Intelligent identification method for dynamic twin body behavior parameters of concrete high dam under physical constraints

Abstract

The application discloses a physical constraint concrete high dam dynamic twin body state parameter intelligent identification method, belongs to the technical field of concrete dam structure health monitoring and safety control, and comprises the following steps: step S1, concrete dam body partitioning and parameter determination; step S2, mathematical model construction; step S3, PINNs construction and training; step S4, parameter optimization is carried out according to PINNs and by using an improved Jaya algorithm, and inversion state parameters are output. The application adopts the above-mentioned physical constraint concrete high dam dynamic twin body state parameter intelligent identification method, effectively reduces the solution difficulty of a multi-objective inversion problem, realizes efficient and accurate inversion of concrete dam partitioned structure state parameters, and provides reliable technical support for concrete dam operation safety monitoring and safety management.

Description

Technical Field

[0001] This invention relates to the field of health monitoring and safety control technology for concrete dam structures, and in particular to an intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints. Background Technology

[0002] In engineering projects, zoning design is usually adopted according to the stress and environmental requirements of different parts of the dam body. Material configuration is optimized by different concrete grades. However, after zoning, the material properties of each area are significantly different, resulting in the dam body exhibiting complex multi-region coupled mechanical properties.

[0003] Performance parameter inversion is a key method for obtaining the true working characteristics of a dam. It requires inverting parameters such as the elastic modulus and Poisson's ratio of concrete by monitoring displacement and stress data, providing a basis for constructing a monitoring and early warning model for the operational performance of concrete dams and for safety assessment feedback. However, the zonal structure complicates the conditions for displacement coordination and stress balance between regions, and traditional inversion methods have significant limitations.

[0004] Early empirical formulas or analytical methods relied on simplifying assumptions, making it difficult to characterize multi-regional coupling behavior and resulting in low accuracy.

[0005] When mainstream numerical methods (such as the finite element method) are combined with traditional optimization algorithms, they are difficult to handle interface coordination conditions, prone to getting trapped in local optima, have low computational efficiency, and rely excessively on complete monitoring data, resulting in poor noise resistance.

[0006] While traditional neural networks improve inversion efficiency, they are "black box" models, lacking mechanical guidance and having insufficient generalization ability and physical rationality.

[0007] While Physical Information Neural Networks (PINNs) can solve "black box" problems by embedding physical constraints and show advantages in engineering inverse problems, existing PINNs are not optimized for the zoning characteristics of concrete dams—they do not fully consider the coordination conditions of regional interfaces, and traditional algorithms cannot meet the requirements when seeking multiple objectives, making it difficult to balance inversion efficiency and accuracy.

[0008] In summary, to address the challenges of partitioned coupling inversion, the inadequacy of traditional methods, and the lack of targeted optimization for existing PINNs, there is an urgent need to construct a physical neural network inversion method that integrates the domain decomposition method, elasticity theory, and improved optimization algorithms. Summary of the Invention

[0009] The purpose of this invention is to provide an intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints, which effectively reduces the difficulty of solving multi-objective inversion problems, realizes efficient and accurate inversion of structural behavior parameters of concrete dam zones, and provides reliable technical support for safe monitoring and management of concrete dam operation.

[0010] To achieve the above objectives, this invention provides an intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints, comprising the following steps:

[0011] Step S1: Dam body zoning and parameter determination. Based on the actual materials of the concrete dam, the dam body-foundation structural system is divided into zones. i Each dam sub-area j Each dam foundation sub-region was identified, and the behavior parameters to be inverted in each region were determined.

[0012] The Latin hypercube sampling method was used to uniformly sample the inversion behavior parameters. The response values ​​of the concrete dam structure under different combinations of behavior parameters were calculated by combining the finite element model. The numerical samples and measured data were fused to form the training dataset of the physical information neural network PINNs, and the parameter samples were normalized.

[0013] Step S2: Mathematical model construction. Based on the region decomposition method, the mathematical equations of the structural response of the concrete dam zone are derived. The displacement vector of the sub-region is decomposed according to the internal degree of freedom and the interface degree of freedom. The stiffness matrix and load vector are processed in blocks, and the control equations of the dam body and dam foundation sub-regions are established.

