Search

CN-121996896-A - High-temperature pneumatic unbalance factor correction method based on microscopic data

CN121996896ACN 121996896 ACN121996896 ACN 121996896ACN-121996896-A

Abstract

The invention provides a high-temperature pneumatic unbalanced factor correction method based on microscopic data, which comprises the steps of firstly obtaining and preprocessing microscopic state-state simulation data, constructing a feature space, utilizing a conditional variation self-encoder to learn data distribution and conduct directional sampling, generating a reconstruction data set conforming to a physical distribution rule in a current temperature interval, then constructing a constraint equation based on an unbalanced factor physical relationship, carrying out inversion solving on a pseudo temperature, obtaining a function framework containing undetermined parameters through symbolic regression mining, carrying out global fitting on the function framework to obtain a reference model, analyzing an error field of the function framework, automatically identifying a high error region according to a set error threshold, dividing the error field into a plurality of physical subareas, extracting analysis boundaries of the subareas, and finally independently optimizing parameters in the function framework aiming at the physical subareas to be integrated into a mixed expert model consisting of a unified function framework, a plurality of subarea boundary equations and a plurality of subarea parameters.

Inventors

  • LIU HUI
  • YUAN WU
  • Liu Xiazhen
  • LIANG PAN
  • ZHANG JIAN

Assignees

  • 中国科学院计算机网络信息中心

Dates

Publication Date
20260508
Application Date
20251231

Claims (10)

  1. 1. A high-temperature pneumatic unbalance factor correction method based on microscopic data comprises the following steps: constructing a basic feature space according to mapping relation data of unbalanced factors in the microscopic state-state simulation data set about translation temperature and vibration temperature; Constructing a condition variation self-encoder, and learning a distribution rule of data in the basic feature space; generating a reconstruction data set conforming to a physical distribution rule through the trained condition variation self-encoder in a target temperature interval; Constructing a nonlinear constraint equation about pseudo temperature according to the physical definition and physical quantity coupling relation of the unbalanced factor about pseudo temperature, translation temperature and vibration temperature, carrying out iterative calculation on the unbalanced factor value in the reconstruction data set by using a numerical optimization method, and reversely solving data of the corresponding pseudo temperature; Based on the function skeleton, global fitting is carried out on a reconstruction data set in a target temperature interval through a constraint optimization algorithm, a group of global optimal reference parameters are obtained, a global reference model is obtained, the relative error of the global reference model in the target temperature interval is calculated, and an error field is constructed; Based on a set error threshold value, identifying an error region in the error field through a connected region analysis algorithm, and dividing the error field into a plurality of physical subregions, wherein the physical subregions comprise a low error region and a high error region; extracting a closed boundary point set between each physical subarea, and obtaining a partition boundary parameter equation between each physical subarea by using a parameterized curve fitting method; And (3) independently optimizing the reference parameters in the corresponding function skeleton aiming at each high-error region to obtain partition reference parameters, and constructing a hybrid expert model according to the function skeleton, partition boundary parameter equations among the physical subregions and the partition reference parameters.
  2. 2. The method of claim 1, wherein the constructing a base feature space from the mapping data of unbalance factors in the microscopic state-state simulation dataset with respect to translational temperature and vibration temperature comprises: Screening translation temperature And vibration temperature Are all greater than or equal to Randomly dividing the screened data into a training set and a testing set according to the proportion of 8:2, and aiming at unbalanced factors Performing logarithmic transformation And for input features And a target variable Performing standardization processing to construct a basic feature space as 。
  3. 3. The method of claim 1, wherein the constructing a conditional variation from an encoder, learning a distribution rule of data in the base feature space, comprises: Constructing an encoder and a decoder, wherein the encoder receives Mapping to 2-dimensional potential space output mean value through 3 full connection layers And logarithmic variance The decoder receives the sampled latent variable And a condition variable Output of the reconstruction Reconstructing error, KL divergence and smoothing regularization term by including MSE Training a conditional variation self-encoder with a smoothing regularization term By calculating L1 norms of adjacent sample gradients, the optimizer adopts Adam, and the initial learning rate is set as And dynamically adjusting strategies in cooperation with the learning rate.
  4. 4. The method of claim 1, wherein the generating a reconstructed data set conforming to a law of physical distribution from an encoder by the trained conditional variation within a target temperature interval comprises: after training, the decoder is used for controlling the target temperature interval to be In-range generation In total 40000 reconstructed data points.
  5. 5. The method of claim 1, wherein constructing a nonlinear constraint equation for the pseudo temperature based on the physical definition and the physical quantity coupling relationship of the unbalance factor for the pseudo temperature, the translation temperature, and the vibration temperature comprises: physical definition based on unbalance factors And a distribution function And pseudo temperature The nonlinear constraint equation for the pseudo temperature U is constructed as: Wherein, the As a result of the non-equilibrium factor, Is an approximate vibration distribution function, For the temperature of the translation, In order to achieve the vibration temperature, the temperature of the vibration, And (3) with Is a combination of two pseudo-temperatures, 、 The known parameters were obtained by fitting the state-state data to the modified Arrhenius.
  6. 6. The method of claim 1, wherein the mining of the functional analytical expression between the pseudo temperature and the translational and vibrational temperatures by the symbolic regression algorithm is: Wherein, the Is a temperature of the pseudo-air at which the temperature is high, In order to achieve the vibration temperature, the temperature of the vibration, For the temperature of the translation, Is the reference parameter.
  7. 7. The method of claim 1, wherein globally fitting the reconstructed data set within the target temperature interval by a constraint optimization algorithm, obtaining a set of globally optimal reference parameters and obtaining a global reference model, calculating a relative error of the global reference model within the target temperature interval, and constructing an error field comprises: the method comprises the steps of carrying out global search in a target temperature interval through a differential evolution algorithm, carrying out fine adjustment through a nonlinear least square algorithm, obtaining a group of global optimal reference parameters and global reference models corresponding to the global optimal reference parameters, calculating relative errors of the global reference models on all training points, mapping discrete errors on a regular grid through a cubic spline interpolation method, and generating a continuous error field.
  8. 8. The method of claim 1, wherein identifying error regions by a connected region analysis algorithm based on a set error threshold and dividing the error field into a plurality of physical sub-regions comprises: Setting an error threshold value to be 2.5%, identifying a communication region exceeding the error threshold value in the error field through a communication region analysis algorithm, removing a micro-fragment region, and dividing the error field into a plurality of physical subregions.
  9. 9. The method of claim 1, wherein extracting the set of closed boundary points between the physical subregions, obtaining a partition boundary parameter equation between the physical subregions using a parameterized curve fitting method, comprises: extracting a closed boundary point set of each physical subarea from an error field through a contour line contour extraction algorithm, and introducing dimensionless parameters Representing the boundary curve as a set of parametric equations The closed boundary point set is aimed at a parameter equation set by using a symbolic regression algorithm And Fitting is carried out, and a partition boundary parameter equation between all the physical sub-areas is obtained.
  10. 10. The method of claim 1, wherein optimizing the reference parameters in the corresponding function skeleton independently for each high error expert region to obtain partitioned reference parameters, constructing a hybrid expert model from the function skeleton, the partitioned boundary parameter equations between the physical sub-regions, and the partitioned reference parameters, comprises: Maintaining the function skeleton unchanged, independently optimizing reference parameters in the function skeleton aiming at each high-error expert region, carrying out global search in a target temperature interval by a differential evolution algorithm in the optimization process, carrying out fine adjustment by a nonlinear least square algorithm, obtaining partition reference parameters for minimizing the mean square logarithm error in a physical subarea, and constructing a mixed expert model consisting of a set of function skeleton, partition boundary parameter equations among the physical subareas and partition reference parameters.

