Rail transit foundation pit support modeling method based on multi-source data fusion

Abstract

This application relates to the field of technical modeling, and in particular to a method for modeling rail transit foundation pit support based on multi-source data fusion, comprising: acquiring multi-source heterogeneous data of the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data, and a geological parameter database; when abnormal data is detected in the real-time monitoring data, the corresponding abnormal grid area in the initial three-dimensional model of the foundation pit is subjected to grid densification processing to obtain a densified three-dimensional model of the foundation pit; based on the densified three-dimensional model of the foundation pit, a foundation pit modeling and analysis model is constructed; the foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during foundation pit construction in real time based on the foundation pit three-dimensional model; the model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the foundation pit three-dimensional model; and updating the foundation pit modeling and analysis model based on the foundation pit modeling and analysis model, the densified three-dimensional model of the foundation pit, and the initial three-dimensional model of the foundation pit.

Description

Technical Field

[0001] This application relates to the field of technical modeling, and in particular to a modeling method for rail transit foundation pit support based on multi-source data fusion. Background Technology

[0002] Multi-source data fusion modeling for rail transit foundation pit support refers to the integration and analysis of data from different sources and of different types in rail transit foundation pit support engineering to construct a more accurate and comprehensive foundation pit support model.

[0003] The existing modeling technology is based on BIM: a three-dimensional model of the foundation pit support structure is created using BIM technology, and the construction process is simulated and optimized.

[0004] However, during construction, the condition of the foundation pit support structure will change continuously, and existing methods are unable to update the model in a timely manner based on real-time monitoring data. Summary of the Invention

[0005] Therefore, it is necessary to provide a modeling method for rail transit foundation pit support based on multi-source data fusion that can perform high-precision real-time tracking and updating of foundation pits, addressing the aforementioned technical problems.

[0006] Firstly, this application provides a modeling method for rail transit foundation pit support based on multi-source data fusion, the method comprising:

[0007] Acquire multi-source heterogeneous data on the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data, and a geological parameter database.

[0008] When abnormal data is detected in the real-time monitoring data, the number of grids in the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit is increased to obtain the encrypted three-dimensional model of the foundation pit.

[0009] Based on the 3D model of the foundation pit, a foundation pit modeling and analysis model is constructed. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during foundation pit construction in real time based on the 3D model of the foundation pit. The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the 3D model of the foundation pit.

[0010] The foundation pit modeling and analysis model is updated based on the foundation pit modeling and analysis model, the foundation pit densified 3D model, and the foundation pit initial 3D model.

[0011] In one embodiment, the multi-source heterogeneous data includes:

[0012] Real-time monitoring data is collected through a distributed sensor network, including three types of time-series signals: pore water pressure data, soil displacement data, and support stress data.

[0013] Static topology data is obtained from the BIM model and includes the spatial coordinates of the support structure and the geometric data of the soil layer interfaces.

[0014] The geological parameter database includes the permeability coefficient, elastic modulus, and Poisson's ratio of each soil layer.

[0015] In one embodiment, determining the abnormal mesh region corresponding to the abnormal data in the initial 3D model of the foundation pit includes:

[0016] Perform spatial gradient analysis on the pore water pressure data to generate a water pressure anomaly distribution map;

[0017] Perform time-domain trend analysis on soil displacement data and output the area where the displacement rate exceeds the limit;

[0018] Statistical outlier detection was performed on the support stress data to mark stress anomaly points;

[0019] Based on the water pressure anomaly distribution map, the displacement rate exceeding the limit area and the stress anomaly point, a multi-dimensional anomaly signal matrix with spatial coordinate markers is generated.

[0020] The multi-dimensional anomaly signal matrix includes the anomaly grid regions corresponding to the anomaly data in the initial three-dimensional model of the foundation pit.

[0021] In one embodiment, the abnormal mesh region corresponding to the abnormal data in the initial 3D model of the foundation pit is subjected to mesh count densification processing, including:

[0022] Based on the multi-dimensional anomaly signal matrix, generate the boundary polygon of the region to be encrypted;

[0023] Inside the boundary polygon of the region to be encrypted, hierarchical mesh subdivision is performed to generate a mesh structure with a core encryption area;

[0024] For the mesh structure with the core encryption area and the boundary polygon of the region to be encrypted, a gradient transition mesh layer is constructed along the extension of the boundary polygon of the region to be encrypted to obtain a mesh region with the core encryption area and the gradient transition mesh layer.

[0025] In one embodiment, a foundation pit modeling and analysis model is constructed based on the foundation pit densification 3D model, including:

[0026] The seepage field control equations of the three-dimensional model of the foundation pit are discretized into mass conservation constraint operators;

[0027] The stress field constitutive relation of the three-dimensional model of the foundation pit is discretized into a mechanical compatibility constraint operator;

[0028] Data-driven constraints are generated based on the 3D model of the foundation pit and real-time monitoring data.

[0029] The mass conservation constraint operator, the mechanical compatibility constraint operator, and the data-driven constraint term are integrated to construct a composite optimization objective function.

[0030] Based on the 3D model of the foundation pit and the composite optimization objective function, a network architecture is selected and the network is trained to construct a foundation pit modeling and analysis model; the foundation pit modeling and analysis model is a deep neural network proxy model with embedded physical mechanisms.

[0031] In one embodiment, based on the 3D model of the foundation pit and a composite optimization objective function, a network architecture is selected and the network is trained to construct a foundation pit modeling and analysis model, including:

[0032] Based on the 3D model of the foundation pit and the composite optimization objective function, the mesh topology relationship is analyzed by a shared feature extraction layer to extract the mesh topology features;

[0033] Based on the grid topology characteristics, the seepage field and stress field distributions are generated using parallel prediction branches, resulting in the predicted distributions of the seepage field and stress field.

[0034] Based on the predicted distribution of the seepage field and the predicted distribution of the stress field, a joint verification of physical conservation across branches is performed to obtain the physical conservation verification results.

[0035] Based on the physical conservation verification results, the model parameters are self-calibrated to obtain the corrected model parameters.

[0036] Based on the corrected model parameters, a foundation pit modeling and analysis model is constructed.

[0037] In one embodiment, the foundation pit modeling and analysis model is updated based on the foundation pit modeling and analysis model, the foundation pit densified 3D model, and the foundation pit initial 3D model, including:

[0038] Based on the foundation pit modeling and analysis model, the three-dimensional model of the foundation pit is analyzed to obtain the first prediction result;

[0039] Based on the foundation pit modeling and analysis model, the initial three-dimensional model of the foundation pit is analyzed to obtain the second prediction result;

[0040] The foundation pit modeling and analysis model is updated based on the first and second prediction results.

[0041] In one embodiment, the foundation pit modeling and analysis model is updated based on the first prediction result and the second prediction result, including:

[0042] For the first and second prediction results, the predicted values ​​of physical quantities at the same spatial location are extracted and a difference quantization matrix is ​​generated to obtain a spatial correlation prediction difference map with labeled difference magnitude and location coordinates.

