A method and system for inverting hydraulic grid parameters based on artificial intelligence

Abstract

This application provides an artificial intelligence-based method and system for inverting hydraulic grid parameters. The method involves acquiring multi-source monitoring data of the target hydraulic engineering area and combining it with a pre-defined hydraulic grid physical mechanism model to establish an AI probabilistic inference framework for parameter inversion in heterogeneous media. It iteratively optimizes the implicit correlation between the spatial distribution of permeability and the dynamic change of hydraulic conductivity in the hydraulic grid physical mechanism model, generating a parameter probability distribution set. Based on this parameter probability distribution set and combined with the historical seepage field evolution pattern of the target hydraulic engineering area, the equivalent parameter values ​​of each grid unit in the hydraulic grid physical mechanism model are dynamically corrected to generate hydraulic grid parameter inversion results. This improves the accuracy and stability of hydraulic grid parameter inversion under complex working conditions, providing reliable support for engineering safety and risk early warning of reservoirs and dams.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method and system for inverting hydraulic grid parameters based on artificial intelligence. Background Technology

[0002] In the field of water conservancy engineering, the alternating operation of reservoir impoundment and flood discharge, as well as the coupled seepage and deformation scenarios of dams, urgently require accurate inversion of hydraulic grid parameters. Due to the heterogeneity of geological structures and the complexity of seepage-deformation coupling effects, traditional methods struggle to accurately capture the implicit correlation between the spatial distribution of permeability and the dynamic changes in hydraulic conductivity. Especially when reservoir water levels fluctuate drastically or dam structures deform, existing technologies cannot effectively balance the accuracy and computational efficiency of parameter inversion. Summary of the Invention

[0003] This application provides a method and system for inverting hydraulic grid parameters based on artificial intelligence, which solves the problems in the prior art that cannot effectively integrate the advantages of physical mechanisms and data-driven approaches, are difficult to capture the seepage-deformation coupling effect under complex working conditions, and lack sufficient utilization of the spatiotemporal characteristics of multi-source monitoring data.

[0004] In a first aspect, embodiments of this application provide an artificial intelligence-based method for inverting hydraulic grid parameters, including:

[0005] Acquire multi-source monitoring data of the target water conservancy project area, wherein the multi-source monitoring data includes groundwater level dynamic sequence, surface deformation field distribution data and hydrogeological boundary constraints, and the hydrogeological boundary constraints are generated by coupling the actual operating conditions and geological structure characteristics of the target water conservancy project;

[0006] Based on the multi-source monitoring data and the preset hydraulic grid physical mechanism model, an artificial intelligence probabilistic inference framework for parameter inversion of heterogeneous media is established. The implicit correlation between the spatial distribution of permeability coefficient and the dynamic change of hydraulic conductivity in the hydraulic grid physical mechanism model is iteratively optimized using the artificial intelligence probabilistic inference framework to generate a parameter probability distribution set that matches the geological structural uncertainty of the target hydraulic engineering area.

[0007] Based on the parameter probability distribution set and combined with the historical seepage field evolution mode of the target water conservancy project area, the equivalent parameter values ​​of each grid unit in the water conservancy grid physical mechanism model are dynamically corrected to generate water conservancy grid parameter inversion results.

[0008] Optionally, based on the multi-source monitoring data and the preset hydraulic grid physical mechanism model, an artificial intelligence probabilistic inference framework for parameter inversion in heterogeneous media is established. This framework iteratively optimizes the implicit correlation between the spatial distribution of permeability and the dynamic change of hydraulic conductivity in the hydraulic grid physical mechanism model, generating a parameter probability distribution set that matches the geological structural uncertainty of the target hydraulic engineering area, including:

[0009] The groundwater level dynamic sequence is decomposed into multi-scale water level fluctuation components in the time dimension, and the surface deformation field distribution data is converted into a deformation gradient tensor in the spatial dimension. Based on the spatiotemporal coupling characteristics of the multi-scale water level fluctuation components and the deformation gradient tensor, a three-dimensional dynamic coupling feature matrix is ​​constructed.

[0010] The initial field of the spatial distribution of the permeability coefficient in the physical mechanism model of the hydraulic grid is bidirectionally embedded with the three-dimensional dynamic coupling feature matrix to generate dynamic weight coefficients.

[0011] Based on the dynamic weight coefficient, an implicit gradient field is introduced into the physical mechanism model of the hydraulic grid, and the implicit gradient field is used to perform probabilistic sparse coding on the correlation between the spatial distribution of the permeability coefficient and the dynamic change of the hydraulic conductivity, thereby generating a set of implicit variables with geological structural uncertainty.

[0012] Based on the set of latent variables, Bayesian Monte Carlo joint sampling is performed within the grid cells of the hydraulic grid physical mechanism model. By simultaneously constraining the periodic boundary conditions of the multi-scale water level fluctuation components and the spatial continuity conditions of the deformation gradient tensor, the joint probability distribution of permeability coefficient and hydraulic conductivity that matches the alternating reservoir water storage and flood discharge conditions and the coupled scenario of dam seepage deformation in the target hydraulic engineering area is iteratively screened, and a set of parameter probability distributions is generated.

[0013] Optionally, based on the set of latent variables, Bayesian Monte Carlo joint sampling is performed within the grid cells of the hydraulic grid physical mechanism model. By simultaneously constraining the periodic boundary conditions of the multi-scale water level fluctuation components and the spatial continuity conditions of the deformation gradient tensor, a joint probability distribution of permeability and hydraulic conductivity matching the alternating reservoir storage and flood discharge conditions and the coupled scenario of dam seepage and deformation in the target hydraulic engineering area is iteratively selected, and a set of parameter probability distributions is generated, including:

[0014] Based on the prior distribution of the spatial distribution of the permeability coefficient in the set of latent variables, and combined with the historical seepage pressure gradient data in the alternating operation of reservoir water storage and flood discharge, a dynamic covariance matrix is ​​constructed.

[0015] The joint probability density field of permeability coefficient and hydraulic conductivity is initialized within the grid cell of the hydraulic grid physical mechanism model, and the joint probability density field of permeability coefficient and hydraulic conductivity is subjected to adaptive covariance adjustment Monte Carlo sampling using the dynamic covariance matrix. During each sampling iteration, the dynamic residual vector between the periodic boundary conditions of the multi-scale water level fluctuation component and the dynamic change of hydraulic conductivity at the current sampling point is calculated synchronously.

[0016] During the convergence phase of the Monte Carlo sampling, a set of sampling points is extracted that satisfy the seepage path topology constraints of the joint probability density field of permeability coefficient and hydraulic conductivity in the dam seepage deformation coupling scenario. The seepage path topology constraints are used to verify the bidirectional coupling of deformation and seepage on the set of sampling points, and the matching degree between the spatial gradient modulus of the deformation gradient tensor and the dynamic change rate of the hydraulic conductivity is calculated to generate the deformation constraint factor.

[0017] Based on the deformation constraint factor, the abnormal sampling points in the sampling point set that deviate from the historical seepage field evolution pattern under the alternating operation of reservoir water storage and flood discharge are probabilistically reweighted, and the spatiotemporal distribution characteristics of the dynamic residual vector and the spatial correlation of the deformation constraint factor are integrated to generate a parameter probability distribution set.

[0018] Optionally, during the convergence phase of the Monte Carlo sampling, a set of sampling points is extracted where the joint probability density field of permeability and hydraulic conductivity satisfies the seepage path topology constraint of the dam seepage deformation coupling scenario. The seepage path topology constraint is used to verify the bidirectional coupling of deformation and seepage on the sampling point set. Furthermore, the matching degree between the spatial gradient modulus of the deformation gradient tensor and the dynamic change rate of the hydraulic conductivity is calculated to generate a deformation constraint factor, including:

[0019] Based on the seepage path topology constraints, the dynamic rate of change of hydraulic conductivity of each sampling point in the sampling point set is mapped to the grid cell boundary of the hydraulic grid physical mechanism model, and the seepage pressure gradient tensor that matches the seepage deformation coupling scenario of the dam is constructed using the triangulation algorithm.

[0020] Within each triangular subdivision unit of the seepage pressure gradient tensor, an alternating solution process for deformation-seepage bidirectional coupling verification is performed. The dynamic change rate of the hydraulic conductivity is used as the input parameter of the seepage field control equation. The seepage pressure distribution field corresponding to the current sampling point is calculated, and a deformation coupling residual field is generated based on the residual between the seepage pressure distribution field and the spatial gradient modulus of the deformation gradient tensor.

[0021] The deformation coupling residual field is spatiotemporally correlated and matched with the time series of the multi-scale water level fluctuation component to invert the deformation displacement increment field corresponding to the current seepage pressure distribution field, and the displacement difference of the deformation displacement increment field at the vertex of the triangular subdivision unit is calculated to generate the modulus matching degree matrix.

[0022] Based on the modulus matching degree matrix, a spatiotemporal alignment window based on the dynamic time warping algorithm is constructed within the triangulation unit, and the spatiotemporal alignment window is used to quantify the phase delay error between the dynamic change rate of hydraulic conductivity and the spatial gradient modulus of the deformation gradient tensor in the seepage deformation coupling process.

[0023] Based on the phase delay error, seepage path branches that satisfy the dam seepage deformation coupling scenario are selected from the seepage pressure gradient tensor. The extreme points of the dynamic rate of change of hydraulic conductivity at the vertices of the triangulation unit corresponding to the seepage path branches are extracted. The extreme points of the dynamic rate of change of hydraulic conductivity are then matched topologically with the extreme points of the spatial gradient magnitude of the deformation gradient tensor to generate deformation constraint factors.

[0024] Optionally, based on the parameter probability distribution set and combined with the historical seepage field evolution pattern of the target water conservancy project area, the equivalent parameter values ​​of each grid cell in the water conservancy grid physical mechanism model are dynamically corrected to generate water conservancy grid parameter inversion results, including:

[0025] The parameter probability distribution set is decomposed into a permeability coefficient heterogeneity component in the spatial dimension and a hydraulic conductivity dynamic response component in the temporal dimension. Based on the topological correlation of the seepage path in the historical seepage field evolution mode of the target water conservancy project area, a spatiotemporal coupled feature fusion matrix is ​​constructed, and the spatiotemporal correlation features of the feature fusion matrix are injected into the grid cell mapping relationship of the water conservancy grid physical mechanism model.

[0026] Within each grid cell of the hydraulic grid physical mechanism model, a bidirectional attention mechanism is used to dynamically assign weights to the heterogeneity component of the permeability coefficient and the dynamic response component of the hydraulic conductivity, thereby generating dynamic weight coefficients that characterize the spatiotemporal correlation of parameters of heterogeneous media.

[0027] Based on the dynamic weight coefficients and the spatiotemporal correlation features of the feature fusion matrix, an implicit regularization constraint term is introduced into the physical mechanism model of the hydraulic grid. The implicit regularization constraint term is used to dynamically iteratively correct the equivalent parameter values ​​of the grid cells. In each iteration, the residual distribution of the seepage field before and after the correction of the seepage path topological correlation is calculated, and the residual distribution of the seepage field is fed back to the feature fusion matrix to update the confidence weight of the spatiotemporal correlation features, thereby obtaining the updated feature fusion matrix.

[0028] Based on the residual distribution of the seepage field and the updated feature fusion matrix, a two-way verification window for seepage deformation is constructed in the physical mechanism model of the hydraulic grid. The two-way verification window is used to match the time rate of change of the dynamic response component of the hydraulic conductivity with the spatial gradient rate of change of the surface deformation field distribution data to generate a verification weight factor. The verification weight factor is then used to screen the equivalent parameter candidate set that satisfies the seepage stability constraint under the alternating operation of reservoir water storage and flood discharge.

[0029] Based on the equivalent parameter candidate set, the spatial distribution confidence of the heterogeneity component of the permeability coefficient and the temporal correlation confidence of the dynamic response component of the hydraulic conductivity are fused to generate a parameter optimization surface. The boundary smoothing process of the parameter optimization surface is then performed by superimposing the permeability mutation characteristics of the fault zone in the hydrogeological boundary constraints, and the hydraulic grid parameter inversion results are output.