[0014] Step S3, PINNs Construction and Training: Based on the mathematical equations of the partitioned structural response and the boundary conditions of the concrete dam stress and displacement, construct the PINNs loss function containing boundary condition loss terms, intra-regional loss terms, and inter-regional interface loss terms; configure the PINNs network structure and train the PINNs model using a piecewise optimization strategy and dynamic learning rate.

[0015] Step S4: Based on PINNs and using the improved Jaya algorithm, perform parameter optimization and output the inversion results of concrete dam behavior parameters.

[0016] Preferably, in step S1, the parameter sample normalization formula is:

[0017] ;

[0018] in, x For the sampling parameters that need to be normalized, i =1,2,…,n; x k,min =min( ) is the first k The minimum value of the data center of the dimensional variable sample. .

[0019] Preferably, the governing equations for the dam sub-region in step S2 are:

[0020] ;

[0021] in, This represents the internal stiffness matrix within a sub-region of the dam body. This is the coupling stiffness matrix between internal nodes and the interface, used to connect internal nodes and interface nodes; This represents the coupling stiffness matrix between interface nodes in the sub-regions of the dam body. The displacement vector of the internal nodes; is the displacement vector of the node at the interface of the sub-region; This is the internal load vector; The interface load vector;

[0022] The governing equations for the dam foundation sub-region are:

[0023] ;

[0024] in, This represents the internal stiffness matrix of the dam foundation sub-region. This is the coupling stiffness matrix between the internal nodes and the interface; This represents the coupling stiffness matrix between interface nodes in the dam foundation sub-region; This represents the displacement vector of the nodes within the sub-region of the dam foundation. is the displacement vector of the node at the interface of the dam foundation sub-region; This represents the load vector within the sub-region of the dam foundation. This represents the interface load vector of the dam foundation sub-region.

[0025] Preferably, in step S2, a coarse space is constructed and the interface displacement is decomposed using the multi-scale finite element method. The decomposition expression for the interface displacement of the sub-region is as follows:

[0026] ;

[0027] ;

[0028] in, It is a coarse basis function matrix; It is a coarse-space coefficient vector, formed by global deformation; , These are the fine spatial components of the dam body and dam foundation interfaces, respectively.

[0029] Furthermore, by introducing displacement transformation matrices and Lagrange multipliers, the coordination and equilibrium equations for three types of interfaces—dam body-dam body, dam body-dam foundation, and dam foundation-dam foundation—are derived, and the global coordination equations for the concrete dam structure are obtained by combining these equations.

[0030] Preferably, in step S3, the loss function for the boundary condition loss term is:

[0031] ;

[0032] ;

[0033] ;

[0034] ;

[0035] in, , , These are the weight values ​​for each loss term; This is the partial differential loss term; This represents the number of integration points within the entire computational domain. , They are respectively , The governing equations for direction; This is the loss term for displacement boundary conditions; This represents the number of displacement boundary nodes; , They are respectively , Directional displacement boundary conditions; This is the stress boundary condition loss term; This represents the number of stress boundary nodes; , They are respectively , Directional stress boundary conditions.

[0036] Preferably, in step S3, the loss function term for the loss term within the region is:

[0037] ;

[0038] in, , To balance the weighting coefficients of the various loss terms by orders of magnitude, a trial-and-error algorithm was used to determine them. This is the term for fitting the total displacement. For physical constraints; This represents the regional displacement water pressure component loss term.

[0039] Preferably, in step S3, the inter-regional interface loss term includes an interface displacement loss term and an interface stress loss term, wherein the interface displacement loss term is:

[0040] ;

[0041] in, , , These refer to the interfaces between dam bodies, between dam bodies and dam foundations, and between dam foundations, resulting from the zoning of the concrete dam structure. The number of discrete points on the interface; , These are the predicted displacement-water pressure components at the interface of the two sub-regions: the dam body and the dam foundation. , The PINN prediction full displacement vectors at the interface of the two sub-regions, the dam body and the dam foundation, are respectively. This is the interface node mapping matrix, used for interpolation to fit mismatched mesh interfaces;

[0042] The interface stress loss term is:

[0043] ;

[0044] in, , The PINN predicted stress tensor for the subregion at the interface; , This is the unit outward normal vector at the interface; This represents the force exerted by the sub-region on the interface, i.e., the surface force vector.

