the X, Y, and Z components of the three-dimensional coordinates with the frequency value to obtain the modulated coordinate value. The modulated coordinate value is then input into sine and cosine functions to generate periodic codes. The sine and cosine coding results for all frequencies are arranged sequentially to form a high-dimensional vector. For example, for a single coordinate component x and frequency f, the generated code contains two values: sin(fx) and cos(fx). When there are N frequencies, a single coordinate component generates 2N coded values; the three coordinate components generate a total of 6N coded values, forming a high-dimensional positional coding vector.

[0047] The parametric model concatenates and combines the high-dimensional location encoding vector with the multi-dimensional attribute information of the original spatial points to form an enhanced feature representation, thus obtaining the encoded point cloud.

[0048] It should be noted that each spatial point in the encoded point cloud contains two parts of information: first, a high-dimensional location encoding vector generated by a parametric model, capturing the multi-scale features of the spatial location; and second, the original multi-dimensional attribute information, retaining the actual measured values ​​of the tunnel boring machine's dynamic operating parameters. Preferably, location encoding enhances the spatial points' ability to represent geological detail changes, providing a sufficient feature base for the high-frequency mapping processing in step B4.

[0049] Step B4: Perform high-frequency mapping processing on the 3D coordinates in the encoded point cloud to generate high-dimensional position features.

[0050] Specifically, the high-frequency mapping process extracts the high-dimensional location encoding vector generated in step B3 from each spatial point in the encoded point cloud, and enhances the expression of high-frequency components through further nonlinear transformation. Specifically, the high-frequency mapping process includes: inputting the high-dimensional location encoding vector into a multilayer perceptron structure, where each layer of the multilayer perceptron structure contains linear transformations and nonlinear activation functions; wherein the nonlinear activation functions amplify the high-frequency components in the encoding vector and suppress low-frequency components; enhancing detailed features while preserving the original encoding information through residual connections; and outputting high-dimensional location features with expanded dimensions.

[0051] It should be noted that the high-dimensional location features are spatial feature representations further extracted and enhanced through nonlinear transformation based on the high-dimensional location encoding vector in step B3. Compared with the original location encoding vector, the high-dimensional location features have a stronger ability to express high-frequency details and can more accurately characterize the local changes and boundary information of geological features. It should also be noted that the generated high-dimensional location features, together with the multi-dimensional attribute information, will constitute a standardized geological feature vector set.

[0052] S3: Integrate the standardized geological feature vector set with the dynamic working parameters of the tunnel boring machine to generate a multidimensional spatiotemporal feature tensor, and perform a correlation analysis between the geological settlement trend and the equipment parameters based on the multidimensional spatiotemporal feature tensor to generate a predicted value of surface settlement.

[0053] In this invention, the steps for generating predicted surface subsidence values ​​include the following steps C1 to C4: Step C1: Perform time-series alignment between the standardized geological feature vector set and the dynamic working parameters of the tunnel boring machine to obtain a joint feature sequence.

[0054] Specifically, by identifying the timestamps of the standardized geological feature vector set and the timestamps of the tunnel boring machine's dynamic working parameters, and based on time series matching, the two are adjusted to a unified time grid to ensure that the standardized geological feature vectors and the tunnel boring machine's dynamic working parameter values ​​corresponding to each time point exist, thereby combining them into a joint feature sequence in chronological order.

[0055] Step C2: Perform spatial recombination on the joint feature sequence to generate a multidimensional spatiotemporal feature tensor; Specifically, spatial recombination processing is performed on the joint feature sequence. By identifying the original spatial point positions corresponding to the standardized geological feature vectors in the joint feature sequence, the standardized geological feature vectors of each time step are assigned to the corresponding grid positions within the spatial range of the point cloud matrix. Each grid unit aggregates the feature values ​​of different time steps at the same spatial position and organizes them along the time dimension to form a four-dimensional tensor structure, generating a multi-dimensional spatiotemporal feature tensor.

[0056] Step C3: Perform eigenvalue decomposition on the multidimensional spatiotemporal feature tensor to obtain the long-term trend tensor of geological subsidence and the equipment parameter tensor.

[0057] Specifically, the feature decomposition method is applied to the multidimensional spatiotemporal feature tensor to decompose it into a series of feature components. From these components, the components representing the slow change process of geological subsidence are identified and separated as the long-term trend tensor of geological subsidence, and the components reflecting the dynamic working parameter change pattern of the tunnel boring machine are extracted as the equipment parameter tensor.