[0030] Optionally, based on the residual distribution of the seepage field and the updated feature fusion matrix, a two-way verification window for seepage deformation is constructed in the hydraulic grid physical mechanism model. This two-way verification window is used to match the time rate of change of the dynamic response component of the hydraulic conductivity with the spatial gradient rate of change of the surface deformation field distribution data, generating verification weight factors. These verification weight factors are then used to screen a candidate set of equivalent parameters that satisfy the seepage stability constraints under alternating reservoir storage and flood discharge conditions, including:

[0031] Based on the residual distribution of the seepage field and the updated feature fusion matrix, the multi-scale seepage pressure residual components in the time dimension and the deformation coupling residual gradient in the spatial dimension are extracted, and the time decay coefficient of the multi-scale seepage pressure residual components and the spatial diffusion coefficient of the deformation coupling residual gradient are fused to generate a spatiotemporal dynamic residual matrix.

[0032] The row vectors of the spatiotemporal dynamic residual matrix are mapped to the time axis of the grid cell of the hydraulic grid physical mechanism model, and the column vectors are mapped to the spatial topology axis. The dynamic time warping algorithm is used to align the time rate of change sequence of the hydraulic conductivity dynamic response component with the spatial gradient rate of change sequence of the surface deformation field distribution data.

[0033] Based on the spatiotemporal alignment path, a dynamic residual propagation network is constructed within the bidirectional verification window of the seepage deformation. The dynamic residual propagation network is used to couple the row and column correlation of the spatiotemporal dynamic residual matrix with the confidence weight of the updated feature fusion matrix to generate a phase synchronization index.

[0034] In the dynamic residual propagation network, an iterative screening of seepage deformation bidirectional coupling verification is performed. The extreme points of the time rate of change of the hydraulic conductivity dynamic response component are mapped to the time axis nodes of the dynamic residual propagation network to generate a time dimension residual vector. The extreme points of the spatial gradient rate of change of the surface deformation field distribution data are mapped to the spatial topology axis edge of the dynamic residual propagation network to generate a spatial dimension residual tensor. The projection matching degree of the time dimension residual vector and the spatial dimension residual tensor on the spatiotemporal alignment path is calculated to generate a dynamic verification score.

[0035] Based on the dynamic verification score and the phase synchronization index, a seepage path stability assessment map is constructed within the seepage deformation bidirectional verification window. The verification weight factor is generated by superimposing the row and column correlation of the spatiotemporal dynamic residual matrix and the confidence weight of the updated feature fusion matrix on the seepage path stability assessment map.

[0036] Optionally, based on the seepage field residual distribution and the updated feature fusion matrix, multi-scale seepage pressure residual components in the time dimension and deformation coupling residual gradients in the spatial dimension are extracted, and the time decay coefficients of the multi-scale seepage pressure residual components and the spatial diffusion coefficients of the deformation coupling residual gradients are fused to generate a spatiotemporal dynamic residual matrix, including:

[0037] The residual distribution of the seepage field is input into a multi-scale decomposition network to extract the low-frequency seepage pressure trend component, the medium-frequency periodic fluctuation component, and the high-frequency transient response component in the time dimension.

[0038] Spatial topological decomposition is performed on the deformation coupling residual gradient to extract the longitudinal gradient component and the transverse gradient component perpendicular to the seepage direction along the seepage dominant path in the updated feature fusion matrix.

[0039] Based on the time decay coefficient of the low-frequency seepage pressure trend component, the phase lag of the mid-frequency periodic fluctuation component, and the energy decay rate of the high-frequency transient response component, a time-dimensional residual energy spectrum is constructed, and based on the spatial diffusion coefficient of the longitudinal gradient component and the anisotropy ratio of the transverse gradient component, a spatial-dimensional residual diffusion tensor is constructed.

[0040] Spatiotemporal coupling is performed between the time-dimensional residual energy spectrum and the spatial-dimensional residual diffusion tensor. The time decay coefficient of the low-frequency seepage pressure trend component and the spatial diffusion coefficient of the longitudinal gradient component are multiplied by a tensor to generate a basic coupling component. The phase lag of the mid-frequency periodic fluctuation component and the anisotropy ratio of the transverse gradient component are multiplied by a Hadamard product to generate a periodic modulation component. By superimposing the basic coupling component and the periodic modulation component, and injecting the energy decay rate of the high-frequency transient response component as a regularization constraint term, a spatiotemporal coupling kernel matrix is ​​generated.

[0041] Secondly, embodiments of this application provide an artificial intelligence-based hydraulic grid parameter inversion system, comprising:

[0042] The acquisition module is used to acquire multi-source monitoring data of the target water conservancy project area. The multi-source monitoring data includes groundwater level dynamic sequence, surface deformation field distribution data and hydrogeological boundary constraints. The hydrogeological boundary constraints are generated by coupling the actual operating conditions and geological structure characteristics of the target water conservancy project.

[0043] The optimization module is used to establish an artificial intelligence probabilistic inference framework for parameter inversion of heterogeneous media based on the multi-source monitoring data and the preset hydraulic grid physical mechanism model. The artificial intelligence probabilistic inference framework is used to iteratively optimize the implicit correlation between the spatial distribution of permeability and the dynamic change of hydraulic conductivity in the hydraulic grid physical mechanism model, and generate a parameter probability distribution set that matches the geological structure uncertainty of the target hydraulic engineering area.

[0044] The correction module is used to dynamically correct the equivalent parameter values ​​of each grid unit in the physical mechanism model of the hydraulic grid based on the parameter probability distribution set and the historical seepage field evolution mode of the target hydraulic engineering area, and generate hydraulic grid parameter inversion results.

[0045] Thirdly, embodiments of this application provide a computing device, including a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute an artificial intelligence-based hydraulic grid parameter inversion method as described in any of the first aspects.

[0046] Fourthly, embodiments of this application provide a computer storage medium storing computer program instructions, which, when executed by a processor, implement the artificial intelligence-based hydraulic grid parameter inversion method described in any one of the first aspects.

[0047] In this embodiment, multi-source monitoring data of the target water conservancy project area is acquired. This multi-source monitoring data includes groundwater level dynamic sequences, surface deformation field distribution data, and hydrogeological boundary constraints. These hydrogeological boundary constraints are generated by coupling the actual operating conditions and geological structural characteristics of the target water conservancy project. Based on the multi-source monitoring data and a pre-set water conservancy grid physical mechanism model, an artificial intelligence probabilistic inference framework for parameter inversion in heterogeneous media is established. This framework iteratively optimizes the implicit correlation between the spatial distribution of permeability and the dynamic change of hydraulic conductivity in the water conservancy grid physical mechanism model, generating a parameter probability distribution set that matches the geological structural uncertainty of the target water conservancy project area. Based on this parameter probability distribution set and combined with the historical seepage field evolution pattern of the target water conservancy project area, the equivalent parameter values ​​of each grid unit in the water conservancy grid physical mechanism model are dynamically corrected, generating water conservancy grid parameter inversion results.

[0048] This application acquires multi-source monitoring data of the target water conservancy project area, including dynamic sequences of groundwater levels, surface deformation field distribution data, and hydrogeological boundary constraints, enabling a comprehensive capture of the seepage-deformation coupling effect under complex operating conditions. Based on the multi-source monitoring data and a pre-set water conservancy grid physical mechanism model, an artificial intelligence probabilistic inference framework for parameter inversion in heterogeneous media is established. This framework effectively integrates the advantages of physical mechanisms and data-driven approaches, iteratively optimizing the implicit correlation between the spatial distribution of permeability and the dynamic changes in hydraulic conductivity, and generating a parameter probability distribution set that matches the geological structural uncertainties of the target water conservancy project area. Furthermore, by combining historical seepage field evolution patterns, the equivalent parameter values ​​of each grid unit in the water conservancy grid physical mechanism model are dynamically corrected, ultimately generating high-precision and highly stable water conservancy grid parameter inversion results. This provides reliable support for engineering safety and risk early warning under alternating reservoir impoundment and flood discharge conditions and dam seepage-deformation coupling scenarios.

[0049] Furthermore, by decomposing the dynamic sequence of groundwater level into multi-scale water level fluctuation components and converting the surface deformation field distribution data into a deformation gradient tensor, and constructing a three-dimensional dynamic coupling feature matrix based on the spatiotemporal coupling characteristics of the two, the spatiotemporal correlation characteristics of multi-source monitoring data are fully explored. Dynamic weight coefficients are generated through bidirectional embedding calculations to accurately characterize the nonlinear response intensity of the spatial distribution of permeability and the dynamic changes in hydraulic conductivity in heterogeneous media. The physical rationality of the parameter optimization direction is ensured by combining the angle constraint term between the fault zone strike and the permeability anisotropy direction. An implicit gradient field is introduced to probabilistically sparsely encode the correlation between permeability and hydraulic conductivity, generating a set of latent variables with geological structural uncertainties. Through Bayesian Monte Carlo joint sampling, a joint probability distribution of permeability and hydraulic conductivity matching the alternating reservoir impoundment-discharge conditions and the dam seepage-deformation coupling scenario is selected, ultimately generating a high-confidence parameter probability distribution set, significantly improving the accuracy and stability of parameter inversion.

[0050] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 A flowchart illustrating an artificial intelligence-based method for inverting hydraulic grid parameters is provided in this application embodiment;

[0053] Figure 2 A schematic diagram of the structure of an artificial intelligence-based hydraulic grid parameter inversion system provided in this application embodiment;

[0054] Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0055] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0056] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0057] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0058] Figure 1 This application provides a flowchart of an artificial intelligence-based method for inverting hydraulic grid parameters, as shown in the following embodiments. Figure 1 As shown, the method includes:

[0059] Step 101: Obtain multi-source monitoring data of the target water conservancy project area. The multi-source monitoring data includes groundwater level dynamic sequence, surface deformation field distribution data, and hydrogeological boundary constraints. The hydrogeological boundary constraints are generated by coupling the actual operating conditions and geological structure characteristics of the target water conservancy project.

[0060] In this step, multi-source monitoring data refers to the collection of various types of monitoring data obtained from the target water conservancy project area, including groundwater level dynamic sequences, surface deformation field distribution data, and hydrogeological boundary constraints. Specifically, the groundwater level dynamic sequence refers to continuous observation data of groundwater level changes over time, reflecting the dynamic response of groundwater under alternating reservoir impoundment and discharge conditions; the surface deformation field distribution data refers to the spatial distribution information of surface deformation obtained through remote sensing or ground monitoring, characterizing the deformation features of the dam structure under the coupling effect of seepage and deformation; and the hydrogeological boundary constraints are boundary conditions generated by coupling the actual operating conditions of the target water conservancy project (such as reservoir water level changes and discharge flow) with geological structural characteristics (such as fault zone distribution and rock permeability), used to constrain the physical rationality of the parameter inversion process.

[0061] In this embodiment, the specific process of acquiring multi-source monitoring data includes: First, by using pressure sensors deployed in groundwater observation wells, the dynamic sequence of groundwater level in the target water conservancy project area is collected. The data sampling frequency is once per hour to ensure that transient water level changes under the alternating conditions of reservoir water storage and flood discharge are captured. Second, surface deformation field distribution data is acquired using synthetic aperture radar (SAR) remote sensing technology, and the deformation gradient information of the dam surface is extracted using differential interferometry, with a spatial resolution of 10 meters × 10 meters. Finally, based on the geological exploration report and actual operation records of the target water conservancy project, geological structural features such as fault zone distribution and rock strata permeability are extracted. Combined with the reservoir water level change curve and flood discharge data, hydrogeological boundary constraints are generated for physical constraints in the subsequent parameter inversion process.

[0062] For example, taking a reservoir dam project as an example, in step 101, the dynamic sequence of groundwater level during the flood season (June to September) of 2023 was obtained through 20 groundwater observation wells. The data recorded the water level fluctuation characteristics under the alternating conditions of reservoir water storage and flood discharge. At the same time, using SAR image data from the Sentinel 1-1 satellite, the surface deformation field distribution data of the dam area was extracted, and a significant deformation gradient anomaly area was found in the middle of the dam. In addition, combined with the geological exploration report, it was determined that there are two main fault zones in the dam area, and based on the reservoir's operation records in 2023 (including water level change curves and flood discharge data), hydrogeological boundary constraints were generated, providing a physical constraint basis for subsequent parameter inversion.