[0043] Based on the complete set of parameters of the predicted difference map and the foundation pit modeling analysis model, the influence of each parameter on the difference map is analyzed and the sensitivity level is quantified. A sensitivity ranking list is generated, and a parameter sensitivity classification table is obtained.

[0044] Based on the parameter sensitivity classification table, set sensitivity thresholds to filter key parameters, divide the optimizable parameter subset and the frozen parameter set, generate a parameter operation instruction set, and obtain a dynamic optimization whitelist containing adjustable parameter ranges and a parameter freezing blacklist that locks non-sensitive parameters.

[0045] The foundation pit modeling and analysis model is updated based on the dynamically optimized whitelist and the parameter-frozen blacklist.

[0046] In one embodiment, the foundation pit modeling and analysis model is updated based on a dynamically optimized whitelist and a parameter-frozen blacklist, including:

[0047] Based on the dynamic optimization whitelist and the current model parameter status, the update step size is calculated and the learning rate is dynamically adjusted in combination with the historical gradient direction. The whitelist parameters are fine-tuned in a targeted manner, while the blacklist parameters are kept constant, and the foundation pit modeling and analysis model is updated.

[0048] Secondly, this application also provides a modeling device for rail transit foundation pit support based on multi-source data fusion, comprising:

[0049] The acquisition module is used to acquire multi-source heterogeneous data of the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data, and a geological parameter database.

[0050] The fine processing module is used to densify the number of grids in the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit when abnormal data is detected in the real-time monitoring data, so as to obtain the densified three-dimensional model of the foundation pit.

[0051] The analysis and modeling module is used to construct a foundation pit modeling and analysis model based on the encrypted 3D model of the foundation pit. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during foundation pit construction in real time based on the 3D model of the foundation pit. The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the 3D model of the foundation pit.

[0052] The modeling update module is used to update the foundation pit modeling analysis model based on the foundation pit modeling analysis model, the foundation pit encrypted 3D model, and the foundation pit initial 3D model.

[0053] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0054] Acquire multi-source heterogeneous data on the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data, and a geological parameter database.

[0055] When abnormal data is detected in the real-time monitoring data, the number of grids in the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit is increased to obtain the encrypted three-dimensional model of the foundation pit.

[0056] Based on the 3D model of the foundation pit, a foundation pit modeling and analysis model is constructed. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during foundation pit construction in real time based on the 3D model of the foundation pit. The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the 3D model of the foundation pit.

[0057] The foundation pit modeling and analysis model is updated based on the foundation pit modeling and analysis model, the foundation pit densified 3D model, and the foundation pit initial 3D model.

[0058] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0059] Acquire multi-source heterogeneous data on the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data, and a geological parameter database.

[0060] When abnormal data is detected in the real-time monitoring data, the number of grids in the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit is increased to obtain the encrypted three-dimensional model of the foundation pit.

[0061] Based on the 3D model of the foundation pit, a foundation pit modeling and analysis model is constructed. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during foundation pit construction in real time based on the 3D model of the foundation pit. The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the 3D model of the foundation pit.

[0062] The foundation pit modeling and analysis model is updated based on the foundation pit modeling and analysis model, the foundation pit densified 3D model, and the foundation pit initial 3D model.

[0063] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0064] Acquire multi-source heterogeneous data on the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data, and a geological parameter database.

[0065] When abnormal data is detected in the real-time monitoring data, the number of grids in the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit is increased to obtain the encrypted three-dimensional model of the foundation pit.

[0066] Based on the 3D model of the foundation pit, a foundation pit modeling and analysis model is constructed. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during foundation pit construction in real time based on the 3D model of the foundation pit. The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the 3D model of the foundation pit.

[0067] The foundation pit modeling and analysis model is updated based on the foundation pit modeling and analysis model, the foundation pit densified 3D model, and the foundation pit initial 3D model.

[0068] The aforementioned modeling method for rail transit foundation pit support based on multi-source data fusion acquires multi-source heterogeneous data of the foundation pit, including static topological data and a geological parameter database. This data provides fundamental information for initial modeling. Based on prior knowledge of geological parameters, a coarse-grained grid can be generated, thereby reducing the initial computational load. When abnormal data appears in the real-time monitoring data, the method triggers a refinement process for the abnormal grid area, resulting in a refined 3D model of the foundation pit. This adaptive grid generation method can dynamically adjust the grid accuracy according to the monitoring data, ensuring that the accuracy in key areas (such as areas of sudden local displacement) is improved to the millimeter level. The foundation pit modeling and analysis model is used to simulate the coupling effect of the seepage field and stress field during foundation pit construction. Its model prediction results include pore water pressure and effective stress. This is consistent with the idea of ​​embedding the finite element equation as a regularization term into the loss function in a physical information neural network, ensuring the consistency between data-driven and physical mechanisms. The model is updated based on the foundation pit modeling and analysis model, the refined 3D model of the foundation pit, and the initial 3D model of the foundation pit. This update method is similar to incremental updates, requiring only adjustments based on changes without the need for full retraining. This improves the accuracy of foundation pit monitoring while ensuring high-precision real-time tracking and updates. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0070] Figure 1 This is a flowchart illustrating a modeling method for rail transit foundation pit support based on multi-source data fusion in one embodiment.

[0071] Figure 2 This is a flowchart illustrating the steps for determining the abnormal mesh region corresponding to abnormal data in the initial three-dimensional model of the foundation pit, as shown in one embodiment.

[0072] Figure 3This is a flowchart illustrating the steps of encrypting the number of grids in the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit in one embodiment.

[0073] Figure 4 A flowchart illustrating the steps involved in constructing a foundation pit modeling and analysis model for one embodiment;

[0074] Figure 5 This is a flowchart illustrating the steps for constructing a foundation pit modeling and analysis model in another embodiment;

[0075] Figure 6 This is a flowchart illustrating the steps for updating the foundation pit modeling and analysis model in one embodiment.

[0076] Figure 7 This is a flowchart illustrating the steps for updating the foundation pit modeling and analysis model in another embodiment;

[0077] Figure 8 This is a structural block diagram of a rail transit foundation pit support modeling device based on multi-source data fusion in one embodiment. Detailed Implementation

[0078] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0079] In one exemplary embodiment, a method for modeling rail transit foundation pit support based on multi-source data fusion is provided, such as... Figure 1 As shown, the method includes:

[0080] S101, acquire multi-source heterogeneous data on the foundation pit.

[0081] Among them, multi-source heterogeneous data includes real-time monitoring data, static topological data, and geological parameter databases.

[0082] Optionally, real-time monitoring data is collected in real time through a sensor network installed around the foundation pit, covering key parameters such as displacement, stress, and seepage, and can promptly reflect the dynamic changes during the construction process. Static topology data describes the initial geometry, structural layout, and spatial relationships between the various parts of the foundation pit, providing a basic framework for 3D modeling. The geological parameter database contains geological information about the area where the foundation pit is located, such as soil layer distribution, lithological characteristics, and permeability coefficients.

[0083] S102, if abnormal data is detected in the real-time monitoring data, the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit is subjected to grid number densification processing to obtain the foundation pit densified three-dimensional model.

