Partitioned rapid inversion method for global structural mechanical parameters of concrete arch dam

Abstract

The invention discloses a concrete arch dam global structural mechanical parameter zoning rapid inversion method, and relates to the technical field of dam operation safety monitoring and management, and the method comprises the steps: building a dam body and foundation three-dimensional finite element model through finite element software according to engineering design and monitoring data, and building a foundation three-dimensional finite element model according to damming material mechanical parameter information; partitioning the three-dimensional finite element model, analyzing the sensitivity of mechanical parameters of each region, determining sensitive mechanical parameters influencing the deformation of the concrete arch dam, and carrying out self-adaptive intelligent sampling on the sensitive mechanical parameters according to a sensitivity analysis result, and constructing a deep learning agent model reflecting a nonlinear relationship between the sensitive mechanical parameters of the dam and the deformation of each monitoring point, and carrying out deep learning inversion on the elastic modulus of the dam body and the deformation modulus of the bedrock. According to the invention, the method can efficiently and accurately invert and determine the structural mechanical parameters of the arch dam in the actual operation period, and provides a good basis for the safety analysis of the dam.

Description

Technical Field

[0001] This invention relates to the field of dam operation safety monitoring and management, and in particular to a method for rapid inversion of the structural mechanical parameters of a concrete arch dam by region. Background Technology

[0002] Health monitoring of large hydraulic structures is receiving increasing attention. Concrete arch dams, as one of the main structural components of hydraulic structures, play a crucial role in water conservancy projects and are widely used in reservoirs, hydropower stations, and flood control projects worldwide. Accidents or collapses in these dams can cause loss of life and property downstream. With increasing service life and the development of inherent quality defects, the actual material parameters of some in-service concrete arch dams will inevitably deviate from the initial design values ​​due to material aging, accumulated damage, and external environmental factors. Therefore, obtaining accurate physical and material parameters is essential for ensuring dam operation and is a key foundation for dam safety risk assessment.

[0003] Currently, the main methods for obtaining dam material and physical parameters are field sampling tests and inversion methods based on monitoring data. Field sampling tests can only partially reveal the dam's material parameters and are affected by factors such as random sampling, sample interference, and damage to the original structure, resulting in high costs and poor representativeness. The idea of ​​using prototype monitoring data to feed back and optimize key mechanical models and parameters has become a consensus in the dam engineering community because prototype monitoring data contains the actual behavioral patterns of the structure over time. Static monitoring for dam health diagnosis and obtaining accurate material parameters has become a hot topic in hydraulic engineering. However, most scholars, in the process of inverting dam material parameters and calculating safety monitoring, often subjectively determine the parameters to be studied. This means that in certain structural engineering projects or structural zones with complex nonlinear relationships, more important material and physical parameters or parameter sets with complex coupling relationships will be ignored, leading to a decrease in the reliability of health monitoring.

[0004] In the process of inverting dam structural parameters, there is a close coupling between two important components: experimental design and machine learning model. A well-designed experiment not only affects the spatial distribution and signal-to-noise ratio of the observed data but also directly determines the stability of the inversion problem. Meanwhile, selecting an appropriate machine learning model plays a crucial role in extracting effective features from complex input-output relationships and establishing a high-precision inversion mapping.

[0005] Therefore, there is an urgent need for a rapid inversion method for the structural mechanical parameters of concrete arch dams across different regions. Summary of the Invention

[0006] To address the aforementioned issues, this application proposes a rapid inversion method for the regional structural mechanical parameters of a concrete arch dam. Based on engineering design and monitoring data, a three-dimensional finite element model of the dam body and foundation is established using finite element software. According to the mechanical parameters of the dam construction materials, the three-dimensional finite element model is divided into regions. The Morris method is used to analyze the sensitivity of mechanical parameters in each region, determining the sensitive mechanical parameters affecting the deformation of the concrete arch dam. Based on these results, adaptive intelligent sampling is performed on the sensitive mechanical parameters. A Bayesian neural network deep learning surrogate model incorporating an attention mechanism is constructed. The optimal combination of elastic modes for each region is inverted using the internal optimizer of the neural network. This method provides a novel approach for the inversion of the regional elastic modes of the dam and foundation during operation, including the following steps: S1. Based on the engineering design and monitoring data, a three-dimensional finite element model of the dam and foundation is established using finite element software, and the three-dimensional finite element model is divided into zones based on the mechanical parameter information of the dam construction materials. S2. Based on the mechanical parameter information of the concrete dam, the Morris method is used to conduct sensitivity analysis on the mechanical parameters of each region in the three-dimensional finite element model to determine the deformation sensitivity mechanical parameters of the concrete arch dam. S3. Based on the deformation-sensitive mechanical parameters of the sensitive concrete arch dam, construct a multi-input-multi-output adaptive intelligent sampling method for the dam's deformation-sensitive mechanical parameters and the deformation at each monitoring point; S4. Construct a deep learning proxy model that reflects the nonlinear relationship between the dam's sensitivity mechanical parameters and the deformation at each monitoring point, and perform deep learning inversion of the dam's elastic modulus.

