CEDAs-based rock-fill dam deformation analysis model parameter updating method and system

Abstract

The invention discloses a rock-fill dam deformation analysis model parameter updating method based on CEDAs, which is characterized in that a multivariate joint distribution relation of model parameters is extracted through triaxial experiment parameters of a plurality of groups of rockfill materials, and a mode of combining a machine learning agent model and a meta-heuristic optimization algorithm is adopted to update the parameters of a rock-fill dam deformation analysis model. More reasonable and reliable deformation analysis model parameters are intelligently and dynamically searched, so that the deformation analysis and prediction precision of the rock-fill dam is improved, and the deformation safety of the rock-fill dam in the full life cycle is guaranteed. According to the model parameter updating method provided by the invention, the optimal model calculation parameter combination can be accurately identified, and the precision of the deformation analysis model is effectively improved when the monitoring data is insufficient.

Description

Technical Field

[0001] The present invention relates to the technical fields of hydraulic engineering and geotechnical engineering, and particularly relates to a method and system for updating parameters of a rockfill dam deformation analysis model based on CEDAs. Background Art

[0002] Accurate and reliable deformation analysis of rockfill dams is crucial for ensuring deformation safety during their entire life cycle. Model parameter updating is an important method to improve the accuracy of rockfill dam deformation analysis. Although geotechnical parameters usually show a certain correlation, previous studies often assumed that a single model parameter follows a normal distribution or directly ignored prior knowledge. This may lead to low fidelity of parameter updating methods and low accuracy of deformation analysis, further restricting applications. Summary of the Invention

[0003] To solve the problems of low fidelity of parameter updating methods and low accuracy of deformation analysis, the present invention provides a method for updating parameters of a rockfill dam deformation analysis model based on CEDAs, which identifies the optimal model parameters by combining the Copula-based Estimation of Distribution Algorithm (EDAs) with a stacked machine learning surrogate model to improve the accuracy of rockfill dam simulation.

[0004] According to one aspect of the specification of the present invention, a method for updating parameters of a rockfill dam deformation analysis model based on CEDAs is provided, including:

[0005] Obtaining experimental parameters of the Duncan-Chang E-B model of the rockfill dam project;

[0006] Based on the obtained experimental parameters, using the Gaussian copula function to construct a copula multivariate joint distribution model of the Duncan-Chang E-B model parameters;

[0007] Based on the constructed copula multivariate joint distribution model, generating parameter combination samples by Latin hypercube sampling and substituting them into the finite element deformation analysis model to obtain finite element calculation samples;

[0008] Based on the stacked machine learning model technology, constructing a mapping relationship between the parameter combination and the deformation analysis result to replace the finite element calculation in iterative optimization;

[0009] Combining the copula multivariate joint distribution model, the stacked surrogate model and CEDAs to identify the optimal model calculation parameters and complete model updating.

[0010] As a further technical solution, constructing a copula multivariate joint distribution model of the Duncan-Chang E-B model parameters includes:

[0011] Identification of the optimal marginal distribution function through the Akaike information criterion;

[0012] Construct a copula multivariate joint distribution model of model parameters by using the Gaussian copula function and combining the optimal marginal distribution functions of each parameter.

[0013] As a further technical solution, the method further includes:

[0014] Perform random sampling on the constructed Gaussian copula function to generate samples.

[0015] As a further technical solution, generate parameter combination samples through Latin hypercube sampling and substitute them into the finite element deformation analysis model to obtain finite element calculation samples, including:

[0016] Assume that there are a parameters in the target space for model parameter update: (1) Divide the dimension corresponding to each parameter into m non-overlapping intervals so that the probability of each interval is the same; (2) Randomly select a point in each interval in the dimension of each parameter; (3) Randomly select the points drawn in step (2) from the dimension of each parameter and obtain the sampled parameter combination; (4) Repeat steps (1) to (3) until k groups of parameter combinations are generated;

[0017] Substitute the obtained k groups of parameter combinations into the finite element analysis model for calculation.