[0058] It should be noted that eigenvalue decomposition is a mathematical technique used to decompose a multidimensional spatiotemporal feature tensor into eigenvectors and eigenvalues. This method extracts the long-term trend tensor of geological subsidence and the equipment parameter tensor for correlation analysis. The component representing the slow change process of geological subsidence is the long-term trend tensor of geological subsidence, which is extracted from the multidimensional spatiotemporal feature tensor through eigenvalue decomposition and represents the long-term trend of geological subsidence.

[0059] Step C4: Based on the correlation between the long-term trend tensor and the equipment parameter tensor, the predicted value of surface subsidence is obtained.

[0060] Specifically, by calculating the covariance matrix or similarity index between the long-term trend tensor and the equipment parameter tensor, and by using the linear dependence between the geological subsidence trend and the equipment operating status, the current value of the equipment parameter tensor is applied as a weighting factor to the corresponding element of the long-term trend tensor. The predicted value of surface subsidence is then derived by using a linear combination or scalar multiplication of the long-term trend tensor and the equipment parameter tensor.

[0061] It should be noted that in this invention, linear dependence refers to a linear relationship between two or more variables, where a change in one variable can be predicted by a linear function of another variable. Linear dependence is reflected in the correlation calculation of the long-term trend tensor and the equipment parameter tensor, used to quantify the degree of linear influence between geological subsidence and equipment parameters. In this invention, linear combination or scalar multiplication quantifies the linear relationship between geological subsidence trends and equipment parameters.

[0062] S4. Construct a physical simulation deviation model, input the predicted surface settlement values ​​into the physical simulation deviation model, output the shield tunnel parameters, and generate a deviation analysis report based on the shield tunnel parameters.

[0063] In this invention, the steps for generating a deviation analysis report include the following steps D1 to D5: Step D1: Based on the interaction between the shield tunnel axis geometric parameters and the strata, integrate the tunnel design axis coordinates, segment assembly parameters and soil mechanical parameters to obtain the basic framework for physical simulation.

[0064] Specifically, based on the interaction between the geometric parameters of the shield tunnel axis and the strata, the coordinate system method is used to align the tunnel design axis coordinates, segment assembly parameters and soil mechanical parameters to the same spatial reference system. Structured data is generated by spatial gridding and reconstruction, and the relationship between parameters is established using a mapping framework to obtain the basic framework for physical simulation.

[0065] It should be noted that the coordinate system method unifies the tunneling parameter sequence with the geological correlation parameters, position and attitude parameters, operating status parameters, and environmental and auxiliary parameters in the shield machine's dynamic working parameters into the same coordinate system through coordinate transformation or alignment operations, thereby generating consistent tunneling status data; segment assembly parameters refer to the key parameters in the segment installation process during shield tunnel construction, including the segment's position coordinates, installation angle, assembly sequence, and joint status; soil mechanics parameters refer to parameters describing the mechanical properties of soil, including shear strength, compression modulus, and permeability coefficient, used to characterize the mechanical behavior of the strata during shield tunnel construction and to simulate the interaction between the strata and the tunnel; within the same spatial reference system, all coordinates and data are based on a unified geometric coordinate system and measurement standard to ensure that the shield machine's dynamic working parameters, geological features, and tunnel design parameters are consistent and comparable in spatial location, thereby accurately integrating and analyzing multidimensional information.

[0066] Step D2: Obtain historical construction deviation data, and perform correction on the stress and deformation parameters of the tunnel segments of the physical simulation foundation frame based on the historical construction deviation data and the multi-condition mechanical response relationship to obtain the physical simulation deviation model.

[0067] Specifically, after obtaining historical construction deviation data, the mechanical behavior under different construction conditions is analyzed using multi-condition mechanical response relationships. By comparing the historical construction deviation data with the output results of the physical simulation foundation frame, the stress and deformation parameters of the segments of the physical simulation foundation frame are optimized and adjusted so that the simulation output matches the historical construction deviation data, thus obtaining the physical simulation deviation model.