[0084] Optionally, real-time analysis can be performed on the acquired monitoring data. Once abnormal data is detected—i.e., data exceeding the preset normal range or exhibiting abrupt changes—the system will automatically identify and locate the corresponding abnormal mesh regions in the initial 3D model of the foundation pit. For these abnormal mesh regions, the mesh count will be increased to generate a refined 3D model of the foundation pit. This refined mesh allows for more detailed capture of changes in the mechanical behavior of abnormal areas, improving the model's accuracy and reliability, and providing a more accurate geometric basis for subsequent analysis.

[0085] S103. Based on the 3D model of the foundation pit, construct a foundation pit modeling and analysis model. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during the foundation pit construction process in real time based on the 3D model of the foundation pit.

[0086] The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the three-dimensional model of the foundation pit.

[0087] Optionally, a foundation pit modeling and analysis model can be constructed based on the encrypted 3D model of the foundation pit. This model can simulate the coupling effect of the seepage field and stress field during foundation pit construction in real time. During foundation pit construction, the seepage field and stress field are two key mechanical fields that influence and restrict each other. Changes in the seepage field will affect the stress state of the soil, while changes in the stress field will in turn affect the seepage path and seepage velocity. By simulating these two fields in a coupled manner, the mechanical behavior during foundation pit construction can be reflected more realistically.

[0088] The model's prediction results include pore water pressure and effective stress in the three-dimensional model of the foundation pit. These two parameters are important indicators for assessing the stability of the foundation pit. By monitoring and predicting these parameters in real time, potential safety hazards can be detected in advance.

[0089] S104. Update the foundation pit modeling and analysis model based on the foundation pit modeling and analysis model, the foundation pit densified 3D model, and the foundation pit initial 3D model.

[0090] During construction, as monitoring data accumulates and construction progresses, new monitoring data and encrypted 3D model information are fed back into the foundation pit modeling and analysis model. Data assimilation technology is employed to compare the new monitoring data with the model's prediction results, calibrate model parameters, and optimize the model structure.

[0091] Based on new monitoring data and encrypted 3D model information, the foundation pit modeling and analysis model is dynamically updated. Updates include adjusting the model's geometry, boundary conditions, and material parameters. The model's predictive accuracy and reliability are evaluated by comparing it with actual monitoring data. If a significant deviation is found between the model's predictions and the actual monitoring results, the model's assumptions and parameters are adjusted promptly to ensure that the model accurately reflects the current state of the foundation pit.

[0092] After the model is updated, the dynamic changes of the foundation pit are monitored in real time, and the updated model is used for real-time simulation and prediction. Based on the model prediction results, potential safety hazards are warned in advance, providing decision support for construction personnel.

[0093] In one exemplary embodiment, the multi-source heterogeneous data includes:

[0094] Real-time monitoring data is collected through a distributed sensor network, including three types of time-series signals: pore water pressure data, soil displacement data, and support stress data.

[0095] Static topology data is obtained from the BIM model and includes the spatial coordinates of the support structure and the geometric data of the soil layer interfaces.

[0096] The geological parameter database includes the permeability coefficient, elastic modulus, and Poisson's ratio of each soil layer.

[0097] Optional data includes: Pore water pressure data: Pore water pressure sensors are installed in the soil to monitor the pore water pressure in real time. Pore water pressure is a key factor affecting the stability of the foundation pit, and its changes reflect changes in the soil's seepage state and mechanical properties. Soil displacement data: Displacement sensors (such as total stations and displacement gauges) are used to monitor the displacement of the soil around the foundation pit. This data includes horizontal displacement, vertical displacement, and deep displacement within the soil, reflecting the deformation trend and stability of the soil during the excavation process. Support stress data: Stress sensors are installed on the support structure (such as diaphragm walls, anchors, and supports) to monitor the stress state of the support structure in real time. This data reflects the stress condition and bearing capacity of the support structure, providing a basis for assessing its safety. Time-series signals: All data are time-series signals, meaning the data changes continuously over time.

[0098] By analyzing time-series signals, the dynamic patterns of changes during foundation pit construction can be captured. Real-time monitoring data is crucial input for foundation pit modeling and analysis. By combining real-time monitoring data with a 3D model of the foundation pit, dynamic simulation and real-time early warning of the construction process can be achieved. For example, pore water pressure data is used to simulate the seepage field of the foundation pit and calculate the pore water pressure distribution; soil displacement data is used to assess the deformation of the foundation pit and predict potential deformation risks; and support stress data is used to analyze the stress state of the support structure and ensure its safety.

[0099] Optionally, static topology data can be obtained by parsing from Building Information Modeling (BIM). BIM is an integrated 3D digital model that contains detailed information such as the geometry, spatial relationships, geographic information, and attributes of building components of a construction project. Spatial coordinates of the support structure: This includes the spatial location, dimensions, and shape of the support structure (such as diaphragm walls, anchors, and braces). This data describes the specific layout of the support structure in the foundation pit, providing an accurate geometric basis for modeling and analysis. Geometric data of soil layer interfaces: This describes the distribution of soil layers in the area where the foundation pit is located, including the location, thickness, and shape of the interfaces of each soil layer. This data reflects the spatial distribution characteristics of the soil layers, providing a basis for assigning geological parameters and numerical simulation.

[0100] Static topology data forms the basis for constructing the initial 3D model of the foundation pit. By importing the spatial coordinates of the support structure and the geometric data of the soil layer interfaces into 3D modeling software, an initial geometric model of the foundation pit can be generated. This model provides an accurate geometric framework for subsequent mesh generation, numerical simulation, and coupled analysis.

[0101] Optionally, the geological parameter database is compiled from geological exploration reports. Geological exploration typically includes methods such as borehole sampling, in-situ testing, and geophysical exploration to obtain geological information about the area where the foundation pit is located. Permeability coefficient: describes the ability of water to permeate the soil layer and is a key parameter for simulating the seepage field of the foundation pit. The magnitude of the permeability coefficient directly affects the distribution and variation of pore water pressure. Elastic modulus: reflects the deformation capacity of soil in the elastic stage and is an important parameter for calculating the stress-strain relationship of soil. The larger the elastic modulus, the greater the stiffness of the soil. Poisson's ratio: describes the ratio of lateral deformation to longitudinal deformation of soil under stress, reflecting the deformation characteristics of the soil. The magnitude of Poisson's ratio affects the stress transfer and deformation behavior of the soil.

[0102] A geological parameter library is an indispensable input data source for foundation pit modeling and analysis. By assigning geological parameters to the corresponding soil layers in the 3D model of the foundation pit, accurate simulation of the seepage field and stress field during foundation pit construction can be achieved. For example, the permeability coefficient is used to calculate the distribution and changes of pore water pressure; the elastic modulus and Poisson's ratio are used to describe the mechanical behavior of the soil and calculate stress and deformation.

[0103] In one exemplary embodiment, such as Figure 2 As shown, the abnormal mesh regions corresponding to the abnormal data in the initial 3D model of the foundation pit are determined, including:

[0104] S201 performs spatial gradient analysis on the pore water pressure data to generate a water pressure anomaly distribution map.

