A data-driven high core wall rockfill dam core wall force-seepage coupling representation method and system

Abstract

The application discloses a kind of high core wall rockfill dam core wall force seepage coupling characterization method and system based on data driving, belong to geotechnical engineering and hydraulic structure safety monitoring technical field.The method includes the following steps: step 1, obtain core wall field monitoring data, build porous elastic state variable tuple;Data classification mapping is carried out based on the integrity evaluation criterion and the noise level determination criterion;Step 2, establish high core wall rockfill dam finite element geometric model;Step 3, data interpolation and / or noise dilution processing;Step 4, all types of porous elastic state variable data are merged into single main data set, and porous elastic state variable data random rearrangement is carried out, output global force seepage coupling state or enter next round iteration.A kind of fluid-solid coupling data driven paradigm is provided, fusion "strain, effective stress, pressure gradient, seepage velocity" constructs closed loop, improves characterization accuracy.

Description

Technical Field

[0001] This invention relates to a data-driven method and system for characterizing the force-permeability coupling of the core wall of a high-core rockfill dam, and more particularly to a method and system for characterizing the force-permeability coupling (fluid-structure coupling) of the core wall of a high-core rockfill dam based on data-driven technology and Biot consolidation theory, belonging to the field of geotechnical engineering and hydraulic structure safety monitoring technology. Background Technology

[0002] During impoundment and operation, high-core rockfill dams exhibit extremely complex coupling effects between the mechanical deformation of the seepage-proof core and pore water seepage. Accurately understanding the force-seepage coupling state within the core is a key issue for dam safety analysis and disaster prevention. Due to the highly nonlinear nature of dam materials, traditional empirical constitutive models often fail to fully capture the fluid-structure interaction under complex real-world conditions.

[0003] Currently, the monitoring data returned by field monitoring instruments (such as earth pressure gauges, piezometers, and water-tube settlement meters) exhibits a certain degree of dispersion, noise, and even data loss due to factors such as burial depth, construction interference, and sensor damage, increasing the difficulty of analysis. Existing data-driven mechanical characterization techniques are mainly limited to single-phase solid mechanics problems. They rely on pre-specified stress-strain state pairs and characterize the mechanical state by minimizing a global penalty function, as disclosed in application number CN202410789326.0. However, this traditional framework cannot incorporate pore water pressure and seepage fields, and cannot handle fully coupled porous elastic multiphysics problems, thus limiting its application in deep geotechnical engineering and earth-rock dam safety evaluation.

[0004] Chinese patent document CN202510478639.9 discloses a method, apparatus, and equipment for dynamic prediction of seepage consolidation in a core-wall rockfill dam using multi-field coupling. The method includes: acquiring spatiotemporal stress data and material data during the seepage consolidation process of the core-wall rockfill dam; constructing a physical model of the internal seepage consolidation process of the core-wall rockfill dam, a random field of the core wall permeability coefficient and compressibility modulus, and a prediction model of the joint random field; constructing a loss function based on the physical model and labeled data; iteratively training the prediction model based on the spatiotemporal stress data and material data, updating the prediction model in reverse iteratively according to the loss function during training, and obtaining a trained prediction model after training; and using the trained prediction model to predict the seepage consolidation process of the core-wall rockfill dam. This solves the problems in related technologies where the finite element method and deep learning methods have certain limitations, making it difficult to fit real monitoring data and leading to inaccurate prediction results.

[0005] However, when the above prediction method is actually applied to complex geotechnical engineering projects such as rockfill dam core walls, it suffers from drawbacks such as high noise in the monitoring data and widespread data gaps. Therefore, the accuracy of the prediction results obtained needs to be further improved. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a data-driven method and system for characterizing the core wall force-permeability coupling of high-core rockfill dams.

[0007] This invention is achieved through the following technical solution: A data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam includes the following steps: Step 1: Obtain on-site monitoring data of the core wall, construct a porous elastic state variable tuple based on the on-site monitoring data of the core wall; establish integrity assessment criteria and noise level judgment criteria, and perform data classification mapping on the porous elastic state variable tuple based on the integrity assessment criteria and the noise level judgment criteria; Step 2: Establish a finite element geometric model of a high-core rockfill dam, and define nodes, topological relationships, shape functions and their derivative matrices and integral point weights in the finite element geometric model of the high-core rockfill dam; Step 3: Perform data interpolation and / or noise dilution on incomplete and high-noise porous elastic state variable data of various types; Step 4: Merge all types of porous elastic state variable data into a single master dataset, randomly rearrange the porous elastic state variable data, establish a global penalty function to evaluate the deviation between the admissible field state variable and the nearest porous elastic state variable tuple, calculate the nodal displacement sequence and pore pressure sequence driven by the current master dataset, calculate the Lagrange multipliers representing the momentum and mass imbalance force, then map the global solution back to the local point to obtain the new physical state, calculate the distance error relative to the data point, and output the global force-permeation coupling state or enter the next iteration based on the distance error.

