Elastic-plastic parameter intelligent inversion method and system based on cloud theory expansion

Abstract

The invention discloses an elastic-plastic parameter intelligent inversion method and system based on cloud theory expansion, and belongs to the field of hydraulic engineering intelligent computation.The inversion method comprises the following specific steps that I, an earth and rockfill dam three-dimensional finite element model is established, an elastic-plastic constitutive model is integrated, and meanwhile a monitoring point is selected as an inversion response reference point; iI, acquiring to-be-identified parameters and a change range of the elastic-plastic constitutive model, and constructing a high-dimensional parameter space; according to the method, a stable high-precision inversion effect can be kept in a high-dimensional parameter space, the problems that a traditional method is low in calculation efficiency and prone to falling into local optimum in a complex parameter space are solved, it is guaranteed that the average error of static inversion is lower than 5%, the goodness of fit between dynamic response prediction and actually measured data reaches 90% or above, and the method is suitable for large-scale popularization and application. And the precision and the efficiency of the full-life-cycle safety evaluation of the earth and rockfill dam are obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent calculation of hydraulic engineering, and particularly relates to an elastoplastic parameter intelligent inversion method and system based on cloud theory expansion. BACKGROUND

[0002] As an important water conservancy project, high earth-rock dam has high structure height, complex filling material and variable working environment. Nonlinear evolution phenomena such as settlement, seepage or material degradation are prone to occur during long-term service, which seriously threatens the overall safety of the dam. Especially with the continuous expansion of the project scale and the increasing complexity of the structure form, the long-term deformation response is controlled by multiple parameters, and under extreme load or earthquake action, it may even cause dam instability, sliding or dam failure, etc. Major accidents may cause extremely serious personnel casualties and property losses. Therefore, accurately grasping the mechanical response characteristics of high earth-rock dam and building a stable and efficient constitutive model parameter identification system are of great practical significance for ensuring the safety of project operation and improving the scientific nature of state evaluation. At present, the acquisition of constitutive model parameters of high earth-rock dam mainly depends on finite element inversion analysis means, and the proxy model inversion method assisted by machine learning gradually replaces direct finite element solution, which has achieved remarkable results in improving the inversion efficiency. However, the traditional machine learning methods (such as BP neural network, support vector machine, etc.) show the "curse of dimensionality" problem in high-dimensional parameter space, and the fitting accuracy and generalization ability decrease significantly, and the required training sample size is large and the calculation cost is high, which is difficult to meet the engineering demand of high performance, fast response inversion under the condition of complex nonlinear of high earth-rock dam. In addition, the current optimization algorithm generally has problems such as easy to fall into local optimum, slow convergence speed and poor robustness in dealing with highly nonlinear and strongly coupled parameter optimization tasks, which is not conducive to efficient identification of constitutive parameters under limited computing resources. Especially when using elastoplastic constitutive model for dam multi-parameter inversion, due to high model dimension, high calculation cost and complex nonlinear relationship between parameters, traditional methods often cannot realize the unification of response speed and algorithm stability while ensuring accuracy, which leads to difficulty in timely feedback of field monitoring data to engineering decision-making, and limits the intelligent process of dam safety evaluation. Therefore, we propose an elastoplastic parameter intelligent inversion method and system based on cloud theory expansion. SUMMARY

[0003] The elastoplastic parameter intelligent inversion method and system based on cloud theory expansion can solve the problems in the prior art.

[0004] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0005] An elastoplastic parameter intelligent inversion method based on cloud theory expansion, the specific steps of the inversion method are as follows:

[0006] Ⅰ: Establish a three-dimensional finite element model of the earth-rock dam and integrate it with an elastoplastic constitutive model, while selecting monitoring points as reference points for inversion response;

[0007] II: Obtain the parameters to be identified and their range of variation for the elastoplastic constitutive model, and construct a high-dimensional parameter space;

[0008] III: Generate a preset number of high-dimensional samples through Latin hypercube sampling, and perform finite element forward modeling on each group of high-dimensional samples to form a sample set;

[0009] IV: Construct a multi-dimensional, multi-layer cloud agent model, and generate global-local feature mappings through virtual cloud droplet layering;

[0010] V: Improve the Jaya algorithm and embed a multi-dimensional, multi-layer cloud proxy model into the improved Jaya algorithm framework to obtain the global constitutive model parameter combination through parameter inversion.

[0011] As a further aspect of the present invention, the specific implementation method of establishing a three-dimensional finite element model of the earth-rock dam and integrating the elastoplastic constitutive model in step I is as follows: Based on the dam's partitioned structural characteristics, a three-dimensional finite element model is constructed by integrating elastoplastic constitutive theory. The mechanical behavior of the core wall-rockfill interface is accurately characterized by non-uniform mesh partitioning. Based on the spatial distribution of monitoring instruments, mesh coupling monitoring points with complete displacement data in key sections of the dam are selected to ensure strict mapping between measured data and numerical nodes. The finite element model is a three-dimensional nonlinear static model, and the selected elastoplastic constitutive model is based on the PZ elastoplastic model, which has the ability to uniformly predict static and dynamic responses and accurately describe the deformation characteristics of the dam material through model coupling.