[0063] Step 102: Based on the multi-source monitoring data and the preset hydraulic grid physical mechanism model, establish an artificial intelligence probabilistic inference framework for parameter inversion of heterogeneous media. Use the artificial intelligence probabilistic inference framework to iteratively optimize the implicit correlation between the spatial distribution of permeability and the dynamic change of hydraulic conductivity in the hydraulic grid physical mechanism model, and generate a parameter probability distribution set that matches the geological structure uncertainty of the target hydraulic engineering area.

[0064] In this step, the hydraulic grid physical mechanism model refers to a numerical model constructed based on the finite element method to simulate the seepage-deformation coupling process in the hydraulic engineering area. Its grid cells can reflect the geological structural characteristics of heterogeneous media. The artificial intelligence probabilistic inference framework refers to a probabilistic modeling framework that combines physical mechanisms and data-driven methods. It generates a set of parameter probability distributions by iteratively optimizing the implicit correlation between the spatial distribution of permeability and the dynamic change of hydraulic conductivity. The set of parameter probability distributions refers to the joint probability distribution of permeability and hydraulic conductivity in the spatial and temporal dimensions obtained through probabilistic inference, which reflects the uncertainty of the geological structure of the target hydraulic engineering area.

[0065] In this embodiment, the specific process of establishing the artificial intelligence probabilistic inference framework includes: First, based on the multi-source monitoring data obtained in step 101, the dynamic sequence of groundwater level is decomposed into multi-scale water level fluctuation components, the surface deformation field distribution data is converted into a deformation gradient tensor, and a three-dimensional dynamic coupling feature matrix is ​​constructed; Second, the initial field of permeability coefficient spatial distribution in the hydraulic grid physical mechanism model is bidirectionally embedded with the three-dimensional dynamic coupling feature matrix to generate dynamic weight coefficients, which are used to characterize the nonlinear response intensity of permeability coefficient and hydraulic conductivity in heterogeneous media; Next, an implicit gradient field is introduced to probabilistically sparsely encode the correlation between permeability coefficient and hydraulic conductivity to generate a set of latent variables; Finally, through Bayesian Monte Carlo joint sampling, a joint probability distribution of permeability coefficient and hydraulic conductivity that matches the alternating operation of reservoir water storage-flood discharge and the coupling scenario of dam seepage-deformation is selected to generate a set of parameter probability distributions.

[0066] For example, taking a reservoir dam project as an example, based on the acquired multi-source monitoring data, the dynamic sequence of groundwater level during the 2023 flood season was decomposed into low-frequency trend, mid-frequency fluctuation, and high-frequency transient components. The surface deformation field distribution data was converted into a deformation gradient tensor, constructing a three-dimensional dynamic coupling feature matrix. Subsequently, the initial permeability coefficient field in the hydraulic grid physical mechanism model was bidirectionally embedded with the three-dimensional dynamic coupling feature matrix to generate dynamic weight coefficients. It was found that the dynamic weight coefficients in the fault zone region in the middle of the dam were significantly higher than those in other regions, indicating that the seepage-deformation coupling effect in this region was more significant. Next, the correlation between permeability and hydraulic conductivity was probabilistically sparsely encoded using an implicit gradient field to generate a set of latent variables. Bayesian Monte Carlo joint sampling was then used to select a joint probability distribution of permeability and hydraulic conductivity that matched the alternating reservoir storage-discharge conditions and the dam seepage-deformation coupling scenario. Finally, a parameter probability distribution set was generated, providing high-confidence input data for subsequent parameter inversion.

[0067] Step 103: Based on the parameter probability distribution set and combined with the historical seepage field evolution mode of the target water conservancy project area, dynamically correct the equivalent parameter values ​​of each grid unit in the water conservancy grid physical mechanism model to generate water conservancy grid parameter inversion results;

[0068] In this step, the historical seepage field evolution model refers to the spatiotemporal variation law of the seepage field in the target water conservancy project area over a period of time, which is usually obtained by comprehensive analysis of historical monitoring data and numerical simulation results; the equivalent parameter value refers to the permeability coefficient and hydraulic conductivity value of each grid cell in the water conservancy grid physical mechanism model after dynamic correction, which can reflect the geological structure characteristics of heterogeneous media and the seepage-deformation coupling effect; the water conservancy grid parameter inversion result refers to the high-precision parameter distribution generated after dynamic correction of the equivalent parameter value, which is used to support the safety assessment and risk warning of water conservancy projects.

[0069] In this embodiment, the specific process of dynamically correcting the equivalent parameter values ​​includes: First, decomposing the generated parameter probability distribution set into a permeability coefficient heterogeneity component in the spatial dimension and a hydraulic conductivity dynamic response component in the temporal dimension, and constructing a spatiotemporally coupled feature fusion matrix in conjunction with historical seepage field evolution patterns; Second, within each grid cell of the hydraulic grid physical mechanism model, a bidirectional attention mechanism is used to dynamically assign weights to the permeability coefficient heterogeneity component and the hydraulic conductivity dynamic response component, generating dynamic weight coefficients characterizing the spatiotemporal correlation of heterogeneous medium parameters; Next, based on the spatiotemporal correlation characteristics of the dynamic weight coefficients and the feature fusion matrix, an implicit regularization constraint term is introduced to dynamically iteratively correct the equivalent parameter values ​​of the grid cells. In each iteration, the seepage path topological correlation is calculated, and the seepage field residual distribution before and after correction is fed back to the feature fusion matrix to update the confidence weights; Finally, based on the updated feature fusion matrix and the seepage field residual distribution, hydraulic grid parameter inversion results adapted to the alternating reservoir storage-discharge conditions and the dam seepage-deformation coupling scenario are generated.

[0070] For example, taking a reservoir dam project as an example, based on the generated parameter probability distribution set and combined with the historical seepage field evolution model during the 2022 flood season, the parameter probability distribution set is decomposed into a heterogeneous component of permeability coefficient and a dynamic response component of hydraulic conductivity, and a spatiotemporally coupled feature fusion matrix is ​​constructed. Subsequently, within each grid cell of the hydraulic grid physical mechanism model, a bidirectional attention mechanism is used to dynamically assign weights to the permeability coefficient and hydraulic conductivity. It is found that the dynamic weight coefficient of the fault zone in the middle of the dam is significantly higher than that of other areas, indicating that the parameter correction in this area should prioritize the seepage-deformation coupling effect. Next, an implicit regularization constraint term is introduced to dynamically iteratively correct the equivalent parameter values. By calculating the residual distribution of the seepage field, it is found that the matching degree between the corrected seepage pressure gradient and the surface deformation field distribution data is significantly improved. The final hydraulic grid parameter inversion results show that the permeability coefficient and hydraulic conductivity values ​​in the fault zone in the middle of the dam are reduced by 15% and 20% respectively compared with the initial values, which is consistent with the actual situation and provides reliable data support for the safe operation of the reservoir.

[0071] In water conservancy projects, parameter inversion of heterogeneous media is a complex and challenging problem, especially under alternating reservoir impoundment and flood discharge conditions and coupled seepage and deformation scenarios in dams. Traditional methods struggle to accurately capture the implicit correlation between the spatial distribution of permeability and the dynamic changes in hydraulic conductivity. Existing technologies typically rely on a single data source or a simplified physical model, resulting in insufficient accuracy in the inversion results and failing to meet the needs of actual engineering projects. Therefore, in some embodiments, as described in step 102, based on the multi-source monitoring data and a preset water conservancy grid physical mechanism model, an artificial intelligence probabilistic inference framework for parameter inversion of heterogeneous media is established. This framework iteratively optimizes the implicit correlation between the spatial distribution of permeability and the dynamic changes in hydraulic conductivity in the water conservancy grid physical mechanism model, generating a parameter probability distribution set that matches the geological structural uncertainties of the target water conservancy project area, including:

[0072] Step 201: Decompose the groundwater level dynamic sequence into multi-scale water level fluctuation components in the time dimension, and convert the surface deformation field distribution data into a deformation gradient tensor in the spatial dimension. Based on the spatiotemporal coupling characteristics of the multi-scale water level fluctuation components and the deformation gradient tensor, construct a three-dimensional dynamic coupling feature matrix.

[0073] In this step, the multi-scale water level fluctuation component refers to the decomposition of the groundwater level dynamic sequence into low-frequency trend component, mid-frequency fluctuation component, and high-frequency transient component through a multi-scale decomposition algorithm, which respectively characterize the long-term trend, periodic fluctuation, and transient response under the alternating conditions of reservoir water storage and flood discharge; the deformation gradient tensor refers to the conversion of the surface deformation field distribution data into longitudinal gradient components along the dominant seepage path and transverse gradient components perpendicular to the seepage direction through spatial topological decomposition, which characterize the deformation characteristics of the dam structure; the three-dimensional dynamic coupling feature matrix is ​​a matrix constructed by fusing the spatiotemporal correlation features of the multi-scale water level fluctuation component and the deformation gradient tensor, which is used to characterize the spatiotemporal distribution law of the seepage-deformation coupling effect.

[0074] In this embodiment, the wavelet transform algorithm is first used to decompose the dynamic sequence of groundwater level into low-frequency trend components, mid-frequency fluctuation components, and high-frequency transient components, and their time decay coefficients, phase lags, and energy decay rates are extracted respectively. Secondly, the spatial topology decomposition algorithm is used to convert the surface deformation field distribution data into a deformation gradient tensor, and the spatial diffusion coefficients of the longitudinal and lateral gradient components are extracted. Finally, based on the spatiotemporal coupling characteristics of the multi-scale water level fluctuation components and the deformation gradient tensor, a three-dimensional dynamic coupling feature matrix is ​​constructed, where the row vector dimension of the matrix corresponds to the time decay sequence, the column vector dimension corresponds to the spatial diffusion sequence, and the matrix elements are generated by fusing the time decay coefficient and the spatial diffusion coefficient.

[0075] Step 202: Perform bidirectional embedding calculation between the initial field of the spatial distribution of the permeability coefficient in the physical mechanism model of the hydraulic grid and the three-dimensional dynamic coupling feature matrix to generate dynamic weight coefficients;

[0076] In this step, bidirectional embedding calculation refers to the process of bidirectionally mapping the initial field of permeability spatial distribution with the three-dimensional dynamic coupling feature matrix to generate dynamic weight coefficients. Dynamic weight coefficients are parameters used to characterize the nonlinear response intensity of permeability and hydraulic conductivity in heterogeneous media. Their values ​​are generated through bidirectional embedding calculation and are corrected by the angle constraint term between the fracture zone orientation and the permeability anisotropy direction.

[0077] In this embodiment, firstly, the initial field of permeability coefficient spatial distribution in the hydraulic grid physical mechanism model is mapped to the time axis of the three-dimensional dynamic coupling feature matrix to generate a time dimension embedding vector; secondly, the three-dimensional dynamic coupling feature matrix is ​​mapped to the spatial axis of the initial field of permeability coefficient spatial distribution to generate a spatial dimension embedding vector; then, the time dimension embedding vector and the spatial dimension embedding vector are fused through bidirectional embedding calculation to generate dynamic weight coefficients, and the angle constraint term between the fault zone orientation and the permeability anisotropy direction is superimposed to correct the update direction of the dynamic weight coefficients, ensuring the physical rationality of the parameter optimization process.

[0078] Step 203: Based on the dynamic weight coefficient, an implicit gradient field is introduced into the physical mechanism model of the hydraulic grid, and the implicit gradient field is used to perform probabilistic sparse coding on the correlation between the spatial distribution of the permeability coefficient and the dynamic change of the hydraulic conductivity, thereby generating a set of implicit variables with geological structural uncertainty.

[0079] In this step, the implicit gradient field refers to the gradient field introduced by dynamic weighting coefficients to characterize the relationship between permeability and hydraulic conductivity, and its value is generated by probabilistic sparse coding; the set of latent variables refers to the set of parameters generated by the implicit gradient field to describe the uncertainty of geological structure, and its value is optimized by probabilistic sparse coding.