[0008] The on-site monitoring data for the core wall includes: total stress. Displacement pore pressure and potential direct traffic; Based on the on-site monitoring data of the core wall, a tuple of porous elastic state variables is constructed, including: Extract four key tensors / vectors at each material point: strain: , where ∇ is the gradient operator, To identify symmetrical parts; Effective stress: ,in These are the Biot-Willis coefficients. Unit tensor; Pressure gradient: ; Darcy flow: ,in For penetration rate, For fluid viscosity, For fluid density, The volumetric force per unit mass of fluid; The four key tensors / vectors extracted from all material points will be combined to form a tuple of porous elastic state variables: .

[0009] The process of establishing integrity assessment criteria in step 1 is as follows: , in, Indicates the first The integrity assessment value of the porous elastic state variable data at each measuring point; , , , They represent the first The data acquisition status of strain, effective stress, pressure gradient and Darcy flow rate at each measuring point is set to 1 when successfully acquired and 0 when missing. when When, it indicates the first Complete data on the porous elastic state variables at each measuring point; when When, it indicates the first The data on the porous elastic state variables at some measuring points are incomplete. The process of establishing the noise level determination criteria in step 1 is as follows: The noise level of porous elastic state variable data at the measuring points is automatically defined using Mahalanobis distance or robust absolute median difference. When using Mahalanobis distance to automatically define the noise level of porous elastic state variable data at measuring points, the following is included: Using the porous elastic state variable data from all complete measurement points, calculate the sample mean vector of the porous elastic state variable tuples. Covariance Matrix ; For the At each measuring point, the porous elastic state variable is denoted as... And calculate its vector to the sample mean. Mahalanobis distance : ; Based on chi-square distribution or kernel density estimation, automatically calculate the critical distance threshold corresponding to the specified noise level confidence threshold. ; like Determine the measuring point Data within the high confidence interval are labeled as low noise for the corresponding porous elastic state variable data. like Determine the measuring point Deviation from sample mean vector The corresponding porous elastic state variable data are marked as high noise.

[0010] The process of classifying and mapping the porous elastic state variable tuples based on the integrity assessment criterion and the noise level determination criterion is as follows: when and When the porous elastic state variable data of the measuring point is classified into Class A, it indicates that the porous elastic state variable data of the measuring point is complete and the noise level is low, serving as a high-quality anchor point for data-driven analysis. when and When the data of the porous elastic state variable at the measuring point is classified into Class B, it indicates that the data of the porous elastic state variable at the measuring point is incomplete and the noise level is low. when and When the porous elastic state variable data of the measuring point is classified into category C, it indicates that the porous elastic state variable data of the measuring point is complete and the noise level is high. when and When the data is incomplete, the porous elastic state variable data of the measuring point is classified as Class D, indicating that the data of the porous elastic state variable of the measuring point is incomplete and the noise level is high.

[0011] When performing step 3, for type B porous elastic state variable data, when effective stress is missing... At that time, according to Perform data interpolation, where This is the local solid stiffness matrix. This refers to the measured strain tensor obtained by the sensor on-site. When it lacks strain At that time, according to Perform data interpolation, where Local solid stiffness matrix The inverse matrix, The measured effective stress tensor is derived from the measured total stress and measured pore pressure of the sensor. ; When it lacks Darcy flow At that time, according to Perform data interpolation, where This is the local flow phase mobility matrix. This is the measured pressure gradient vector calculated from the measured pore water pressure. The fluid density of pore water, It is the gravity vector; When it lacks a pressure gradient At that time, according to Perform data interpolation, where Local flow phase mobility matrix The inverse matrix, This is the Darcy flow vector actually measured on-site.

[0012] When performing step 3, for the C-type porous elastic state variable data, random permeation and stress boundary conditions are applied to the Biot porous elastic FEM forward solver using the estimated global parameters to generate theoretical state tuples. ,in, To synthesize strain, To synthesize effective stress, To synthesize the pressure gradient, To synthesize the flow rate, the number of synthesized data in the theoretical state tuple is 1 to 2 times the number of C-type porous elastic state variable data. Then, all the synthesized data in the theoretical state tuple are appended to the C-type porous elastic state variable dataset and randomly shuffled to dilute the noise in the C-type porous elastic state variable data.