[0018] As a further technical solution, combine the copula multivariate joint distribution model, the stacked surrogate model and CEDAs to identify the optimal model calculation parameters and complete model update, including:

[0019] Construct the following optimization function required by CEDAs:

[0020]

[0021] In the formula: f(X) is the objective function, X is a vector composed of model parameters {x1,…,x m}, and satisfies [X min ≤X≤X max ; n is the number of deformation monitoring points participating in model update; TSF i and are the time series characteristics of the finite element analysis and monitoring data of the i-th measurement point respectively, that is, the deformation analysis result;

[0022] When using CEDAs to iteratively optimize the set objective function, the deformation analysis result is generated by the trained stacked surrogate model each time, and the optimization function is rewritten as:

[0023]

[0024] wherein: is the output of the trained stacked surrogate model;

[0025] The improved CEDAs are used for iterative optimization. An initial population is generated by an initial copula model constructed from real data, and high-fidelity individuals are added in each iteration.

[0026] As a further technical solution, the improved CEDAs are used for iterative optimization, including:

[0027] Collect and process real data;

[0028] Based on the processed real data, construct an initial Gaussian copula model;

[0029] Generate an initial population within the search space through the Gaussian copula model;

[0030] When any termination condition is not met, perform the following operations:

[0031] Evaluate and rank all individuals within the population;

[0032] According to the truncation coefficient, divide the population into dominant individuals and inferior individuals;

[0033] Based on the inferior individuals, construct a rebel Gaussian copula model;

[0034] Generate rebel individuals based on the rebel Gaussian copula model;

[0035] Generate high-fidelity individuals based on the initial Gaussian copula model;

[0036] Combine the three types of individuals into the next generation population: dominant individuals, rebel individuals, and high-fidelity individuals;

[0037] The iteration terminates and the optimal solution is output.

[0038] As a further technical solution, based on the stacked machine learning model technology, construct a mapping relationship between parameter combinations and deformation analysis results, including:

[0039] Use the stacking method to integrate multiple machine learning models;

[0040] Use Extreme Gradient Boosting (XGBoost), Gaussian Process Regression (GPR), and Multilayer Perceptron (MLP) as base models, Gradient Boosting Decision Tree (GBDT) as the final model, and use random search for hyperparameter optimization to enable the stacked surrogate model to perform regression training on parameter combinations and time series.

[0041] According to one aspect of the specification of the present invention, there is provided a system for updating parameters of a rockfill dam deformation analysis model based on CEDAs, including:

[0042] A first main module for obtaining experimental parameters of the Duncan-Chang E-B model of the rockfill dam project;

[0043] A second main module for constructing a copula multivariate joint distribution model of the Duncan-Chang E-B model parameters by using the Gaussian copula function based on the obtained experimental parameters;

[0044] A third main module for generating parameter combination samples by Latin hypercube sampling based on the constructed copula multivariate joint distribution model and substituting them into the finite element deformation analysis model to obtain finite element calculation samples;

[0045] A fourth main module for constructing a mapping relationship between parameter combinations and deformation analysis results based on the stacked machine learning model technology to replace the finite element calculation in iterative optimization;

[0046] A fifth main module for combining the copula multivariate joint distribution model, the stacked surrogate model and CEDAs to identify the optimal model calculation parameters and complete model update.

[0047] According to one aspect of the specification of the present invention, there is provided a device for updating parameters of a rockfill dam deformation analysis model based on CEDAs, including a memory and a processor. The memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the method for updating parameters of the rockfill dam deformation analysis model based on CEDAs.

[0048] According to one aspect of the specification of the present invention, there is provided a non-transitory computer-readable storage medium. The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the method for updating parameters of the rockfill dam deformation analysis model based on CEDAs.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0050] (1) Prior knowledge support. This method comprehensively integrates multi-source information. The key innovation is to extract the multivariate joint distribution of experimental parameters through copula and fuse it into model parameter update as prior knowledge to guide model parameter update. And it improves CEDAs so that it can maintain the high fidelity and diversity of the population throughout the iterative calculation process. This method can effectively identify model parameters with both high fidelity and high precision, improving the accuracy of rockfill dam deformation analysis. To our knowledge, this is the first time to fuse parameter correlation as prior knowledge in the parameter update method of the numerical simulation model of the dam structure.