[0105] Optionally, the collected pore water pressure data is cleaned and formatted to remove noise and invalid data. The spatial gradient of pore water pressure is calculated using the finite difference method or numerical differentiation method. Specifically, for each monitoring point, the water pressure difference between it and its adjacent monitoring points is calculated and divided by the spatial distance between the two points. Based on the magnitude of the spatial gradient, areas with abnormal water pressure changes are identified. A threshold is typically set; when the gradient exceeds this threshold, the area is considered to have an abnormal water pressure. The identified abnormal areas are visualized in the initial 3D model of the foundation pit to generate a water pressure anomaly distribution map.

[0106] Spatial gradient analysis is a method for identifying anomalous regions by calculating the rate of change of a physical quantity in space. In pore water pressure monitoring, the spatial gradient of water pressure reflects the variation of water pressure at different locations. Under normal circumstances, water pressure changes are relatively gradual, with a small gradient; however, when there are seepage anomalies or soil deformation, the water pressure gradient will increase significantly. By calculating and analyzing the spatial gradient of water pressure, areas of abnormal water pressure can be quickly located.

[0107] S202 performs time-domain trend analysis on soil displacement data and outputs the area where the displacement rate exceeds the limit.

[0108] Optionally, preprocessing of the soil displacement data before time series analysis can be performed, including data smoothing and handling of missing values. Time-domain trend analysis: Time series analysis methods (such as moving average, linear regression, etc.) are used to calculate the trend of the soil displacement data in the time domain. Displacement rate calculation: The soil displacement rate is obtained by calculating the difference or derivative of the displacement data. Exceeding-limit region identification: A threshold for the displacement rate is set; when the displacement rate exceeds this threshold, the region is considered to have exceeded the displacement rate limit. Output of exceeding-limit regions: Regions with exceeding displacement rate limits are marked and relevant information is output.

[0109] Time-domain trend analysis is a method for identifying abnormal changes by analyzing trends in time-series data. In soil displacement monitoring, displacement rate is a crucial indicator reflecting the speed of soil deformation. Under normal circumstances, the displacement rate is low and changes smoothly; however, when soil deformation intensifies during foundation pit construction, the displacement rate increases significantly. Through time-domain trend analysis, areas with excessive displacement rates can be identified in a timely manner, thus providing early warning of potential deformation risks.

[0110] S203 performs statistical outlier detection on support stress data and marks stress anomaly points.

[0111] Optional steps include: Data preprocessing: Cleaning and standardizing the support stress data to remove outliers and noise. Outlier detection method selection: Choosing a suitable statistical outlier detection method, such as the Z-score method, IQR method, or clustering-based outlier detection method. Outlier index calculation: Calculating the outlier index for each monitoring point according to the selected method. For example, when using the Z-score method, calculating the Z value for each data point. Outlier identification: Setting a threshold for the outlier index; when the outlier index exceeds this threshold, the point is considered a stress anomaly. Outlier location marking: Marking the identified stress anomalies in the initial 3D model of the foundation pit.

[0112] Statistical outlier detection is a method for identifying abnormal data based on the statistical distribution characteristics of the data. In support stress monitoring, normal stress data usually follows a certain statistical distribution (such as a normal distribution). When a data point deviates from the normal range of this distribution, it can be considered an outlier. By calculating outlier indices and setting thresholds, stress anomalies can be effectively identified, thus providing a basis for the safety assessment of the support structure.

[0113] S204 generates a multi-dimensional anomaly signal matrix with spatial coordinate markers based on the water pressure anomaly distribution map, the displacement rate exceeding the limit area, and the stress anomaly point.

[0114] The multi-dimensional anomaly signal matrix includes the anomaly grid regions corresponding to the anomaly data in the initial three-dimensional model of the foundation pit.

[0115] Optionally, the water pressure anomaly areas, displacement rate exceeding limits areas, and stress anomaly points identified in steps S201, S202, and S203 are integrated. Spatial coordinate matching: The anomaly data is spatially matched with the grid areas in the initial 3D model of the foundation pit to determine the grid position corresponding to each anomaly data. Multi-dimensional anomaly signal matrix construction: A multi-dimensional anomaly signal matrix is ​​constructed, containing the spatial coordinates, anomaly type (water pressure anomaly, displacement rate exceeding limits, stress anomaly), and anomaly degree of each anomaly data. Visualization and marking: In the initial 3D model of the foundation pit, the anomaly areas and points in the anomaly signal matrix are visualized and marked, generating a multi-dimensional anomaly signal matrix with spatial coordinate markings.

[0116] A multi-dimensional anomaly signal matrix is ​​a data structure that integrates anomaly information from multiple dimensions to comprehensively reflect abnormal situations during foundation pit construction. By integrating different types of anomaly data and combining them with spatial coordinates in the three-dimensional model of the foundation pit, precise location and multi-dimensional analysis of anomaly areas can be achieved. This method helps construction personnel understand the anomalies in the foundation pit more intuitively, enabling them to take timely and effective countermeasures.

[0117] Furthermore, such as Figure 3 As shown, the abnormal mesh regions corresponding to the abnormal data in the initial 3D model of the foundation pit are subjected to mesh count densification processing, including:

[0118] S301, Generate the boundary polygon of the region to be encrypted based on the multi-dimensional anomaly signal matrix.

[0119] Optionally, extract anomaly signals: extract the spatial coordinate information of all anomaly data from the multi-dimensional anomaly signal matrix. Determine boundary points: determine the boundary points of the region to be encrypted based on the spatial distribution of the anomaly points using geometric algorithms such as the Convex Hull algorithm or the Alpha Shape algorithm. Construct a polygon: connect the determined boundary points to form a closed polygon, which serves as the boundary of the region to be encrypted.

[0120] By utilizing the spatial coordinate information in the multi-dimensional anomaly signal matrix, the specific location of the anomaly data within the foundation pit model can be clearly identified. Generating boundary polygons using geometric algorithms accurately defines the area requiring encryption, providing clear boundary constraints for subsequent mesh encryption.

[0121] S302, within the boundary polygon of the region to be encrypted, perform hierarchical mesh subdivision to generate a mesh structure with a core encryption area.

[0122] Optionally, initialize the mesh: using the polygonal boundary of the region to be encrypted as the initial boundary, generate a coarse mesh structure. Hierarchical subdivision: recursively subdivide the initial mesh using a hierarchical mesh subdivision algorithm (such as the Loop subdivision algorithm). During the subdivision process, the mesh density is gradually increased based on the density or importance of the abnormal data. Determine the core encryption area: during the subdivision process, the region with the densest or most critical abnormal signals is designated as the core encryption area, generating a higher-density mesh.

[0123] Hierarchical mesh subdivision algorithms can progressively increase mesh density at different levels, thereby generating finer mesh structures in anomalous regions. The establishment of a core refinement zone ensures sufficient mesh density in critical areas to capture subtle changes, improving the model's simulation accuracy in anomalous regions.