[0013] When performing step 3, for D-type porous elastic state variable data, data interpolation is first performed according to the processing method for B-type porous elastic state variable data, and then noise dilution is performed according to the processing method for C-type porous elastic state variable data. The number of synthetic data in the theoretical state tuple generated when processing D-type porous elastic state variable data is 0.8 to 0.9 times the number of synthetic data in the theoretical state tuple generated when processing C-type porous elastic state variable data.

[0014] The permissible field state variable is denoted as Then it is related to the nearest porous elastic state variable tuple. The deviation between them is calculated using the following global penalty function: , in, For global penalty function, This represents the total number of material points. For the first The integral weight of each material point; , , , They represent the first The theoretical allowable strain, effective stress, pressure gradient, and Darcy flow rate of a material point satisfying the physical conservation equations; , , , They represent the first The strain, effective stress, pressure gradient, and Darcy flow rate corresponding to each material point in the dataset; Local solid stiffness matrix Local solid stiffness matrix inverse matrix Local flow phase mobility matrix Local flow phase mobility matrix The inverse matrix; superscript This indicates the matrix transpose.

[0015] The node displacement sequence and pore pressure sequence driven by the current master dataset are calculated according to the following formula: , in, Here is the stiffness matrix of the solid. The solid-fluid coupling matrix, For fluid permeability matrix, For fluid compressibility matrix, To constrain water storage rate, The time step is the implicit time discrete step. It is a sequence of node displacements. It is a pore pressure sequence; For external node load vectors, For external fluid source-sink vectors; The matrix of derivatives of displacement shape functions. The derivative matrix of the pressure shape function; The Lagrange multipliers are denoted as ( ),in This is a virtual displacement. For virtual pressure, Lagrange multipliers ( Calculate according to the following formula: , in, It is a virtual shift sequence. It is a virtual pressure sequence; The new physical state is represented as ( ),in, For strain sequence, For effective stress sequence, It is a pressure gradient sequence. It is a flow sequence.

[0016] The distance error between the local point and the data point is calculated according to the following formula: , in, For distance error, For the first In the next iteration, the global driving field is mapped back to the first... A tuple of the physical allowable states of a local material point, namely ; For the first In the nth iteration, it is the... The nearest dataset state tuple matched by each material point in the global dataset, i.e. ; If distance error If the tolerance limit is met, the global force-interference coupling state is output; if the distance error is not met, the global force-interference coupling state is output. If the tolerance limit is not met, a nearest neighbor global search is performed in the main dataset to find a new data point for each material point and proceed to the next iteration.

[0017] A high-core rockfill dam core wall force-seepage coupling system includes: The data acquisition and classification module is used to acquire on-site monitoring data of the core wall, construct a porous elastic state variable tuple based on the on-site monitoring data of the core wall, establish integrity assessment criteria and noise level judgment criteria, and perform data classification mapping on the porous elastic state variable tuple based on the integrity assessment criteria and the noise level judgment criteria. The finite element geometric model establishment module is used to establish a finite element geometric model of a high-center-wall rockfill dam, and to define nodes, topological relationships, shape functions and their derivative matrices and integral point weights in the finite element geometric model of the high-center-wall rockfill dam. The data interpolation and noise reduction module is used to perform data interpolation and / or noise dilution processing on incomplete and high-noise porous elastic state variable data of various types. The data fusion-driven module is used to merge all types of porous elastic state variable data into a single master dataset, randomly rearrange the porous elastic state variable data, establish a global penalty function to evaluate the deviation between the admissible field state variable and the nearest porous elastic state variable tuple, calculate the nodal displacement sequence and pore pressure sequence driven by the current master dataset, calculate the Lagrange multipliers representing the momentum and mass imbalance force, then map the global solution back to local points to obtain the new physical state, calculate the distance error relative to the data points, and output the global force-permeation coupling state or enter the next iteration based on the distance error.

[0018] The beneficial effects of this invention are as follows: it breaks through the limitations of traditional single-phase consolidation mechanisms and provides a data-driven paradigm for fluid-structure interaction that conforms to the fundamental principles of computational mechanics. Addressing the common problems of high noise and missing data in monitoring data in complex geotechnical engineering projects such as rockfill dam core walls, it can directly integrate "strain, effective stress, pressure gradient, and seepage velocity" to construct a closed loop, greatly improving the accuracy of characterizing the force-seepage coupling law under actual service conditions of high-core rockfill dams. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0020] The technical solution of the present invention is further described below, but the scope of protection is not limited to what is described.