[0007] Preferably, in S1, a three-dimensional finite element model of the dam and foundation is established using finite element software based on engineering design and monitoring data. Specific details include: Define the direction of water flow as X Direction: downstream is positive, perpendicular to the direction of water flow is... Y Direction: pointing to the left bank is the positive direction, and the vertical direction is... Z Direction: Vertically upward is positive; In the three-dimensional finite element model, the simulated depth of the dam foundation is taken as twice the dam height, and is extended by one dam height to both banks along the dam axis, and is taken as 1.5 times the dam height in the upstream and downstream directions.

[0008] The three-dimensional finite element model mainly uses hexahedral and tetrahedral meshes.

[0009] Based on the dam material properties in the mechanical parameters of the dam construction materials, the finite element model is divided into several regions.

[0010] Preferably, in S2, based on the mechanical parameter information of the concrete dam, the Morris method is used to perform sensitivity analysis on the mechanical parameters of each region in the three-dimensional finite element model to determine the deformation sensitivity mechanical parameters of the concrete arch dam. The specific content is as follows: The mechanical parameters of the concrete dam are obtained by analyzing and extracting engineering design and monitoring data. The mechanical parameters of the concrete dam include the mechanical parameters of the dam body and the dam foundation. Mechanical design elements are defined based on the mechanical parameters of the concrete dam. These mechanical design elements include the elastic modulus of the dam body, the deformation modulus of the dam foundation, Poisson's ratio, and unit weight. The mechanical parameters of each region in the three-dimensional finite element model are assigned values ​​using the design values ​​of the dam body and foundation parameters, and the upper and lower limits of the parameter values ​​for each region are determined, i.e., the range of parameter values. Based on the Morris sensitivity analysis principle, target group samples are extracted as experimental parameter input sets within the range of each parameter value; The experimental parameter set was input into the three-dimensional finite element model to calculate the response value, and the displacement water pressure component value in the direction of water flow at the dam monitoring point was obtained. The sensitivity parameters were obtained by analyzing the sensitivity of the elastic modulus of the dam body, the deformation modulus of the dam foundation, Poisson's ratio and unit weight to the deformation of the concrete arch dam in each region using the Morris sensitivity analysis principle. The deformation sensitivity mechanical parameters of concrete arch dams were obtained by screening sensitive parameters; The deformation-sensitive mechanical parameters of concrete arch dams are defined as the parameters to be inverted, that is, the parameters that have a high sensitivity to the deformation of concrete arch dams are selected as the parameters to be inverted.

[0011] Preferably, the specific content of the multi-input-multi-output adaptive intelligent sampling method for constructing the dam's deformation sensitivity mechanical parameters and the deformation at each monitoring point based on the deformation sensitivity mechanical parameters of the concrete arch dam includes: In terms of Design of Experiments (DoE), the response values ​​of the dam monitoring points are defined as output parameters, the target to be inverted is set as random input parameters, and the remaining parameters are used as deterministic input parameters. Using the Sobol sampling method, a target random sample group is generated based on the range of values ​​of the random input parameters. This sample group is then substituted into the established finite element model to calculate the output and obtain the sample pool. N A ; Using the Kennard-Stone algorithm in N A The initial sample set is obtained by selecting representative samples. N 1. Sample Pool N A The remaining unselected samples automatically become the candidate data pool. NH .

[0012] Preferably, the accuracy evaluation index of the adaptive intelligent sampling method is the correlation coefficient of each monitoring point. R 2 Correlation coefficient R 2 It is inversely correlated with the weight; In each iteration from N H The sample with the highest diversity score among the target samples was selected as the filler sample to supplement the target sample. N 1. Continue iterating until the accuracy evaluation index requirements are met.

[0013] Preferably, the surrogate model of the adaptive intelligent sampling method is a radial basis function network (RBFN). from N A Several sets of sample pool data were selected as training and validation sets respectively to train and validate the proxy model; RBFN uses a grid search method to optimize kernel parameters, employs 5-fold cross-validation for balanced evaluation, and uses root mean square error as the optimization index. The root mean square error expression is: ; in This is the actual value. This is the predicted value, and n is the number of data points.