[0068] It should be noted that in this invention, the multi-condition mechanical response relationship refers to the correlation between the mechanical behavior of shield tunnel segments under various construction conditions, different geological conditions, tunneling parameters, and environmental factors, and the resulting stress and deformation. This relationship is established through historical construction deviation data analysis and mechanical simulation, and is used to correct the parameters in the physical simulation model. In this invention, the process of comparing historical construction deviation data with the physical simulation framework involves comparing the historical construction deviation data with the simulated deviation values ​​calculated by the physical simulation framework under the same input conditions point by point, and calculating the distance between the actual deviation value and the simulated deviation value at each spatial point. Through the multi-condition mechanical response relationship, the stress and deformation parameters of the tunnel segments in the physical simulation framework are mathematically optimized and numerically adjusted, so that the physical simulation deviation model can accurately reflect the mechanical response behavior under various actual construction conditions.

[0069] Step D3: Input the predicted surface settlement value into the physical simulation deviation model, and perform local deformation correction on the spatial coordinates of the shield tunnel based on the physical simulation deviation model to obtain the shield tunnel parameters.

[0070] Specifically, the predicted surface settlement value is input into the physical simulation deviation model. Based on the physical simulation deviation model, the tunnel design axis coordinates, segment assembly parameters and soil mechanical parameters are integrated. Combined with historical construction deviation data and multi-condition mechanical response relationships, the spatial coordinates of the shield tunnel are locally deformed and corrected. The influence of the predicted surface settlement value is reflected by the coordinate adjustment, and the shield tunnel parameters are obtained.

[0071] It should be noted that the multi-condition mechanical response relationship refers to the response relationship between mechanical parameters such as segment stress and deformation and geological conditions, equipment parameters, etc., during shield tunnel construction under various construction conditions. It is established based on historical construction deviation data and is used to correct the segment stress and deformation parameters of the physical simulation basic framework. Local deformation correction is the process of adjusting the local deformation of the shield tunnel spatial coordinates by the physical simulation deviation model. By correcting the small deformations of the coordinate points caused by geological settlement or mechanical response, more accurate shield tunnel parameters are output.

[0072] Step D4: Perform a full-domain scan of the shield tunnel parameters to generate the original deviation dataset.

[0073] Specifically, when performing a full-domain scan of the shield tunnel parameters, each data point of the tunnel design axis coordinates, segment assembly parameters, and soil mechanics parameters contained in the shield tunnel parameters is compared with the corresponding tunnel design axis coordinates, segment assembly parameters, and soil mechanics parameters in the physical simulation framework. The deviation value of each data point is obtained, and all deviation values ​​are summarized to generate the original deviation dataset.

[0074] It should be noted that point-by-point difference calculation refers to calculating the difference between each spatial point or data point of the shield tunnel parameters and the tunnel design axis coordinates or expected parameter values ​​during the generation of the original deviation dataset, thereby quantifying the deviation value of each point.

[0075] Step D5: Perform multidimensional correlation analysis on the original deviation dataset, and generate a deviation analysis report based on the analysis results of the multidimensional correlation analysis.

[0076] Specifically, a multidimensional correlation analysis was performed on the shield tunnel parameter deviation values ​​in the original deviation dataset. The analysis included: identifying the correlations between deviations in tunnel design axis coordinates, segment assembly parameters, and soil mechanical parameters; grouping data points with similar deviation characteristics using cluster analysis (K-means algorithm, dividing data points into several categories based on Euclidean distance of deviation values); each cluster center representing a typical deviation characteristic, revealing the inherent distribution structure of the deviation data; and employing association rule mining to discover the correlations between different deviation patterns. Association rule mining identifies the occurrence of a deviation pattern by analyzing frequently occurring pattern combinations in the deviation dataset. If the probability of another deviation pattern occurring simultaneously, for example, when the horizontal deviation of the tunnel axis exceeds a preset horizontal deviation threshold (e.g., 50 mm), the probability that the segment assembly angle deviation also exceeds a preset angle deviation threshold (e.g., 2 degrees) reaches a set confidence level (e.g., 85%), then an association rule between the two is established to identify the potential causes and propagation paths of the deviation. The above correlation identification results are used as the main deviation feature description, the inherent distribution structure revealed by cluster analysis is used as the spatial distribution law of the deviation, the correlation relationship obtained by association rule mining is used as the correlation evaluation result between parameters, and the identified potential causes and propagation paths are used as the deviation causal analysis, and integrated to form a deviation analysis report.