[0124] S303, for the mesh structure with the core encryption area and the boundary polygon of the region to be encrypted, a gradient transition mesh layer is constructed along the extension of the boundary polygon of the region to be encrypted to obtain a mesh region with the core encryption area and the gradient transition mesh layer.

[0125] Optionally, define a transition region: Outside the boundary polygon of the region to be encrypted, determine the width of a transition region as needed. Gradient mesh generation: Gradually decrease the mesh density outwards from the core encrypted region to generate a gradient transition mesh layer. Linear or non-linear gradient strategies can be used to smoothly transition the mesh density from the high density of the core encrypted region to the low density of the background mesh. Integrate the mesh region: Integrate the mesh structure of the core encrypted region, the gradient transition mesh layer, and the background mesh to form a complete mesh region.

[0126] Setting a gradually transitioning mesh layer can avoid abrupt changes in mesh density between the core refinement region and the background mesh, thereby reducing errors and instabilities in numerical simulations. By rationally designing the mesh density variation pattern in the transition region, a smooth transition of mesh density can be achieved, improving the consistency and simulation accuracy of the entire mesh region.

[0127] Furthermore, such as Figure 4 As shown, based on the 3D model of the foundation pit, a foundation pit modeling and analysis model is constructed, including:

[0128] S401 discretizes the seepage field control equations of the three-dimensional model of the foundation pit into mass conservation constraint operators.

[0129] Optionally, define the governing equations for the seepage field: Based on Darcy's law and the law of conservation of mass, establish the governing equations for the seepage field in the foundation pit. The governing equations are typically partial differential equations (PDEs), describing the changes in pore water pressure over time and space. Discretization: Use numerical methods such as the finite difference method, finite element method, or finite volume method to discretize the continuous seepage field governing equations. Divide the three-dimensional model of the foundation pit into multiple discrete grid elements, and approximately solve the governing equations on each grid element. Generate mass conservation constraint operators: The discretized equations can be represented as a set of algebraic equations that reflect the mass conservation relationships within each grid element. Integrate these equations into mass conservation constraint operators for subsequent model solving.

[0130] The governing equations of the seepage field, based on the law of conservation of mass and Darcy's law, describe the flow behavior of pore water in soil. Through discretization, the continuous partial differential equations can be transformed into a discrete system of algebraic equations, facilitating numerical solutions. The mass conservation constraint operator ensures that the mass conservation relationship of pore water is satisfied in the numerical simulation.

[0131] S402 discretizes the stress field constitutive relation of the three-dimensional model of the foundation pit into a mechanical compatibility constraint operator.

[0132] Optionally, define the stress field constitutive relation: Based on the mechanical properties of the soil, select a suitable constitutive model (such as a linear elastic model, Mohr-Coulomb model, etc.) to describe the stress-strain relationship of the soil. Discretization: Use numerical methods such as the finite element method to discretize the stress field constitutive relation. Establish the discrete relationship between stress and strain in each grid element of the three-dimensional model of the foundation pit. Generate mechanical compatibility constraint operators: The discretized stress-strain relationship can be represented as a set of algebraic equations, which reflect the mechanical compatibility relationship within each grid element. Integrate these equations into mechanical compatibility constraint operators.

[0133] The constitutive relations of the stress field describe the deformation and stress distribution of soil under stress. Discretization transforms the continuous constitutive relations into a discrete system of algebraic equations, facilitating numerical solutions. Mechanical compatibility constraint operators ensure that the mechanical behavior of the soil conforms to actual physical laws in numerical simulations.

[0134] S403 generates data-driven constraint terms based on the 3D model of the foundation pit and real-time monitoring data.

[0135] Optional, data integration: Integrate the mesh information in the 3D model of the foundation pit with real-time monitoring data, mapping the monitoring data onto the corresponding mesh cells. Data-driven constraint generation: Generate data-driven constraint terms based on the difference between the monitoring data and the model predictions. These constraints can be error correction terms or penalty terms in the objective function, used to adjust model parameters to make the model predictions closer to the actual monitoring data.

[0136] The introduction of data-driven constraints aims to combine actual monitoring data with the numerical model, improving the model's prediction accuracy through a data feedback mechanism. This method can effectively compensate for potential errors in the numerical model, making the model more closely reflect reality.

[0137] S404 integrates mass conservation constraint operators, mechanical compatibility constraint operators, and data-driven constraint terms to construct a composite optimization objective function.

[0138] Optional, multi-constraint integration: Integrate mass conservation constraint operators, mechanical compatibility constraint operators, and data-driven constraint terms into a unified framework to form multi-constraint terms. Construct a composite optimization objective function: Define a composite optimization objective function, incorporating the multi-constraint terms as components of the objective function. The objective function can be in the form of a sum of squared errors, a weighted sum of penalty terms, etc., used to measure the difference between model predictions and actual data.

[0139] The construction of a composite optimization objective function aims to simultaneously satisfy multiple physical constraints and data-driven constraints in numerical simulations. By optimizing the objective function, optimal model parameters can be found under multiple constraints, thereby improving the overall performance of the model.

[0140] S405, based on a 3D model of the foundation pit and a composite optimization objective function, selects a network architecture and trains the network to construct a foundation pit modeling and analysis model. The foundation pit modeling and analysis model is a deep neural network surrogate model embedding physical mechanisms.

[0141] Optional, select network architecture: Based on the needs of pit modeling and analysis, select a suitable deep neural network architecture, such as convolutional neural network (CNN), recurrent neural network (RNN), or hybrid neural network.

[0142] Network Training: Based on the grid data of the 3D model of the foundation pit and the composite optimization objective function, a deep neural network is trained. The network parameters are adjusted using backpropagation and optimization algorithms (such as gradient descent) to optimize the network's prediction results. Construction of the Foundation Pit Modeling and Analysis Model: The trained deep neural network is embedded into the foundation pit modeling and analysis model, forming a deep neural network surrogate model that incorporates the physical mechanisms.

[0143] Deep neural networks possess powerful nonlinear fitting and data learning capabilities, making them suitable for numerical simulations of complex physical problems. By combining physical mechanism constraints with data-driven constraints, deep neural networks can be trained to construct foundation pit modeling and analysis models that both conform to physical laws and closely resemble actual data.

[0144] Furthermore, such as Figure 5 As shown, based on the 3D model of the foundation pit and the composite optimization objective function, a network architecture is selected and the network is trained to construct a foundation pit modeling and analysis model, including:

[0145] S501, based on the 3D model of the pit densification and the composite optimization objective function, extracts the mesh topology features by analyzing the mesh topology relationship through a shared feature extraction layer.

[0146] Optionally, a shared feature extraction layer can be constructed: Design a shared feature extraction layer for a neural network that can simultaneously process the mesh data of the pit-refined 3D model and related information of the composite optimization objective function. Analyze mesh topological relationships: Utilize graph neural networks (GNNs) or mesh processing algorithms to analyze the topological relationships between meshes, such as connectivity and adjacency. Extract mesh topological features: Extract the topological features of the mesh by calculating its geometric and topological properties (such as curvature, normals, and connectivity). Morse theory can be used to identify key points (such as maxima, minima, and saddle points) and extract the mesh's skeleton structure.