[0014] Preferably, the specific content of constructing a deep learning surrogate model reflecting the nonlinear relationship between the dam's sensitivity mechanical parameters and the deformation at each monitoring point, and performing deep learning inversion of the dam's elastic modulus, is as follows: An adaptive intelligent sampling method was constructed, and a deep learning sample set was obtained through adaptive sampling. N S ; A self-attention mechanism is introduced to construct an Attention-BNN proxy model for the inversion of mechanical parameters in multiple zones of concrete arch dams; In the deep learning inversion process of the dam's elastic modulus, the deep learning sample set is used. N S The data is fed into the Attention-BNN for training and from the candidate data pool. N H Randomly select several sample points from the unscreened samples as the training set; The loss function for minimizing the objective function is calculated by using the actual deformation value of the arch dam and the predicted value of the deep learning surrogate model. The expression for minimizing the loss function is: ; in, It is the optimization objective; yes KL Divergence (a complexity penalty term) is used to limit computation by minimizing this term. The distribution must not deviate too far from the prior distribution, thus introducing regularization. To approximate the posterior distribution, it is usually assumed to be a Gaussian distribution, where the parameter... Control the shape of this distribution (such as the mean and variance). For the prior distribution, These are the weights of each layer in the neural network; It is the likelihood term, and this term is related to... KL The divergence term balances the fitted data with maintaining reasonable uncertainty, among which N The total number of samples in the deep learning sample set. The output (prediction) of a neural network depends on the input. and weight , It is a model prediction Generate real labels in time The probability of.

[0015] The parameter inversion values ​​of the deep learning model are obtained by using the internal optimizer of the neural network.

[0016] In summary, this invention presents a method for rapid inversion of structural mechanical parameters of a concrete arch dam by region. Compared with traditional techniques, this invention proposes a method for rapid inversion of structural mechanical parameters of a concrete arch dam by region. It utilizes Morris sensitivity analysis to obtain parameters that are sensitive to deformation at each monitoring point. Furthermore, it establishes an adaptive intelligent sampling method based on the RBFN model for a multi-input multi-output problem to obtain a high-precision, non-redundant sample set. Finally, it constructs a deep learning model that integrates Bayesian neural networks and self-attention mechanisms to perform uncertainty inversion of sensitive mechanical parameters, thereby improving inversion accuracy and efficiency. This provides a new approach for the regional material parameter inversion of dams and foundations during operation.

[0017] The technical method of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] Figure 1 This is a flowchart of a rapid inversion method for the whole-domain structural mechanical parameters of a concrete arch dam according to the present invention. Figure 2 Layout diagram of arch dam monitoring facilities; Figure 3This is a finite element model diagram of a concrete arch dam. Figure 3 (a) in the diagram represents the dam foundation material zoning. Figure 3 (b) in the diagram represents the dam material zoning; Figure 4 This is a graph showing the results of a partial sensitivity analysis of the performance parameters of the arch dam structure. Figure 4 (a) in the figure shows the sensitivity analysis results of the vertical monitoring point PLA1 on the elastic modulus, Poisson's ratio and unit weight of the dam body and dam foundation; Figure 4 (b) in the figure shows the sensitivity analysis results of the vertical monitoring point PLA2 on the elastic modulus, Poisson's ratio and unit weight of the dam body and dam foundation; Figure 4 (c) in the figure shows the sensitivity analysis results of the inverted plumb line monitoring point IPA1 on the elastic modulus, Poisson's ratio and unit weight of the dam body and dam foundation; Figure 5 Graph showing the RBF kernel parameter optimization process; Figure 6 This is a comparison chart of the initial prediction results of the RBF surrogate model for some monitoring points. Figure 6 (a) in the figure is a comparison of the prediction results of the vertical monitoring point PLA1 in the validation set; Figure 6 (b) in the diagram is a scatter plot of the actual values ​​and corresponding predicted values ​​of each value in the validation set of the vertical monitoring point PLA1. Figure 6 (c) in the figure is a comparison of the prediction results of the vertical monitoring point PLA2 in the validation set; Figure 6 (d) in the figure is a scatter plot of the actual values ​​and corresponding predicted values ​​of each value in the validation set of the vertical line monitoring point PLA2; Figure 6 (e) in the figure is a comparison of the prediction results of the inverted plumb line monitoring point IPA1 in the validation set; Figure 6 (f) in the diagram is a scatter plot of the predicted values ​​of each actual value and the corresponding predicted value in the IPA1 validation set of the inverted plumb line monitoring points; Figure 7 Correlation coefficient of monitoring points Infographic Figure 7 In the figure, (a) represents the curves showing the changes in the average correlation coefficient R2 and the minimum correlation coefficient R2 of all monitoring points of the arch dam as the number of iterations increases during the iteration process; Figure 7 (b) shows the curves of the correlation coefficient R2 of each monitoring point of the arch dam as the number of iterations increases during the iteration process, and the threshold dotted line obtained according to the stopping criterion. Figure 8 Correlation coefficients of monitoring points in multiple models Result comparison chart Figure 8 (a) in the figure represents the correlation coefficient of Attention-BNN. Result image, Figure 8 (b) in the figure represents the BNN correlation coefficient. Result image, Figure 8(c) in the equation represents the N / A correlation coefficient. Resulting image; Figure 9 Scatter plots of multi-sample prediction results for each model. Figure 9 (a) in the image is a scatter plot of the Attention-BNN multi-sample prediction results. Figure 9 (b) in the figure is a scatter plot of the BNN multi-sample prediction results. Figure 9 (c) in the figure is a scatter plot of the NN multi-sample prediction results. Detailed Implementation

[0019] The technical method of the present invention will be further described below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application.

