CN-121980882-A - Large model driven Bayesian model correction method, system and device

CN121980882ACN 121980882 ACN121980882 ACN 121980882ACN-121980882-A

Abstract

The invention provides a large model driven Bayesian model correction method, a large model driven Bayesian model correction system and a large model driven Bayesian model correction device, and relates to the technical field of Bayesian model correction. The method comprises the steps of establishing a finite element model of a target building, carrying out sensitivity analysis on each layer group, inputting attribute information, evolution information and environment information of the target building into a damage prediction large model, outputting damage degree proportion of each layer group, combining a corresponding relation, converting the damage degree proportion into a reference mean value of correction material parameters, constructing prior distribution corrected by a Bayesian model, constructing Bayesian posterior distribution, executing an MCMC sampling method, extracting optimal estimated values of the correction material parameters of each layer group based on statistical characteristics of the posterior distribution, and carrying out double validity verification on the correction model adopting the optimal estimated values, wherein the optimal estimated values are used as final correction material parameters to be updated to the finite element model if verification is passed. The scheme can enable the priori knowledge to be more fit with the actual structural state, and improves the physical rationality and the precision of parameter correction.

Inventors

LI YANNA
YAO ZHIDONG
LI YUMENG
WANG JIAHAO

Assignees

中冶建筑研究总院（深圳）有限公司
中冶建筑研究总院有限公司

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. A large model driven bayesian model correction method, comprising: step 1, acquiring attribute information, evolution information and environment information of a target building, dividing the target building into a plurality of independent layer groups according to a preset floor grouping method based on the attribute information; Step 2, establishing an initial finite element model of a target building, setting material parameters of the finite element model according to the strength grade of the concrete, giving the material parameters to each layer group, wherein the material parameters comprise elastic modulus Density and density of ; Step 3, carrying out sensitivity analysis on the material parameters of each layer group to obtain the relative sensitivity coefficient of the natural frequency to the material parameters, and selecting the material parameters with the relative sensitivity coefficient larger than a preset threshold value as corrected material parameters ; Step 4, inputting attribute information, evolution information and environment information of the target building into a damage prediction large model, and outputting damage degree proportion of each layer group; Step 5, target building front The order actual measurement natural frequency is used as a correction target, a weighted Gaussian likelihood function is constructed, a weighted strategy of the damage degree proportion of the layer group is introduced, the prior distribution is fused with the weighted Gaussian likelihood function, and Bayesian posterior distribution is constructed; step 6, based on Bayesian posterior distribution, selecting and executing a corresponding MCMC sampling method according to the dimension of the corrected material parameter, and synchronously and iteratively verifying the convergence of the sampling result until the sampling result converges to obtain the posterior distribution of the corrected material parameter of each layer group; And 7, extracting the optimal estimated value of the corrected material parameter of each layer group from the posterior distribution of the corrected material parameter, carrying out double validity verification on the finite element model corrected by adopting the optimal estimated value, and updating the optimal estimated value to the finite element model as the final corrected material parameter if verification is passed.
2. The bayesian model correcting method according to claim 1, wherein the floor grouping method comprises: step 11, dividing floors with the same strength grade of the concrete into a group to obtain an initial material layer group; step 12, using total layer number Can be scaled by the optimal group The integer division is the principle, and one value is selected from the values 2,3 and 4 to be used as And performing integer division operation to obtain an average layer group; and step 13, adjusting the equipartition group based on the building height grouping requirement.
3. The bayesian model correction method according to claim 2, wherein the building height grouping requirement comprises: 1-3 layers, each layer being grouped separately; 4-9 layers, at least 3 groups; 10-27 layers, at least 4 groups; Over 27 layers, at least 6 groups.
4. The bayesian model correcting method according to claim 1, wherein the attribute information comprises construction year, number of floors, structure type, position of functional layer and concrete strength level of each floor, the evolution information comprises service year, load accumulation change amount and load change rate, and the environment information comprises regional annual precipitation, average humidity and earthquake intensity.
5. The bayesian model correcting method according to claim 1, wherein the specific construction method of the damage prediction large model comprises: Step a1, searching attribute information, evolution information and environment information of a plurality of target structures which are the same as the structures of the target buildings in a sample library, and constructing standardized feature vectors; Step a2, unifying the data formats of the standardized feature vectors, using the standardized feature vectors as a data set, calculating a structural aging index based on the design service life, the service life and the structural type, calculating a load change coefficient based on the design load and the load change quantity of each layer group, and counting the damage degree proportion of each layer group ; And a step a3 of fusing the standardized feature vector, the structural aging index and the load change coefficient to construct a multisource fusion feature vector as model input, taking the damage degree proportion as a supervision tag, and performing supervised fine tuning on the open-source large model to obtain a damage prediction large model of the adaptive target building.