[0045] Preferably, in step S3, the PINNs network structure is configured with an input layer and hidden layers. The input layer is responsible for receiving the raw data and preset parameters required for the fitting problem, specifically including the spatial coordinates of the monitoring points. x , y , z ), based on the water level in front of the dam H External effects characterized by equivalent water pressure load P, and highly sensitive parameter sequences to be inverted ( X i , X j , X k ,….);

[0046] The hidden layers are set to 4-6 layers, and the hidden layers use the Swish function, which is as follows:

[0047] ;

[0048] in, Input vector or matrix; It is the sigmoid activation function; To adjust the parameters, the default value is 1; when using the Swish function, first calculate the sigmoid function, then combine it with... The output result of multiplication is:

[0049] .

[0050] Preferably, in step S3, the segmented optimization strategy first uses the Adam algorithm for initial training to make the loss function decrease rapidly and tend to stabilize; then it switches to the L-BFGS algorithm to further fine-tune the network parameters.

[0051] Use a low learning rate in the initial stage, and increase the learning rate after the warm-up process to avoid gradient explosion.

[0052] Preferably, step S4 specifically includes the following sub-steps:

[0053] Step S401: Initialize the parameter population, list the regional behavior parameters to be identified and set their dimension as D according to their quantity, the initial population size as N, and set the search range of each regional behavior parameter as [[] based on the concrete dam design reference value and engineering practice experience. ] ;

[0054] Step S402: Randomly generate the initial population: ,in , r for Random numbers between;

[0055] Step S403: Select the measured displacement data and combine each parameter individually. Substitute into PINNs to calculate the objective function value Record the best individual in the current population. and the worst individual ;

[0056] Step S404: Update individuals using the individual update formula, optimize the population by combining mutation and crossover operations, and output the behavior parameters to be inverted.

[0057] Preferably, step S404 specifically includes:

[0058] ;

[0059] in, As a new individual, r 1. r 2 is Random numbers between; new individual performance shows a tendency to move towards the best individual while moving away from the worst individual;

[0060] The mutation and crossover operation specifically involves: randomly updating new individuals. One dimension parameter in the equation is used to examine the fitness value of the new individual after crossover. If there is... If the original individual is replaced by the new individual, then the original individual is replaced by the new individual; otherwise, the original individual is retained.

[0061] Repeat the iteration until the optimal fitness value is less than a pre-set threshold or the number of iterations reaches the maximum number of iterations, and then output the behavior parameters to be identified.

[0062] Therefore, the present invention employs the above-mentioned intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints, and the technical effect is as follows:

[0063] By comprehensively applying the domain decomposition method to construct the mathematical equations of the structural response of concrete dam zones and the loss function of physical neural networks (PINNs), an optimization method for the uncertain mapping relationship of concrete dam deformation response considering the inter-zone equilibrium coordination is proposed. On this basis, a multi-objective inverse analysis model of concrete dam performance parameters considering the deformation coordination of blocks is constructed, and a multi-objective performance parameter inversion method for concrete dam systems based on improved Jaya-PINNs is proposed. This method can effectively reduce the difficulty of solving multi-objective inversion problems, achieve efficient and accurate inversion of the structural performance parameters of concrete dam zones, and provide reliable technical support for the safe monitoring and management of concrete dam operation.

[0064] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0065] Figure 1 This is a flowchart of an embodiment of the intelligent identification method for dynamic twin behavior parameters of high concrete dams under mixed physical constraints according to the present invention;

[0066] Figure 2 This is a schematic diagram of the safety monitoring equipment layout for a concrete gravity dam body in an embodiment of the intelligent identification method for dynamic twin behavior parameters of a high concrete dam under physical constraints according to the present invention.

[0067] Figure 3 This is an embodiment of the intelligent identification method for dynamic twin behavior parameters of concrete high dams under physical constraints, which is a calculation sub-region decomposition and finite element model of a concrete dam No. 3 section.

