CN-121980867-A - Concretization-heat-humidity multi-field coupling parameter inversion method and system based on BP neural network-genetic algorithm

CN121980867ACN 121980867 ACN121980867 ACN 121980867ACN-121980867-A

Abstract

The invention provides a concretization-heat-humidity multi-field coupling parameter inversion method and system based on BP neural network-genetic algorithm, wherein the method comprises the steps of obtaining actually measured humidity evolution data of the interior of a roller compacted concrete test block along with the growth of an age; the method comprises the steps of obtaining the size and actual maintenance conditions of a roller compacted concrete test block, establishing a multi-field coupling numerical model of the roller compacted concrete test block by adopting finite element analysis software, generating a parameter set to be inverted of the multi-field coupling numerical model of the roller compacted concrete test block by utilizing a Latin hypercube sampling method, substituting the parameter set to be inverted into the multi-field coupling numerical model of the roller compacted concrete test block, performing forward calculation to obtain a relative humidity curve, constructing a training data set by using the parameter set to be inverted and a forward calculation result thereof, training a BP neural network by using the training data set, taking the trained BP neural network as a proxy model, inverting and optimizing concrete-heat-humidity multi-field coupling parameters by adopting a genetic algorithm based on actual measured humidity evolution data.

Inventors

WANG QIAO
LIU CHANG
ZHOU WEI
CHANG XIAOLIN
ZHANG YONGZHEN
LI YUCHEN

Assignees

武汉大学

Dates

Publication Date: 20260505
Application Date: 20260127

Claims (10)

1. The concretization-heat-humidity multi-field coupling parameter inversion method based on BP neural network-genetic algorithm is characterized by comprising the following steps: obtaining actually measured humidity evolution data of the interior of a roller compacted concrete test block along with the growth of the age; acquiring the size and the actual maintenance condition of a roller compacted concrete test block, and establishing a multi-field coupling numerical model of the roller compacted concrete test block by adopting finite element analysis software; generating a parameter set to be inverted of a multi-field coupling numerical model of the roller compacted concrete test block by using a Latin hypercube sampling method; substituting the parameter set to be inverted into a multi-field coupling numerical model of the roller compacted concrete test block, and performing forward calculation to obtain a relative humidity curve, so as to construct a training data set according to the parameter set to be inverted and a forward calculation result thereof; Training a BP neural network by using a training data set, taking the trained BP neural network as a proxy model, inputting the proxy model into a relative humidity curve, and outputting the proxy model into a parameter set to be inverted; Based on the actually measured humidity evolution data, inverting and optimizing the concrete hydration-heat-humidity multi-field coupling parameters by utilizing a proxy model and combining a genetic algorithm.
2. The method for inverting the concretization-heat-humidity multi-field coupling parameters based on the BP neural network-genetic algorithm as set forth in claim 1, wherein the process for constructing the multi-field coupling numerical model of the roller compacted concrete test block comprises the following steps: Constructing a concrete-thermal-wet coupling model, which comprises constructing control equations of a hydration reaction field, a temperature field and a humidity field and initial conditions; based on the concrete chemical-thermal-wet coupling model, combining the actual size of the roller compacted concrete test block, establishing a concrete multi-field coupling numerical model in equal proportion with the test block by means of finite element analysis software, and setting boundary conditions of a temperature field and a humidity field in the concrete multi-field coupling numerical model according to the actual curing condition of the roller compacted concrete test block.
3. The inversion method of concretization-heat-humidity multi-field coupling parameters based on BP neural network-genetic algorithm according to claim 1, wherein the process of generating a plurality of parameter sets to be inverted of a multi-field coupling numerical model of a roller compacted concrete test block by using Latin hypercube sampling method comprises: And selecting a plurality of parameters to be inverted from parameters related to a multi-field coupling numerical model of the roller compacted concrete test block, determining the value range of each parameter to be inverted, and sequentially sampling a plurality of times from the value range of each parameter to be inverted by using a Latin hypercube sampling method to obtain a plurality of parameter groups to be inverted.
4. The method for inverting the concrete-thermal-wet multi-field coupling parameters based on the BP neural network-genetic algorithm as claimed in claim 1, wherein the BP neural network, the output layer is provided with the quantity of neurons corresponding to the parameter types of the parameter sets to be inverted, the parameter sets to be inverted in the training data set are taken as sample output sets, the multi-field coupling numerical model is forward-developed to obtain a relative humidity curve as an input set, the training data set is learned by using the BP neural network model, cross validation is carried out, and the trained BP neural network is taken as a proxy model for fitting the implicit function relation between the parameters to be inverted and the relative humidity.
5. The inversion method of the concretization-thermal-wet multi-field coupling parameter based on the BP neural network-genetic algorithm as set forth in claim 1, wherein the process of calculating the optimized concretization-thermal-wet multi-field coupling parameter includes: inputting a parameter set to be inverted into a proxy model, extracting an output relative humidity curve, calculating the accumulated humidity value deviation of actually measured humidity evolution data and the relative humidity curve, and constructing an objective function by taking the minimum accumulated humidity value deviation as a target; and solving an objective function through a genetic algorithm by using the proxy model inversion, and outputting an optimized parameter set to be inverted obtained by solving as an optimized concrete-thermal-wet multi-field coupling parameter.
6. The method for inverting the concretization-thermal-wet multi-field coupling parameters based on the BP neural network-genetic algorithm according to claim 5, wherein the expression of the objective function is as follows: ; In the formula, The method comprises the steps of integrating humidity value deviation, wherein x is a parameter set to be inverted, i is a relative humidity measuring point, m is the number of the relative humidity measuring points, j is a time point serial number, and n is the number of time points of the duration of a relative humidity curve of a single measuring point; For the point in time Calculated relative humidity values in the relative humidity curve output by the proxy model, For the point in time Is a measured relative humidity value.
7. The method for inverting the concretization-heat-humidity multi-field coupling parameter based on the BP neural network-genetic algorithm as set forth in claim 1, further comprising substituting the optimized concretization-heat-humidity multi-field coupling parameter value into a multi-field coupling numerical model of a roller compacted concrete test block, performing forward calculation to obtain a concrete relative humidity curve, comparing the relative humidity curve with test values and measured humidity evolution data, and verifying the accuracy of the optimized concretization-heat-humidity multi-field coupling parameter value.
8. The concretization-heat-humidity multi-field coupling parameter inversion system based on BP neural network-genetic algorithm is characterized by comprising The humidity evolution module is used for acquiring actually measured humidity evolution data of the interior of the roller compacted concrete test block along with the growth of the age; The multi-field coupling numerical model construction module is used for acquiring the size and the actual maintenance condition of the roller compacted concrete test block and adopting finite element analysis software to establish a multi-field coupling numerical model of the roller compacted concrete test block; The parameter set to be inverted design module is used for generating a parameter set to be inverted of a multi-field coupling numerical model of the roller compacted concrete test block by using a Latin hypercube sampling method; The training data preparation module is used for substituting the parameter set to be inverted into a multi-field coupling numerical model of the roller compacted concrete test block, obtaining a relative humidity curve through forward calculation, and constructing a training data set according to the parameter set to be inverted and a forward calculation result thereof; The agent model building module is used for training the BP neural network by using the training data set, taking the trained BP neural network as an agent model, inputting the agent model into a relative humidity curve, and outputting the agent model into a parameter set to be inverted; And the concrete parameter inversion module is used for inverting and optimizing the hydration-heat-humidity multi-field coupling parameters of the concrete by utilizing the agent model and combining a genetic algorithm based on the actually measured humidity evolution data.
9. An electronic device comprising a memory and a processor, the memory storing program instructions for execution by the processor, the processor invoking the program instructions to perform the method of any of claims 1-7.
10. A non-transitory computer readable storage medium, wherein the non-transitory computer readable storage medium stores computer instructions, the computer instructions cause the computer to perform the method of any one of claims 1 to 7.