[0080] In this embodiment, firstly, an implicit gradient field is introduced into the hydraulic grid physical mechanism model based on dynamic weight coefficients to generate a correlation matrix between permeability and hydraulic conductivity; secondly, the correlation matrix is ​​optimized using a probabilistic sparse coding algorithm to extract a set of latent variables with geological structural uncertainties. The sparse coding process is achieved by constraining the angle range between the topological correlation of the seepage path and the anisotropy of permeability.

[0081] Step 204: Based on the set of latent variables, perform Bayesian Monte Carlo joint sampling within the grid cells of the hydraulic grid physical mechanism model. By simultaneously constraining the periodic boundary conditions of the multi-scale water level fluctuation components and the spatial continuity conditions of the deformation gradient tensor, iteratively screen the joint probability distribution of permeability coefficient and hydraulic conductivity that matches the alternating reservoir water storage and flood discharge conditions and the coupled scenario of dam seepage deformation in the target hydraulic engineering area, and generate a set of parameter probability distributions.

[0082] In this step, Bayesian Monte Carlo joint sampling refers to the process of jointly sampling the set of latent variables based on the Bayesian probabilistic framework and the Monte Carlo sampling method; the parameter probability distribution set refers to the joint probability distribution of permeability and hydraulic conductivity generated through joint sampling, which is used to describe the geological structural uncertainty of the target water conservancy project area.

[0083] In this embodiment, firstly, the joint probability density field of permeability coefficient and hydraulic conductivity is initialized within the grid cells of the hydraulic grid physical mechanism model based on the set of latent variables; secondly, the probability density field is iteratively optimized through Bayesian Monte Carlo joint sampling, simultaneously constraining the periodic boundary conditions of the multi-scale water level fluctuation components and the spatial continuity conditions of the deformation gradient tensor; finally, the joint probability distribution of permeability coefficient and hydraulic conductivity that matches the alternating operation of reservoir water storage and flood discharge and the coupling scenario of dam seepage and deformation is selected to generate a set of parameter probability distributions.

[0084] For example, taking a reservoir dam project as an example, based on the dynamic sequence of groundwater levels during the 2023 flood season, a wavelet transform algorithm was used to decompose low-frequency trend components, mid-frequency fluctuation components, and high-frequency transient components. Combined with the deformation gradient tensor generated from surface deformation field distribution data, a three-dimensional dynamic coupling feature matrix was constructed. The initial field of permeability coefficient spatial distribution was bidirectionally embedded into the three-dimensional dynamic coupling feature matrix to generate dynamic weight coefficients. It was found that the dynamic weight coefficients in the fault zone region in the middle of the dam were significantly higher than those in other regions. Based on the dynamic weight coefficients, an implicit gradient field was introduced to generate a set of latent variables. It was found that the latent variable values ​​in the middle region of the dam deviated significantly from the mean, indicating a high degree of uncertainty in the geological structure of this region. Through Bayesian Monte Carlo joint sampling to generate a parameter probability distribution set, it was found that the permeability and hydraulic conductivity values ​​in the fault zone region in the middle of the dam decreased compared to their initial values, consistent with the actual situation.

[0085] This embodiment achieves efficient fusion of multi-source monitoring data and physical mechanism models through multi-scale decomposition, bidirectional embedding computation, implicit gradient field introduction, and Bayesian Monte Carlo joint sampling. It accurately captures the implicit correlation between the spatial distribution of permeability and the dynamic changes in hydraulic conductivity, generating a parameter probability distribution set that matches the geological structural uncertainties of the target water conservancy project area. This method offers advantages of high accuracy and high stability under complex operating conditions, providing reliable support for engineering safety and risk early warning in alternating reservoir impoundment and flood discharge scenarios and dam seepage-deformation coupling scenarios.

[0086] In the process of inverting parameters from a hydraulic grid, traditional Monte Carlo sampling methods often struggle to effectively handle the complex relationship between permeability and hydraulic conductivity in heterogeneous media, especially under alternating reservoir storage and discharge conditions and dam seepage-deformation coupling scenarios, resulting in low sampling efficiency and insufficient accuracy. Furthermore, existing methods lack explicit consideration of seepage path topological constraints, leading to poor applicability of the inversion results in regions with complex geological structures. Therefore, as another embodiment, according to step 204, Bayesian Monte Carlo joint sampling is performed within the grid cells of the hydraulic grid physical mechanism model based on the set of latent variables. By simultaneously constraining the periodic boundary conditions of the multi-scale water level fluctuation components and the spatial continuity conditions of the deformation gradient tensor, a joint probability distribution of permeability and hydraulic conductivity matching the alternating reservoir storage and discharge conditions and the dam seepage-deformation coupling scenario of the target hydraulic engineering area is iteratively selected, generating a parameter probability distribution set, including:

[0087] Step 301: Based on the prior spatial distribution of the permeability coefficient in the set of latent variables, and combined with the historical seepage pressure gradient data in the alternating operation of reservoir water storage and flood discharge, construct a dynamic covariance matrix.

[0088] In this step, the dynamic covariance matrix refers to the covariance matrix generated by fusing the prior distribution of the spatial distribution of the permeability coefficient with historical seepage pressure gradient data. It is used to describe the spatiotemporal correlation characteristics of the permeability coefficient and hydraulic conductivity in heterogeneous media, and its value is dynamically updated through an adaptive covariance adjustment algorithm.

[0089] In this embodiment, firstly, based on the prior distribution of the spatial distribution of the permeability coefficient in the latent variable set, the heterogeneity characteristics of the permeability coefficient in the spatial dimension are extracted, and combined with the historical seepage pressure gradient data under the alternating operation of reservoir water storage and flood discharge, an initial covariance matrix is ​​generated. Secondly, the initial covariance matrix is ​​dynamically updated by an adaptive covariance adjustment algorithm. The update process is achieved by constraining the angle range between the topological correlation of the seepage path and the anisotropy of permeability, ensuring that the covariance matrix can reflect the spatiotemporal correlation characteristics of parameters under complex operating conditions.

[0090] Step 302: Initialize the joint probability density field of permeability coefficient and hydraulic conductivity within the grid cell of the hydraulic grid physical mechanism model, and perform adaptive covariance adjustment Monte Carlo sampling on the joint probability density field of permeability coefficient and hydraulic conductivity using the dynamic covariance matrix. In each sampling iteration, synchronously calculate the dynamic residual vector between the periodic boundary conditions of the multi-scale water level fluctuation component and the dynamic change of hydraulic conductivity at the current sampling point.

[0091] In this step, the joint probability density field of permeability and hydraulic conductivity refers to the field initialized within the grid cells of the hydraulic grid physical mechanism model to describe the joint probability distribution of permeability and hydraulic conductivity, and its value is generated through Monte Carlo sampling; the adaptive covariance-adjusted Monte Carlo sampling refers to the process of iteratively sampling the joint probability density field based on the dynamic covariance matrix, and its sampling step size is adaptively adjusted through the dynamic update of the covariance matrix.

[0092] In this embodiment, firstly, the joint probability density field of permeability and hydraulic conductivity is initialized within the grid cells of the hydraulic grid physical mechanism model, where the initial values ​​are generated through the prior distribution in the latent variable set; secondly, Monte Carlo sampling is performed on the joint probability density field using a dynamic covariance matrix, where the sampling step size is adaptively adjusted through dynamic updates of the covariance matrix to ensure efficient convergence of the sampling process; finally, the joint probability distribution of permeability and hydraulic conductivity is generated through iterative sampling, providing high-confidence input data for subsequent parameter selection.

[0093] Step 303: During the convergence phase of the Monte Carlo sampling, extract the set of sampling points for which the joint probability density field of permeability coefficient and hydraulic conductivity satisfies the topological constraints of the seepage path in the dam seepage deformation coupling scenario. Use the topological constraints of the seepage path to perform bidirectional coupling verification of deformation and seepage on the set of sampling points, and calculate the matching degree between the spatial gradient modulus of the deformation gradient tensor and the dynamic change rate of the hydraulic conductivity to generate a deformation constraint factor.

[0094] In this step, the seepage path topology constraint refers to the constraint conditions generated by the angle range between the seepage path direction and the permeability anisotropy direction, which is used to screen the set of sampling points that meet the seepage-deformation coupling scenario of the dam; the deformation constraint factor refers to the parameter generated by calculating the matching degree between the spatial gradient modulus of the deformation gradient tensor and the dynamic change rate of hydraulic conductivity, which is used to characterize the physical rationality of the set of sampling points under the seepage-deformation coupling scenario.

[0095] In this embodiment, firstly, during the convergence phase of Monte Carlo sampling, the set of sampling points that satisfy the seepage path topology constraints in the joint probability density field of permeability coefficient and hydraulic conductivity is extracted; secondly, the deformation-seepage bidirectional coupling verification of the sampling point set is performed using the seepage path topology constraints, and the matching degree between the spatial gradient modulus of the deformation gradient tensor and the dynamic change rate of hydraulic conductivity is calculated; finally, a deformation constraint factor is generated based on the matching degree to screen the set of sampling points that satisfy the seepage-deformation coupling scenario of the dam.

[0096] Step 304: Based on the deformation constraint factor, perform probability reweighting on the abnormal sampling points in the sampling point set that deviate from the historical seepage field evolution pattern under the alternating operation of reservoir water storage and flood discharge, and integrate the spatiotemporal distribution characteristics of the dynamic residual vector with the spatial correlation of the deformation constraint factor to generate a parameter probability distribution set.

[0097] In this step, probability reweighting refers to the process of adjusting the probability weights of abnormal sampling points based on deformation constraint factors, which is used to improve the accuracy and stability of the parameter probability distribution set; dynamic residual vector refers to the vector generated by calculating the residual between the sampling point set and the historical seepage field evolution mode, which is used to describe the error distribution of the parameter inversion results.

[0098] In this embodiment, firstly, the abnormal sampling points in the sampling point set that deviate from the historical seepage field evolution pattern are probabilistically reweighted according to the deformation constraint factor. The reweighting process is achieved by constraining the angle range between the topological correlation of the seepage path and the direction of permeability anisotropy. Secondly, the spatiotemporal distribution characteristics of the dynamic residual vector and the spatial correlation of the deformation constraint factor are fused to generate a parameter probability distribution set. The fusion process is achieved by superimposing the spatial continuity conditions of the seepage pressure gradient characteristics and the deformation gradient tensor.

[0099] Taking a reservoir dam project as an example, based on the prior spatial distribution of permeability coefficient in the latent variable set and historical seepage pressure gradient data during the 2023 flood season, a dynamic covariance matrix is ​​constructed. It is found that the covariance value in the fault zone region in the middle of the dam is significantly higher than in other regions. A joint probability density field of permeability coefficient and hydraulic conductivity is initialized, and Monte Carlo sampling is performed using the dynamic covariance matrix to generate a joint probability distribution. It is found that the sampling point density in the middle region of the dam is significantly higher than in other regions. A set of sampling points satisfying the seepage path topology constraints is extracted, and a deformation constraint factor is generated. It is found that the deformation constraint factor value in the middle region of the dam deviates significantly from the mean. Based on the deformation constraint factor, the abnormal sampling points are probabilistically reweighted to generate a set of parameter probability distributions.

[0100] This embodiment achieves high accuracy and stability in hydraulic grid parameter inversion under complex operating conditions by constructing a dynamic covariance matrix, performing Monte Carlo sampling with adaptive covariance adjustment, generating deformation constraint factors, and applying probability reweighting. This method effectively captures the spatiotemporal correlation characteristics of permeability and hydraulic conductivity in heterogeneous media, generating a parameter probability distribution set that matches the geological structural uncertainties of the target hydraulic engineering area. This provides reliable support for engineering safety and risk early warning under alternating reservoir impoundment and flood discharge conditions, as well as dam seepage and deformation coupling scenarios.