[0068] Figure 4 This is the PINNs loss function interface coordination verification process line of the embodiment of the intelligent identification method for dynamic twin behavior parameters of concrete high dams under physical constraints of the present invention; Figure 4 (a) in the figure represents the loss function and interface coordination verification for region 1; Figure 4 (b) in the figure represents the verification of the loss function and interface coordination in region 2; Figure 4 (c) in the figure represents the verification of the loss function and interface coordination in region 3; Figure 4 (d) in the figure represents the verification of the loss function and interface coordination in region 4; Figure 4 (e) in the figure represents the loss function and interface coordination verification for region 5; Figure 4 (f) in the figure represents the loss function and interface coordination verification for region 6; Figure 4In the table, (g) represents the PINNs loss function for each region;

[0069] Figure 5 This is the optimization process of regional performance parameters in an embodiment of the intelligent identification method for dynamic twin performance parameters of high concrete dams under physical constraints of the present invention; Figure 5 (a) in the figure represents the process line for optimizing the elastic modulus in region 4; Figure 5 (b) in the figure represents the process line for optimizing the elastic modulus in region 3; Figure 5 (c) in the figure represents the process line for optimizing the elastic modulus in region 2; Figure 5 (d) in the figure represents the process line for optimizing the elastic modulus in region 1; Figure 5 (e) in the figure represents the Poisson's ratio optimization process line in region 1; Figure 5 (f) in the figure represents the Poisson's ratio optimization process line in region 2; Figure 5 (g) in the figure represents the process line for optimizing the elastic modulus in region 5; Figure 5 (h) in the figure represents the process line for optimizing the elastic modulus in region 6;

[0070] Figure 6 This is a comparison and verification diagram of the inversion results of an embodiment of the intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints of the present invention; Figure 6 (a) in the figure represents a comparison of the relative values ​​of horizontal displacement; Figure 6 (b) in the figure represents a comparison of error coefficients; Figure 6 (c) in the figure represents the comparison of MAPE (mean absolute percentage error). Detailed Implementation

[0071] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0072] 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.

[0073] Example 1

[0074] like Figure 1 As shown, this invention provides an intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints, comprising the following steps:

[0075] Step S1: Dam body zoning and parameter determination. Based on the actual material load conditions of the concrete dam in different regions, the dam body is divided into... i Each sub-region and dam foundation are j Each sub-region is selected, and the sequence of behavior parameters to be inverted in each region is chosen.

[0076] Multiple sets of input parameter combinations were extracted using the Latin hypercube sampling method. The structural response values ​​were calculated using a finite element model, and the numerical samples and measured data were fused to form a training dataset for the Physical Information Neural Network (PINNs). The parameter samples were then normalized; the parameter sample normalization formula is as follows:

[0077] ;

[0078] in, x For the sampling parameters that need to be normalized, i =1,2,…,n; x k,min =min( ) is the first k The minimum value of the data center of the dimensional variable sample. .

[0079] Step S2: Mathematical Model Construction. Based on the domain decomposition method, the mathematical equations for the structural response of the concrete dam in different zones are derived. The displacement vectors of the sub-regions are decomposed according to the internal and interface degrees of freedom. The stiffness matrix and load vector are processed in blocks to establish the control equations for the dam body and dam foundation sub-regions. The control equations for the dam body sub-regions are as follows:

[0080] ;

[0081] in, This represents the internal stiffness matrix within a sub-region of the dam body. This is the coupling stiffness matrix between internal nodes and the interface, used to connect internal nodes and interface nodes; This represents the coupling stiffness matrix between interface nodes in the sub-regions of the dam body. The displacement vector of the internal nodes; is the displacement vector of the node at the interface of the sub-region; This is the internal load vector; The interface load vector;

[0082] The governing equations for the dam foundation sub-region are:

[0083] ;

[0084] in, This represents the internal stiffness matrix of the dam foundation sub-region. This is the coupling stiffness matrix between the internal nodes and the interface; This represents the coupling stiffness matrix between interface nodes in the dam foundation sub-region; This represents the displacement vector of the nodes within the sub-region of the dam foundation. is the displacement vector of the node at the interface of the dam foundation sub-region; This represents the load vector within the sub-region of the dam foundation. This represents the interface load vector of the dam foundation sub-region.