[0021] like Figure 1 As shown, the data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam according to the present invention includes the following steps: Step 1: Obtain on-site monitoring data of the core wall, and construct a porous elastic state variable tuple based on the on-site monitoring data of the core wall; establish integrity assessment criteria and noise level judgment criteria, and perform data classification mapping on the porous elastic state variable tuple based on the integrity assessment criteria and the noise level judgment criteria.

[0022] The on-site monitoring data for the core wall includes: total stress. Displacement pore pressure and potential direct traffic; Based on the on-site monitoring data of the core wall, a tuple of porous elastic state variables is constructed, including: Extract four key tensors / vectors at each material point: strain: , where ∇ is the gradient operator, This is an identifier for the symmetrical part, indicating that the displacement gradient has been symmetrically processed. That is... ; Effective stress: ,in These are the Biot-Willis coefficients. Unit tensor; Pressure gradient: ; Darcy flow: ,in For penetration rate, For fluid viscosity, For fluid density, The volumetric force per unit mass of fluid; The four key tensors / vectors extracted from all material points will be combined to form a tuple of porous elastic state variables: .

[0023] Specifically, parameters marked with an asterisk (*) represent parameters of the dataset's state points, such as... These represent the strain, effective stress, pressure gradient, and Darcy flow rate at the state points in the dataset, respectively. They originate from known data measured in the field by the sensor, or from discrete material data points stored in a database after interpolation / noise reduction processing.

[0024] Parameters without an asterisk (*) represent parameters of the physically admissible state field, such as... These represent the strain, effective stress, pressure gradient, and Darcy flow rate of the physical allowable field, respectively. They are the theoretical mechanical field solutions calculated by the finite element global solver under strict constraints of the continuity equation, momentum balance, mass conservation, and boundary conditions.

[0025] The relationship between parameters with and without * is as follows: The ultimate goal of the data-driven algorithm is to find a set of physically permissible state fields without * through iterative calculation, so that the "distance" (global penalty function error) between the state points of the dataset with * in phase space is minimized.

[0026] Specifically, existing technologies only specify a stress-strain state pair at each material point, which cannot cover the solid-fluid phase that is crucial in geotechnical engineering. Based on this, the present invention needs to expand the state space into a four-variable porous elastic state variable tuple that includes both solid and fluid phases, so as to upgrade single-phase solid mechanics to a fully coupled porous elastic physical field.

[0027] The process of establishing integrity assessment criteria in step 1 is as follows: , in, Indicates the first The integrity assessment value of the porous elastic state variable data at each measuring point; They represent the first The data acquisition status of strain, effective stress, pressure gradient and Darcy flow rate at each measuring point is recorded. The value is 1 when the data is successfully acquired and 0 when the data is missing.

[0028] when When, it indicates the first Complete data on the porous elastic state variables at each measuring point; when When, it indicates the first The data on the porous elastic state variables at some measuring points are incomplete. The process of establishing the noise level determination criteria in step 1 is as follows: Since the solid and fluid phase state variables in a porous elastic field have different dimensions and orders of magnitude and are physically correlated with each other, this invention uses Mahalanobis distance or robust absolute median (MAD) to automatically define the noise level of the porous elastic state variable data at the measuring points. When using Mahalanobis distance to automatically define the noise level of porous elastic state variable data at measuring points, the following is included: Using the porous elastic state variable data from all complete measurement points, calculate the sample mean vector of the porous elastic state variable tuples. Covariance Matrix ; For the At each measuring point, the porous elastic state variable is denoted as... And calculate its vector to the sample mean. Mahalanobis distance : ; According to the chi-square distribution ( The system can automatically calculate the critical distance threshold corresponding to a specified noise level confidence threshold (e.g., setting the noise level confidence threshold to 95%) using either distribution or kernel density estimation (KDE). ; like Determine the measuring point Data within the high confidence interval are labeled as low noise for the corresponding porous elastic state variable data. like Determine the measuring point Deviation from sample mean vector The corresponding porous elastic state variable data are marked as high noise.

[0029] The process of classifying and mapping the porous elastic state variable tuples based on the integrity assessment criterion and the noise level determination criterion is as follows: when and When the porous elastic state variable data of the measuring point is classified into Class A, it indicates that the porous elastic state variable data of the measuring point is complete and the noise level is low, serving as a high-quality anchor point for data-driven analysis. when and When the data of the porous elastic state variable at the measuring point is classified into Class B, it indicates that the data of the porous elastic state variable at the measuring point is incomplete and the noise level is low. when and When the porous elastic state variable data of the measuring point is classified into category C, it indicates that the porous elastic state variable data of the measuring point is complete and the noise level is high. when and When the data is incomplete, the porous elastic state variable data of the measuring point is classified as Class D, indicating that the data of the porous elastic state variable of the measuring point is incomplete and the noise level is high.