[0020] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this application and its application or use.

[0021] Techniques, systems, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, they should be considered part of the instruction manual.

[0022] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0023] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0024] The present application proposes a method for rapid inversion of structural mechanical parameters of a concrete arch dam by region. First, based on engineering design and monitoring data, a three-dimensional finite element model of the dam body and foundation is established using finite element software. Based on the mechanical parameter information of the dam construction materials, the three-dimensional finite element model is divided into regions, and the Morris method is used to analyze the sensitivity of mechanical parameters in each region to determine the sensitive mechanical parameters that affect the deformation of the concrete arch dam.

[0025] Based on the sensitivity analysis results, adaptive intelligent sampling is performed on the sensitive mechanical parameters to effectively select a high-quality, non-redundant sample set required for constructing a high-precision deep learning proxy model. Based on this sample set, a deep learning proxy model integrating a self-attention mechanism and an improved Bayesian neural network is constructed, providing an equivalent inverse analysis method for the uncertainty of sensitive physical parameters of arch dams. Numerical examples and engineering cases demonstrate that the inverse method for the uncertainty of multi-input-multi-output sensitive mechanical parameters of arch dams is effective and accurate. The computational accuracy and efficiency are significantly improved compared to traditional methods, with a lower overall error. It can efficiently and accurately inversely determine the mechanical parameters of arch dam structures during actual operation, providing a solid foundation for dam safety analysis.

[0026] Example 1 A method for rapid inversion of the structural mechanical parameters of a concrete arch dam by region includes the following steps: S1. Based on the engineering design and monitoring data, a three-dimensional finite element model of the dam and foundation is established using finite element software, and the three-dimensional finite element model is divided into zones based on the mechanical parameter information of the dam construction materials. Furthermore, in S1, based on engineering design and monitoring data, a three-dimensional finite element model of the dam and foundation is established using finite element software. The specific details of partitioning the three-dimensional finite element model using the mechanical parameters of the dam construction materials include: Since the concrete arch dam in the selected engineering example was poured and compacted in layers, it has obvious zoning characteristics after forming. In order to more realistically and accurately invert the physical and mechanical parameters of the dam body and dam foundation, the dam foundation is divided into two regions according to the geological survey data of the key project, and the dam body is divided into four regions according to the construction data of the dam body pouring. Define the direction of water flow as X Direction: downstream is positive, perpendicular to the direction of water flow is... Y Direction: pointing to the left bank is the positive direction, and the vertical direction is... Z The direction is vertically upward as positive; in the finite element model, the simulated depth of the dam foundation is taken as twice the dam height, extending one dam height to each bank along the dam axis, and 1.5 times the dam height in the upstream and downstream directions. The model mainly uses hexahedral and tetrahedral meshes.

[0027] Based on the properties of the dam construction materials, the finite element model is divided into several regions.

[0028] S2. Based on the mechanical parameter information of the concrete dam, the Morris method is used to conduct sensitivity analysis on the mechanical parameters of each region in the three-dimensional finite element model to determine the deformation sensitivity mechanical parameters of the concrete arch dam. Furthermore, in S2, based on the mechanical parameter information of the concrete dam, the Morris method is used to conduct sensitivity analysis on the mechanical parameters of each region in the three-dimensional finite element model, and the specific contents of the deformation sensitivity mechanical parameters of the concrete arch dam are determined as follows: The Morris method was used to perform sensitivity analysis on the mechanical parameters of each region in the three-dimensional finite element model to determine the deformation sensitivity mechanical parameters of the concrete arch dam. The specific details are as follows: The mechanical parameters of the concrete dam are obtained by analyzing and extracting engineering design and monitoring data. The mechanical parameters of the concrete dam include the mechanical parameters of the dam body and the dam foundation. Mechanical design elements are defined based on the mechanical parameters of the concrete dam. These mechanical design elements include the elastic modulus of the dam body, the deformation modulus of the dam foundation, Poisson's ratio, and unit weight. The mechanical parameters of each region in the three-dimensional finite element model are assigned values ​​using the design values ​​of the dam body and foundation parameters, and the upper and lower limits of the parameter values ​​for each region are determined, i.e., the range of parameter values. Based on the Morris sensitivity analysis principle, 950 sets of samples were extracted as the experimental parameter input set within the range of each parameter value. The experimental parameter set was input into the three-dimensional finite element model to calculate the response value, and the displacement water pressure component values ​​in the direction of water flow were obtained for all 10 positive and negative vertical monitoring points of the dam. The experimental parameter set was input into the three-dimensional finite element model to calculate the response value. The displacement and water pressure components in the direction of water flow at each monitoring point of the dam were obtained as the experimental parameter output set. Based on this, the Morris sensitivity analysis principle was used to analyze the sensitivity of the elastic modulus, Poisson's ratio and unit weight (experimental parameter input set) of the dam body and foundation to the deformation of the concrete arch dam (experimental parameter output set, which is also another expression of the displacement and water pressure components in the direction of water flow). The parameters with high sensitivity to the deformation of the concrete arch dam were selected as the targets to be inverted.