[0051] (2) Intelligence and dynamics. This method effectively integrates the machine learning surrogate model and the intelligent optimization algorithm. By stacking machine learning models to construct the mapping relationship between parameter combinations and deformation analysis results, it can effectively combine the advantages of different types of machine learning models, enabling the surrogate model to have high precision, high robustness, and strong generalization ability. By combining the improved CEDAs with the stacked surrogate model, it can dynamically identify the optimal model parameters, and the total time for single inversion and deformation analysis is within 2 hours. Description of the Drawings

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings used in the description of the embodiments or the prior art. Obviously, the following-described drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0053] Figure 1 Flowchart of a method for updating the parameters of a rockfill dam deformation analysis model based on CEDAs provided by an embodiment of the present invention;

[0054] Figure 2 Recognition result diagram of the marginal distribution of AIC according to an embodiment of the present invention;

[0055] Figure 3 511 experimental parameters and 1000 sampling parameters from the constructed copula model according to an embodiment of the present invention;

[0056] Figure 4 Model prediction effect diagram of stacking multiple machine learning models such as GPR, MLP, and XGBoost according to an embodiment of the present invention;

[0057] Figure 5 Convergence of CEDAs based on different truncation coefficients β according to an embodiment of the present invention;

[0058] Figure 6 Scatter plot of the initial population of parameters K and Kb and experimental data obtained by different construction methods according to an embodiment of the present invention (ρ = 0.62);

[0059] Figure 7 Settlement cloud diagram of the L0+009.200 section at the completion period (June 2019) of the Jiangpinghe rockfill dam project according to an embodiment of the present invention;

[0060] Figure 8 Deformation time history curve diagram of the characteristic measuring point obtained after model update according to an embodiment of the present invention (measuring point number: SG3b-04(A));

[0061] Figure 9After the model update for the embodiments of the present invention, it is the deformation time history curve graph of the characteristic measurement points obtained (measurement point number: SG1d-04(A)). Specific embodiments

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. In addition, the technical features in each embodiment or individual embodiment provided by the present invention can be combined with each other arbitrarily to form a new technical solution. Such combination is not restricted by the order of steps and / or the mode of structural composition, but must be based on what can be achieved by those of ordinary skill in the art. When the combination of technical solutions results in contradictions or cannot be achieved, it should be considered that such combination of technical solutions does not exist and is not within the protection scope required by the present invention.

[0063] The present invention proposes a method for updating the parameters of a high-fidelity rockfill dam deformation analysis model, using parameter correlation as prior knowledge to guide the model parameter update. The present invention uses a Copula-based estimation of distribution algorithm (EDAs) combined with a stacked machine learning surrogate model to identify the optimal model parameters. Specifically, multiple groups of indoor triaxial test parameters collected from rockfill dam projects are used to construct a Gaussian copula model to extract the multivariate joint distribution between model parameters; by sampling the copula model, an initial population is created, and high-fidelity individuals are added generation by generation for iterative optimization to identify model parameters with both high accuracy and high fidelity; the rockfill dam deformation analysis based on the updated parameters can accurately reflect the nonlinear evolution and distribution pattern of the dam body deformation. The finite element results obtained by calculating the deformation analysis model updated by the method proposed by the present invention are in good agreement with the monitoring values, effectively improving the prediction, evaluation, and control levels of rockfill dam deformation.

[0064] A method for updating the parameters of a rockfill dam deformation analysis model based on CEDAs provided by the present invention, as Figure 1 The main steps include:

[0065] S1: Collect the experimental parameters of the Duncan-Chang E-B model obtained from indoor triaxial tests of the rockfill dam project from engineering technical reports or literature.

[0066] In the step S1, the sources of the experimental parameters of the Duncan-Chang E-B model include but are not limited to engineering technical reports and literature, and may also include other relevant materials of other geotechnical engineering accumulated by the experimental unit and other channels.