[0147] The purpose of mesh topology feature extraction is to capture the shape and structural information of the mesh model, which is crucial for subsequent predictions of seepage and stress fields. By analyzing the topological relationships of the mesh, a better understanding of the model's geometry can be achieved, thus providing more accurate input for the prediction of physical fields.

[0148] S502, based on the grid topology characteristics, uses parallel prediction branches to generate the seepage field and stress field distribution, and obtains the predicted distribution of the seepage field and the predicted distribution of the stress field.

[0149] Optionally, design parallel prediction branches: Design two parallel prediction branches in the neural network architecture, one for seepage field prediction and the other for stress field prediction. Input mesh topology features: Input the extracted mesh topology features into the prediction branches for the seepage field and stress field, respectively. Generate prediction distributions: Generate the predicted distributions for the seepage field and stress field using the trained neural network model.

[0150] The parallel prediction branch design allows for simultaneous prediction of seepage and stress fields, improving computational efficiency. By using mesh topology features as input, the model can better capture the distribution patterns of the physical fields, thereby improving prediction accuracy.

[0151] S503, based on the predicted distribution of the seepage field and the predicted distribution of the stress field, performs joint verification of physical conservation across branches, and obtains the physical conservation verification results.

[0152] Optionally, define physical conservation verification rules: Based on physical laws (such as mass conservation, energy conservation, momentum conservation, etc.), define verification rules to check whether the predicted distributions of the seepage field and stress field satisfy physical conservation. Cross-branch joint verification: Jointly verify the predicted distributions of the seepage field and stress field to check for contradictions or non-compliance with physical conservation. Generate verification results: Based on the verification rules, generate physical conservation verification results, identifying the parts of the predicted distribution that do not conform to physical conservation.

[0153] The purpose of joint verification of physical conservation laws is to ensure that the model's predictions conform to the laws of physics. Through cross-branch joint verification, unreasonable aspects in the predictions can be identified and corrected in a timely manner, improving the reliability and accuracy of the model.

[0154] S504, based on the physical conservation verification results, triggers model parameter self-calibration to obtain the calibrated model parameters.

[0155] Optionally, a self-calibration mechanism can be designed: This mechanism automatically adjusts model parameters based on the physical conservation verification results. Self-calibration is triggered when the physical conservation verification results indicate that the predicted distribution does not conform to physical conservation. Model parameters are then adjusted using optimization algorithms (such as gradient descent) to make the predicted distribution more consistent with physical conservation.

[0156] The model parameter self-calibration mechanism can automatically adjust the model parameters based on the validation results, thereby improving the model's prediction accuracy. By continuously calibrating the model parameters, the model's prediction results can be made closer to actual physical phenomena.

[0157] S505, based on the corrected model parameters, constructs a foundation pit modeling and analysis model.

[0158] Optional, update model parameters: Update the corrected model parameters into the neural network model. Build the final model: Use the updated model parameters to build the final pit modeling and analysis model. Model validation and optimization: Validate and optimize the built model to ensure its accuracy and reliability in different scenarios.

[0159] By constructing a foundation pit modeling and analysis model using corrected model parameters, it can be ensured that the model's predictions both conform to physical laws and accurately reflect actual monitoring data. This method improves the overall performance of the model and provides a reliable tool for safety monitoring and analysis of foundation pit construction.

[0160] In one exemplary embodiment, such as Figure 6 As shown, based on the foundation pit modeling and analysis model, the foundation pit densified 3D model, and the foundation pit initial 3D model, the foundation pit modeling and analysis model is updated, including:

[0161] S601, based on the foundation pit modeling and analysis model, the three-dimensional model of the foundation pit is analyzed to obtain the first prediction result.

[0162] Optional, Model Preparation: Import the refined 3D model of the foundation pit into the foundation pit modeling and analysis model. The refined 3D model contains more detailed meshing, especially in abnormal areas. Parameter Settings: Adjust the physical parameters and boundary conditions in the model according to the mesh characteristics of the refined 3D model. Simulation Analysis: Run the foundation pit modeling and analysis model to perform coupled analysis of the seepage field and stress field on the refined 3D model. Result Extraction: Extract key parameters from the simulation results, such as pore water pressure, effective stress, and displacement distribution, to form the first prediction results.

[0163] Encrypted 3D models provide more refined geometric and topological information, enabling a more accurate reflection of the physical state of the foundation pit. Analysis of these encrypted models allows for more precise prediction of the mechanical behavior during foundation pit construction.

[0164] S602, based on the foundation pit modeling and analysis model, the initial three-dimensional model of the foundation pit is analyzed to obtain the second prediction result.

[0165] Optional, Model Import: Import the initial 3D model of the foundation pit into the foundation pit modeling and analysis model. The initial 3D model reflects the initial geometric state of the foundation pit. Parameter Adjustment: Set the physical parameters and boundary conditions in the model according to the mesh characteristics of the initial 3D model. Simulation Analysis: Run the foundation pit modeling and analysis model to perform coupled analysis of the seepage field and stress field on the initial 3D model. Result Extraction: Extract key parameters from the simulation results, such as pore water pressure, effective stress, and displacement distribution, to form a second prediction result.

[0166] The initial 3D model provides information on the initial state of the foundation pit before construction, and analysis of this model can predict the mechanical behavior of the pit before construction. This helps in comparing changes during the construction process.

[0167] S603, based on the first and second prediction results, update the foundation pit modeling and analysis model.

[0168] Optional, result comparison: Compare the first prediction result (prediction result of the encrypted 3D model) with the second prediction result (prediction result of the initial 3D model). Error analysis: Calculate the difference between the two prediction results and analyze the sources of error, such as the impact of mesh refinement on the prediction results. Model parameter adjustment: Based on the error analysis results, adjust the parameters in the foundation pit modeling and analysis model, such as physical parameters and boundary conditions. Model update: Update the adjusted parameters into the foundation pit modeling and analysis model to complete the model update.

[0169] By comparing the prediction results of the encrypted 3D model and the initial 3D model, we can identify the differences in predictions under different grid densities. Updating the model using these differences can improve its prediction accuracy and reliability.

[0170] Furthermore, such as Figure 7 As shown, based on the first and second prediction results, the foundation pit modeling and analysis model is updated, including:

[0171] S701, For the first prediction result and the second prediction result, extract the predicted values ​​of physical quantities at the same spatial location and generate a difference quantization matrix to obtain a spatial correlation prediction difference map with labeled difference magnitude and location coordinates.

[0172] Optional steps include: Data extraction: Extracting predicted physical quantities such as pore water pressure, effective stress, and displacement at the same spatial location from the first prediction result (based on the encrypted 3D model) and the second prediction result (based on the initial 3D model). Difference calculation: Calculating the differences between the predicted physical quantity values ​​at the same location to generate a difference quantization matrix. Map generation: Combining the difference quantization matrix with spatial location coordinates to generate a spatial correlation prediction difference map annotated with the magnitude of the difference and the location coordinates.