[0029] The Morris sensitivity analysis principle was used to analyze the sensitivity of the elastic modulus of the dam body, the deformation modulus of the dam foundation, Poisson's ratio and unit weight of the concrete arch dam to the deformation of the concrete arch dam in each region. Sensitive parameters were obtained by screening the sensitive parameters to obtain the deformation sensitivity mechanical parameters of the concrete arch dam.

[0030] The deformation-sensitive mechanical parameters of concrete arch dams are defined as the parameters to be inverted, that is, the parameters that have a high sensitivity to the deformation of concrete arch dams are selected as the parameters to be inverted.

[0031] S3. Based on the sensitivity analysis parameter screening results, construct a multi-input-multi-output adaptive intelligent sampling method for the dam's sensitivity mechanical parameters and the deformation of each monitoring point; Furthermore, based on the sensitivity analysis parameter screening results, the specific content of constructing a multi-input-multi-output adaptive intelligent sampling method for the dam's sensitivity mechanical parameters and deformation at each monitoring point includes: In terms of Design of Experiments (DoE), the response values ​​of all 10 positive and negative vertical monitoring points of the dam were set as output parameters, the target to be inverted was set as random input parameters, and the remaining parameters were set as deterministic input parameters. Using the Sobol sampling method, 2000 random sample groups are generated based on the range of values ​​for the random input parameters. These samples are then substituted into the established finite element model to calculate the output and obtain the sample pool. N A ; Using the Kennard-Stone algorithm in N A The initial sample set is obtained by selecting representative samples. N 1. The initial sample set size proposed in this invention is 50, that is... N 1=50, Sample Pool N A The remaining unselected samples automatically become the candidate data pool. N H .

[0032] Furthermore, the accuracy evaluation index of the adaptive intelligent sampling method is the correlation coefficient of each monitoring point. R 2 Correlation coefficient R 2 It is inversely correlated with the weight, that is, when a certain monitoring point R 2 When the value is lower, it will receive a higher weight, enabling adaptive sampling to dynamically identify and prioritize outputs that are more difficult to predict; In each iteration from N H Two samples with the highest diversity scores were selected as filler samples to supplement the sample. N 1. Continue iterating until the accuracy evaluation index requirements are met.

[0033] Furthermore, the surrogate model of the adaptive intelligent sampling method is a radial basis function network (RBFN). from N A 80 and 20 groups were selected as training and validation sets, respectively, to train and validate the proxy model. The performance of the RBFN model is highly dependent on the selection of kernel function parameters; that is, kernel parameter optimization is a key step in constructing a high-precision RBFN model. This invention uses a grid search method for kernel parameter optimization, employs 5-fold cross-validation to balance and evaluate reliability, and incorporates the root mean square error. RMSE As an optimization metric.

[0034] S4. Construct a deep learning proxy model that reflects the nonlinear relationship between the dam's sensitivity mechanical parameters and the deformation at each monitoring point, and perform deep learning inversion of the dam's elastic modulus.

[0035] The deep learning sample set obtained through adaptive sampling and filtering N S This study lays a solid foundation for model training. In the parameter inversion stage, traditional surrogate models and classic deep learning methods are mostly limited to point estimation, failing to effectively express prediction uncertainty and thus restricting their application in high-risk fields such as safety monitoring. To address this bottleneck, this research combines Bayesian deep learning technology to enhance the model's ability to quantify parameter uncertainty, significantly improving the reliability and interpretability of predictions. Furthermore, a self-attention mechanism is introduced to construct a multi-zone concrete arch dam mechanical parameter inversion model, Attention-BNN. This model excels in capturing complex spatial heterogeneity and multi-scale features, providing solid data support and theoretical basis for risk assessment and decision-making in engineering practice. Furthermore, a deep learning surrogate model reflecting the nonlinear relationship between the dam's sensitivity mechanical parameters and the deformation at each monitoring point is constructed. The specific content of the deep learning inversion of the dam's elastic modulus is as follows: The deep learning sample set obtained through S3 adaptive sampling and filtering N S ; A self-attention mechanism is introduced to construct an Attention-BNN proxy model for the inversion of mechanical parameters in multiple zones of concrete arch dams; In the deep learning inversion process of the dam's elastic modulus, the deep learning sample set is used. N S The data is fed into the Attention-BNN for training and from the candidate data pool. N H Twenty sample points were randomly selected from the unscreened samples to serve as the training set. The loss function for minimizing the objective function is calculated by using the actual deformation value of the arch dam and the predicted value of the deep learning surrogate model. The parameter inversion value under the deep learning model is obtained by using the internal optimizer of the neural network.

[0036] We selected a standard Bayesian neural network (BNN) without self-attention mechanism, a traditional neural network (NN), a random forest ensemble learning algorithm (RF), and a support vector regression algorithm (SVR) to perform parameter inversion together, thereby verifying the fitting and prediction effects of the deep learning model.