[0067] S2: Use the Gaussian copula function to construct a copula multivariate joint distribution model for the Duncan-Chang E-B model parameters. In step S2, the multivariate joint distribution of the Duncan-Chang E-B model parameters is mainly extracted, and the main steps include:

[0068] S2.1: Identify the optimal marginal distribution function through the Akaike Information Criterion (AIC). The alternative marginal distribution functions include Beta, Gaussian, Gaussian KDE, Student t, Gamma, and truncated Gaussian, etc. The calculation method of AIC is as shown in formula (1):

[0069]

[0070] In the formula: x i (i = 1,..., N) are the experimental parameters of the Duncan-Chang E-B model, and N is the number of experimental parameter combinations; f(x; p, q) is the probability density function of the selected marginal distribution, and p and q are the parameters of the corresponding distribution; k1 is the number of selected marginal distribution functions;

[0071] S2.2: Then, through the Gaussian copula function, combined with the optimal marginal distribution functions of each parameter, construct a multivariate joint distribution model of the model parameters. Among them, constructing the parameter multivariate joint distribution through the Gaussian copula function mainly estimates the linear correlation matrix of the Gaussian copula function by parametric or non-parametric methods. The multivariate Gaussian copula function C G (·,...,·; θ) is constructed as follows:

[0072] C G (x1,..., x n ; θ) = Φ n (Φ -1 (x1),..., Φ -1 (x n ); θ) * (2)

[0073] In the formula, Φ n (·,...,·; θ) is the standard n-dimensional multivariate normal distribution function, and its linear correlation coefficient matrix is Φ is the standard normal distribution function, and Φ -1 is its inverse function.

[0074] S2.3: Randomly sample the constructed Gaussian copula function to generate samples. The specific steps are as follows:

[0075] (1) Use the Cholesky decomposition method to calculate the matrix Σ ∈ R n×n , and this matrix satisfies ΣΣ T= θ, ∑ is a lower triangular matrix. And ∑ can be replaced by θ 1 / 2 where θ 1 / 2 satisfies θ 12 θ 12 = θ;

[0076] (2) Generate a vector consisting of independent random variables from a uniform distribution and then apply the inverse cumulative distribution function (CDF) of the standard normal distribution to each element in to obtain the vector (3) Obtain a vector with normally correlated elements

[0077] (3) Then apply the cumulative distribution function of the standard normal distribution to each element of<> i.e.,

[0078] (4) Apply the inverse transformation of the marginal distribution to each element of the vector(4) Apply the inverse transformation of the marginal distribution to each element of the vector to calculate the sample vector i.e.,

[0079] (5) Repeat steps (2) to (4) until n samples are generated.(5) Repeat steps (2) to (4) until n samples are generated.

[0080] S3: Generate parameter combination samples through Latin hypercube sampling and substitute them into the finite element deformation analysis model to obtain finite element calculation samples.

[0081] In the step S3, parameter combination samples are generated through Latin hypercube sampling and substituted into the finite element deformation analysis model to obtain samples for surrogate model training. The specific steps are as follows:

[0082] S3.1: The method of approximate random sampling in the Latin hypercube sampling of multivariate parameter distributions belongs to the stratified sampling technique. Assume that there are a parameters in the target space for model parameter update

[0083] (1) Divide the dimension corresponding to each parameter into m non-overlapping intervals so that the probability of each interval is the same.

[0084] (2) Randomly select a point in each interval in the dimension of each parameter;

[0085] (3) Then randomly select the points drawn in step (2) from the dimension of each parameter and obtain the parameter combination of the sampling.

[0086] (4) Repeat steps (1) to (4) until k groups of parameter combinations are generated;

[0087] S3.2: Substitute the obtained k groups of parameter combinations into the finite element analysis model for calculation. The models for realizing finite element analysis include, but are not limited to, the instantaneous deformation constitutive model mainly based on the Duncan-Chang E-B model, the rheological constitutive model mainly based on the South-to-North Water seven-parameter model, and the wetting constitutive model mainly based on the Shen Zhujiang three-parameter wetting model. In addition, the software for deformation analysis includes, but is not limited to, commercial software represented by ABAQUS and ANSYS, etc., and self-developed finite element calculation software.

[0088] S4: Based on the stacked machine learning model technology, construct the mapping relationship between the parameter combination and the deformation analysis result to replace the finite element calculation in the iterative optimization.