[0030] Step 2: Establish a finite element geometric model of a high-center-wall rockfill dam, and define nodes, topological relationships, shape functions and their derivative matrices and integral point weights in the finite element geometric model of the high-center-wall rockfill dam.

[0031] Specifically, the local solid-state stiffness matrix must be initialized before performing data interpolation and Lagrange multiplier calculations. and local flow phase mobility matrix .

[0032] Local solid stiffness matrix For high-core riprap and clay-core walls, their mechanical behavior exhibits significant nonlinearity. This invention is not limited to linear elasticity and can be seamlessly integrated into the Duncan-Chang EB model to dynamically update the initial local solid-phase stiffness matrix. The tangent modulus is calculated by extracting the current stress level. and bulk modulus This allows for the assembly of isotropic or orthotropic local solid stiffness matrices that vary with stress state. This makes the interpolation of missing data in both Type B and Type D porous elastic state variable data more consistent with the true laws of geotechnical mechanics.

[0033] Local flow phase mobility matrix According to Darcy's law, the mobility of a fluid in a porous medium is defined as follows: ,in The inherent permeability of the material point (obtained by on-site pressure water test or empirical formula). The dynamic viscosity coefficient of pore water. It is a unit tensor.

[0034] Step 3: Perform data interpolation and / or noise dilution on incomplete and high-noise porous elastic state variable data of various types.

[0035] When performing step 3, for type B porous elastic state variable data, when effective stress is missing... At that time, according to Perform data interpolation, where This is the local solid stiffness matrix. This refers to the measured strain tensor obtained by the sensor on-site. When it lacks strain At that time, according to Perform data interpolation, where Local solid stiffness matrix The inverse matrix, The measured effective stress tensor is derived from the measured total stress and measured pore pressure of the sensor. ; When it lacks Darcy flow At that time, according to Perform data interpolation, where This is the local flow phase mobility matrix. This is the measured pressure gradient vector calculated from the measured pore water pressure. The fluid density of pore water, It is the gravity vector; When it lacks a pressure gradient At that time, according to Perform data interpolation, where Local flow phase mobility matrix The inverse matrix, This is the Darcy flow vector actually measured on-site.

[0036] Specifically, in practical engineering, it is difficult to perfectly measure all state variables simultaneously. , , and State variables are often missing, therefore, targeted data imputation is necessary: ​​when effective stress is missing... or / and strain When the solid phase is missing a state variable, it is necessary to interpolate it; when the Darcy flow rate is missing... or / and pressure gradient In this case, it is necessary to interpolate the missing state variables of the flow phase. The mobility matrix is ​​obtained by dividing the permeability by the fluid viscosity.

[0037] When performing step 3, for the C-type porous elastic state variable data, random permeation and stress boundary conditions are applied to the Biot porous elastic FEM forward solver using the estimated global parameters to generate theoretical state tuples. ,in, To synthesize strain, To synthesize effective stress, To synthesize the pressure gradient, To synthesize the flow rate, the number of synthesized data in the theoretical state tuple is 1 to 2 times the number of C-type porous elastic state variable data. Then, all the synthesized data in the theoretical state tuple are appended to the C-type porous elastic state variable dataset and randomly shuffled to dilute the noise in the C-type porous elastic state variable data.

[0038] Specifically, random permeation and stress boundary conditions are random and physically real fluid-force boundary conditions (such as changing reservoir water levels).

[0039] When performing step 3, for D-type porous elastic state variable data, data interpolation is first performed according to the processing method for B-type porous elastic state variable data, and then noise dilution is performed according to the processing method for C-type porous elastic state variable data. The number of synthetic data in the theoretical state tuple generated when processing D-type porous elastic state variable data is 0.8 to 0.9 times the number of synthetic data in the theoretical state tuple generated when processing C-type porous elastic state variable data.

[0040] Specifically, noise reduction processing is performed on both type C and type D porous elastic state variable data to effectively prevent high noise from leading the global Lagrangian solution to a non-physical divergence state, thereby improving the characterization accuracy.

[0041] Step 4: Merge all types of porous elastic state variable data into a single master dataset, randomly rearrange the porous elastic state variable data, establish a global penalty function to evaluate the deviation between the admissible field state variable and the nearest porous elastic state variable tuple, calculate the nodal displacement sequence and pore pressure sequence driven by the current master dataset, calculate the Lagrange multipliers representing the momentum and mass imbalance force, then map the global solution back to the local point to obtain the new physical state, calculate the distance error relative to the data point, and output the global force-permeation coupling state or enter the next iteration based on the distance error.