6. The Bayesian model correction method according to claim 5, wherein, The specific construction method of the corresponding relation comprises the following steps: Step b1, constructing a reference mean value calculation formula of the corrected material parameters, wherein the specific formula comprises the following steps: ; ; Wherein, the Representing a reference mean value of elastic modulus; an initial value of elastic modulus is shown; Represents the variation of the elastic modulus; Representing the modulus of elasticity correction coefficient; Representing a density reference mean; Representing an initial value of density; represents the amount of density change; Representing a density correction coefficient; An index of deterioration of elastic modulus; Represents a density deterioration index; Step b2, regarding 1/2 of the damage degree ratio as And Obtaining a correspondence between the damage degree ratio and the material degradation index; Step b3, setting according to the position of the layer group And 。
7. The bayesian model correcting method according to claim 6, wherein the specific expression of the weighted gaussian likelihood function comprises: ; Wherein, the Represent the first Natural frequency of the order finite element calculation; represent the first The step actual measurement natural frequency; represent the first The variance of the natural frequency calculated by the order finite element; represent the first The specific expression of the damage weight of the order natural frequency comprises the following steps: ; Wherein P j represents the first The damage degree ratio of the layer group.
8. The bayesian model correcting method according to claim 7, wherein the specific verification conditions for the double validity verification include: For the first 3 rd order natural frequency, the relative error of the natural frequency calculated by the finite element and the measured natural frequency ≤5%; The posterior material degradation rate falls within a degradation interval corresponding to the damage level.
9. A large model driven Bayesian model correction system is characterized by comprising a data acquisition module, a data processing module and a result generation module; The data acquisition module is used for acquiring attribute information, evolution information and environment information of a target building, wherein the attribute information comprises a construction year, the number of floors, a structure type, a functional layer position and concrete strength grades of all floors, the evolution information comprises a use year, a load accumulated change amount and a load change rate, and the environment information comprises regional annual precipitation, average humidity and earthquake intensity; the data processing module comprises a grouping unit, a finite element unit, a sensitivity analysis unit, a priori unit, a correction unit, a post-test unit and a calibration unit; The grouping unit divides the target building into a plurality of independent groups of layers according to a preset floor grouping method based on the attribute information; The finite element unit is used for establishing an initial finite element model of a target building, setting material parameters of the finite element model according to the strength grade of concrete, endowing the material parameters to each layer group, and the material parameters comprise elastic modulus Density and density of ; The sensitivity analysis unit is used for carrying out sensitivity analysis on the material parameters of each layer group to obtain the relative sensitivity coefficient of the natural frequency to the material parameters, and selecting the material parameters with the relative sensitivity coefficient larger than a preset threshold value as corrected material parameters ; The prior unit is used for inputting attribute information, evolution information and environment information of a target building into the damage prediction large model, outputting the damage degree proportion of each layer group, converting the damage degree proportion into a reference mean value of corrected material parameters by combining with a preset corresponding relation, and setting prior variance to construct prior distribution corrected by the Bayesian model; The correction unit is used for targeting the front of a building The order actual measurement natural frequency is used as a correction target, a weighted strategy of fusion layer group damage degree proportion is introduced, a weighted Gaussian likelihood function is constructed, the prior distribution is fused with the weighted Gaussian likelihood function, and Bayesian posterior distribution is constructed; the posterior unit is used for selecting and executing a corresponding MCMC sampling method according to the dimensionality of the corrected material parameter based on Bayesian posterior distribution, synchronously and iteratively verifying the convergence of the sampling result until the sampling result converges to obtain corrected material parameter posterior distribution of each layer group; The calibration unit extracts the optimal estimated value of the correction material parameter of each layer group from the posterior distribution of the correction material parameter, performs double validity verification on the finite element model corrected by adopting the optimal estimated value, and updates the optimal estimated value to the finite element model as the final correction material parameter if the verification is passed; And the result generation module is used for issuing the finite element model.
10. A large model driven bayesian model correcting device, comprising a processor, a memory and a bus, wherein the memory stores instructions and data read by the processor, the processor is used for calling the instructions and data in the memory to execute the bayesian model correcting method according to any one of claims 1-8, and the bus is connected between functional components and is used for transmitting information.