[0089] In step S4, through the stacked machine learning model technology, integrate multiple machine learning models, establish the mapping relationship between the parameter combination and the deformation analysis result, and replace the finite element calculation in the CEDAs optimization. Specifically:

[0090] Use the stacking method to integrate multiple machine learning models;

[0091] By using XGBoost, GPR, and MLP as the base models, GBDT as the final model, and performing hyperparameter optimization by using the random search method, make the integrated surrogate model perform regression training on the parameter combination and the time series.

[0092] S5: Combine the copula multivariate joint distribution model, the stacked surrogate model, and CEDAs to identify the optimal model calculation parameters and complete the model update.

[0093] In step S5, combine the copula multivariate joint distribution model, the stacked surrogate model, and CEDAs, and accurately identify the model calculation parameters through iterative optimization. The specific steps are as follows:

[0094] S5.1: Construct the optimization function required by CEDAs, as shown in the following formula:

[0095]

[0096] In the formula: f(X) is the objective function, X is a vector composed of model parameters {x1,..., x m}, and satisfies [X min ≤ X ≤ X max ; n is the number of deformation monitoring points participating in the model update; TSF i and are the time series characteristics of the finite element analysis and the monitoring data of the i-th measurement point respectively, that is, the deformation analysis result.

[0097] S5.2: When iteratively optimizing the objective function set in Equation (3) using CEDAs, the deformation analysis results are generated by the trained stacked surrogate model each time, so as to greatly reduce the computational burden. Equation (3) can be rewritten as:

[0098]

[0099] Where: is the output of the trained stacked surrogate model, and TSF in Equation (3) i is replaced.

[0100] S5.3: When using the improved CEDAs for iterative optimization, its calculation method can be seen in the pseudocode in Table 1:

[0101]

[0102] S5.4: Different from the conventional CEDAs, the improved CEDAs is characterized in that: the initial population is generated by the initial copula model constructed by real data, and high-fidelity individuals are added in each iteration to ensure the diversity of the algorithm and the high fidelity of individuals.

[0103] The core of the present invention is to extract the multivariate joint distribution relationship of model parameters through the triaxial experimental parameters of multiple groups of rockfill materials, and adopt the method of combining machine learning surrogate model and meta-heuristic optimization algorithm to intelligently and dynamically search for more reasonable and reliable deformation analysis model parameters, so as to improve the deformation analysis and prediction accuracy of rockfill dams and ensure the deformation safety of rockfill dams throughout their life cycle. Specifically, collect the triaxial experimental parameters of the Duncan-Chang E-B model from different rockfill dam projects, and extract the multivariate joint distribution among model parameters through the Gaussian copula function. Use stacked machine learning technology to construct the mapping relationship between parameter combinations and deformation prediction results, and reduce the computational load of multiple finite element analyses in the optimization algorithm. And improve the copula-based estimation of distribution algorithm (CEDAs), randomly sample the constructed copula model to generate the initial population of CEDAs, and add high-fidelity individuals generation by generation. Combine the stacked surrogate model with the improved CEDAs to search for the optimal model calculation parameters generation by generation, and obtain model parameters with both high fidelity and high precision. The research results show that the proposed model parameter update method can accurately identify the optimal model calculation parameter combination and effectively improve the accuracy of the deformation analysis model when the monitoring data is insufficient.

[0104] Example 1

[0105] Example 1 uses a method for updating the parameters of the deformation analysis model of a rockfill dam based on CEDAs provided by the present invention to update the deformation analysis model of the Jiangpinghe concrete face rockfill dam.

[0106] 1. Search and obtain 511 groups of indoor triaxial test parameter combinations of the Duncan-Chang E-B model from the literature and engineering technical reports related to 48 rockfill dams, as the data basis for constructing the copula model;

[0107] 2. Through the SHAP (SHapley Additive exPlanations) interpretable machine learning method, obtain the influence degree of model parameters on the model output, and accordingly determine the parameters to be updated, namely the parameters K, n, K b , m in the Duncan-Chang E-B model, the parameter α of the exponential decay empirical creep model, and the parameters c w , n w and b w ;

[0108] 3. Since the experimental parameters of the Duncan-Chang E-B model are very rich, construct the multivariate joint distribution of its model parameters through the Gaussian copula function. The results prove that the correlation between the parameters K and K b , n and m is the most significant, and their linear correlation coefficients ρ are 0.80 and 0.64 respectively( Figure 2 and Figure 3 );