[0173] By comparing the prediction results of the two models, the differences between the encrypted model and the initial model can be visually reflected, thereby determining the distribution of the model's prediction bias. The difference map helps identify which regions exhibit significant prediction bias, providing a basis for subsequent model updates.

[0174] S702. Based on the complete set of parameters of the predicted difference map and the foundation pit modeling analysis model, analyze the degree of influence of each parameter on the difference map and quantify the sensitivity level, generate a sensitivity ranking list, and obtain a parameter sensitivity classification table for sensitivity classification.

[0175] Optional, parameter analysis: Combining the prediction difference map and the complete parameter set of the foundation pit modeling analysis model, analyze the degree of influence of each parameter on the prediction difference. Sensitivity quantification: Quantify the sensitivity level of each parameter through correlation analysis or sensitivity analysis methods. Ranking and grading: Sort the parameters according to their sensitivity levels, generate a sensitivity ranking list, and divide the parameters into different sensitivity levels to form a parameter sensitivity grading table.

[0176] Sensitivity analysis can identify key parameters that significantly impact model predictions, and these parameters require close monitoring during model updates. By quantifying sensitivity levels, resources for model optimization can be allocated more rationally, improving the efficiency of model updates.

[0177] S703, based on the parameter sensitivity classification table, sets sensitivity thresholds to filter key parameters, divides the optimizable parameter subset and the frozen parameter set, generates a parameter operation instruction set, and obtains a dynamic optimization whitelist containing adjustable parameter ranges and a parameter freezing blacklist that locks non-sensitive parameters.

[0178] Optionally, set a threshold: Based on the parameter sensitivity grading table, set a sensitivity threshold to filter out key parameters. Parameter segmentation: Divide key parameters into a subset of optimizable parameters and other non-sensitive parameters into a frozen parameter set. Instruction set generation: Generate parameter operation instruction sets for the subset of optimizable parameters, clarifying the range of adjustable parameters; generate locking instructions for the frozen parameter set to keep these parameters constant. List generation: Create a dynamic optimization whitelist (containing the range of adjustable parameters) and a parameter freezing blacklist (locking non-sensitive parameters).

[0179] By setting sensitivity thresholds to filter key parameters, model updates can focus on parameters that have a significant impact on prediction results. Freezing insensitive parameters can reduce the computational load during model updates and improve their efficiency.

[0180] S704 updates the foundation pit modeling and analysis model based on the dynamically optimized whitelist and the parameter-frozen blacklist.

[0181] Optionally, the foundation pit modeling and analysis model is updated based on the dynamically optimized whitelist and the parameter freezing blacklist, including: calculating the update step size and dynamically adjusting the learning rate based on the dynamically optimized whitelist and the current model parameter status, combined with the historical gradient direction, performing directional fine-tuning of whitelist parameters, keeping blacklist parameters constant, and updating the foundation pit modeling and analysis model.

[0182] Specifically, parameter adjustment: Based on the adjustable parameter range in the dynamic optimization whitelist, combined with the current model parameter state and historical gradient direction, calculate the update step size and dynamically adjust the learning rate. Whitelist parameter fine-tuning: Perform targeted fine-tuning on the parameters in the dynamic optimization whitelist to optimize model performance. Blacklist parameter constancy: Keep the parameters in the frozen blacklist unchanged to ensure model stability. Model update: Apply the adjusted parameters to the foundation pit modeling and analysis model to complete the model update.

[0183] The combined use of dynamic optimization whitelists and parameter freezing blacklists can maintain the overall stability of the model while optimizing key parameters. By dynamically adjusting the learning rate and update step size, model parameters can be optimized more effectively, improving the model's prediction accuracy.

[0184] Based on the same inventive concept, this application also provides a modeling device for rail transit foundation pit support based on multi-source data fusion, used to implement the above-mentioned modeling method for rail transit foundation pit support based on multi-source data fusion. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the modeling device for rail transit foundation pit support based on multi-source data fusion provided below can be found in the limitations of the rail transit foundation pit support modeling method based on multi-source data fusion described above, and will not be repeated here.

[0185] In one exemplary embodiment, a modeling device for rail transit foundation pit support based on multi-source data fusion is provided, such as... Figure 8 As shown, the device includes:

[0186] Module 11 is used to acquire multi-source heterogeneous data of the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data and geological parameter database;

[0187] The fine processing module 12 is used to perform grid density processing on the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit when abnormal data is detected in the real-time monitoring data, so as to obtain the encrypted three-dimensional model of the foundation pit.

[0188] The analysis and modeling module 13 is used to construct a foundation pit modeling and analysis model based on the foundation pit densified 3D model. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during foundation pit construction in real time based on the foundation pit 3D model. The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the foundation pit 3D model.

[0189] The modeling update module 14 is used to update the foundation pit modeling analysis model based on the foundation pit modeling analysis model, the foundation pit encrypted 3D model, and the foundation pit initial 3D model.

[0190] The modules in the above-mentioned rail transit foundation pit support modeling device based on multi-source data fusion can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0191] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0192] Acquire multi-source heterogeneous data on the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data, and a geological parameter database.

[0193] When abnormal data is detected in the real-time monitoring data, the number of grids in the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit is increased to obtain the encrypted three-dimensional model of the foundation pit.

[0194] Based on the 3D model of the foundation pit, a foundation pit modeling and analysis model is constructed. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during foundation pit construction in real time based on the 3D model of the foundation pit. The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the 3D model of the foundation pit.

[0195] The foundation pit modeling and analysis model is updated based on the foundation pit modeling and analysis model, the foundation pit densified 3D model, and the foundation pit initial 3D model.

[0196] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0197] Acquire multi-source heterogeneous data on the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data, and a geological parameter database.

[0198] When abnormal data is detected in the real-time monitoring data, the number of grids in the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit is increased to obtain the encrypted three-dimensional model of the foundation pit.

[0199] Based on the 3D model of the foundation pit, a foundation pit modeling and analysis model is constructed. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during foundation pit construction in real time based on the 3D model of the foundation pit. The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the 3D model of the foundation pit.

[0200] The foundation pit modeling and analysis model is updated based on the foundation pit modeling and analysis model, the foundation pit densified 3D model, and the foundation pit initial 3D model.

[0201] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0202] Acquire multi-source heterogeneous data on the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data, and a geological parameter database.

[0203] When abnormal data is detected in the real-time monitoring data, the number of grids in the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit is increased to obtain the encrypted three-dimensional model of the foundation pit.

[0204] Based on the 3D model of the foundation pit, a foundation pit modeling and analysis model is constructed. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during foundation pit construction in real time based on the 3D model of the foundation pit. The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the 3D model of the foundation pit.

[0205] The foundation pit modeling and analysis model is updated based on the foundation pit modeling and analysis model, the foundation pit densified 3D model, and the foundation pit initial 3D model.