Description

Concretization-heat-humidity multi-field coupling parameter inversion method and system based on BP neural network-genetic algorithm Technical Field The invention belongs to the technical field of concrete parameter inversion, and particularly relates to a concrete-thermal-wet multi-field coupling parameter inversion method and system based on a BP neural network-genetic algorithm. Background The hydration heat release, heat transfer, humidity migration and other physical field coupling behaviors of the early-age concrete are main reasons for inducing structural stress concentration and causing shrinkage cracks and durability reduction. For such problems, establishing a chemical-thermal-wet multi-field coupling numerical model has become an important means for predicting the evolution of concrete performance. In simulating the relative humidity change process of early-age concrete, the accuracy of the model mainly depends on the humidity diffusion coefficient and the setting of key parameters in a chemical affinity function. These parameters are commonly affected by material inhomogeneities, environmental interactions, and age variations, and experimentally obtained parameter values often deviate significantly from the actual situation. When the theoretical model is adopted for calculation, the problems that the model is various, different model results are inconsistent, part of parameters are difficult to accurately determine and the like are also faced. These difficulties lead to significant deviation between the model prediction result and the measured data, which restricts the application of the model prediction result in engineering fine management and control. And optimizing model input parameters by a parameter inversion method, for example, performing parameter identification based on intelligent algorithms such as genetic algorithm, particle swarm optimization and the like. Based on the excellent nonlinear mapping characteristic of the neural network, the algorithm can autonomously establish a complex association mechanism of parameters and measured data without an explicit relation between predefined variables. By constructing a reverse analysis framework driven by deep learning, intelligent identification of concrete key parameters can be realized, and the method remarkably improves the parameter inversion precision while avoiding the limitations of the traditional model. Disclosure of Invention In order to overcome the defect that the prior art has obvious deviation with actual conditions due to the common influence of material nonuniformity, environmental interaction and age change on model parameters when simulating the relative humidity change process of early-age concrete, and the obvious deviation between a model prediction result and measured data caused by difficult accurate determination of theoretical model parameters, the invention provides a concretization-heat-humidity multi-field coupling parameter inversion method and a system based on BP neural network-genetic algorithm, the method comprises the steps of outputting a relative humidity curve through a multi-field coupling numerical model of a roller compacted concrete test block which is constructed through finite element analysis software and takes into account chemical-thermal-wet multi-field coupling, constructing a proxy model for subsequent inversion through a BP neural network, fitting a nonlinear relation between key material parameters and the relative humidity curve by considering the common influence of material non-uniformity, environment interaction and age change, and finally inverting through a genetic algorithm to obtain the optimized concrete chemical-thermal-wet multi-field coupling parameters. According to an aspect of the present disclosure, there is provided a method for inversion of a concretization-thermal-wet multi-field coupling parameter based on a BP neural network-genetic algorithm, including: obtaining actually measured humidity evolution data of the interior of a roller compacted concrete test block along with the growth of the age; acquiring the size and the actual maintenance condition of a roller compacted concrete test block, and establishing a multi-field coupling numerical model of the roller compacted concrete test block by adopting finite element analysis software; generating a parameter set to be inverted of a multi-field coupling numerical model of the roller compacted concrete test block by using a Latin hypercube sampling method; substituting the parameter set to be inverted into a multi-field coupling numerical model of the roller compacted concrete test block, and performing forward calculation to obtain a relative humidity curve, so as to construct a training data set according to the parameter set to be inverted and a forward calculation result thereof; Training a BP neural network by using a training data set, taking the trained BP neural network as a proxy model, inputting the proxy mo