[0109] 4. Generate 300 groups of parameter combination samples through Latin hypercube sampling and substitute them into the finite element model for calculation to obtain training samples;

[0110] 5. Based on the stacked machine learning model technology, construct the mapping relationship between the parameter combination and the deformation analysis result, and the model prediction accuracy is excellent( Figure 4 );

[0111] 6. Combine the stacked surrogate model, the initial copula model and CEDAs for iterative optimization. Since the truncation coefficient β determines the ratio of rebel individuals to high-fidelity individuals during population iteration, design experiments to explore. The results prove that when β = 0.3, the iterative optimization result is the best( Figure 5 );

[0112] 7. To verify that the model parameter update method proposed in the present invention can maintain the high fidelity of the population, assume that a single parameter follows a normal distribution or a uniform distribution during iteration, and compare and display the constructed initial population. The sampling results of the parameters K and K b prove that the initial population obtained based on the method proposed in the present invention can not only effectively fill the search space, but also ensure that the parameters have high fidelity( Figure 6 );

[0113] 8. To verify that the present invention can stably support the identification of optimal model parameters by fusing parameter correlations, thereby ensuring the deformation safety of the rockfill dam throughout its life cycle, the number of measuring points was randomly reduced, and the effects of model update schemes with and without prior knowledge guidance were compared. The specific experimental settings are shown in Table 2. Further, based on the parameters obtained from different experimental settings, a finite element deformation analysis of Jiangping River was carried out, and the settlement contour map ( Figure 7 ) at the completion stage (June 2019) of the L0+009.200 interface was compared, and the time series curves of the calculated and monitored settlements of typical measuring points SG3b-04(A) and SG1d-04(A) were extracted for comparison ( Figure 8 and Figure 9 ). The deformation simulation results based on the present invention are significantly more accurate than other methods.

[0114] 9. In a rockfill dam, due to the limitations of monitoring technology and the complexity of the monitoring environment, after years of operation, the deformation monitoring instruments are significantly damaged. This leads to a large reduction in monitoring data. Table 3 shows the comparison of relative errors of some measuring points. When updating parameters using 12, 6, and 3 measuring points, the relative errors are reduced by 4.1%, 3.8%, and 6.2% respectively. Since the data basis of parameter correlation comes from multiple rockfill dams and there is no data missing phenomenon, the improvement effect on the accuracy of deformation analysis and prediction is relatively stable.

[0115] Table 2 Experimental settings

[0116]

[0117] Table 3 Comparison of relative errors of some measuring points

[0118]

[0119] The implementation basis of each embodiment of the present invention is achieved through programmed processing by a device with processor functions. Therefore, in engineering practice, the technical solutions and functions of each embodiment of the present invention are encapsulated into various modules. Based on this actual situation, on the basis of the above embodiments, an embodiment of the present invention provides a system for updating parameters of a rockfill dam deformation analysis model based on CEDAs, which is used to execute the method for updating parameters of a rockfill dam deformation analysis model based on CEDAs in the above method embodiments.

[0120] The system includes: a first main module for obtaining experimental parameters of the Duncan-Chang E-B model for rockfill dam projects; a second main module for constructing a copula multivariate joint distribution model of the Duncan-Chang E-B model parameters by using the Gaussian copula function based on the obtained experimental parameters; a third main module for generating parameter combination samples through Latin hypercube sampling based on the constructed copula multivariate joint distribution model and substituting them into the finite element deformation analysis model to obtain finite element calculation samples; a fourth main module for constructing a mapping relationship between parameter combinations and deformation analysis results based on the stacked machine learning model technology to replace the finite element calculation in iterative optimization; and a fifth main module for combining the copula multivariate joint distribution model, the stacked surrogate model, and CEDAs to identify the optimal model calculation parameters and complete model update.

[0121] The system for updating parameters of the rockfill dam deformation analysis model based on CEDAs provided by the embodiments of the present invention aims at the problems of low fidelity of the parameter update method and low accuracy of deformation analysis. By using the foregoing several modules, the best model parameters are identified through the Copula-based distribution estimation algorithm CEDAs combined with the stacked machine learning surrogate model to improve the simulation accuracy of the rockfill dam.