Description

Large model driven Bayesian model correction method, system and device Technical Field The invention relates to the technical field of Bayesian model correction, in particular to a large model driven Bayesian model correction method, system and device. Background The prior Bayesian model correction method mainly regards parameters of a building structure (to be corrected) as random variables based on Bayesian theorem, the parameters are integrated into engineering information through priori distribution, the priori distribution of the parameters of the materials in the prior method is dependent on design specifications or experience values, performance degradation characteristics of the structure caused by environmental effect and load evolution in a long-term service process are difficult to fully reflect, the correction result deviation is large, the prior method mainly uses structure dynamic response data (such as natural frequency) only, and lacks effective fusion on multi-source data such as building attribute information, service evolution information and environmental factors, the indication effect of structural performance degradation cannot be fully mined, the prior method generally does not consider the differential influence of different layer group damage degrees on structural dynamic characteristics, and the weight of each parameter is unreasonable to set when a likelihood function is constructed, so that correction precision is reduced. In addition, the publication number is CN118862565A, the name is a patent application of a structural damage identification method and system based on a convolutional neural network proxy model and a Bayesian model, which discloses that a priori, likelihood and posterior distribution of damage parameters are created through Bayesian modeling, a sample is generated by using an inverse sampling strategy, finite element analysis is performed, a data set is constructed, a CNN proxy model is designed and trained, a AdaMH algorithm model is formed by combining a Metropolis-Hasting algorithm, the posterior sample is extracted by using the model, and deviation is corrected through an error enhancement model. However, the scheme does not correlate with the actual damage state of the structure, so that prior information is disjointed with engineering reality, and the same punishment is given to the group frequency deviation of different damage degrees, so that the parameter sensitivity of a high damage region cannot be highlighted. In addition, the patent application with the publication number of CN106897717A and the name of Bayesian model correction method under multiple tests based on environmental excitation data discloses that the structural acceleration data under the environmental excitation collected under multiple tests is analyzed to obtain the natural frequency and the vibration mode of the structure measured by each test, the uncertainty of the modal parameters is calculated and represented by a covariance matrix, an objective function is constructed based on the structural modal parameters and the covariance matrix obtained by multiple tests and based on the Bayesian theory, and the optimal value of the model parameters of the finite element model to be corrected is obtained by optimizing the objective function. However, the prior distribution of the scheme is artificially assumed, and different prior distribution assumptions can obtain distinct correction results, which easily influence the reliability of the posterior results. Furthermore, the patent application with the publication number of CN114282398A and the name of a bridge health monitoring system and method based on big data discloses that a hyperbolic model and a power function model are adopted to obtain characteristic parameters of a concrete constitutive relation, and big data analysis software WEKA is used for determining mass data of bridge monitoring. However, the scheme has high computational complexity, is completely dependent on data quality and has poor adaptability. Therefore, a new bayesian model correction method is needed to improve the model accuracy and practicality. Disclosure of Invention The invention aims to provide a large model driven Bayesian model correction method, a large model driven Bayesian model correction system and a large model driven Bayesian model correction device, so as to solve at least one of the technical problems in the prior art. In order to solve the above technical problems, the present invention provides a large model driven bayesian model correction method, comprising the following steps: The method comprises the steps of 1, collecting attribute information, evolution information and environment information of a target building, dividing the target building into a plurality of independent layer groups according to a preset floor grouping method based on the attribute information to ensure that stress states of all the layer groups are consistent with d