[0042] The permissible field state variable is denoted as Then it is related to the nearest porous elastic state variable tuple. The deviation between them is calculated using the following global penalty function: , in, It is a global penalty function (or distance functional); This represents the total number of material points (or integration points); For the first The integral weight (or volume) of each material point; , , , They represent the first The theoretical allowable strain, effective stress, pressure gradient, and Darcy flow rate of a material point satisfying the physical conservation equations; , , , They represent the first The corresponding (measured, interpolated, or synthesized) strain, effective stress, pressure gradient, and Darcy flow rate for each material point in the dataset; This is the local solid stiffness matrix. Local solid stiffness matrix The inverse matrix; This is the local flow phase mobility matrix. Local flow phase mobility matrix The inverse matrix; superscript This indicates the matrix transpose.

[0043] The node displacement sequence and pore pressure sequence driven by the current master dataset are calculated according to the following formula: , in, Here is the stiffness matrix of the solid. The solid-fluid coupling matrix, For fluid permeability matrix, This is the fluid compressibility matrix (or water storage matrix). To constrain the water storage ratio (or specific water supply). The time step is the implicit time discrete step. It is a sequence of node displacements. It is a pore pressure sequence; For external node load vectors, For external fluid source-sink vectors; It is the derivative matrix of the displacement shape function (strain-displacement matrix). Let be the derivative matrix of the pressure shape function.

[0044] The Lagrange multipliers are denoted as ( ),in This is a virtual displacement. For virtual pressure, Lagrange multipliers ( Calculate according to the following formula: , in, The sequence of Lagrange multipliers (i.e., the sequence of virtual displacements) representing the momentum imbalance force. The sequence of Lagrange multipliers representing unbalanced mass forces (i.e., the sequence of virtual pressures); The new physical state is represented as ,in, For strain sequence, For effective stress sequence, It is a pressure gradient sequence. It is a flow sequence.

[0045] The distance error between the local point and the data point is calculated according to the following formula: , in, For distance error, For the first In the next iteration, the global driving field is mapped back to the first... A tuple of the physical permissible states of a local material point (i.e.) ); For the first In the nth iteration, it is the... The nearest dataset state tuple matched by each material point in the global dataset (i.e. ).

[0046] If distance error If the tolerance limit is met, the global force-interference coupling state is output; if the distance error is not met, the global force-interference coupling state is output. If the tolerance limit is not met, a nearest neighbor global search is performed in the main dataset to find a new data point for each material point and proceed to the next iteration.

[0047] A high-core rockfill dam core wall force-seepage coupling system includes: The data acquisition and classification module is used to acquire on-site monitoring data of the core wall, construct a porous elastic state variable tuple based on the on-site monitoring data of the core wall, establish integrity assessment criteria and noise level judgment criteria, and perform data classification mapping on the porous elastic state variable tuple based on the integrity assessment criteria and the noise level judgment criteria. The finite element geometric model establishment module is used to establish a finite element geometric model of a high-center-wall rockfill dam, and to define nodes, topological relationships, shape functions and their derivative matrices and integral point weights in the finite element geometric model of the high-center-wall rockfill dam. The data interpolation and noise reduction module is used to perform data interpolation and / or noise dilution processing on incomplete and high-noise porous elastic state variable data of various types. The data fusion-driven module is used to merge all types of porous elastic state variable data into a single master dataset, randomly rearrange the porous elastic state variable data, establish a global penalty function to evaluate the deviation between the admissible field state variable and the nearest porous elastic state variable tuple, calculate the nodal displacement sequence and pore pressure sequence driven by the current master dataset, calculate the Lagrange multipliers representing the momentum and mass imbalance force, then map the global solution back to local points to obtain the new physical state, calculate the distance error relative to the data points, and output the global force-permeation coupling state or enter the next iteration based on the distance error.

Claims

1. A data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam, characterized in that: Includes the following steps: Step 1: Obtain on-site monitoring data of the core wall, and construct a porous elastic state variable tuple based on the on-site monitoring data of the core wall; Establish integrity assessment criteria and noise level determination criteria, and perform data classification mapping on the porous elastic state variable tuples based on the integrity assessment criteria and the noise level determination criteria; Step 2: Establish a finite element geometric model of a high-core rockfill dam, and define nodes, topological relationships, shape functions and their derivative matrices and integral point weights in the finite element geometric model of the high-core rockfill dam; Step 3: Perform data interpolation and / or noise dilution on incomplete and high-noise porous elastic state variable data of various types; Step 4: Merge all types of porous elastic state variable data into a single master dataset, randomly rearrange the porous elastic state variable data, establish a global penalty function to evaluate the deviation between the admissible field state variable and the nearest porous elastic state variable tuple, calculate the nodal displacement sequence and pore pressure sequence driven by the current master dataset, calculate the Lagrange multipliers representing the momentum and mass imbalance force, then map the global solution back to the local point to obtain the new physical state, calculate the distance error relative to the data point, and output the global force-permeation coupling state or enter the next iteration based on the distance error.