[0012] As a further aspect of the present invention, the specific steps for constructing the high-dimensional parameter space in step II are as follows: Combining the indoor triaxial test law and material gradation characteristics, and screening the core constitutive parameter set controlling the deformation of the dam body, then based on the physical meaning of the parameters and engineering experience values, delineating the engineering feasible variation domain of each parameter, forming a constraint space with continuous boundaries, mapping the parameters to be identified to a multi-dimensional space, constructing a hypercube search domain considering parameter interaction effects, quantifying the coupling relationship between parameters through the covariance matrix, providing a mathematical basis for subsequent intelligent sampling, wherein the dimension of the parameter space is set according to the selected constitutive model and the inversion accuracy requirements, and the dimension can be extended to 17 dimensions, while ensuring that the fluctuation range of each parameter is within reasonable engineering limits; the number of samples is determined by calculating the single simulation time and the acceptable training time to meet the requirements of high-precision modeling with small samples.

[0013] As a further scheme of the present application, the specific steps of forming the sample set in step III are as follows: using Latin hypercube sampling technology, a low correlation sample set is generated in the parameter hypercube, through stratified equal probability sampling strategy, it is ensured that each parameter dimension independently covers the entire variation range, the boundary aggregation effect is avoided, for each group of parameter samples, static finite element forward calculation is performed, displacement response data of monitoring points are extracted, parameter combination and corresponding displacement values are systematically integrated, a structured training data set is constructed, and a parameter-deformation mapping relationship consistent with physics is provided for the proxy model, wherein the Latin hypercube sampling is used to ensure the uniform distribution of input parameters in the multi-dimensional space, the sample number is 100, and the settlement data of key monitoring points obtained through finite element forward calculation are used as training output.

[0014] As a further scheme of the present application, the specific steps of constructing the multi-dimensional multi-layer cloud proxy model in step IV are as follows:

[0015] S1.1: Obtain sample set data, and divide it into a training set and a test set according to a ratio of 7:3, and then standardize each item of data in the sample set data to have a zero mean and a unit variance;

[0016] S1.2: Pre-set the sample set data as X=x1,x2,...,x n , and each input x i in it is taken as a group of m-dimensional vectors x i =[x i 1,x i 2,...,x im ], and each input x i corresponds to an output response y i , and the sample expectation of the center position of the sample set generated by the subsequent virtual cloud droplet is calculated through , and the covariance matrix of the disturbance control in the generation of the subsequent virtual cloud droplet is calculated through , wherein x i ∈R m ,

[0017] S1.3: Based on the calculated Ex and the covariance matrix CoV, the statistical characteristics of each item of data in each dimension of the sample set data are extracted to construct a one-dimensional normal cloud model, and En and He are calculated through and , wherein He can be replaced by the proportion of entropy En, specifically He=a·En;

[0018] S1.4: In the multi-dimensional input space, based on the calculated sample expectation and covariance matrix, it is pre-set that the lth layer of cloud droplets contains N l cloud droplets, wherein each cloud droplet can be represented as is expressed as and He is introduced into the covariance matrix to control the overall disturbance amplitude, so that CoV l = He l ·CoV, then, in each layer, the cloud droplets are virtually generated by the statistical characteristics of each item of data, and the number of cloud droplets generated by each layer can be the same or different, and different degrees of uncertainty are simulated by setting the disturbance intensity, and the core of the multi-layer structure is formed by each virtual cloud droplet, and each layer represents a different expression scale.

[0019] S1.5: The certainty of each cloud droplet in each layer of the cloud agent model is calculated by determining the degree function, so as to obtain the feasibility or typicality of each cloud droplet to the cloud agent model, wherein the specific form of the certainty function is as follows:

[0020]

[0021] In the formula, μ l (x i ) represents the certainty function thereof; D 2 (x i ) represents the Mahalanobis distance square of the cloud droplet x i and Ex in the agent model; represents the average entropy value of the lth layer, which is used as the denominator adjustment parameter for certainty calculation; wherein the specific calculation formula of the Mahalanobis distance square is as follows:

[0022] D 2 (x i )=(x i -Ex) T ∑ -1 (x i -Ex);

[0023] The specific calculation formula of the denominator adjustment parameter is as follows:

[0024]

[0025] In the formula, En l,j represents the disturbance entropy of the jth dimension in the lth layer, wherein the specific calculation formula of the disturbance entropy is

[0026] S1.6: Each layer of cloud droplets respectively predicts the input sample, and at the same time, according to the certainty of each layer of cloud droplets, the corresponding prediction result is given a corresponding weight, and the final output result after superposition of the prediction results of all levels is obtained by weighted fusion After the multi-layer weighted fusion is completed, the cloud agent model is used to output the prediction value.