[0206] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0207] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0208] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A modeling method for rail transit foundation pit support based on multi-source data fusion, characterized in that, The method includes: Acquire multi-source heterogeneous data on the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data, and a geological parameter database. If abnormal data is detected in the real-time monitoring data, the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit is subjected to grid number densification processing to obtain a densified three-dimensional model of the foundation pit. Based on the 3D model of the foundation pit, a foundation pit modeling and analysis model is constructed. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during the foundation pit construction process in real time based on the 3D model of the foundation pit. The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the 3D model of the foundation pit. The foundation pit modeling and analysis model is updated based on the foundation pit modeling and analysis model, the foundation pit encrypted 3D model, and the foundation pit initial 3D model.

2. The method according to claim 1, characterized in that, The multi-source heterogeneous data includes: The real-time monitoring data is collected through a distributed sensor network and includes three types of time-series signals: pore water pressure data, soil displacement data, and support stress data. The static topology data is obtained from the BIM model and includes the spatial coordinates of the support structure and the geometric data of the soil layer interface. The geological parameter database includes the permeability coefficient, elastic modulus, and Poisson's ratio of each soil layer.

3. The method according to claim 2, characterized in that, Determining the abnormal mesh region corresponding to the abnormal data in the initial 3D model of the foundation pit includes: Spatial gradient analysis is performed on the pore water pressure data to generate a water pressure anomaly distribution map; Perform time-domain trend analysis on the soil displacement data and output the region where the displacement rate exceeds the limit; Statistical outlier detection was performed on the support stress data to mark stress anomaly points; Based on the water pressure anomaly distribution map, the displacement rate exceeding the limit area, and the stress anomaly point, a multi-dimensional anomaly signal matrix with spatial coordinate markers is generated. The multi-dimensional anomaly signal matrix includes the anomaly grid region corresponding to the anomaly data in the initial three-dimensional model of the foundation pit.

4. The method according to claim 3, characterized in that, The step of densifying the mesh count of the abnormal mesh region corresponding to the abnormal data in the initial 3D model of the foundation pit includes: Based on the multi-dimensional anomaly signal matrix, generate the boundary polygon of the region to be encrypted; Inside the boundary polygon of the region to be encrypted, hierarchical mesh subdivision is performed to generate a mesh structure with a core encryption area; For the mesh structure with a core encryption area and the boundary polygon of the region to be encrypted, a gradient transition mesh layer is constructed along the extension of the boundary polygon of the region to be encrypted to obtain a mesh region with a core encryption area and a gradient transition mesh layer.

5. The method according to claim 2, characterized in that, The step of constructing a foundation pit modeling and analysis model based on the 3D model of the foundation pit includes: The seepage field control equations of the three-dimensional model of the foundation pit are discretized into mass conservation constraint operators; The stress field constitutive relation of the three-dimensional model of the foundation pit is discretized into a mechanical compatibility constraint operator; Based on the 3D model of the foundation pit and the real-time monitoring data, data-driven constraint terms are generated. The mass conservation constraint operator, mechanical compatibility constraint operator, and data-driven constraint term are integrated into a composite optimization objective function by combining multiple constraint terms. Based on the 3D model of the foundation pit and the composite optimization objective function, a network architecture is selected and the network is trained to construct a foundation pit modeling and analysis model; the foundation pit modeling and analysis model is a deep neural network proxy model with embedded physical mechanisms.

6. The method according to claim 5, characterized in that, The process of constructing a foundation pit modeling and analysis model based on the 3D model of the foundation pit and the composite optimization objective function, including selecting a network architecture, training the network, and building the foundation pit modeling and analysis model, includes: Based on the 3D model of the foundation pit and the composite optimization objective function, the mesh topology relationship is analyzed through a shared feature extraction layer to extract mesh topology features; Based on the grid topology features, the seepage field and stress field distributions are generated using parallel prediction branches, resulting in the predicted distributions of the seepage field and stress field. Based on the predicted distribution of the seepage field and the predicted distribution of the stress field, a joint verification of physical conservation across branches is performed to obtain the physical conservation verification results. Based on the physical conservation verification results, the model parameters are self-calibrated to obtain the calibrated model parameters. Based on the corrected model parameters, a foundation pit modeling and analysis model is constructed.

7. The method according to claim 2, characterized in that, The step of updating the foundation pit modeling and analysis model based on the foundation pit modeling and analysis model, the foundation pit refined 3D model, and the foundation pit initial 3D model includes: Based on the foundation pit modeling and analysis model, the foundation pit densified 3D model is analyzed to obtain the first prediction result; Based on the foundation pit modeling and analysis model, the initial three-dimensional model of the foundation pit is analyzed to obtain a second prediction result; The foundation pit modeling and analysis model is updated based on the first prediction result and the second prediction result.

8. The method according to claim 7, characterized in that, The step of updating the foundation pit modeling and analysis model based on the first prediction result and the second prediction result includes: For the first prediction result and the second prediction result, the predicted values ​​of physical quantities at the same spatial location are extracted and a difference quantization matrix is ​​generated to obtain a spatial correlation prediction difference map with labeled difference magnitude and location coordinates. Based on the predicted difference map and the complete set of parameters of the foundation pit modeling and analysis model, the influence of each parameter on the difference map is analyzed and the sensitivity level is quantified. A sensitivity ranking list is generated to obtain a parameter sensitivity classification table. Based on the parameter sensitivity classification table, a sensitivity threshold is set to filter key parameters, divide the optimizable parameter subset and the frozen parameter set, generate a parameter operation instruction set, and obtain a dynamic optimization whitelist containing adjustable parameter ranges and a parameter freezing blacklist that locks non-sensitive parameters. The foundation pit modeling and analysis model is updated based on the dynamically optimized whitelist and the parameter-frozen blacklist.

9. The method according to claim 8, characterized in that, The step of updating the foundation pit modeling and analysis model based on the dynamically optimized whitelist and the parameter-frozen blacklist includes: Based on the dynamically optimized whitelist and the current model parameter status, the update step size is calculated and the learning rate is dynamically adjusted in combination with the historical gradient direction. The whitelist parameters are fine-tuned in a targeted manner, while the blacklist parameters are kept constant, and the foundation pit modeling and analysis model is updated.

10. A modeling device for rail transit foundation pit support based on multi-source data fusion, characterized in that, The device includes: The acquisition module is used to acquire multi-source heterogeneous data of the foundation pit; the multi-source heterogeneous data includes real-time monitoring data, static topology data, and a geological parameter database. The fine processing module is used to densify the number of grids in the abnormal grid area corresponding to the abnormal data in the initial three-dimensional model of the foundation pit when abnormal data is detected in the real-time monitoring data, so as to obtain a densified three-dimensional model of the foundation pit. The analysis and modeling module is used to construct a foundation pit modeling and analysis model based on the encrypted 3D model of the foundation pit. The foundation pit modeling and analysis model is used to simulate the coupling effect of seepage field and stress field during foundation pit construction in real time based on the 3D model of the foundation pit. The model prediction results of the foundation pit modeling and analysis model include the pore water pressure and effective stress of the 3D model of the foundation pit. The modeling update module is used to update the foundation pit modeling analysis model based on the foundation pit modeling analysis model, the foundation pit encrypted 3D model, and the foundation pit initial 3D model.

Images