[0077] S5: Using the deviation value of the shield tunnel parameters as the optimization target, adjust the model parameters of the physical simulation deviation model to generate the adjusted model parameters.

[0078] Specifically, based on the shield tunnel parameters and deviation analysis report output in step S4, the deviation value between each spatial point and the tunnel design axis coordinates in the shield tunnel parameters is extracted as the optimization target. The segment stress and deformation parameters in the physical simulation deviation model are adjusted and optimized through step-by-step training. The training process uses the gradient descent method, with the objective function being to minimize the sum of deviation values. The optimal adjustment amount for each model parameter is iteratively calculated. In each iteration, the parameter values ​​are updated based on the deviation gradient and the set learning rate. After multiple iterations, the deviation values ​​converge to a preset threshold, and finally, the adjusted model parameters are output.

[0079] It should be noted that step-by-step training refers to breaking down the model parameter adjustment process into multiple training steps, each optimizing parameters of a different category. For example, the first step optimizes the segment stress parameters, including the segment axial force coefficient, bending moment coefficient, and shear force coefficient; the second step optimizes the deformation parameters, including the segment radial deformation coefficient, longitudinal deformation coefficient, and joint rotation coefficient; and the third step optimizes the soil reaction parameters, including the foundation reaction coefficient and lateral earth pressure coefficient. This invention, through step-by-step training, first adjusts the segment stress parameters that have a significant impact on deviation, then adjusts the deformation-related parameters, and finally fine-tunes the soil reaction parameters, thereby improving training efficiency and convergence speed. The adjusted model parameters include updated segment stress coefficients, deformation moduli, soil reaction coefficients, and other key physical simulation parameters.

[0080] S6: Apply the adjusted model parameters to the physical simulation deviation model, and generate updated shield tunnel parameters through iterative processing.

[0081] In this invention, step S6 specifically includes the following steps E1 to E4: Step E1: Replace the adjusted model parameters (including segment stress coefficient and deformation parameters, etc.) generated in step S5 with the original parameter values ​​in the physical simulation deviation model to complete the update of the model parameters and form the updated physical simulation deviation model.

[0082] Specifically, the main deviation characteristics and correlation results between parameters are obtained through multidimensional correlation analysis. The stress and deformation parameters of the tunnel segments in the physical simulation deviation model are adjusted, and the gradient descent method is used to minimize the deviation between the shield tunnel parameters and the tunnel design axis coordinates, thereby generating the adjusted model parameters. These adjusted model parameters are then updated into the physical simulation deviation model to form the updated physical simulation deviation model.

[0083] Step E2: Based on the updated physical simulation deviation model, perform deviation calculations on the shield tunnel parameters to obtain the deviation values.

[0084] Specifically, based on the updated physical simulation deviation model, and using the corrected segment stress and deformation parameters, the deviation verification calculation of the shield tunnel parameters generated in step S4 is performed. The deviation calculation process is as follows: extract the three-dimensional coordinates of each spatial point in the shield tunnel parameters, including the actual position coordinates of the tunnel axis, the actual assembly position coordinates of the segments, and the actual deformed coordinates due to geological settlement; compare each actual coordinate with the theoretical target coordinates in the tunnel design axis coordinates, and calculate the Euclidean distance between the actual coordinates of each spatial point and the corresponding theoretical target coordinates, i.e., the straight-line distance between two points in three-dimensional space; sum the distance values ​​of all spatial points to obtain the deviation value characterizing the overall deviation. The larger the deviation value, the greater the deviation between the shield tunnel parameters and the design requirements, requiring further optimization and adjustment.

[0085] Step E3: Perform gradient adjustment on the shield tunnel parameters based on the deviation value to generate the objective function of the shield tunnel parameters, and perform physical simulation deviation calculation through the finite difference method to obtain the gradient direction of the shield tunnel parameters.

[0086] Specifically, the objective function of the shield tunnel parameters is defined as minimizing the deviation value. Then, the finite difference method is used to perform small perturbations on each component of the shield tunnel parameters. The perturbed shield tunnel parameters are then input into the physical simulation deviation model to recalculate the deviation value. By comparing the changes in the deviation value before and after the perturbation, the partial derivatives of the objective function with respect to the shield tunnel parameters are estimated, thereby obtaining the gradient direction of the shield tunnel parameters.