[0101] Because existing methods often struggle to effectively verify the physical rationality of the sampling point set in the seepage-deformation coupling scenario during the convergence phase of Monte Carlo sampling, especially lacking a quantitative assessment of the matching degree between the dynamic change rate of hydraulic conductivity and the spatial gradient magnitude of the deformation gradient tensor, and because traditional methods lack explicit modeling of the deformation-seepage bidirectional coupling effect within the triangularly partitioned unit during the construction process of the seepage pressure gradient tensor, resulting in insufficient reliability of the inversion results, as another embodiment, according to step 303, during the convergence phase of the Monte Carlo sampling, a sampling point set is extracted where the joint probability density field of permeability and hydraulic conductivity satisfies the seepage path topology constraint of the dam seepage-deformation coupling scenario. The seepage path topology constraint is used to verify the bidirectional coupling of deformation and seepage on the sampling point set, and the matching degree between the spatial gradient magnitude of the deformation gradient tensor and the dynamic change rate of hydraulic conductivity is calculated to generate a deformation constraint factor, including:

[0102] Step 401: Based on the seepage path topology constraints, the dynamic rate of change of hydraulic conductivity of each sampling point in the sampling point set is mapped to the grid cell boundary of the hydraulic grid physical mechanism model, and a seepage pressure gradient tensor matching the seepage deformation coupling scenario of the dam is constructed using the triangulation algorithm.

[0103] In this step, the seepage pressure gradient tensor is a tensor constructed by the triangulation algorithm to describe the spatial distribution of seepage pressure. Its value is generated by mapping the dynamic rate of change of hydraulic conductivity to the grid cell boundary.

[0104] In this embodiment, firstly, based on the topological constraints of the seepage path, the dynamic rate of change of hydraulic conductivity of each sampling point in the sampling point set is mapped to the grid cell boundary of the hydraulic grid physical mechanism model to generate an initial seepage pressure distribution. Secondly, the initial seepage pressure distribution is spatially partitioned using a triangulation algorithm to construct a seepage pressure gradient tensor, wherein the vertices of the triangulation cells are determined by constraining the angle range between the seepage path direction and the permeability anisotropy direction. Finally, by superimposing the spatial continuity conditions of the seepage pressure gradient features and the deformation gradient tensor, a seepage pressure gradient tensor matching the seepage-deformation coupling scenario of the dam is generated.

[0105] Step 402: In each triangular subdivision unit of the seepage pressure gradient tensor, perform an alternating solution process for deformation-seepage bidirectional coupling verification, use the dynamic change rate of hydraulic conductivity as the input parameter of the seepage field control equation, calculate the seepage pressure distribution field corresponding to the current sampling point, and generate a deformation coupling residual field based on the residual between the seepage pressure distribution field and the spatial gradient modulus of the deformation gradient tensor.

[0106] In this step, the deformation coupling residual field refers to the field generated by calculating the residual between the seepage pressure distribution field and the spatial gradient modulus of the deformation gradient tensor, which is used to describe the error distribution of the seepage-deformation coupling effect.

[0107] In this embodiment, firstly, within each triangulated unit of the seepage pressure gradient tensor, the dynamic rate of change of hydraulic conductivity is used as the input parameter of the seepage field control equation to calculate the seepage pressure distribution field corresponding to the current sampling point. Secondly, based on the residual between the seepage pressure distribution field and the spatial gradient modulus of the deformation gradient tensor, a deformation-coupled residual field is generated. The residual calculation process is achieved by constraining the angle range between the topological correlation of the seepage path and the direction of permeability anisotropy. Finally, by superimposing the seepage pressure gradient characteristics and the spatial continuity condition of the deformation gradient tensor, a deformation-coupled residual field is generated.

[0108] Step 403: Spatiotemporally correlate and match the time series of the deformation coupling residual field with the multi-scale water level fluctuation component to invert the deformation displacement increment field corresponding to the current seepage pressure distribution field, and calculate the displacement difference of the deformation displacement increment field at the vertex of the triangular subdivision unit to generate the modulus matching degree matrix.

[0109] In this step, the modulus matching degree matrix refers to the matrix generated by calculating the displacement difference at the vertices of the triangularly partitioned element by the deformation displacement increment field, which is used to describe the spatiotemporal matching degree of the seepage-deformation coupling effect.

[0110] In this embodiment, firstly, the time series of the deformation coupling residual field and the multi-scale water level fluctuation component are spatiotemporally correlated and matched to invert the deformation displacement increment field corresponding to the current seepage pressure distribution field; secondly, the displacement difference of the deformation displacement increment field at the vertices of the triangular partitioning unit is calculated to generate the modulus matching degree matrix, wherein the displacement difference calculation process is achieved by constraining the angle range between the topological correlation of the seepage path and the permeability anisotropy direction; finally, the modulus matching degree matrix is ​​generated by superimposing the spatial continuity conditions of the seepage pressure gradient characteristics and the deformation gradient tensor.

[0111] Step 404: Based on the modulus matching degree matrix, construct a spatiotemporal alignment window based on the dynamic time warping algorithm within the triangulation unit, and use the spatiotemporal alignment window to quantify the phase delay error between the dynamic change rate of hydraulic conductivity and the spatial gradient modulus of the deformation gradient tensor during the seepage deformation coupling process.

[0112] In this step, the spatiotemporal alignment window refers to a window constructed by the dynamic time warping algorithm to quantify the phase delay error between the dynamic rate of change of hydraulic conductivity and the spatial gradient magnitude of the deformation gradient tensor. Its value is generated by the magnitude matching degree matrix.

[0113] In this embodiment, firstly, a spatiotemporal alignment window is constructed within the triangularly partitioned unit based on the modulus matching degree matrix, wherein the window size is determined by constraining the angle range between the topological correlation of the seepage path and the direction of permeability anisotropy; secondly, the phase delay error between the dynamic change rate of hydraulic conductivity and the spatial gradient modulus of the deformation gradient tensor is quantified using the spatiotemporal alignment window, wherein the quantization process is achieved by superimposing the spatial continuity conditions of the seepage pressure gradient characteristics and the deformation gradient tensor; finally, the phase delay error is generated by superimposing the spatial continuity conditions of the seepage pressure gradient characteristics and the deformation gradient tensor.

[0114] Step 405: Based on the phase delay error, filter the seepage path branches that satisfy the dam seepage deformation coupling scenario in the seepage pressure gradient tensor, extract the extreme points of the dynamic rate of change of hydraulic conductivity at the vertices of the triangular partitioning unit corresponding to the seepage path branches, and perform topological correlation matching between the extreme points of the dynamic rate of change of hydraulic conductivity and the extreme points of the spatial gradient modulus of the deformation gradient tensor to generate deformation constraint factors.

[0115] In this step, the seepage path branch refers to the seepage path that satisfies the seepage-deformation coupling scenario of the dam, which is selected by the phase delay error. Its value is generated by the seepage pressure gradient tensor. The topology correlation matching refers to the process of associating and matching the extreme points of the dynamic rate of change of hydraulic conductivity with the extreme points of the spatial gradient magnitude of the deformation gradient tensor, which is used to generate the deformation constraint factor.

[0116] In this embodiment, firstly, seepage path branches that satisfy the seepage-deformation coupling scenario of the dam are screened in the seepage pressure gradient tensor based on the phase delay error; secondly, the extreme points of the dynamic rate of change of hydraulic conductivity at the vertices of the triangular subdivision unit corresponding to the seepage path branches are extracted, and they are topologically matched with the extreme points of the spatial gradient modulus of the deformation gradient tensor; finally, the deformation constraint factor is generated by superimposing the seepage pressure gradient features and the spatial continuity conditions of the deformation gradient tensor.

[0117] When dynamically correcting the equivalent parameter values ​​of the hydraulic grid physical mechanism model, existing methods often struggle to effectively integrate the spatiotemporal correlation characteristics of the parameter probability distribution set and the historical seepage field evolution pattern, resulting in insufficient applicability of the corrected parameter values ​​under complex working conditions. Furthermore, traditional methods lack a dynamic feedback mechanism for the distribution of seepage path topological correlations in the seepage field residuals before and after correction, making it difficult to achieve closed-loop optimization of the parameter correction process. Therefore, as another embodiment, according to step 103, based on the parameter probability distribution set and combined with the historical seepage field evolution pattern of the target hydraulic engineering area, the equivalent parameter values ​​of each grid unit in the hydraulic grid physical mechanism model are dynamically corrected to generate hydraulic grid parameter inversion results, including:

[0118] Step 501: Decompose the parameter probability distribution set into a permeability coefficient heterogeneity component in the spatial dimension and a hydraulic conductivity dynamic response component in the time dimension. Based on the topological correlation of the seepage path in the historical seepage field evolution mode of the target water conservancy project area, construct a spatiotemporal coupled feature fusion matrix, and inject the spatiotemporal correlation features of the feature fusion matrix into the grid cell mapping relationship of the water conservancy grid physical mechanism model.

[0119] In this step, the heterogeneity component of the permeability coefficient refers to the component of the parameter probability distribution set in the spatial dimension, which is used to describe the spatial distribution characteristics of the permeability coefficient in heterogeneous media; the dynamic response component of the hydraulic conductivity refers to the component of the parameter probability distribution set in the time dimension, which is used to describe the dynamic change characteristics of the hydraulic conductivity under complex working conditions; the feature fusion matrix is ​​a matrix constructed by fusing the spatiotemporal correlation characteristics of the heterogeneity component of the permeability coefficient and the dynamic response component of the hydraulic conductivity, which is used to characterize the spatiotemporal distribution law of the parameters in heterogeneous media.

[0120] In this embodiment, the parameter probability distribution set is first decomposed into a permeability coefficient heterogeneity component in the spatial dimension and a hydraulic conductivity dynamic response component in the temporal dimension, and their spatial distribution characteristics and temporal variation characteristics are extracted respectively. Second, based on the topological correlation of the seepage path in the historical seepage field evolution mode of the target hydraulic engineering area, a spatiotemporally coupled feature fusion matrix is ​​constructed, where the row vector dimension of the matrix corresponds to the spatial distribution characteristics and the column vector dimension corresponds to the temporal variation characteristics. Finally, the spatiotemporal correlation characteristics of the feature fusion matrix are injected into the grid cell mapping relationship of the hydraulic grid physical mechanism model to ensure that the parameter inversion process can reflect the spatiotemporal correlation characteristics under complex working conditions.

[0121] Step 502: In each grid cell of the hydraulic grid physical mechanism model, a bidirectional attention mechanism is used to dynamically assign weights to the heterogeneity component of the permeability coefficient and the dynamic response component of the hydraulic conductivity, thereby generating dynamic weight coefficients that characterize the spatiotemporal correlation of parameters of heterogeneous media.

[0122] In this step, the bidirectional attention mechanism refers to an algorithm that dynamically assigns weights to the heterogeneity component of the permeability coefficient and the dynamic response component of the hydraulic conductivity by simultaneously considering the correlation between the spatial and temporal dimensions. The dynamic weight coefficient is a parameter generated by the bidirectional attention mechanism to characterize the spatiotemporal correlation of parameters in heterogeneous media, and its value is generated through dynamic weight assignment.

[0123] In this embodiment, firstly, within each grid cell of the hydraulic grid physical mechanism model, a bidirectional attention mechanism is used to dynamically assign weights to the heterogeneity component of the permeability coefficient and the dynamic response component of the hydraulic conductivity. The weight assignment process is achieved by constraining the angle range between the topological correlation of the seepage path and the direction of permeability anisotropy. Secondly, dynamic weight coefficients characterizing the spatiotemporal correlation of the parameters of the heterogeneous medium are generated. The coefficient values ​​are generated by superimposing the spatial continuity conditions of the seepage pressure gradient characteristics and the deformation gradient tensor.

[0124] Step 503: Based on the dynamic weight coefficients and the spatiotemporal correlation features of the feature fusion matrix, an implicit regularization constraint term is introduced into the physical mechanism model of the hydraulic grid. The implicit regularization constraint term is used to dynamically iteratively correct the equivalent parameter values ​​of the grid cells. In each iteration, the residual distribution of the seepage field before and after the correction of the seepage path topology correlation is calculated, and the residual distribution of the seepage field is fed back to the feature fusion matrix to update the confidence weight of the spatiotemporal correlation features, thereby obtaining the updated feature fusion matrix.

[0125] In this step, the implicit regularization constraint term refers to the term introduced by the spatiotemporal correlation features of the dynamic weight coefficient and the feature fusion matrix to constrain the parameter correction process, and its value is generated by dynamic iterative correction; the seepage field residual distribution refers to the distribution generated by calculating the difference in seepage path topological correlation before and after correction, which is used to describe the error distribution of the parameter correction process.