[0085] A coarse space is constructed using the multi-scale finite element method, and the interface displacements are decomposed. The decomposition expression for the interface displacements in the sub-regions is as follows:

[0086] ;

[0087] ;

[0088] in, It is a coarse basis function matrix; It is a coarse-space coefficient vector, formed by global deformation; , These are the fine spatial components of the dam body and dam foundation interfaces, respectively.

[0089] Furthermore, by introducing displacement transformation matrices and Lagrange multipliers, the coordination and equilibrium equations for three types of interfaces—dam body-dam body, dam body-dam foundation, and dam foundation-dam foundation—are derived, and the global coordination equations for the concrete dam structure are obtained by combining these equations.

[0090] Step S3, PINNs Construction and Training: Based on the mathematical equations of the partitioned structural response and the stress and displacement boundary conditions of the concrete dam, a PINNs loss function is constructed, including boundary condition loss terms, intra-regional loss terms, and inter-regional interface loss terms. The PINNs network structure is configured, and a piecewise optimization strategy and dynamic learning rate are used to train the PINNs. The loss function for the boundary condition loss term is:

[0091] ;

[0092] ;

[0093] ;

[0094] ;

[0095] in, , , These are the weight values ​​for each loss term; This is the partial differential loss term; This represents the number of integration points within the entire computational domain. , They are respectively , The governing equations for direction; This is the loss term for displacement boundary conditions; This represents the number of displacement boundary nodes; , They are respectively , Directional displacement boundary conditions; This is the stress boundary condition loss term; This represents the number of stress boundary nodes; , They are respectively , Directional stress boundary conditions.

[0096] The loss function term for the intra-region loss term is:

[0097] ;

[0098] in, , To balance the weighting coefficients of the various loss terms by orders of magnitude, a trial-and-error algorithm was used to determine them. This is the term for fitting the total displacement. For physical constraints; This represents the regional displacement water pressure component loss term.

[0099] The inter-region interface loss term includes an interface displacement loss term and an interface stress loss term. The interface displacement loss term is as follows:

[0100] ;

[0101] in, , , These refer to the interfaces between dam bodies, between dam bodies and dam foundations, and between dam foundations, resulting from the zoning of the concrete dam structure. The number of discrete points on the interface; , These are the predicted displacement-water pressure components at the interface of the two sub-regions: the dam body and the dam foundation. , The PINN prediction full displacement vectors at the interface of the two sub-regions, the dam body and the dam foundation, are respectively. This is the interface node mapping matrix, used for interpolation to fit mismatched mesh interfaces;

[0102] The interface stress loss term is:

[0103] ;

[0104] in, , The PINN predicted stress tensor for the subregion at the interface; , This is the unit outward normal vector at the interface; This represents the force exerted by the sub-region on the interface, i.e., the surface force vector.

[0105] Configure the PINNs network structure with an input layer and hidden layers. The input layer is responsible for receiving the raw data and preset parameters required for the fitting problem, specifically including the spatial coordinates of the monitoring points. x , y , z ), based on the water level in front of the dam H External effects characterized by equivalent water pressure load P, and highly sensitive parameter sequences to be inverted ( X i , X j , X k ,….);

[0106] The hidden layers are set to 4-6 layers, and the hidden layers use the Swish function, which is as follows:

[0107] ;

[0108] in, Input vector or matrix; It is the sigmoid activation function; To adjust the parameters, the default value is 1; when using the Swish function, first calculate the sigmoid function, then combine it with... The output result of multiplication is:

[0109] .

[0110] The segmented optimization strategy first uses the Adam algorithm for initial training, which causes the loss function to decrease rapidly and stabilize; then it switches to the L-BFGS algorithm for further fine-tuning of the network parameters.

[0111] Use a low learning rate in the initial stage, and increase the learning rate after the warm-up process to avoid gradient explosion.