[0037] Example 2 This invention provides a method for rapid inversion of the structural mechanical parameters of a concrete arch dam by region, the flowchart of which is shown below. Figure 1As shown, this embodiment takes a roller-compacted concrete double-curvature arch dam as an example. The dam has a maximum height of 99.1m, a crest elevation of 247.6m, a minimum excavation elevation of 148.5m, and a total reservoir capacity of 1.0481 × 10⁸ m³. It began impounding water in 2012 and has been operating safely for over 10 years. To ensure the safe operation of the dam during its service life, a large number of monitoring facilities were designed and installed simultaneously to detect dam deformation, seepage, and environmental parameters. The dam's interior is equipped with three vertical lines with six monitoring points each and three inverted vertical lines with four monitoring points each. The monitoring points are located in the dam observation corridor at elevations of 163.0m, 195.0m, and 220.0m, respectively. The key location for dam displacement monitoring is the section of the dam where the arch crown beam is located. This section has two vertical observation points (PLA1 and PLA2) and two inverted vertical observation points (IPA1 and IPA2). Figure 2 As shown, Figure 2 In this diagram, PL (Plumb Line) is the upright vertical line, IP (Inverted Plumb Line) is the inverted vertical line, M (Bedrock displacement meter) is the bedrock displacement meter, K (Pore water pressure gauge) is the pore water pressure gauge, J (joint meter) is the joint gauge, and T (thermometer) is the thermometer. The process is as follows: S11. Since the concrete arch dam in the selected engineering example was poured and compacted in layers, it has obvious zoning characteristics after forming. In order to more realistically and accurately invert the physical and mechanical parameters of the dam body and dam foundation, the dam foundation is divided into two regions according to the geological survey data of the key project, and the dam body is divided into four regions according to the construction data of the dam body pouring. S12. Define the direction of water flow as... X Direction: downstream is positive, perpendicular to the direction of water flow is... Y Direction: pointing to the left bank is the positive direction, and the vertical direction is... Z The direction is vertically upwards as positive; in the finite element model, the simulated depth of the dam foundation is twice the dam height, extending one dam height along the dam axis to both banks, and 1.5 times the dam height in the upstream and downstream directions. The model mainly uses hexahedral and tetrahedral meshes. Based on the dam construction material properties, the finite element model is divided into several regions. The constructed finite element model contains 151,200 mesh elements, and the three-dimensional model is as follows: Figure 3 As shown.

[0038] S21. Based on the mechanical parameters of the dam body and foundation provided in the engineering data, the elastic modulus of the dam body, the deformation modulus of the dam foundation, Poisson's ratio and unit weight are selected as the research objects. The mechanical parameters of each region in the three-dimensional finite element model are assigned values ​​using the design values ​​of the dam body and foundation parameters, and the upper and lower limits of the parameter values ​​of each region are determined, i.e. the parameter value range, as shown in Table 1 and Table 2. Table 1. Sampling range of dam body test values ;

[0039] Table 2 Sampling range of dam foundation test values ;

[0040] S22. Following the Morris sensitivity analysis principle, 950 sets of samples were extracted within the range of each parameter value as the experimental parameter input set. Then, the parameters were input into the established finite element model to calculate the response value and obtain the displacement water pressure component values ​​in the direction of water flow at all 10 positive and negative vertical monitoring points of the dam. S23. The Morris method was used to analyze the sensitivity of each parameter to the deformation of the concrete arch dam. Parameters with high sensitivity to the deformation of the concrete arch dam were selected as parameters to be inverted. Some sensitivity analysis results are shown below. Figure 4 As shown, to facilitate the comparison of parameter sensitivity, only the top 6 mechanical parameters (i.e., ...) are observed at each measuring point. Figure 4 The yellow highlighted area is the key area for observation and comparison.

[0041] S31. Regarding DoE, the response values ​​of all 10 monitoring points for both positive and negative plumb lines are designated as output parameters. The predetermined mechanical parameters to be inverted are set as random input parameters with a range of values. All other mechanical parameters are taken as design values ​​as deterministic input parameters. Using the Sobol sampling method, 2000 random sample groups are generated based on the range of random input parameter values. These samples are then substituted into the established finite element model to calculate the output and obtain the sample pool. N A ; S32. Using the Kennard-Stone algorithm in N A The initial sample set is obtained by selecting representative samples. N 1. The initial sample set size proposed in this invention is 50, that is... N 1=50, Sample Pool N A The remaining unselected samples automatically become the candidate data pool. N H The subsequent adaptive sampling will determine the correlation coefficients of each monitoring point. R 2 As an accuracy assessment indicator, when a certain monitoring point R 2 Lower values ​​will receive higher weights, enabling adaptive sampling to dynamically identify and prioritize outputs that are more difficult to predict, and to adjust the weights in each iteration. N H Two samples with the highest diversity scores were selected as filler samples to supplement the sample. N 1. Continue iterating until the evaluation criteria are met; S33. The RBFN model is selected as the surrogate model required for adaptive sampling iteration, and from... N A We randomly select 80 and 20 sets as training and validation sets, respectively, to pre-validate the applicability and effectiveness of the surrogate model. The performance of the RBFN surrogate model is highly dependent on the selection of kernel function parameters; therefore, kernel parameter optimization is a crucial step in constructing a high-precision RBFN model. This invention uses the common grid search method for kernel parameter optimization, employs 5-fold cross-validation to balance and evaluate reliability, and incorporates the root mean square error... RMSE As an optimization metric, the optimization process is as follows: Figure 5 As shown, RBFN is used for training and prediction based on the optimal kernel parameters. The initial prediction fitting effect and prediction scatter plot of each monitoring point on the validation set are shown below. Figure 6 As shown, adaptive sampling then begins, stopping at the 187th iteration. Evaluation metrics for some monitoring points are as follows: Figure 7 As shown.