[0122] It should be noted that the system embodiments provided by the present invention are used not only to implement the methods in the above method embodiments, but also to implement the methods in other method embodiments provided by the present invention. The difference is only in setting the corresponding functional modules. The principle is basically the same as that of the above system embodiments provided by the present invention. As long as those skilled in the art, based on the above system embodiments, refer to the specific technical solutions in other method embodiments, obtain the corresponding technical means by combining technical features, and the technical solutions composed of these technical means, and on the premise of ensuring the practicability of the technical solutions, improve the modules in the above system embodiments to obtain the corresponding system-like embodiments for implementing the methods in other method-like embodiments.

[0123] Based on the same inventive concept as the foregoing embodiments, the embodiments of the present invention further provide a device for updating parameters of the rockfill dam deformation analysis model based on CEDAs, including a memory and a processor. The memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the method for updating parameters of the rockfill dam deformation analysis model based on CEDAs.

[0124] Based on the same inventive concept as the foregoing embodiments, the embodiments of the present invention further provide a non-transitory computer-readable storage medium. The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the method for updating parameters of the rockfill dam deformation analysis model based on CEDAs.

[0125] In summary of the above embodiments, the embodiment of the present invention adopts a method for updating the parameters of a rockfill dam deformation analysis model based on CEDAs. The main steps include: collecting experimental parameters from multiple sources, constructing a Gaussian copula multivariate joint distribution model to extract the parameter correlation of the model. Through the LHS method, generating parameter combination samples and performing finite element analysis to obtain training samples, and constructing a mapping relationship between parameters and deformation results based on a stacked machine learning model. Improving CEDAs, introducing a new truncation coefficient β and high-fidelity individuals to keep the population with high fidelity. Combining the copula model, the stacked surrogate model and CEDAs to optimize and identify the optimal model parameters, so as to improve the accuracy of rockfill dam deformation analysis. The present invention has the advantages of being dynamic and intelligent, and can accurately identify model parameters with the advantages of high fidelity and high precision. By updating the model parameters of the Jiangpinghe concrete face rockfill dam, the accuracy of the rockfill dam deformation analysis is significantly improved, the analysis results are accurate and reasonable, and the degree of agreement with the actual monitoring results is high. The examples of the present invention improve the safety status assessment level of the rockfill dam, promote the development of the rockfill dam deformation analysis and prediction method, and can contribute to the deformation safety assessment and control of the rockfill dam.

[0126] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0127] The terms "including" and "having" and any variations thereof in the specification and claims of the present invention and the above drawings are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for updating parameters of a deformation analysis model of a rockfill dam based on CEDAs, characterized in that, Including: Obtain the experimental parameters of the Duncan-Chang E-B model for the rockfill dam project; Based on the obtained experimental parameters, use the Gaussian copula function to construct a copula multivariate joint distribution model of the Duncan-Chang E-B model parameters; Based on the constructed copula multivariate joint distribution model, generate parameter combination samples through Latin hypercube sampling and substitute them into the finite element deformation analysis model to obtain finite element calculation samples; Based on the stacked machine learning model technology, construct the mapping relationship between the parameter combination and the deformation analysis result to replace the finite element calculation in the iterative optimization; Combine the copula multivariate joint distribution model, the stacked surrogate model and CEDAs to identify the optimal model calculation parameters and complete the model update.

2. The method for updating the parameters of the rockfill dam deformation analysis model based on CEDAs according to claim 1, characterized in that Construct a copula multivariate joint distribution model of the Duncan-Chang E-B model parameters, including: Identify the optimal marginal distribution function through the Akaike information criterion; Through the Gaussian copula function, combine the optimal marginal distribution functions of each parameter to construct a copula multivariate joint distribution model of the model parameters.

3. The method for updating parameters of the rockfill dam deformation analysis model based on CEDAs according to claim 2, wherein The method further includes: Perform random sampling on the constructed Gaussian copula function to generate samples.