[0112] Step S4: Based on PINNs and using the improved Jaya algorithm, perform parameter optimization and output the inversion behavior parameters. This specifically includes the following sub-steps:

[0113] Step S401: Initialize the parameter population, list the regional behavior parameters to be identified and set their dimension as D according to their quantity, the initial population size as N, and set the search range of each regional behavior parameter as [[] based on the concrete dam design reference value and engineering practice experience. ] ;

[0114] Step S402: Randomly generate the initial population: ,in , r for Random numbers between;

[0115] Step S403: Select the measured displacement data and combine each parameter individually. Substitute into PINNs to calculate the objective function value Record the best individual in the current population. and the worst individual ;

[0116] Step S404: Update individuals using the individual update formula, optimize the population using mutation and crossover operations, and output the phenotype parameters to be inverted, specifically:

[0117] ;

[0118] in, As a new individual, r 1. r 2 is Random numbers between; new individual performance shows a tendency to move towards the best individual while moving away from the worst individual;

[0119] The mutation and crossover operation specifically involves: randomly updating new individuals. One dimension parameter in the equation is used to examine the fitness value of the new individual after crossover. If there is... If the original individual is replaced by the new individual, then the original individual is replaced by the new individual; otherwise, the original individual is retained.

[0120] Repeat the iteration until the optimal fitness value is less than a pre-set threshold or the number of iterations reaches the maximum number of iterations, and then output the behavior parameters to be identified.

[0121] Based on specific engineering examples, the specific implementation process of this invention is as follows:

[0122] (1) Determine the number of zones for the concrete dam and select the sequence of behavior parameters to be inverted. A concrete gravity dam in western China is selected as an engineering case study. The selected project is a concrete gravity dam with a maximum height of 66m, a crest length of 250.0m, a crest width of 9.2m, a normal water level of 2785m, and a design flood level of 2788m. The average annual temperature at the dam site is 3.0℃, the extreme maximum temperature is 28.6℃, and the extreme minimum temperature is -30.9℃. The locations of some monitoring instruments arranged inside the dam body are as follows: Figure 2 As shown, a vertical line system is installed on each of the two monitoring sections of dam #2 and #3. Each vertical line system consists of inverted and upright vertical lines and is used to monitor the displacement of the dam foundation and dam body.

[0123] The dam body is divided into upper and lower parts along the initial design interface. The lower part of the dam body is further divided into a heel region and a toe region (corresponding to Region 3 and Region 2, respectively). A region with a width of 1 / 3 of the dam base width is designated as Region 1. The dam foundation is divided into 3 sub-regions. The region division diagram and finite element model are shown below. Figure 3 As shown.

[0124] Based on practical engineering experience and literature review, the top 8 parameter sequences that contribute significantly to the deformation behavior of concrete dams were identified as follows: E 4. E 3. E 2. E 1. E 5. μ 1. μ 2. E 6. Use it as the inversion target.

[0125] (2) Select concrete dam monitoring data to establish a displacement statistical model; separate the water pressure component in the monitored deformation value and relative it to obtain the measured relative displacement value of each area center under the corresponding water pressure H on the monitoring date. Considering the eight parameter combinations to be inverted, the LHS method is used, and 600 sets of PINNs training data are extracted. Each partition's PINNs network is trained independently, and a global coordination equation is used to constrain the inter-block responses. The input layer parameters are 6-dimensional: representing the region center coordinates (X, Y, Z) and the converted external loads. P i Regional elastic modulus E i Compared to Poisson μ i The hidden layers are set to 3 layers, with 256, 128, and 64 neurons respectively. The Swish function is used as the activation function in all hidden layers. The output layer has 3 types of parameters. These represent the predicted regional displacement value, total displacement vector, and stress tensor, respectively, using PINNs. Fifty sampling points are set at the interface of each sub-region, and the displacement at these sampling points is used as the calculation point for the compatibility equation. The interface stress is constrained by the derivative of the displacement variable at the sampling points, achieving displacement constraint and transfer across the interval domain. The global loss function value and the regional loss function value are calculated, and interface compatibility verification is performed. The displacement values ​​at the interface sampling points are selected for verification to ensure interface compatibility. The interface displacement overlap threshold is set to 0.1 mm, and the stress imbalance threshold is set to 1 kPa. The compatibility verification of the loss function and interface displacement for each region is as follows: Figure 4 As shown.

[0126] (3) The improved Jaya algorithm combined with PINNs was used to carry out the inversion of regional structural parameters. The measured data selected were the 20 water pressure component values ​​corresponding to the dates when the water level in front of the dam was higher than 60m from 2011 to 2013. The parameter identification results were obtained by iterative calculation, as shown in Table 1. The identification process of each regional behavior parameter is as follows: Figure 5 As shown.