[0042] S41. Deep learning sample set obtained through adaptive sampling and filtering N S This study lays a solid foundation for model training. In the parameter inversion stage, traditional surrogate models and classic deep learning methods are mostly limited to point estimation, failing to effectively express prediction uncertainty and thus restricting their application in high-risk fields such as safety monitoring. To address this bottleneck, this research combines Bayesian deep learning technology to enhance the model's ability to quantify parameter uncertainty, significantly improving the reliability and interpretability of predictions. Furthermore, a self-attention mechanism is introduced to construct a multi-zone concrete arch dam mechanical parameter inversion model, Attention-BNN. This model excels in capturing complex spatial heterogeneity and multi-scale features, providing solid data support and theoretical basis for risk assessment and decision-making in engineering practice. S42. In the process of inverting the performance parameters of the arch dam structure, the deep learning sample set will be used. N S The data is fed into the Attention-BNN for training and from the candidate data pool. N H Twenty sample points were randomly selected from the unscreened samples as the training set. The inversion objective function was constructed using the actual deformation values ​​of the arch dam and the predicted values ​​of the deep learning model. The calculation was performed using the internal optimizer of the neural network, and the parameter inversion values ​​under the deep learning model were finally obtained. The inversion results are shown in Table 3. Table 3. Parameter Inversion Results of Attention-BNN ;

[0043] S43. A standard Bayesian neural network (BNN) without self-attention mechanism and a traditional neural network (NN) were used together for parameter inversion to verify the fitting and prediction effects of the deep learning model. The parameter inversion results are shown in Table 4, using correlation coefficients. R 2 The correlation coefficients of some monitoring points in each model were used as an evaluation indicator for verification. R 2 Values ​​such as Figure 8 As shown in the figure, the scatter plots of the multi-sample prediction results for each model are as follows: Figure 9 As shown.

[0044] Table 4. Parameter Inversion Results Using Multiple Methods ;

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical methods of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical methods of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical methods to deviate from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for rapid inversion of structural mechanical parameters of a concrete arch dam by region, characterized in that, Includes the following steps: S1. Based on the engineering design and monitoring data, a three-dimensional finite element model of the dam and foundation is established using finite element software, and the three-dimensional finite element model is divided into zones based on the mechanical parameter information of the dam construction materials. S2. Based on the mechanical parameter information of the concrete dam, the Morris method is used to conduct sensitivity analysis on the mechanical parameters of each region in the three-dimensional finite element model to determine the deformation sensitivity mechanical parameters of the concrete arch dam. S3. Based on the deformation sensitivity mechanical parameters of sensitive concrete arch dams, construct a multi-input-multi-output adaptive intelligent sampling method for the dam's deformation sensitivity mechanical parameters and the deformation of each monitoring point of the dam. S4. Construct a deep learning proxy model that reflects the nonlinear relationship between the dam's sensitivity mechanical parameters and the deformation at each monitoring point, and perform deep learning inversion of the dam's elastic modulus.

2. The method for rapid inversion of structural mechanical parameters of a concrete arch dam by region according to claim 1, characterized in that, The specific content of S1, which involves establishing a three-dimensional finite element model of the dam and foundation using finite element software based on engineering design and monitoring data, includes: Define the direction of water flow as X Direction: downstream is positive, perpendicular to the direction of water flow is... Y Direction: pointing to the left bank is the positive direction, and the vertical direction is... Z Direction: Vertically upward is positive; In the three-dimensional finite element model, the simulated dimension of the dam foundation depth is taken as twice the dam height, and is extended by one dam height to both banks along the dam axis, and is taken as 1.5 times the dam height in the upstream and downstream directions. The three-dimensional finite element model uses hexahedral and tetrahedral meshes; Based on the dam material properties in the mechanical parameters of the dam construction materials, the three-dimensional finite element model is divided into several regions.