[0027] As a further scheme of the present application, the specific steps of improving the Jaya algorithm in step V are as follows:

[0028] S2.1: Construct a target function based on the termination requirement for the selection and judgment of solutions, wherein the specific form of the target function is as follows:

[0029]

[0030] In the formula, fitness represents the target function; n represents the number of monitoring points; d gen,i is the predicted displacement value of the algorithm; d real,i is the measured displacement value; the target function quantifies the closeness of the simulation value and the measured value, and a fitness value < 5% is considered as an acceptable solution in engineering, and the smaller the value, the better the parameter group;

[0031] S2.2: Generate an initial population in a high-dimensional parameter space through Latin hypercube sampling, and determine the number of particles, the number of iterations and the initial position in the search space through basic settings, calculate the change of the position of the particle in each iteration process, and further improve the position update strategy by introducing a dynamic convergence factor;

[0032] S2.3: Calculate the fitness value of the new position of the particle in each iteration, and simultaneously maintain the elite solution library, constantly generate new solutions based on the improved position update strategy, and compare the new and old solutions based on the target function in the next iteration process, if the target function value of the new solution is higher than that of the old solution, replace the original solution to enter the next iteration, otherwise keep the original solution position.

[0033] As a further scheme of the present application, the specific calculation formula of the change of the position of the particle in each iteration process in S2.2 is as follows:

[0034] A(i+1,j,k)=A(i,j,k)+r(i,j,1)(A(i,j,b)-|A(i,j,k)|)

[0035] -r(i,j,2)(A(i,j,b)-|A(i,j,w)|)

[0036] In the formula, i, j, k represent the iteration variable, individual solution variable and solution in the population respectively;

[0037] The specific calculation formula of the dynamic convergence factor in S2.2 is as follows:

[0038]

[0039] In the formula, j represents the current iteration number; MaxIter represents the maximum iteration number; a i and a fare the initial value and final value of convergence factor, respectively, which are set to 0 and 1, respectively; n represents a descending index, which is usually taken as 1, to ensure that the step length is consistent with the last iteration, thereby ensuring the continuity of the search;

[0040] The specific calculation formula of the improved position updating strategy in S2.2 is as follows:

[0041] A(i+1,j,k)=A(i,j,k)+a(j)+[r(i,j,1)(A(i,j,b)-|A(i,j,k)|)

[0042] -r(i,j,2)(A(i,j,b)-|A(i,j,w)|)]

[0043] In the formula, each unknown quantity is explained above.

[0044] An elasto-plastic parameter intelligent inversion system based on cloud theory expansion comprises a modeling selection module, a space definition module, a sample generation module, a forward calculation module, a data processing module, an agent calculation module, an optimization evaluation module, an inversion identification module and a verification reconstruction module.

[0045] The modeling selection module is configured to establish a three-dimensional finite element model of the earth-rock dam and select corresponding monitoring points from the three-dimensional finite element model of the earth-rock dam according to the engineering layout.

[0046] The space definition module is configured to set a reasonable value range of the elasto-plastic parameters according to experimental data and engineering experience to construct a high-dimensional parameter space.

[0047] The sample generation module is configured to generate uniformly distributed sample points in the high-dimensional parameter space by using a Latin hypercube sampling method.

[0048] The forward calculation module is configured to perform finite element forward analysis on the parameters of each sample point to obtain displacement and deformation data of each monitoring point.

[0049] The data processing module is configured to perform standardization processing on the displacement and deformation data of each monitoring point and divide the displacement and deformation data into a training set and a test set.

[0050] The agent calculation module is configured to construct a multi-dimensional and multi-layer agent model and evaluate the credibility of virtual cloud droplets by using a Mahalanobis distance weighting mechanism.

[0051] The optimization evaluation module is configured to iteratively optimize the prediction results of the multi-dimensional and multi-layer agent model and evaluate the pros and cons of the current parameter solution by using the measured monitoring data and the prediction results.

[0052] The inversion identification module is used for combining a multi-dimensional multi-layer agent model with an improved Jaya algorithm, obtaining a parameter combination with the highest global matching degree, and outputting an inversion result.

[0053] The verification reconstruction module is used for comparing the prediction result with the measured data, and outputting a whole dam deformation field and a response of a key position by using the parameters obtained by the inversion identification.

[0054] As a further scheme of the present application, the specific steps of the inversion identification module outputting an inversion result are as follows:

[0055] S3.1: According to engineering background and sensitivity analysis, key constitutive parameters needing inversion identification are screened out, an initial parameter sample is generated by using an LHS method, and dam body response data corresponding to each parameter sample in the initial parameter sample is obtained by performing finite element forward modeling, so as to construct a training set and a test set;

[0056] S3.2: One-dimensional cloud model features of each parameter response are extracted on the basis of the training set, and the cloud features are layered and developed to generate multi-layer virtual cloud droplets, a complete multi-dimensional multi-layer cloud agent model is formed by multi-layer weighted fusion, and the parameters to be inverted are input into the improved Jaya algorithm as decision variables to perform iterative optimization;

[0057] S3.3: When the maximum number of iterations is reached or the fitness reaches the preset accuracy requirement, the iterative algorithm stops, otherwise, the iterative search step is returned to continue optimization, and after the iteration is completed, the global optimal particle position is output as the final inversion identification result.