[0127] Table 1. Multi-parameter inversion results of concrete gravity dams

[0128]

[0129] (4) To verify the rationality and accuracy of the inversion results, the equivalent elastic modulus values ​​of the dam body and dam foundation were obtained by inversion using the traditional method, which were 23.54 GPa and 18.47 GPa, respectively. The inversion results obtained in this invention and the inversion results obtained by the traditional method were substituted into the finite element model of the concrete dam structure, and the water level values ​​of 11 monitoring dates in 2013 were selected as the working conditions for finite element calculation. The relative value of the horizontal displacement at monitoring point PL2-1 was obtained. The result was compared with the actual relative value of the horizontal displacement. Figure 6 As shown in (a) above, RMSE, R 2 Error coefficients such as MAPE and MAE are used as standards for error analysis to obtain error pairs. Figure 6 (b) Figure 6 As shown in (c) in the figure.

[0130] Depend on Figure 6 Analysis shows that after substituting the inversion results obtained by the two methods into the finite element model, the trend of the horizontal relative displacement values ​​is consistent with the measured values. This indicates that the behavior parameters obtained by the present invention are reasonable. Moreover, compared with the inversion results obtained by the traditional method, the horizontal relative displacement values ​​of the behavior parameters obtained by the present invention have a higher degree of fit with the measured relative values ​​after being substituted into the FEM model. The coefficients of determination of the present invention and the traditional method are 0.904 and 0.698, respectively, which indicates that the accuracy of the inversion results of the present invention is better than that of the inversion results of the traditional method.

[0131] Therefore, this invention adopts the above-mentioned intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints, which effectively reduces the difficulty of solving multi-objective inversion problems, realizes efficient and accurate inversion of the structural behavior parameters of gravity dam zones, and provides reliable technical support for the operation safety monitoring and safety management of concrete dams.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions 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 solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent identification of dynamic twin behavior parameters of high concrete dams under physical constraints, characterized in that, Includes the following steps: Step S1: Dam body zoning and parameter determination. Based on the actual materials of the concrete dam, the dam body-foundation structural system is divided into zones. i Each dam sub-area j The dam foundation sub-regions were identified, and the behavior parameters to be inverted in each region were determined. The Latin hypercube sampling method was used to uniformly sample the behavior parameters to be inverted. The response values ​​of the concrete dam structure under different combinations of behavior parameter sampling were calculated by combining the finite element model. The numerical samples and measured data were fused to form the training dataset of the physical information neural network PINNs, and the parameter samples were normalized. Step S2: Mathematical Model Construction. Based on the domain decomposition method, the mathematical equations for the structural response of the concrete dam in different zones are derived. The displacement vectors of the sub-regions are decomposed according to the internal and interface degrees of freedom. The stiffness matrix and load vector are processed in blocks to establish the control equations for the dam body and dam foundation sub-regions. The control equations for the dam body sub-regions are as follows: ； in, This represents the internal stiffness matrix within a sub-region of the dam body. This is the coupling stiffness matrix between internal nodes and the interface, used to connect internal nodes and interface nodes; This represents the coupling stiffness matrix between interface nodes in the sub-regions of the dam body. The displacement vector of the internal nodes; is the displacement vector of the node at the interface of the sub-region; This is the internal load vector; The interface load vector; The governing equations for the dam foundation sub-region are: ； in, This represents the internal stiffness matrix of the dam foundation sub-region. This is the coupling stiffness matrix between the internal nodes and the interface; This represents the coupling stiffness matrix between interface nodes in the dam foundation sub-region; This represents the displacement vector of the nodes within the sub-region of the dam foundation. is the displacement vector of the node at the interface of the dam foundation sub-region; This represents the load vector within the sub-region of the dam foundation. The interface load vector of the dam foundation sub-region; Step S3, PINNs Construction and Training: Based on the partitioned structural response mathematical equations and stress and displacement boundary conditions, construct a PINNs loss function containing boundary condition loss terms, intra-regional loss terms, and inter-regional interface loss terms; configure the PINNs network structure and train the PINNs model using a piecewise optimization strategy and dynamic learning rate. Step S4: Optimize parameters using the improved Jaya algorithm and output the inversion results of the concrete dam behavior parameters; specifically including the following sub-steps: Step S401: Initialize the parameter population, list the regional morphological parameters to be identified and set their dimension to D according to the quantity, and set the initial population size to N; Step S402: Randomly generate the initial population: ,in , r for Random numbers between; Step S403: Select the measured displacement data and combine each parameter. Substitute into PINNs to calculate the objective function value Record the best individual in the current population. and the worst individual ; Step S404: Update individuals using the individual update formula, optimize the population by combining mutation and crossover operations, and output the behavior parameters to be inverted; Step S404 is as follows: ； in, As a new individual, r 1. r 2 is Random numbers between; The mutation and crossover operation specifically involves: randomly updating new individuals. One dimension parameter in the equation is used to examine the fitness value of the new individual after the crossover. If there is a new individual with a positive objective function value... If the result is positive, replace the original individual with the new individual; otherwise, retain the original individual. Repeat the iteration until the optimal fitness value is less than the preset threshold or the number of iterations reaches the maximum number of iterations, and output the behavior parameters to be identified.