[0126] In this embodiment, firstly, based on the spatiotemporal correlation characteristics of the dynamic weight coefficients and the feature fusion matrix, an implicit regularization constraint term is introduced into the physical mechanism model of the hydraulic grid, where the value of the constraint term is generated through dynamic iterative correction. Secondly, the equivalent parameter values ​​of the grid cells are dynamically iteratively corrected using the implicit regularization constraint term, and the residual distribution of the seepage field before and after correction is calculated in each iteration. Finally, the residual distribution of the seepage field is fed back to the feature fusion matrix to update the confidence weights of the spatiotemporal correlation characteristics, resulting in the updated feature fusion matrix.

[0127] Step 504: Based on the residual distribution of the seepage field and the updated feature fusion matrix, construct a two-way verification window for seepage deformation in the hydraulic grid physical mechanism model. Use the two-way verification window to match the time rate of change of the dynamic response component of hydraulic conductivity with the spatial gradient rate of change of the surface deformation field distribution data to generate a verification weight factor. Use the verification weight factor to screen the equivalent parameter candidate set that satisfies the seepage stability constraint under the alternating operation of reservoir water storage and flood discharge.

[0128] In this step, the seepage deformation bidirectional verification window refers to a window constructed by the seepage field residual distribution and the updated feature fusion matrix to verify the seepage-deformation coupling effect. Its value is generated by matching the spatiotemporal correlation features of the hydraulic conductivity dynamic response component and the surface deformation field distribution data. The verification weight factor refers to a parameter generated by the seepage deformation bidirectional verification window to screen the equivalent parameter candidate set. Its value is generated by matching the time change rate and the spatial gradient change rate.

[0129] In this embodiment, firstly, based on the residual distribution of the seepage field and the updated feature fusion matrix, a two-way verification window for seepage deformation is constructed in the hydraulic grid physical mechanism model. The window size is determined by the range of the angle between the topological correlation of the seepage path and the direction of permeability anisotropy. Secondly, the two-way verification window is used to match the time rate of change of the dynamic response component of hydraulic conductivity with the spatial gradient rate of change of the surface deformation field distribution data to generate a verification weight factor. Finally, the verification weight factor is used to screen the equivalent parameter candidate set that satisfies the seepage stability constraint under the alternating reservoir impoundment-discharge condition.

[0130] Step 505: Based on the equivalent parameter candidate set, the spatial distribution confidence of the heterogeneity component of the permeability coefficient and the temporal correlation confidence of the dynamic response component of the hydraulic conductivity are fused to generate a parameter optimization surface. The boundary smoothing process of the parameter optimization surface is then performed by superimposing the fracture zone permeability abrupt change characteristics in the hydrogeological boundary constraints, and the hydraulic grid parameter inversion results are output.

[0131] In this step, the parameter optimization surface refers to the surface generated by fusing the spatial distribution confidence of the heterogeneity component of the permeability coefficient with the temporal correlation confidence of the dynamic response component of the hydraulic conductivity, which is used to describe the optimized distribution of the parameter inversion results; the boundary smoothing process refers to the process of smoothing the parameter optimization surface by superimposing the permeability abrupt change characteristics of the fault zone in the hydrogeological boundary constraints, which is used to improve the stability of the inversion results.

[0132] In this embodiment, firstly, based on the equivalent parameter candidate set, the spatial distribution confidence of the heterogeneity component of the permeability coefficient and the temporal correlation confidence of the dynamic response component of the hydraulic conductivity are fused to generate a parameter optimization surface; secondly, the boundary smoothing process of the parameter optimization surface is performed by superimposing the permeability abrupt change characteristics of the fault zone in the hydrogeological boundary constraints, wherein the smoothing process is achieved by constraining the angle range between the topological correlation of the seepage path and the permeability anisotropy direction; finally, the hydraulic grid parameter inversion results are output.

[0133] When constructing a two-way verification window for seepage deformation, existing methods often struggle to effectively integrate the spatiotemporal correlation features of the seepage field residual distribution and the feature fusion matrix. This results in a lack of precise quantification of the spatiotemporal matching degree between the dynamic response component of hydraulic conductivity and the surface deformation field distribution data during the generation of verification weight factors. Furthermore, traditional methods lack explicit constraints on the phase synchronization index during the construction of the dynamic residual propagation network, making it difficult to ensure the stability and reliability of the parameter selection process. Therefore, as another embodiment, according to step 504, based on the seepage field residual distribution and the updated feature fusion matrix, a two-way verification window for seepage deformation is constructed in the hydraulic grid physical mechanism model. The two-way verification window is used to match the temporal rate of change of the dynamic response component of hydraulic conductivity with the spatial gradient rate of change of the surface deformation field distribution data to generate verification weight factors. These verification weight factors are then used to select a set of equivalent parameter candidates that satisfy the seepage stability constraints under alternating reservoir storage and flood discharge conditions, including:

[0134] Step 601: Based on the residual distribution of the seepage field and the updated feature fusion matrix, extract the multi-scale seepage pressure residual components in the time dimension and the deformation coupling residual gradient in the spatial dimension, and fuse the time decay coefficient of the multi-scale seepage pressure residual components with the spatial diffusion coefficient of the deformation coupling residual gradient to generate a spatiotemporal dynamic residual matrix.

[0135] In this step, the spatiotemporal dynamic residual matrix refers to the matrix generated by fusing the time decay coefficient of the multi-scale seepage pressure residual components with the spatial diffusion coefficient of the deformation coupling residual gradient, which is used to describe the spatiotemporal distribution law of the seepage-deformation coupling effect.

[0136] In this embodiment, firstly, based on the residual distribution of the seepage field and the updated feature fusion matrix, multi-scale seepage pressure residual components in the time dimension are extracted, including low-frequency trend components, mid-frequency fluctuation components, and high-frequency transient components, and their time decay coefficients are extracted respectively; secondly, deformation coupling residual gradients in the spatial dimension are extracted, including longitudinal gradient components and transverse gradient components, and their spatial diffusion coefficients are extracted respectively; finally, by fusing the time decay coefficients and spatial diffusion coefficients, a spatiotemporal dynamic residual matrix is ​​generated, where the row vector dimension of the matrix corresponds to the time decay sequence, and the column vector dimension corresponds to the spatial diffusion sequence.

[0137] Step 602: Map the row vectors of the spatiotemporal dynamic residual matrix to the time axis of the grid cell of the hydraulic grid physical mechanism model, and map the column vectors to the spatial topology axis. Use the dynamic time warping algorithm to align the time rate of change sequence of the hydraulic conductivity dynamic response component with the spatial gradient rate of change sequence of the surface deformation field distribution data.

[0138] In this step, the dynamic time warping algorithm refers to an algorithm that generates a spatiotemporal alignment path by aligning the time rate of change sequence and the spatial gradient rate of change sequence, which is used to quantify the spatiotemporal matching degree between the dynamic response component of hydraulic conductivity and the distribution data of surface deformation field.

[0139] In this embodiment, firstly, the row vectors of the spatiotemporal dynamic residual matrix are mapped to the time axis of the grid cells in the hydraulic grid physical mechanism model, and the column vectors are mapped to the spatial topology axis to generate an initial alignment path. Secondly, the dynamic time warping algorithm is used to align the time rate of change sequence of the hydraulic conductivity dynamic response component with the spatial gradient rate of change sequence of the surface deformation field distribution data to generate a spatiotemporal alignment path. The alignment process is achieved by constraining the angle range between the topological correlation of the seepage path and the direction of permeability anisotropy. Finally, the spatiotemporal alignment path is generated by superimposing the spatial continuity conditions of the seepage pressure gradient characteristics and the deformation gradient tensor.

[0140] Step 603: Based on the spatiotemporal alignment path, a dynamic residual propagation network is constructed within the bidirectional verification window of the seepage deformation. The dynamic residual propagation network is used to couple the row and column correlation of the spatiotemporal dynamic residual matrix with the confidence weight of the updated feature fusion matrix to generate a phase synchronization index.

[0141] In this step, the dynamic residual propagation network refers to the network constructed through the spatiotemporal alignment path for propagating residual information. Its value is generated by coupling the row and column correlation of the spatiotemporal dynamic residual matrix with the confidence weight of the feature fusion matrix. The phase synchronization index is a parameter generated by the dynamic residual propagation network for quantifying the phase synchronization between the dynamic response component of hydraulic conductivity and the distribution data of the surface deformation field.

[0142] In this embodiment, firstly, a dynamic residual propagation network is constructed within the bidirectional verification window of seepage deformation based on the spatiotemporal alignment path. The network nodes are determined by constraining the angle range between the topological correlation of the seepage path and the direction of permeability anisotropy. Secondly, the phase synchronization index is generated by coupling the row and column correlation of the spatiotemporal dynamic residual matrix with the confidence weight of the updated feature fusion matrix using the dynamic residual propagation network. The index value is generated by superimposing the spatial continuity conditions of the seepage pressure gradient features and the deformation gradient tensor.

[0143] Step 604: Perform iterative screening for seepage deformation bidirectional coupling verification in the dynamic residual propagation network. Map the extreme points of the time rate of change of the hydraulic conductivity dynamic response component to the time axis nodes of the dynamic residual propagation network to generate a time dimension residual vector. Map the extreme points of the spatial gradient rate of change of the surface deformation field distribution data to the spatial topology axis edges of the dynamic residual propagation network to generate a spatial dimension residual tensor. Calculate the projection matching degree between the time dimension residual vector and the spatial dimension residual tensor on the spatiotemporal alignment path to generate a dynamic verification score.

[0144] In this step, the dynamic verification score refers to the score generated by calculating the projection matching degree of the time dimension residual vector and the spatial dimension residual tensor on the spatiotemporal alignment path, which is used to describe the verification results of the seepage-deformation coupling effect.

[0145] In this embodiment, firstly, an iterative screening for bidirectional coupling verification of seepage deformation is performed in the dynamic residual propagation network, mapping the extreme points of the time rate of change of the hydraulic conductivity dynamic response component to time axis nodes to generate a time-dimensional residual vector; secondly, the extreme points of the spatial gradient rate of change of the surface deformation field distribution data are mapped to spatial topological axis edges to generate a spatial-dimensional residual tensor; finally, the projection matching degree between the time-dimensional residual vector and the spatial-dimensional residual tensor on the spatiotemporal alignment path is calculated to generate a dynamic verification score.

[0146] Step 605: Based on the dynamic verification score and the phase synchronization index, construct a seepage path stability evaluation map within the seepage deformation bidirectional verification window, and use the seepage path stability evaluation map to superimpose the row and column correlation of the spatiotemporal dynamic residual matrix and the confidence weight of the updated feature fusion matrix to generate a verification weight factor.

[0147] In this step, the seepage path stability assessment map refers to a map constructed by dynamic verification score and phase synchronization index to assess the stability of seepage path. Its value is generated by superimposing the row and column correlation of the spatiotemporal dynamic residual matrix and the confidence weight of the feature fusion matrix. The verification weight factor refers to a parameter generated by the seepage path stability assessment map to screen the equivalent parameter candidate set. Its value is generated by superimposing the spatial continuity condition of seepage pressure gradient features and deformation gradient tensor.

[0148] In this embodiment, firstly, a seepage path stability evaluation map is constructed within the seepage deformation bidirectional verification window based on the dynamic verification score and the phase synchronization index. The nodes of the evaluation map are determined by constraining the angle range between the topological correlation of the seepage path and the direction of permeability anisotropy. Secondly, the verification weight factor is generated by superimposing the row and column correlation of the spatiotemporal dynamic residual matrix and the confidence weight of the updated feature fusion matrix on the seepage path stability evaluation map.

[0149] Existing methods often struggle to effectively fuse the spatiotemporal correlation features of multi-scale seepage pressure residual components and deformation-coupled residual gradients when generating spatiotemporal dynamic residual matrices, particularly lacking collaborative modeling of low-frequency trend components, mid-frequency fluctuation components, and high-frequency transient response components. Furthermore, traditional methods lack explicit separation of the fundamental coupling components and periodic modulation components during the construction of the spatiotemporal coupling kernel matrix, resulting in unclear physical meaning of the matrix generation process. Therefore, as another embodiment, according to step 601, based on the seepage field residual distribution and the updated feature fusion matrix, multi-scale seepage pressure residual components in the time dimension and deformation-coupled residual gradients in the spatial dimension are extracted. The time decay coefficient of the multi-scale seepage pressure residual components and the spatial diffusion coefficient of the deformation-coupled residual gradients are then fused to generate the spatiotemporal dynamic residual matrix, including:

[0150] Step 701: Input the residual distribution of the seepage field into a multi-scale decomposition network to extract the low-frequency seepage pressure trend component, the medium-frequency periodic fluctuation component, and the high-frequency transient response component in the time dimension.