2. The data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam as described in claim 1, characterized in that: The on-site monitoring data for the core wall includes: total stress. Displacement pore pressure and potential direct traffic; Based on the on-site monitoring data of the core wall, a tuple of porous elastic state variables is constructed, including: Extract four key tensors / vectors at each material point: strain: , where ∇ is the gradient operator, To identify symmetrical parts; Effective stress: ,in These are the Biot-Willis coefficients. Unit tensor; Pressure gradient: ; Darcy flow: ,in For penetration rate, For fluid viscosity, For fluid density, The volumetric force per unit mass of fluid; The four key tensors / vectors extracted from all material points will be combined to form a tuple of porous elastic state variables: .

3. The data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam as described in claim 2, characterized in that: The process of establishing integrity assessment criteria in step 1 is as follows: ， in, Indicates the first The integrity assessment value of the porous elastic state variable data at each measuring point; , , , They represent the first The data acquisition status of strain, effective stress, pressure gradient and Darcy flow rate at each measuring point is set to 1 when successfully acquired and 0 when missing. when When, it indicates the first Complete data on the porous elastic state variables at each measuring point; when When, it indicates the first The data on the porous elastic state variables at some measuring points are incomplete. The process of establishing the noise level determination criteria in step 1 is as follows: The noise level of porous elastic state variable data at the measuring points is automatically defined using Mahalanobis distance or robust absolute median difference. When using Mahalanobis distance to automatically define the noise level of porous elastic state variable data at measuring points, the following is included: Using the porous elastic state variable data from all complete measurement points, calculate the sample mean vector of the porous elastic state variable tuples. Covariance Matrix ; For the At each measuring point, the porous elastic state variable is denoted as... And calculate its vector to the sample mean. Mahalanobis distance : ； Based on chi-square distribution or kernel density estimation, automatically calculate the critical distance threshold corresponding to the specified noise level confidence threshold. ; like Determine the measuring point Data within the high confidence interval are labeled as low noise for the corresponding porous elastic state variable data. like Determine the measuring point Deviation from sample mean vector The corresponding porous elastic state variable data are marked as high noise.

4. The data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam as described in claim 3, characterized in that: The process of classifying and mapping the porous elastic state variable tuples based on the integrity assessment criterion and the noise level determination criterion is as follows: when and When the porous elastic state variable data of the measuring point is classified into Class A, it indicates that the porous elastic state variable data of the measuring point is complete and the noise level is low, serving as a high-quality anchor point for data-driven analysis. when and When the data of the porous elastic state variable at the measuring point is classified into Class B, it indicates that the data of the porous elastic state variable at the measuring point is incomplete and the noise level is low. when and When the porous elastic state variable data of the measuring point is classified into category C, it indicates that the porous elastic state variable data of the measuring point is complete and the noise level is high. when and When the data is incomplete, the porous elastic state variable data of the measuring point is classified as Class D, indicating that the data of the porous elastic state variable of the measuring point is incomplete and the noise level is high.

5. The data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam as described in claim 4, characterized in that: When performing step 3, for type B porous elastic state variable data, when effective stress is missing... At that time, according to Perform data interpolation, where This is the local solid stiffness matrix. This refers to the measured strain tensor obtained by the sensor on-site. When it lacks strain At that time, according to Perform data interpolation, where Local solid stiffness matrix The inverse matrix, The measured effective stress tensor is derived from the measured total stress and measured pore pressure of the sensor. ; When it lacks Darcy flow At that time, according to Perform data interpolation, where This is the local flow phase mobility matrix. This is the measured pressure gradient vector calculated from the measured pore water pressure. The fluid density of pore water, It is the gravity vector; When it lacks a pressure gradient At that time, according to Perform data interpolation, where Local flow phase mobility matrix The inverse matrix, This is the Darcy flow vector actually measured on-site.