[0058] Compared with the prior art, the present application has the following beneficial effects:

[0059] The application firstly acquires sample set data, and divides it into a training set and a test set according to 7:3, then carries out standardization processing on the sample set data, so that it has zero mean and unit variance, each input is a group of dimension vectors, corresponding to the output response, and the center position sample expectation of the cognitive center and the virtual cloud drop is calculated, and the covariance matrix required for disturbance control is calculated, then the features of each dimension are extracted through the sample expectation and the covariance matrix, a one-dimensional normal cloud model is constructed, and a multi-layer virtual cloud drop structure is generated in the multi-dimensional input space, the number of cloud drops in each layer is variable, and the disturbance control overall uncertainty is introduced, each layer of cloud drops is generated based on statistical characteristics, and the feasibility or typicality is calculated through the certainty function to weight the prediction result, and finally the overall output is obtained through weighted fusion, and after fusion, the cloud agent model is used for nonlinear mapping to obtain the prediction value, further, the target function is constructed based on the termination requirement as the judgment standard of the solution, Latin hypercube sampling is used to generate an initial population in a high-dimensional space, and the number of particles, the number of iterations and the initial position are set, in the iteration process, the particle position update strategy is improved by introducing a dynamic convergence factor, the fitness is calculated each time, and the elite solution library is maintained, and the new solution is compared with the old solution, and the better one is reserved, and better solutions are generated continuously until the convergence is completed, the stable high-precision inversion effect can be maintained in the high-dimensional parameter space, the problems of low calculation efficiency and easy to fall into local optimum in the complex parameter space of the traditional method are solved, the average error of static inversion is ensured to be less than 5%, the coincidence degree of dynamic response prediction and measured data is more than 90%, and the precision and efficiency of the whole life cycle safety evaluation of the earth-rock dam are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0060] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the specification, illustrate the application together with the embodiments thereof, and are used to explain the application, and do not constitute a limitation on the application.

[0061] Figure 1 A flow chart of an intelligent inversion method of elastoplastic parameters based on cloud theory expansion is provided for the application.

[0062] Figure 2 A system block diagram of an intelligent inversion system of elastoplastic parameters based on cloud theory expansion is provided for the application.

[0063] Figure 3 An algorithm flow chart of an intelligent inversion method of elastoplastic parameters based on cloud theory expansion is provided for the application.

[0064] Figure 4 A high earth-rock dam three-dimensional finite element mesh division schematic diagram of an intelligent inversion method of elastoplastic parameters based on cloud theory expansion is provided for the application.

[0065] Figure 5A reference point setting and selection schematic diagram of an elastic-plastic parameter intelligent inversion method based on cloud theory expansion is provided for the present application.

[0066] Figure 6 A cloud drop layer evolution schematic diagram in a three-dimensional space of an elastic-plastic parameter intelligent inversion method based on cloud theory expansion is provided for the present application.

[0067] Figure 7 An inversion error schematic diagram of different methods at each monitoring point of an elastic-plastic parameter intelligent inversion method based on cloud theory expansion is provided for the present application.

[0068] Figure 8 A multi-condition single-rule process schematic diagram of an elastic-plastic parameter intelligent inversion method based on cloud theory expansion is provided for the present application.

[0069] Figure 9 A fitness change trajectory comparison schematic diagram of different optimization algorithms of an elastic-plastic parameter intelligent inversion method based on cloud theory expansion is provided for the present application. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments.

[0071] Embodiment 1

[0072] Reference Figure 1 , Figures 3-9 The present embodiment discloses an elastic-plastic parameter intelligent inversion method based on cloud theory expansion, and the specific steps of the inversion method are as follows:

[0073] I: Establish a three-dimensional finite element model of the earth-rock dam, integrate the elastic-plastic constitutive model, and select monitoring points as inversion response reference points.

[0074] It should be further pointed out that the specific implementation method of step I of establishing a three-dimensional finite element model of the earth-rock dam and integrating the elastic-plastic constitutive model is as follows: based on the zoning structural characteristics of the dam body, a three-dimensional finite element model is constructed by fusing the elastic-plastic constitutive theory, the mechanical behavior of the core wall-rockfill body interface is accurately characterized by non-uniform mesh division, and the monitoring points are selected by coupling the grid with complete displacement data in the key section of the dam body according to the spatial distribution of the monitoring instruments, so as to ensure that the measured data and the numerical nodes are strictly mapped, wherein the finite element model is a three-dimensional nonlinear static model, the selected elastic-plastic constitutive model is based on the P-Z elastic-plastic model, which has the ability to uniformly predict static and dynamic responses, and the material deformation characteristics of the dam body are accurately described by model coupling.

[0075] II: Obtain the parameters to be identified and the range of variation of the elastic-plastic constitutive model, and construct a high-dimensional parameter space.