[0151] In this step, the multi-scale decomposition network refers to a network that decomposes the residual distribution of the seepage field into low-frequency trend components, mid-frequency fluctuation components, and high-frequency transient components through a multi-scale decomposition algorithm, in order to capture the seepage pressure variation characteristics at different time scales.

[0152] In this embodiment, the residual distribution of the seepage field is first input into a multi-scale decomposition network, and the wavelet transform algorithm is used to decompose it into low-frequency trend components, mid-frequency fluctuation components, and high-frequency transient components. Second, the time decay coefficient of the low-frequency trend component, the phase lag of the mid-frequency fluctuation component, and the energy decay rate of the high-frequency transient component are extracted to characterize the long-term trend, periodic fluctuation, and transient response of the seepage pressure, respectively. Finally, the multi-scale seepage pressure residual components are generated by superimposing the spatial continuity conditions of the seepage pressure gradient characteristics and the deformation gradient tensor.

[0153] Step 702: Perform spatial topological decomposition on the deformation coupling residual gradient to extract the longitudinal gradient component along the seepage dominant path direction and the transverse gradient component perpendicular to the seepage direction in the updated feature fusion matrix.

[0154] In this step, spatial topology decomposition refers to the process of decomposing the deformation coupling residual gradient into longitudinal gradient components along the dominant seepage path and transverse gradient components perpendicular to the seepage direction using a spatial decomposition algorithm, in order to capture the spatial distribution characteristics of the deformation field.

[0155] In this embodiment, firstly, spatial topological decomposition is performed on the deformation-coupled residual gradient to extract the longitudinal gradient component along the dominant seepage path and the transverse gradient component perpendicular to the seepage direction; secondly, based on the orientation of the dominant seepage path in the updated feature fusion matrix, the spatial distribution characteristics of the longitudinal and transverse gradient components are determined; finally, by superimposing the spatial continuity conditions of the seepage pressure gradient characteristics and the deformation gradient tensor, the spatial decomposition result of the deformation-coupled residual gradient is generated.

[0156] Step 703: Based on the time decay coefficient of the low-frequency seepage pressure trend component, the phase lag of the mid-frequency periodic fluctuation component, and the energy decay rate of the high-frequency transient response component, construct the time-dimensional residual energy spectrum, and based on the spatial diffusion coefficient of the longitudinal gradient component and the anisotropy ratio of the transverse gradient component, construct the spatial-dimensional residual diffusion tensor.

[0157] In this step, the time-dimensional residual energy spectrum refers to the spectrum constructed by the time decay coefficient of the low-frequency trend component, the phase lag of the mid-frequency fluctuation component, and the energy decay rate of the high-frequency transient component, which is used to describe the energy distribution of seepage pressure in the time dimension; the spatial-dimensional residual diffusion tensor refers to the tensor constructed by the spatial diffusion coefficient of the longitudinal gradient component and the anisotropy ratio of the transverse gradient component, which is used to describe the diffusion characteristics of the deformation field in the spatial dimension.

[0158] In this embodiment, firstly, a time-dimensional residual energy spectrum is constructed based on the time decay coefficient of the low-frequency trend component, the phase lag of the mid-frequency fluctuation component, and the energy decay rate of the high-frequency transient component; secondly, a spatial-dimensional residual diffusion tensor is constructed based on the spatial diffusion coefficient of the longitudinal gradient component and the anisotropy ratio of the transverse gradient component; finally, the time-dimensional residual energy spectrum and the spatial-dimensional residual diffusion tensor are generated by superimposing the spatial continuity conditions of the seepage pressure gradient characteristics and the deformation gradient tensor.

[0159] Step 704: Spatiotemporally couple the time-dimensional residual energy spectrum with the spatial-dimensional residual diffusion tensor; perform tensor product operation on the time decay coefficient of the low-frequency seepage pressure trend component and the spatial diffusion coefficient of the longitudinal gradient component to generate a basic coupling component; perform Hadamard product operation on the phase lag of the mid-frequency periodic fluctuation component and the anisotropy ratio of the transverse gradient component to generate a periodic modulation component; generate a spatiotemporal coupling kernel matrix by superimposing the basic coupling component and the periodic modulation component and injecting the energy decay rate of the high-frequency transient response component as a regularization constraint term.

[0160] In this step, the spatiotemporal coupling kernel matrix refers to the matrix generated by fusing the residual energy spectrum of the time dimension and the residual diffusion tensor of the spatial dimension, which is used to describe the spatiotemporal distribution law of the seepage-deformation coupling effect; the basic coupling component refers to the coupling component generated by the time decay coefficient of the low-frequency trend component and the spatial diffusion coefficient of the longitudinal gradient component; the periodic modulation component refers to the modulation component generated by the phase lag of the mid-frequency fluctuation component and the anisotropy ratio of the transverse gradient component.

[0161] In this embodiment, firstly, the time-dimensional residual energy spectrum and the spatial-dimensional residual diffusion tensor are spatiotemporally coupled. The time decay coefficient of the low-frequency trend component and the spatial diffusion coefficient of the longitudinal gradient component are tensor-producted to generate the basic coupling component. Secondly, the phase lag of the mid-frequency fluctuation component and the anisotropy ratio of the transverse gradient component are Hadamard-producted to generate the periodic modulation component. Finally, the spatiotemporal coupling kernel matrix is ​​generated by superimposing the basic coupling component and the periodic modulation component and injecting the energy decay rate of the high-frequency transient component as a regularization constraint.

[0162] Figure 2 This application provides a schematic diagram of the structure of an artificial intelligence-based hydraulic grid parameter inversion system, as shown in the embodiment. Figure 2 As shown, the system includes:

[0163] The acquisition module 21 is used to acquire multi-source monitoring data of the target water conservancy project area. The multi-source monitoring data includes groundwater level dynamic sequence, surface deformation field distribution data and hydrogeological boundary constraints. The hydrogeological boundary constraints are generated by coupling the actual operating conditions and geological structure characteristics of the target water conservancy project.

[0164] Optimization module 22 is used to establish an artificial intelligence probabilistic inference framework for parameter inversion of heterogeneous media based on the multi-source monitoring data and the preset hydraulic grid physical mechanism model. The artificial intelligence probabilistic inference framework is used to iteratively optimize the implicit correlation between the spatial distribution of permeability and the dynamic change of hydraulic conductivity in the hydraulic grid physical mechanism model, and generate a parameter probability distribution set that matches the geological structure uncertainty of the target hydraulic engineering area.

[0165] The correction module 23 is used to dynamically correct the equivalent parameter values ​​of each grid unit in the hydraulic grid physical mechanism model based on the parameter probability distribution set and the historical seepage field evolution mode of the target hydraulic engineering area, and generate hydraulic grid parameter inversion results.

[0166] Figure 2 The aforementioned AI-based hydraulic grid parameter inversion system can perform... Figure 1The implementation principle and technical effects of the AI-based hydraulic grid parameter inversion method described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit of the AI-based hydraulic grid parameter inversion system in the above embodiments are executed have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0167] In one possible design, Figure 2 The illustrated embodiment of an AI-based hydraulic grid parameter inversion system can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0168] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0169] The processing component 32 is used to acquire multi-source monitoring data of the target water conservancy project area. The multi-source monitoring data includes groundwater level dynamic sequences, surface deformation field distribution data, and hydrogeological boundary constraints. These hydrogeological boundary constraints are generated by coupling the actual operating conditions and geological structural characteristics of the target water conservancy project. Based on the multi-source monitoring data and a preset water conservancy grid physical mechanism model, an artificial intelligence probabilistic inference framework for parameter inversion in heterogeneous media is established. This framework iteratively optimizes the implicit correlation between the spatial distribution of permeability and the dynamic change of hydraulic conductivity in the water conservancy grid physical mechanism model, generating a parameter probability distribution set that matches the geological structural uncertainty of the target water conservancy project area. Based on the parameter probability distribution set and combined with the historical seepage field evolution pattern of the target water conservancy project area, the equivalent parameter values ​​of each grid unit in the water conservancy grid physical mechanism model are dynamically corrected, generating water conservancy grid parameter inversion results.

[0170] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0171] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0172] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0173] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0174] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0175] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0176] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is an artificial intelligence-based method for inverting hydraulic grid parameters.

[0177] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0178] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0179] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0180] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for inverting hydraulic grid parameters based on artificial intelligence, characterized in that, include: Acquire multi-source monitoring data of the target water conservancy project area, wherein the multi-source monitoring data includes groundwater level dynamic sequence, surface deformation field distribution data and hydrogeological boundary constraints, and the hydrogeological boundary constraints are generated by coupling the actual operating conditions and geological structure characteristics of the target water conservancy project; Based on the multi-source monitoring data and the preset hydraulic grid physical mechanism model, an artificial intelligence probabilistic inference framework for parameter inversion of heterogeneous media is established. The implicit correlation between the spatial distribution of permeability coefficient and the dynamic change of hydraulic conductivity in the hydraulic grid physical mechanism model is iteratively optimized using the artificial intelligence probabilistic inference framework to generate a parameter probability distribution set that matches the geological structural uncertainty of the target hydraulic engineering area. Based on the parameter probability distribution set and combined with the historical seepage field evolution mode of the target water conservancy project area, the equivalent parameter values ​​of each grid unit in the water conservancy grid physical mechanism model are dynamically corrected to generate water conservancy grid parameter inversion results. Specifically, based on the multi-source monitoring data and a preset hydraulic grid physical mechanism model, an artificial intelligence probabilistic inference framework for parameter inversion in heterogeneous media is established. This framework iteratively optimizes the implicit correlation between the spatial distribution of permeability and the dynamic change of hydraulic conductivity in the hydraulic grid physical mechanism model, generating a parameter probability distribution set that matches the geological structural uncertainty of the target hydraulic engineering area. This includes: The groundwater level dynamic sequence is decomposed into multi-scale water level fluctuation components in the time dimension, and the surface deformation field distribution data is converted into a deformation gradient tensor in the spatial dimension. Based on the spatiotemporal coupling characteristics of the multi-scale water level fluctuation components and the deformation gradient tensor, a three-dimensional dynamic coupling feature matrix is ​​constructed. The initial field of the spatial distribution of the permeability coefficient in the physical mechanism model of the hydraulic grid is bidirectionally embedded with the three-dimensional dynamic coupling feature matrix to generate dynamic weight coefficients. Based on the dynamic weight coefficient, an implicit gradient field is introduced into the physical mechanism model of the hydraulic grid, and the implicit gradient field is used to perform probabilistic sparse coding on the correlation between the spatial distribution of the permeability coefficient and the dynamic change of the hydraulic conductivity, thereby generating a set of implicit variables with geological structural uncertainty. Based on the set of latent variables, Bayesian Monte Carlo joint sampling is performed within the grid cells of the hydraulic grid physical mechanism model. By simultaneously constraining the periodic boundary conditions of the multi-scale water level fluctuation components and the spatial continuity conditions of the deformation gradient tensor, the joint probability distribution of permeability coefficient and hydraulic conductivity that matches the alternating reservoir water storage and flood discharge conditions and the coupled scenario of dam seepage deformation in the target hydraulic engineering area is iteratively screened, and a set of parameter probability distributions is generated.