[0087] It should be noted that in this invention, the objective function is a mathematical expression centered on minimizing the deviation value. The gradient direction of the shield tunnel parameters relative to the deviation value is calculated using the finite difference method, which guides parameter updates to approximate the tunnel design axis coordinates. The finite difference method is a numerical method that approximates the derivative by calculating the difference of function values ​​under small changes. The gradient direction is estimated by changing the input parameters in small steps and observing the output changes.

[0088] Step E4: Iteratively update the shield tunnel parameters based on the gradient direction of the shield tunnel parameters to generate the updated shield tunnel parameters.

[0089] Specifically, based on the gradient direction obtained in step E3, the shield tunnel parameters are updated according to a set step size (e.g., 0.01). After each update, steps E2 and E3 are re-executed to calculate the new deviation value and gradient direction. The iterative process is repeated until the deviation value converges to an acceptable range (e.g., less than 10 mm) or reaches the maximum number of iterations (e.g., 100 times). Finally, the updated shield tunnel parameters are output.

[0090] This embodiment also provides a shield tunnel parameter prediction system based on deep learning, including: The parameter processing module is used to acquire the dynamic working parameters of the tunnel boring machine and generate a point cloud matrix; The feature encoding module is used to perform position encoding and mapping processing on the three-dimensional coordinates and multi-dimensional attribute information in the point cloud matrix to generate a standardized set of geological feature vectors. The predictive analysis module is used to integrate the standardized geological feature vector set with the dynamic working parameters of the tunnel boring machine to generate a multidimensional spatiotemporal feature tensor, and to perform a correlation analysis between the geological settlement trend and the equipment parameters based on the multidimensional spatiotemporal feature tensor to generate a predicted value of surface settlement. The model analysis module is used to construct a physical simulation deviation model, input the predicted surface subsidence value into the physical simulation deviation model, generate shield tunnel parameters, and generate a deviation analysis report based on the shield tunnel parameters. The parameter calibration module is used to adjust the model parameters of the physical simulation deviation model through step-by-step training processing based on the shield tunnel parameters and the deviation analysis report, and generate the adjusted model parameters. The parameter update module is used to apply the adjusted model parameters to the physical simulation deviation model and generate updated shield tunnel parameters through iterative processing.

[0091] This embodiment also provides a computer device applicable to the deep learning-based shield tunnel parameter prediction method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the deep learning-based shield tunnel parameter prediction method proposed in the above embodiment.

[0092] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0093] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the deep learning-based shield tunnel parameter prediction method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0094] In summary, this invention achieves high-dimensional spatiotemporal correlation representation of geological attributes and equipment status by: integrating a standardized set of geological feature vectors with the dynamic working parameters of the tunnel boring machine to generate a multidimensional spatiotemporal feature tensor, thereby improving the accuracy and reliability of settlement trend prediction; and by constructing and iteratively optimizing a physical simulation deviation model, it achieves dynamic correction and deviation control of tunnel parameters, thereby enhancing the system's adaptability and decision support capabilities under complex geological conditions.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting shield tunnel parameters, characterized in that, include: Collect dynamic operating parameters of the tunnel boring machine and generate a point cloud matrix; The three-dimensional coordinates and multi-dimensional attribute information in the point cloud matrix are subjected to position encoding and mapping to generate a standardized set of geological feature vectors; By integrating the standardized geological feature vector set with the dynamic working parameters of the tunnel boring machine, a multidimensional spatiotemporal feature tensor is generated. Based on the multidimensional spatiotemporal feature tensor, a correlation analysis between geological subsidence trend and equipment parameters is performed to generate a predicted value of surface subsidence. A physical simulation deviation model is constructed, the predicted surface settlement value is input into the physical simulation deviation model, the shield tunnel parameters are output, and a deviation analysis report is generated based on the shield tunnel parameters. Using the deviation value of the shield tunnel parameters as the optimization target, the model parameters of the physical simulation deviation model are adjusted to generate the adjusted model parameters; The adjusted model parameters are applied to the physical simulation deviation model, and updated shield tunnel parameters are generated through iterative processing. The specific steps for performing position encoding and mapping processing on the three-dimensional coordinates and multi-dimensional attribute information in the point cloud matrix are as follows: Based on the spatial distribution of the point cloud matrix, a continuous mapping function from spatial points to geological attribute values ​​is established. Based on the mapping relationship from spatial points to geological attribute values ​​defined by the continuous mapping function, the high-frequency response range is expanded by enhancing the density and range of frequency components in the location coding, and a parameterized model that continuously represents geological features is constructed. Based on the parameterized model, position encoding is performed on the three-dimensional coordinates of each spatial point in the point cloud matrix to obtain the encoded point cloud; each spatial point in the encoded point cloud contains a high-dimensional position encoding vector and the original multi-dimensional attribute information. For each spatial point in the encoded point cloud, a high-dimensional location encoding vector is extracted. The expression of high-frequency components is enhanced through further nonlinear transformation to generate high-dimensional location features. The high-dimensional location features and multi-dimensional attribute information together constitute a standardized geological feature vector set. The process of fusing the standardized geological feature vector set with the dynamic operating parameters of the tunnel boring machine to generate a multidimensional spatiotemporal feature tensor involves the following steps: The standardized geological feature vector set is time-series aligned with the dynamic operating parameters of the tunnel boring machine to obtain a joint feature sequence; Spatial recombination processing is performed on the joint feature sequence. By identifying the original spatial point positions corresponding to the standardized geological feature vectors in the joint feature sequence, the standardized geological feature vectors of each time step are assigned to the corresponding grid positions within the spatial range of the point cloud matrix. Each grid cell aggregates the feature values ​​of different time steps at the same spatial position and organizes them along the time dimension to form a four-dimensional tensor structure, generating a multi-dimensional spatiotemporal feature tensor.