[0076] It should be further pointed out that the specific steps of constructing the high-dimensional parameter space in step II are as follows: combining the indoor triaxial test law and the material gradation characteristics, and screening the core constitutive parameter set controlling the deformation of the dam body, and then based on the physical meaning of the parameters and the engineering experience value, the engineering feasible variation domain of each parameter is determined to form a constrained space with continuous boundary, the parameters to be identified are mapped to a multi-dimensional space, a hypercube search domain considering the interaction effect of parameters is constructed, and the coupling relationship between parameters is quantified through the covariance matrix, which provides a mathematical basis for subsequent intelligent sampling. Among them, the dimension of the parameter space is determined according to the selected constitutive model and the inversion accuracy requirement, and the dimension can be expanded to 17 dimensions, and it is ensured that the fluctuation range of each parameter is within the reasonable engineering limit; the number of samples is determined by calculating the single simulation time and the acceptable training time together to meet the demand of small sample high precision modeling.

[0077] In addition, it should be pointed out that the deterministic values of each parameter of the P-Z model participating in the simulation analysis, and the finite element calculation results of the vertical deformation of the reference point of the high earth-rock dam are as follows:

[0078] Table 1

[0079]

[0080] Table 2

[0081]

[0082]

[0083] The above table 1 is the setting of the determined value of the P-Z model parameter; table 2 is the finite element calculation result of the vertical deformation of the reference point of the high earth-rock dam.

[0084] III: Generate a predetermined number of high-dimensional samples through Latin hypercube sampling, and perform finite element forward calculation on each group of high-dimensional samples to form a sample set.

[0085] It needs to be further explained that the specific steps of step III for forming the sample set are as follows: using Latin hypercube sampling technology, a low correlation sample set is generated in the parameter hypercube, through stratified equal probability sampling strategy, it is ensured that each parameter dimension independently covers the entire variation range, the boundary aggregation effect is avoided, static finite element forward calculation is performed on each group of parameter samples, displacement response data of monitoring points are extracted, parameter combination and corresponding displacement values are systematically integrated, a structured training data set is constructed, and a parameter-deformation mapping relationship consistent with physics is provided for the proxy model, wherein Latin hypercube sampling is used to ensure uniform distribution of input parameters in multi-dimensional space, the number of samples is 100, and settlement data of key monitoring points obtained by finite element forward calculation are used as training output.

[0086] IV: Construct a multi-dimensional and multi-layer cloud proxy model, and generate global-local feature mapping through virtual cloud droplet stratification.

[0087] Specifically, the sample set data is obtained, and is divided into a training set and a test set according to a ratio of 7:3, and each data in the sample set data is standardized to have a zero mean and a unit variance, and the sample set data is X=x1, x2, …, x n , and each input x i is taken as a group of m-dimensional vectors x i =[x i 1,x i 2,...,x im ], and each input x i corresponds to an output response y i , and the sample expectation of the center position of the subsequent virtual cloud droplet generated sample set is calculated, and the covariance matrix of the disturbance control in the subsequent virtual cloud droplet generation is calculated, the statistical characteristics of each data in each dimension in the sample set data are extracted based on the calculated Ex and covariance matrix CoV, and En and He are calculated to construct a one-dimensional normal cloud model, in a multi-dimensional input space, based on the calculated sample expectation and covariance matrix, it is assumed that the lth layer of cloud droplets contains N l cloud droplets, and He is introduced into the covariance matrix to control the overall disturbance amplitude, so that CoV l =He l·CoV, subsequently, in each layer, cloud droplets are virtually generated based on the statistical characteristics of various data. The number of cloud droplets generated in each layer can be the same or different. Different degrees of uncertainty are simulated by setting the perturbation intensity. At the same time, the core of the multi-layer structure is formed by these virtual cloud droplets, with each layer representing a different expression scale. The certainty function is used to calculate the certainty of each cloud droplet in each layer of the cloud proxy model to obtain the feasibility or typicality of each cloud droplet for the cloud proxy model center. Each layer of cloud droplets predicts the input sample, and the prediction result is assigned a corresponding weight based on the certainty of each layer of cloud droplet. The final output result is obtained by weighted fusion of the prediction results of all layers. After completing the multi-layer weighted fusion, the cloud agent model is used to perform nonlinear mapping on the output results, and finally output the predicted value.

[0088] It should be further explained that the specific formula for calculating the sample expectation is as follows:

[0089]

[0090] The specific formula for calculating the covariance matrix is ​​as follows:

[0091]

[0092] Where, x i ∈R m ,

[0093] The specific formula for calculating En is as follows:

[0094]

[0095] The specific formula for calculating He is as follows:

[0096]

[0097] Here, He can be replaced by the proportion of entropy En, specifically He = a·En;

[0098] The specific form of the degree of determination function is as follows:

[0099]

[0100] In the formula, μ l (x i ) represents its degree of determination function; D 2 (x i ) indicates cloud droplet x i Squared Mahalanobis distance from Ex in the surrogate model; The average entropy value of the l-th layer is used as the denominator adjustment parameter in the determination calculation; the specific formula for calculating the squared Mahalanobis distance is:

[0101] D 2 (x i )=(x i -Ex) T ∑ -1 (x i -Ex);

[0102] The specific calculation formula for the denominator adjustment parameter is as follows:

[0103]

[0104] In the formula, En l,j Let represent the perturbation entropy in the j-th dimension of the l-th layer, where the specific formula for calculating the perturbation entropy is:

[0105] It should be further noted that each cloud droplet is available express.