2. The method according to claim 1, characterized in that, Based on the set of latent variables, Bayesian Monte Carlo joint sampling is performed within the grid cells of the hydraulic grid physical mechanism model. By simultaneously constraining the periodic boundary conditions of the multi-scale water level fluctuation components and the spatial continuity conditions of the deformation gradient tensor, a joint probability distribution of permeability coefficient and hydraulic conductivity matching the alternating reservoir storage and flood discharge conditions and the coupled scenario of dam seepage and deformation in the target hydraulic engineering area is iteratively screened, and a set of parameter probability distributions is generated, including: Based on the prior spatial distribution of the permeability coefficient in the set of latent variables, and combined with the historical seepage pressure gradient data in the alternating operation of reservoir water storage and flood discharge, a dynamic covariance matrix is ​​constructed. The joint probability density field of permeability coefficient and hydraulic conductivity is initialized within the grid cell of the hydraulic grid physical mechanism model, and the joint probability density field of permeability coefficient and hydraulic conductivity is subjected to adaptive covariance adjustment Monte Carlo sampling using the dynamic covariance matrix. During each sampling iteration, the dynamic residual vector between the periodic boundary conditions of the multi-scale water level fluctuation component and the dynamic change of hydraulic conductivity at the current sampling point is calculated synchronously. During the convergence phase of the Monte Carlo sampling, a set of sampling points is extracted that satisfy the seepage path topology constraints of the joint probability density field of permeability coefficient and hydraulic conductivity in the dam seepage deformation coupling scenario. The seepage path topology constraints are used to verify the bidirectional coupling of deformation and seepage on the set of sampling points, and the matching degree between the spatial gradient modulus of the deformation gradient tensor and the dynamic change rate of the hydraulic conductivity is calculated to generate the deformation constraint factor. Based on the deformation constraint factor, the abnormal sampling points in the sampling point set that deviate from the historical seepage field evolution pattern under the alternating operation of reservoir water storage and flood discharge are probabilistically reweighted, and the spatiotemporal distribution characteristics of the dynamic residual vector and the spatial correlation of the deformation constraint factor are integrated to generate a parameter probability distribution set.

3. The method according to claim 2, characterized in that, During the convergence phase of the Monte Carlo sampling, a set of sampling points is extracted where the joint probability density field of permeability and hydraulic conductivity satisfies the seepage path topology constraints of the dam seepage deformation coupling scenario. The seepage path topology constraints are used to verify the bidirectional coupling of deformation and seepage on the sampling point set. Furthermore, the matching degree between the spatial gradient magnitude of the deformation gradient tensor and the dynamic change rate of the hydraulic conductivity is calculated to generate a deformation constraint factor, including: Based on the seepage path topology constraints, the dynamic rate of change of hydraulic conductivity of each sampling point in the sampling point set is mapped to the grid cell boundary of the hydraulic grid physical mechanism model, and the seepage pressure gradient tensor that matches the seepage deformation coupling scenario of the dam is constructed using the triangulation algorithm. Within each triangular subdivision unit of the seepage pressure gradient tensor, an alternating solution process for deformation-seepage bidirectional coupling verification is performed. The dynamic change rate of the hydraulic conductivity is used as the input parameter of the seepage field control equation. The seepage pressure distribution field corresponding to the current sampling point is calculated, and a deformation coupling residual field is generated based on the residual between the seepage pressure distribution field and the spatial gradient modulus of the deformation gradient tensor. The deformation coupling residual field is spatiotemporally correlated and matched with the time series of the multi-scale water level fluctuation component to invert the deformation displacement increment field corresponding to the current seepage pressure distribution field, and the displacement difference of the deformation displacement increment field at the vertex of the triangular subdivision unit is calculated to generate the modulus matching degree matrix. Based on the modulus matching degree matrix, a spatiotemporal alignment window based on the dynamic time warping algorithm is constructed within the triangulation unit, and the spatiotemporal alignment window is used to quantify the phase delay error between the dynamic change rate of hydraulic conductivity and the spatial gradient modulus of the deformation gradient tensor in the seepage deformation coupling process. Based on the phase delay error, seepage path branches that satisfy the dam seepage deformation coupling scenario are selected from the seepage pressure gradient tensor. The extreme points of the dynamic rate of change of hydraulic conductivity at the vertices of the triangulation unit corresponding to the seepage path branches are extracted. The extreme points of the dynamic rate of change of hydraulic conductivity are then matched topologically with the extreme points of the spatial gradient magnitude of the deformation gradient tensor to generate deformation constraint factors.

4. The method according to claim 1, characterized in that, Based on the parameter probability distribution set and combined with the historical seepage field evolution pattern of the target water conservancy project area, the equivalent parameter values ​​of each grid cell in the water conservancy grid physical mechanism model are dynamically corrected to generate water conservancy grid parameter inversion results, including: The parameter probability distribution set is decomposed into a permeability coefficient heterogeneity component in the spatial dimension and a hydraulic conductivity dynamic response component in the temporal dimension. Based on the topological correlation of the seepage path in the historical seepage field evolution mode of the target water conservancy project area, a spatiotemporal coupled feature fusion matrix is ​​constructed, and the spatiotemporal correlation features of the feature fusion matrix are injected into the grid cell mapping relationship of the water conservancy grid physical mechanism model. Within each grid cell of the hydraulic grid physical mechanism model, a bidirectional attention mechanism is used to dynamically assign weights to the heterogeneity component of the permeability coefficient and the dynamic response component of the hydraulic conductivity, thereby generating dynamic weight coefficients that characterize the spatiotemporal correlation of parameters of heterogeneous media. Based on the dynamic weight coefficients and the spatiotemporal correlation features of the feature fusion matrix, an implicit regularization constraint term is introduced into the physical mechanism model of the hydraulic grid. The implicit regularization constraint term is used to dynamically iteratively correct the equivalent parameter values ​​of the grid cells. In each iteration, the residual distribution of the seepage field before and after the correction of the seepage path topological correlation is calculated, and the residual distribution of the seepage field is fed back to the feature fusion matrix to update the confidence weight of the spatiotemporal correlation features, thereby obtaining the updated feature fusion matrix. Based on the residual distribution of the seepage field and the updated feature fusion matrix, a two-way verification window for seepage deformation is constructed in the physical mechanism model of the hydraulic grid. The two-way verification window is used to match the time rate of change of the dynamic response component of the hydraulic conductivity with the spatial gradient rate of change of the surface deformation field distribution data to generate a verification weight factor. The verification weight factor is then used to screen the equivalent parameter candidate set that satisfies the seepage stability constraint under the alternating operation of reservoir water storage and flood discharge. Based on the equivalent parameter candidate set, the spatial distribution confidence of the heterogeneity component of the permeability coefficient and the temporal correlation confidence of the dynamic response component of the hydraulic conductivity are fused to generate a parameter optimization surface. The boundary smoothing process of the parameter optimization surface is then performed by superimposing the permeability mutation characteristics of the fault zone in the hydrogeological boundary constraints, and the hydraulic grid parameter inversion results are output.

5. The method according to claim 4, characterized in that, Based on the residual distribution of the seepage field and the updated feature fusion matrix, a two-way verification window for seepage deformation is constructed in the hydraulic grid physical mechanism model. This two-way verification window is used to match the time rate of change of the dynamic response component of the hydraulic conductivity with the spatial gradient rate of change of the surface deformation field distribution data, generating verification weight factors. These verification weight factors are then used to screen a candidate set of equivalent parameters that satisfy the seepage stability constraints under alternating reservoir storage and flood discharge conditions, including: Based on the residual distribution of the seepage field and the updated feature fusion matrix, the multi-scale seepage pressure residual components in the time dimension and the deformation coupling residual gradient in the spatial dimension are extracted, and the time decay coefficient of the multi-scale seepage pressure residual components and the spatial diffusion coefficient of the deformation coupling residual gradient are fused to generate a spatiotemporal dynamic residual matrix. The row vectors of the spatiotemporal dynamic residual matrix are mapped to the time axis of the grid cell of the hydraulic grid physical mechanism model, and the column vectors are mapped to the spatial topology axis. The dynamic time warping algorithm is used to align the time rate of change sequence of the hydraulic conductivity dynamic response component with the spatial gradient rate of change sequence of the surface deformation field distribution data. Based on the spatiotemporal alignment path, a dynamic residual propagation network is constructed within the bidirectional verification window of the seepage deformation. The dynamic residual propagation network is used to couple the row and column correlation of the spatiotemporal dynamic residual matrix with the confidence weight of the updated feature fusion matrix to generate a phase synchronization index. In the dynamic residual propagation network, an iterative screening of seepage deformation bidirectional coupling verification is performed. The extreme points of the time rate of change of the hydraulic conductivity dynamic response component are mapped to the time axis nodes of the dynamic residual propagation network to generate a time dimension residual vector. The extreme points of the spatial gradient rate of change of the surface deformation field distribution data are mapped to the spatial topology axis edge of the dynamic residual propagation network to generate a spatial dimension residual tensor. The projection matching degree of the time dimension residual vector and the spatial dimension residual tensor on the spatiotemporal alignment path is calculated to generate a dynamic verification score. Based on the dynamic verification score and the phase synchronization index, a seepage path stability assessment map is constructed within the seepage deformation bidirectional verification window. The verification weight factor is generated by superimposing the row and column correlation of the spatiotemporal dynamic residual matrix and the confidence weight of the updated feature fusion matrix on the seepage path stability assessment map.

6. The method according to claim 5, characterized in that, Based on the seepage field residual distribution and the updated feature fusion matrix, multi-scale seepage pressure residual components in the time dimension and deformation coupling residual gradients in the spatial dimension are extracted. The time decay coefficients of the multi-scale seepage pressure residual components and the spatial diffusion coefficients of the deformation coupling residual gradients are then fused to generate a spatiotemporal dynamic residual matrix, including: The residual distribution of the seepage field is input into a multi-scale decomposition network to extract the low-frequency seepage pressure trend component, the medium-frequency periodic fluctuation component, and the high-frequency transient response component in the time dimension. Spatial topological decomposition is performed on the deformation coupling residual gradient to extract the longitudinal gradient component and the transverse gradient component perpendicular to the seepage direction along the seepage dominant path in the updated feature fusion matrix. Based on the time decay coefficient of the low-frequency seepage pressure trend component, the phase lag of the mid-frequency periodic fluctuation component, and the energy decay rate of the high-frequency transient response component, a time-dimensional residual energy spectrum is constructed, and based on the spatial diffusion coefficient of the longitudinal gradient component and the anisotropy ratio of the transverse gradient component, a spatial-dimensional residual diffusion tensor is constructed. Spatiotemporal coupling is performed between the time-dimensional residual energy spectrum and the spatial-dimensional residual diffusion tensor. The time decay coefficient of the low-frequency seepage pressure trend component and the spatial diffusion coefficient of the longitudinal gradient component are multiplied by a tensor to generate a basic coupling component. The phase lag of the mid-frequency periodic fluctuation component and the anisotropy ratio of the transverse gradient component are multiplied by a Hadamard product to generate a periodic modulation component. By superimposing the basic coupling component and the periodic modulation component, and injecting the energy decay rate of the high-frequency transient response component as a regularization constraint term, a spatiotemporal coupling kernel matrix is ​​generated.

7. An artificial intelligence-based hydraulic grid parameter inversion system, applied to the artificial intelligence-based hydraulic grid parameter inversion method according to any one of claims 1-6, characterized in that, include: The acquisition module is used to acquire multi-source monitoring data of the target water conservancy project area. The multi-source monitoring data includes groundwater level dynamic sequence, surface deformation field distribution data and hydrogeological boundary constraints. The hydrogeological boundary constraints are generated by coupling the actual operating conditions and geological structure characteristics of the target water conservancy project. The optimization module is used to establish an artificial intelligence probabilistic inference framework for parameter inversion of heterogeneous media based on the multi-source monitoring data and the preset hydraulic grid physical mechanism model. The artificial intelligence probabilistic inference framework is used to iteratively optimize the implicit correlation between the spatial distribution of permeability and the dynamic change of hydraulic conductivity in the hydraulic grid physical mechanism model, and generate a parameter probability distribution set that matches the geological structure uncertainty of the target hydraulic engineering area. The correction module is used to dynamically correct the equivalent parameter values ​​of each grid unit in the physical mechanism model of the hydraulic grid based on the parameter probability distribution set and the historical seepage field evolution mode of the target hydraulic engineering area, and generate hydraulic grid parameter inversion results.

8. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the artificial intelligence-based hydraulic grid parameter inversion method as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements an artificial intelligence-based method for inverting hydraulic grid parameters as described in any one of claims 1 to 6.

Images