3. The method for rapid inversion of structural mechanical parameters of a concrete arch dam by region according to claim 2, characterized in that, Based on the mechanical parameters of the concrete dam, S2 uses the Morris method to perform sensitivity analysis on the mechanical parameters of each region in the three-dimensional finite element model, determining the specific contents of the deformation sensitivity mechanical parameters of the concrete arch dam as follows: The mechanical parameters of the concrete dam are obtained by analyzing and extracting engineering design and monitoring data. The mechanical parameters of the concrete dam include the mechanical parameters of the dam body and the dam foundation. Mechanical design elements are defined based on the mechanical parameters of the concrete dam. These mechanical design elements include the elastic modulus of the dam body, the deformation modulus of the dam foundation, Poisson's ratio, and unit weight. The mechanical parameters of each region in the three-dimensional finite element model are assigned values ​​using the design values ​​of the dam body and foundation parameters, and the upper and lower limits of the parameter values ​​for each region are determined, i.e., the range of parameter values. Based on the Morris sensitivity analysis principle, target group samples are extracted as experimental parameter input sets within the range of each parameter value; The experimental parameter set was input into the three-dimensional finite element model to calculate the response value, and the displacement water pressure component value in the direction of water flow at the dam monitoring point was obtained. The sensitivity parameters were obtained by analyzing the sensitivity of the elastic modulus of the dam body, the deformation modulus of the dam foundation, Poisson's ratio and unit weight to the deformation of the concrete arch dam in each region using the Morris sensitivity analysis principle. The deformation sensitivity mechanical parameters of concrete arch dams were obtained by screening sensitive parameters.

4. The method for rapid inversion of structural mechanical parameters of a concrete arch dam by region according to claim 3, characterized in that, The specific content of constructing a multi-input, multi-output adaptive intelligent sampling method for the deformation-sensitive mechanical parameters of concrete arch dams and the deformation at each monitoring point includes: Define the deformation sensitivity mechanical parameters of concrete arch dams as the parameters to be inverted; In terms of the Design of Experiments (DoE), the response values ​​of the dam monitoring points are defined as output parameters, the target to be inverted is set as random input parameters, and the remaining parameters are used as deterministic input parameters. Using the Sobol sampling method, a target random sample group is generated based on the range of values ​​of the random input parameters. This sample group is then substituted into the established finite element model to calculate the output and obtain the sample pool. N A ; Using the Kennard-Stone algorithm in N A The initial sample set is obtained by selecting representative samples. N 1. Sample Pool N A The remaining unselected samples automatically become the candidate data pool. N H .

5. The method for rapid inversion of structural mechanical parameters of a concrete arch dam by region according to claim 4, characterized in that, The accuracy evaluation index of the adaptive intelligent sampling method is the correlation coefficient of each monitoring point. R 2 Correlation coefficient R 2 It is inversely correlated with the weight; In each iteration from N H The sample with the highest diversity score among the target samples was selected as the filler sample to supplement the target sample. N 1. Continue iterating until the accuracy evaluation index requirements are met.

6. The method for rapid inversion of structural mechanical parameters of a concrete arch dam by region according to claim 5, characterized in that, The surrogate model for the adaptive intelligent sampling method is a radial basis neural network; from N A Several sets of sample pool data were selected as training and validation sets respectively to train and validate the proxy model; RBFN uses a grid search method to optimize kernel parameters, employs 5-fold cross-validation for balanced evaluation, and uses root mean square error as the optimization index. The root mean square error expression is: ； in, This is the actual value. This is the predicted value, and n is the number of data points.

7. The method for rapid inversion of structural mechanical parameters of a concrete arch dam by region according to claim 6, characterized in that, The specific content of constructing a deep learning surrogate model reflecting the nonlinear relationship between the dam's sensitivity mechanical parameters and the deformation at each monitoring point, and performing deep learning inversion of the dam's elastic modulus, is as follows: An adaptive intelligent sampling method was constructed, and a deep learning sample set was obtained through adaptive sampling. N S ; A self-attention mechanism is introduced to construct an Attention-BNN proxy model for the inversion of mechanical parameters in multiple zones of concrete arch dams; In the deep learning inversion process of the dam's elastic modulus, the deep learning sample set is used. N S The data is fed into the Attention-BNN for training and from the candidate data pool. N H Randomly select several sample points from the unscreened samples as the training set; The loss function for minimizing the objective function is calculated by using the actual deformation value of the arch dam and the predicted value of the deep learning surrogate model. The expression for minimizing the loss function is: ； in, It is the optimization objective; yes KL Divergence, or complexity penalty term, is used to limit computation by minimizing this term. The distribution must not deviate too far from the prior distribution, thus introducing regularization. For an approximate posterior distribution, where the parameter is... Controlling the shape of this distribution, For the prior distribution, These are the weights of each layer in the neural network; It is the likelihood term, and the likelihood term and KL The divergence term balances the fitted data with maintaining reasonable uncertainty, among which N The total number of samples in the deep learning sample set. The output of a neural network depends on its input. and weight , It is a model prediction Generate real labels in time The probability of; The parameter inversion values ​​of the deep learning model are obtained by using the internal optimizer of the neural network.

Images