4. The method for updating the parameters of the rockfill dam deformation analysis model based on CEDAs according to claim 1, characterized in that Generate parameter combination samples through Latin hypercube sampling and substitute them into the finite element deformation analysis model to obtain finite element calculation samples, including: Assume that there are a parameters in the target space of model parameter update: (1) Divide the dimension corresponding to each parameter into m non-overlapping intervals so that the probability of each interval is the same; (2) Randomly select a point in each interval in the dimension of each parameter; (3) Randomly select the points extracted in step (2) from the dimension of each parameter and obtain the sampled parameter combination; (4) Repeat steps (1) to (3) until k groups of parameter combinations are generated; Substitute the obtained k groups of parameter combinations into the finite element analysis model for calculation.

5. The method for updating the parameters of the rockfill dam deformation analysis model based on CEDAs according to claim 1, characterized in that Combine the copula multivariate joint distribution model, the stacked surrogate model and CEDAs to identify the optimal model calculation parameters and complete the model update, including: Construct the following optimization function required by CEDAs: Where: f(X) is the objective function, X is a vector composed of model parameters {x1, …, x m}, and satisfies [X min ≤ X ≤ X max ; n is the number of deformation monitoring points participating in model update; TSF i and TSF i * are the time series features of the finite element analysis and monitoring data of the i-th measurement point, respectively, that is, the deformation analysis results; When using the objective function set by CEDAs for iterative optimization, the deformation analysis result is generated by the trained stacked surrogate model each time, and the optimization function is rewritten as: where: TSF i Sur is the output of the trained stacked surrogate model; Use the improved CEDAs for iterative optimization, generate the initial population through the initial copula model constructed by real data, and add high-fidelity individuals in each iteration.

6. The method for updating the parameters of the rockfill dam deformation analysis model based on CEDAs according to claim 5, wherein Use the improved CEDAs for iterative optimization, including: Collect and process real data; Based on the processed real data, construct an initial Gaussian copula model; Generate the initial population within the search space through the Gaussian copula model; When any termination condition is not met, perform the following operations: Evaluate and rank all individuals in the population; According to the truncation coefficient, divide the population into dominant individuals and inferior individuals; Based on the inferior individuals, construct a rebel Gaussian copula model; Based on the rebel Gaussian copula model, generate rebel individuals; Based on the initial Gaussian copula model, generate high-fidelity individuals; Combine three types of individuals into the next-generation population: dominant individuals, rebellious individuals, and high-fidelity individuals; The iteration terminates, and the optimal solution is output.

7. The method for updating the parameters of the rockfill dam deformation analysis model based on CEDAs according to claim 1, characterized in that, Based on the stacked machine learning model technology, construct the mapping relationship between parameter combinations and deformation analysis results, including: Use the stacking method to integrate multiple machine learning models; Use Extreme Gradient Boosting XGBoost, Gaussian Process Regression GPR, and Multi-Layer Perceptron MLP as the base models, Gradient Boosting Decision Tree GBDT as the final model, and use random search to optimize hyperparameters, so that the stacked surrogate model performs regression training on parameter combinations and time series.

8. The parameter updating system for the deformation analysis model of rockfill dams based on CEDAs, characterized in that, Including: The first main module is used to obtain the experimental parameters of the Duncan-Chang E-B model of the rockfill dam project; The second main module is used to construct a copula multivariate joint distribution model of the Duncan-Chang E-B model parameters by using the Gaussian copula function based on the obtained experimental parameters; The third main module is used to generate parameter combination samples by Latin hypercube sampling based on the constructed copula multivariate joint distribution model, and substitute them into the finite element deformation analysis model to obtain finite element calculation samples; The fourth main module is used to construct the mapping relationship between parameter combinations and deformation analysis results based on the stacked machine learning model technology, replacing the finite element calculation in the iterative optimization; The fifth main module is used to combine the copula multivariate joint distribution model, the stacked surrogate model, and CEDAs to identify the optimal model calculation parameters and complete the model update.

9. A device for updating parameters of a deformation analysis model of a rockfill dam based on CEDAs, characterized in that, It includes a memory and a processor. The memory stores program instructions executed by the processor. The processor calls the program instructions to execute the method for updating the parameters of the rockfill dam deformation analysis model based on CEDAs according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the method for updating the parameters of the rockfill dam deformation analysis model based on CEDAs according to any one of claims 1 to 7.

Images