[0106] V: Improve the Jaya algorithm and embed the multi-dimensional, multi-layer cloud proxy model into the improved Jaya algorithm framework to obtain the global constitutive model parameter combination through parameter inversion.

[0107] Specifically, an objective function is constructed based on the termination requirement to determine the selection of solutions. An initial population is generated in a high-dimensional parameter space through Latin hypercube sampling. The number of particles, the number of iterations, and their initial positions in the search space are determined through basic settings. The position changes of particles in each iteration are calculated, and the position update strategy is further improved by introducing a dynamic convergence factor. The fitness value of the new position of the particles is calculated in each iteration, and an elite solution library is maintained simultaneously. New solutions are continuously generated based on the improved position update strategy. In the next iteration, the superiority of the new and old solutions is compared based on the objective function. If the objective function value of the new solution is higher than that of the old solution, the original solution is replaced and enters the next round of iteration; otherwise, the original solution's position is retained.

[0108] It should be further explained that the objective function is expressed in the following form:

[0109]

[0110] In the formula, fitness represents the objective function; n represents the number of monitoring points; d gen,i The algorithm predicts the displacement value; d real,i The measured displacement value is used; the objective function quantifies the closeness between the simulated value and the measured value. A fitness value of <5% is considered an engineering-acceptable solution, and the smaller the value, the better the parameter set.

[0111] The specific calculation formula of the change of the position of the particle in each iteration process is as follows:

[0112] A(i+1,j,k) = A(i,j,k) + r(i,j,1)(A(i,j,b) - |A(i,j,k)|)

[0113] -r(i,j,2)(A(i,j,b) - |A(i,j,w)|)

[0114] In the formula, i, j, and k represent iteration variables, individual solution variables, and solutions in the population, respectively;

[0115] The specific calculation formula of the dynamic convergence factor is as follows:

[0116]

[0117] In the formula, j represents the current iteration number; MaxIter represents the maximum iteration number; a i and a f are the initial value and the final value of the convergence factor, respectively, and are set to 0 and 1, respectively; n represents a descending index, and is usually taken as 1 to ensure that the step size is consistent with the last iteration, thereby ensuring the continuity of the search;

[0118] The specific calculation formula of the improved position update strategy is as follows:

[0119] A(i+1,j,k) = A(i,j,k) + a(j) + [r(i,j,1)(A(i,j,b) - |A(i,j,k)|)

[0120] -r(i,j,2)(A(i,j,b) - |A(i,j,w)|)]

[0121] In the formula, the unknown quantities are as explained above.

[0122] In addition, it should be noted that the inversion analysis results of different monitoring points, as well as the inversion analysis results of different monitoring points, are shown in the following tables, respectively:

[0123] Table 3

[0124]

[0125]

[0126] Table 4

[0127]

[0128] The above Table 3 is the inversion analysis result of different monitoring points, and Table 4 is the inversion analysis result of different monitoring points.

[0129] Embodiment 2

[0130] With reference to Figure 2 The embodiment discloses an intelligent inversion system of elastoplastic parameters based on cloud theory expansion, comprising a modeling selection module, a space definition module, a sample generation module, a forward calculation module, a data processing module, a proxy calculation module, an optimization evaluation module, an inversion identification module and a verification reconstruction module.

[0131] The modeling selection module is used for establishing a three-dimensional finite element model of the earth-rock dam and selecting corresponding monitoring points from the three-dimensional finite element model of the earth-rock dam according to the engineering layout.

[0132] The sample generation module generates uniformly distributed sample points in the high-dimensional parameter space by using the Latin hypercube sampling method; and the forward calculation module is used for performing finite element forward analysis on the parameters of each sample point to obtain displacement and deformation data of each monitoring point.

[0133] The data processing module is used for standardizing the displacement and deformation data of each monitoring point and dividing the training set and the test set; the proxy calculation module is used for constructing a multi-dimensional multi-layer proxy model and evaluating the credibility of virtual cloud droplets by using a Mahalanobis distance weighting mechanism; and the optimization evaluation module is used for iteratively optimizing the prediction results of the multi-dimensional multi-layer proxy model and evaluating the pros and cons of the current parameter solution by using the measured monitoring data and the prediction results.

[0134] The inversion identification module is used for combining the multi-dimensional multi-layer proxy model with the improved Jaya algorithm, obtaining the parameter combination with the highest global matching degree, and outputting the inversion result.