6. The data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam as described in claim 5, characterized in that: When performing step 3, for the C-type porous elastic state variable data, random permeation and stress boundary conditions are applied to the Biot porous elastic FEM forward solver using the estimated global parameters to generate theoretical state tuples. ,in, For synthetic strain, To synthesize effective stress, To synthesize the pressure gradient, To synthesize the flow rate, the number of synthesized data in the theoretical state tuple is 1 to 2 times the number of C-type porous elastic state variable data. Then, all the synthesized data in the theoretical state tuple are appended to the C-type porous elastic state variable dataset and randomly shuffled to dilute the noise in the C-type porous elastic state variable data.

7. The data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam as described in claim 6, characterized in that: When performing step 3, for D-type porous elastic state variable data, data interpolation is first performed according to the processing method for B-type porous elastic state variable data, and then noise dilution is performed according to the processing method for C-type porous elastic state variable data. The number of synthetic data in the theoretical state tuple generated when processing D-type porous elastic state variable data is 0.8 to 0.9 times the number of synthetic data in the theoretical state tuple generated when processing C-type porous elastic state variable data.

8. The data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam as described in claim 2, characterized in that: The permissible field state variable is denoted as Then it is related to the nearest porous elastic state variable tuple. The deviation between them is calculated using the following global penalty function: ， in, For global penalty function, This represents the total number of material points. For the first The integral weight of each material point; , , , They represent the first The theoretical allowable strain, effective stress, pressure gradient, and Darcy flow rate of a material point satisfying the physical conservation equations; , , , They represent the first The strain, effective stress, pressure gradient, and Darcy flow rate corresponding to each material point in the dataset; Local solid stiffness matrix Local solid stiffness matrix inverse matrix Local flow phase mobility matrix Local flow phase mobility matrix The inverse matrix; superscript This indicates the matrix transpose.

9. The data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam as described in claim 8, characterized in that: The node displacement sequence and pore pressure sequence driven by the current master dataset are calculated according to the following formula: ， in, Here is the stiffness matrix of the solid. The solid-fluid coupling matrix, For fluid permeability matrix, For fluid compressibility matrix, To constrain water storage rate, The time step is the implicit time discrete step. It is a sequence of node displacements. It is a pore pressure sequence; For external node load vectors, For external fluid source-sink vectors; The matrix of derivatives of displacement shape functions. The derivative matrix of the pressure shape function; The Lagrange multipliers are denoted as ( ),in This is a virtual displacement. For virtual pressure, Lagrange multipliers ( Calculate according to the following formula: ， in, It is a virtual shift sequence. It is a virtual pressure sequence; The new physical state is represented as ( ),in, For strain sequence, For effective stress sequence, It is a pressure gradient sequence. It is a flow sequence.

10. The data-driven method for characterizing the core wall force-seepage coupling of a high-core rockfill dam as described in claim 9, characterized in that: The distance error between the local point and the data point is calculated according to the following formula: ， in, For distance error, For the first In the next iteration, the global driving field is mapped back to the first... A tuple of the physical allowable states of a local material point, namely ; For the first In the nth iteration, it is the... The nearest dataset state tuple matched by each material point in the global dataset, i.e. ; If distance error If the tolerance limit is met, the global force-interference coupling state is output; if the distance error is not met, the global force-interference coupling state is output. If the tolerance limit is not met, a nearest neighbor global search is performed in the main dataset to find a new data point for each material point and proceed to the next iteration.

11. A high-core rockfill dam core wall force-seepage coupling system, employing the high-core rockfill dam core wall force-seepage coupling characterization method as described in any one of claims 1 to 10, characterized in that: include: The data acquisition and classification module is used to acquire on-site monitoring data of the core wall and construct a porous elastic state variable tuple based on the on-site monitoring data of the core wall. Establish integrity assessment criteria and noise level determination criteria, and perform data classification mapping on the porous elastic state variable tuples based on the integrity assessment criteria and the noise level determination criteria; The finite element geometric model establishment module is used to establish a finite element geometric model of a high-center-wall rockfill dam, and to define nodes, topological relationships, shape functions and their derivative matrices and integral point weights in the finite element geometric model of the high-center-wall rockfill dam. The data interpolation and noise reduction module is used to perform data interpolation and / or noise dilution processing on incomplete and high-noise porous elastic state variable data of various types. The data fusion-driven module is used to merge all types of porous elastic state variable data into a single master dataset, randomly rearrange the porous elastic state variable data, establish a global penalty function to evaluate the deviation between the admissible field state variable and the nearest porous elastic state variable tuple, calculate the nodal displacement sequence and pore pressure sequence driven by the current master dataset, calculate the Lagrange multipliers representing the momentum and mass imbalance force, then map the global solution back to local points to obtain the new physical state, calculate the distance error relative to the data points, and output the global force-permeation coupling state or enter the next iteration based on the distance error.

Images