2. The shield tunnel parameter prediction method as described in claim 1, characterized in that, The specific steps for generating the point cloud matrix are as follows: The tunneling parameters in the dynamic operating parameters of the tunnel boring machine are time-aligned to obtain a tunneling parameter sequence; The tunneling parameter sequence is combined with the geological correlation parameters, position and attitude parameters, operating status parameters, and environmental and auxiliary parameters in the dynamic working parameters of the tunnel boring machine to generate tunneling status data by applying coordinate system one. The tunneling status data is spatially gridded and reorganized to generate the point cloud matrix.

3. The shield tunnel parameter prediction method as described in claim 1, characterized in that, The process of performing correlation analysis between geological subsidence trends and equipment parameters based on the multidimensional spatiotemporal feature tensor to generate predicted surface subsidence values ​​involves the following steps: After performing eigenvalue decomposition on the multidimensional spatiotemporal feature tensor, the long-term trend tensor of geological subsidence and the equipment parameter tensor are obtained; The predicted value of land subsidence is obtained based on the correlation calculation between the long-term trend tensor and the equipment parameter tensor.

4. The shield tunnel parameter prediction method as described in claim 1, characterized in that, The construction of the physical simulation deviation model includes the following steps: Based on the interaction between the shield tunnel axis geometric parameters and the strata, the tunnel design axis coordinates, segment assembly parameters and soil mechanical parameters are integrated to obtain the basic framework for physical simulation; Historical construction deviation data is obtained, and the stress and deformation parameters of the segments of the physical simulation foundation frame are corrected based on the historical construction deviation data and the multi-condition mechanical response relationship to obtain the physical simulation deviation model.

5. A shield tunnel parameter prediction system, based on the shield tunnel parameter prediction method according to any one of claims 1 to 4, characterized in that, include: The parameter processing module is used to acquire the dynamic working parameters of the tunnel boring machine and generate a point cloud matrix; The feature encoding module is used to perform position encoding and mapping processing on the three-dimensional coordinates and multi-dimensional attribute information in the point cloud matrix to generate a standardized set of geological feature vectors. The predictive analysis module is used to integrate the standardized geological feature vector set with the dynamic working parameters of the tunnel boring machine to generate a multidimensional spatiotemporal feature tensor, and to perform a correlation analysis between the geological settlement trend and the equipment parameters based on the multidimensional spatiotemporal feature tensor to generate a predicted value of surface settlement. The model analysis module is used to construct a physical simulation deviation model, input the predicted surface subsidence value into the physical simulation deviation model, generate shield tunnel parameters, and generate a deviation analysis report based on the shield tunnel parameters. The parameter calibration module is used to adjust the model parameters of the physical simulation deviation model with the deviation value of the shield tunnel parameters as the optimization target, and generate the adjusted model parameters. The parameter update module is used to apply the adjusted model parameters to the physical simulation deviation model and generate updated shield tunnel parameters through iterative processing.

Images