[0135] Specifically, according to the engineering background and sensitivity analysis, the key constitutive parameters for inversion identification are selected, the initial parameter samples are generated by using the LHS method, the dam body response data corresponding to each group of parameter samples in the initial parameter samples are obtained by performing finite element forward simulation, the training set and the test set are constructed, the one-dimensional cloud model features of each parameter response are extracted based on the training set, the cloud features are layered and unfolded to generate multi-layer virtual cloud droplets, a complete multi-dimensional multi-layer cloud proxy model is formed by multi-layer weighted fusion, the parameters to be inverted are input into the improved Jaya algorithm as decision variables for iterative optimization, and when the maximum number of iterations is reached or the fitness reaches the preset accuracy requirement, the iterative algorithm stops, otherwise, the iterative search step is returned to continue optimization. After the iteration is completed, the global optimal particle position is output as the final inversion identification result.

[0136] The verification reconstruction module is used for comparing the prediction results with the measured data, and using the parameters obtained by inversion identification to output the full-dam deformation field and the response of the key parts.

Claims

1. An elasto-plastic parameter intelligent inversion method based on cloud theory expansion, characterized in that, The specific steps of the inversion method are as follows: I: Establish a three-dimensional finite element model of the earth-rock dam, integrate the elastoplastic constitutive model, and select the monitoring points as the inversion response reference points; II: Obtain the parameters to be identified of the elastoplastic constitutive model and the variation range, and construct a high-dimensional parameter space; III: Generate a predetermined number of high-dimensional samples through Latin hypercube sampling, and perform finite element forward calculation on each group of high-dimensional samples to form a sample set; IV: Construct a multi-dimensional and multi-layer cloud agent model, and generate global-local feature mapping through virtual cloud droplet layering; V: Improve the Jaya algorithm, and embed the multi-dimensional and multi-layer cloud agent model into the improved Jaya algorithm framework to obtain the global constitutive model parameter combination through parameter inversion.

2. The intelligent inversion method of elastic-plastic parameters based on the cloud theory expansion according to claim 1, characterized in that, The specific steps of constructing the multi-dimensional and multi-layer cloud agent model in step IV are as follows: S1.1: Obtain the sample set data, and divide it into training set and test set according to the ratio of 7:3, and then standardize the data in the sample set to make the data in the sample set have zero mean and unit variance; S1.2: The pre-set sample set data is X = x1, x2,..., xN n , and each input x i in it is taken as a set of m-dimensional vectors x i = [x i 1, x i 2,..., x im ], and each input x i corresponds to the output response y i , and the sample expectation of the center position of the sample set of the cognitive center and the subsequent virtual cloud droplet generation is calculated, and the covariance matrix of the disturbance control in the subsequent virtual cloud droplet generation is calculated, wherein x i ∈ R m , S1.3: Based on the calculated Ex and the covariance matrix CoV, the statistical characteristics of each dimension of the sample set data are extracted to construct a one-dimensional normal cloud model, and the statistical characteristics of each dimension are calculated by and En and He are calculated, wherein He can be replaced by the ratio of entropy En, that is, He=a·En. S1.4: In the multi-dimensional input space, based on the calculated sample expectation and covariance matrix, preset cloud droplet layer l contains N l cloud droplets, wherein each cloud droplet can be represented by x ji ~ N(Ex, CoV l ), and He is introduced into the covariance matrix to control the overall disturbance amplitude, so that CoV l = He l · CoV, then in each of them, the cloud droplets are virtually generated by the statistical characteristics of each item of data, and the number of cloud droplets generated by each layer can be the same or different, and different degrees of uncertainty are simulated by setting the disturbance intensity, and the core of the multi-layer structure is formed by each virtual cloud droplet, and each layer represents a different expression scale; S1.5: Calculate the certainty of each cloud droplet in each layer of the cloud agent model through the certainty function to obtain the feasibility or typicality of each cloud droplet to the center of the cloud agent model, wherein the specific form of the certainty function is as follows: where μ l (x i ) represents its degree of certainty function; D 2 (x i ) represents the Mahalanobis distance square of cloud drop x i to Ex in the proxy model; represents the average entropy value of the lth layer, which is used as the denominator adjustment parameter for the degree of certainty calculation; and the specific calculation formula of the Mahalanobis distance square is: D 2 (x i -Ex) i (x T -Ex) -1 (x i -Ex) The specific calculation formula of the denominator adjustment parameter is as follows: In the formula, En l,j represents the perturbation entropy of the jth dimension of the ith layer, and the specific calculation formula of the perturbation entropy is S1.6: Each layer of cloud droplets respectively predicts the input sample, and at the same time, according to the certainty of each layer of cloud droplets, the corresponding prediction result is given a corresponding weight, and through the weighted fusion mode, the final output result of the superposition of all levels of prediction results is obtained After completing the multi-layer weighted fusion, the cloud agent model is used to output the result for nonlinear mapping, and finally the prediction value is output. 3.The intelligent inversion method of elastic-plastic parameters based on cloud theory expansion according to claim 1, characterized in that, The specific steps of improving the Jaya algorithm in step V are as follows: S2.1: Construct a target function based on the termination requirement for solution selection and judgment, wherein the specific form of the target function is as follows: Where fitness represents the objective function; n represents the number of monitoring points; d gen,i is the predicted displacement value; d real,i is the measured displacement value; the objective function quantifies the closeness of the simulated value to the measured value, and a fitness value < 5% is considered an acceptable engineering solution, and the smaller the value, the better the parameter set; S2.2: Generate an initial population in the high-dimensional parameter space through Latin hypercube sampling, and determine the number of particles, iteration times and initial position in the search space through basic settings, calculate the change of particle position in each iteration process, and further improve the position update strategy by introducing a dynamic convergence factor; S2.3: Calculate the fitness value of the new position of the particle in each iteration, and simultaneously maintain the elite solution library, constantly generate new solutions based on the improved position update strategy, and compare the new and old solutions based on the target function in the next iteration process, if the target function value of the new solution is higher than that of the old solution, replace the original solution to enter the next iteration, otherwise keep the original solution position.