2. The intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints according to claim 1, characterized in that, In step S1, the parameter sample normalization formula is: ； in, For the sampling parameters that need to be normalized, i =1,2,…,n; x k,min =min( ), x k,min For the first k The minimum value of the sample data of the dimensional variable. .

3. The intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints according to claim 2, characterized in that, In step S2, a coarse space is constructed and the interface displacement is decomposed using the multi-scale finite element method. The decomposition expression for the interface displacement of the sub-region is as follows: ； ； in, It is a coarse basis function matrix; It is a coarse-space coefficient vector, formed by global deformation; , These are the fine spatial components of the dam body and dam foundation interfaces, respectively.

4. The intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints according to claim 3, characterized in that, In step S3, the loss function for the boundary condition loss term is: ； ； ； ； in, , , These are the weight values ​​for each loss term; This is the partial differential loss term; This represents the number of integration points within the entire computational domain. , They are respectively , The governing equations for direction; This is the loss term for displacement boundary conditions; This represents the number of displacement boundary nodes; , They are respectively , Directional displacement boundary conditions; This is the stress boundary condition loss term; This represents the number of stress boundary nodes; , They are respectively , Directional stress boundary conditions.

5. The intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints according to claim 4, characterized in that, In step S3, the loss function term for the loss term within the region is: ； in, , To balance the weighting coefficients of the various loss terms by orders of magnitude, a trial-and-error algorithm was used to determine them. This is the term for fitting the total displacement. For physical constraints; This represents the regional displacement water pressure component loss term.

6. The intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints according to claim 5, characterized in that, In step S3, the inter-regional interface loss term includes an interface displacement loss term and an interface stress loss term. The interface displacement loss term is: ； in, , , These refer to the interfaces between dam bodies, between dam bodies and dam foundations, and between dam foundations, resulting from the zoning of the concrete dam structure. The number of discrete points on the interface; , These are the predicted displacement-water pressure components at the interface of the two sub-regions: the dam body and the dam foundation. , The PINN prediction full displacement vectors at the interface of the two sub-regions, the dam body and the dam foundation, are respectively. This is the interface node mapping matrix, used for interpolation to fit mismatched mesh interfaces; The interface stress loss term is: ； in, , The PINN predicted stress tensor for the subregion at the interface; , This is the unit outward normal vector at the interface.

7. The intelligent identification method for dynamic twin behavior parameters of high concrete dams under physical constraints according to claim 6, characterized in that, In step S3, the PINNs network structure is configured with an input layer and hidden layers. The input layer is responsible for receiving the raw data and preset parameters required for the fitting problem, specifically including the spatial coordinates of the monitoring points. x , y , z ), based on the water level in front of the dam H External effects characterized by equivalent water pressure load P, and highly sensitive parameter sequences to be inverted ( X i , X j , X k ,….); The hidden layers are set to 4-6 layers, and the hidden layers use the Swish function, which is as follows: ； in, Input vector or matrix; It is the sigmoid activation function; To adjust the parameters, the default value is 1; when using the Swish function, first calculate the sigmoid function, then combine it with... The output result of multiplication is: 。

Images