4. The intelligent inversion method of elastic-plastic parameters based on the cloud theory expansion according to claim 3, characterized in that, The specific calculation formula of the change of particle position in each iteration process in S2.2 is as follows: A(i+1,j,k)=A(i,j,k)+r(i,j,1)(A(i,j,b)-|A(i,j,k)| -r(i,j,2)(A(i,j,b)-|A(i,j,w)| Where i, j, k represent iteration variable, individual solution variable and solution in population respectively; The specific calculation formula of the dynamic convergence factor in S2.2 is as follows: Where j represents the current iteration number; MaxIter represents the maximum number of iterations; a i and a f are the initial value and the final value of the convergence factor, respectively, which are set to 0 and 1, respectively; n represents the descending index, which is usually taken as 1 to ensure that the step size is consistent with the last iteration at the last iteration, thereby ensuring the continuity of the search; The specific calculation formula of the improved position update strategy in S2.2 is as follows: A(i+1,j,k)=A(i,j,k)+a(j)+[r(i,j,1)(A(i,j,b)-|A(i,j,k)| -r(i,j,2)(A(i,j,b)-|A(i,j,w)| Wherein each unknown quantity is explained above.

5. An intelligent inversion system of elastic-plastic parameters based on cloud theory expansion, used to implement the intelligent inversion method of elastic-plastic parameters based on cloud theory expansion of any one of claims 1-4, characterized in that, The modeling selection module is configured to establish a three-dimensional finite element model of the earth-rock dam, and select corresponding monitoring points from the three-dimensional finite element model of the earth-rock dam according to engineering layout conditions. The space definition module is configured to set a reasonable value range of the elastoplasticity parameters according to experimental data and engineering experience, so as to construct a high-dimensional parameter space. The sample generation module is configured to generate uniformly distributed sample points in the high-dimensional parameter space by using a Latin hypercube sampling method. The forward calculation module is configured to perform finite element forward analysis on the parameters of each sample point, so as to obtain displacement and deformation data of each monitoring point. The data processing module is configured to perform standardization processing on the displacement and deformation data of each monitoring point, and divide a training set and a test set. The proxy calculation module is configured to construct a multi-dimensional and multi-layer proxy model, and evaluate the credibility of virtual cloud droplets by using a Mahalanobis distance weighting mechanism. The optimization evaluation module is configured to iteratively optimize the prediction results of the multi-dimensional and multi-layer proxy model, and evaluate the pros and cons of the current parameter solution by using measured monitoring data and the prediction results. The inversion identification module is configured to combine the multi-dimensional and multi-layer proxy model with an improved Jaya algorithm, obtain a parameter combination with the highest global matching degree, and output an inversion result. The verification and reconstruction module is configured to compare the prediction results with measured data, and output a full-dam deformation field and responses of key positions by using the parameters obtained through inversion identification. The inversion identification module outputs the inversion result in the following specific steps:

6. The intelligent inversion system of elastic-plastic parameters based on the cloud theory expansion according to claim 5, characterized in that, S3.1: According to the engineering background and sensitivity analysis, the key constitutive parameters that need to be inverted and identified are screened out, the initial parameter samples are generated by using the LHS method, and the corresponding dam response data are obtained by performing finite element forward simulation on each parameter sample in the initial parameter samples, so as to construct a training set and a test set; S3.2: The one-dimensional cloud model features of each parameter response are extracted on the basis of the training set, the cloud features are hierarchically unfolded to generate multi-layer virtual cloud droplets, a complete multi-dimensional and multi-layer cloud proxy model is formed by multi-layer weighted fusion, and the parameters to be inverted are input into the improved Jaya algorithm as decision variables to perform iterative optimization; S3.3: When the maximum number of iterations is reached or the fitness reaches the preset accuracy requirement, the iterative algorithm stops, otherwise, the iterative search step is returned to continue optimization, and after the iteration is completed, the global optimal particle position is output as the final inversion and identification result. ​

Images