Description

High-temperature pneumatic unbalance factor correction method based on microscopic data Technical Field The invention relates to the technical field of hypersonic aerodynamics, computational Fluid Dynamics (CFD) and artificial intelligence assisted scientific computation intersection, in particular to a high-temperature pneumatic unbalance factor correction method based on microscopic data. Background When the hypersonic aircraft flies at Mach numbers of 10-30 in a 30-70 km airspace, the gas in the shock wave layer at the head is strongly compressed, and the temperature can be increased to be more than 20000K, so that a remarkable thermochemical unbalanced effect is caused. In this state, the flow field exhibits typical thermodynamic imbalance characteristics, namely translational-rotational temperatureWith vibration temperatureSignificant deviations occur and the resulting vibration-dissociation coupling effect directly affects shock deagglomeration distance, wall heat flow, and aerodynamic physical properties. Currently, a Park dual temperature model and its CVDV (Coupled Vibration-Dissociation-absorption) framework widely adopted in engineering Computational Fluid Dynamics (CFD) codes depend on chemical reaction rate parameters which are mainly obtained by fitting shock tube experimental data in the fifties to sixties of the twentieth century. Limited by the current wind tunnel experimental conditions, the measured dissociation rate data is only applicable to a temperature range of about 10000K at the highest, resulting in a significant reduction in the prediction reliability due to the fact that the reaction rate has to be extrapolated from the experience of low temperature data under extremely high temperature conditions. Related researches indicate that the empirical model has larger prediction deviation in a high-temperature area and the unbalance factor adopted by the empirical modelThe non-Boltzmann effect in microscopic level dynamics is difficult to reflect due to the pure empirical formula, so that obvious differences are easy to occur in peak heat flow prediction of an aircraft with a complex appearance. Therefore, a high temperature pneumatic unbalance factor correction method based on microscopic data is required. Disclosure of Invention The invention aims to overcome the defects of poor precision of an empirical model in a high-temperature area and insufficient physical interpretability and generalization capability of a pure data driving model in the prior art, and provides a high-temperature pneumatic unbalanced factor correction method based on microscopic data, which has definite physical meaning, high prediction precision and high calculation efficiency. In order to achieve the above object, the present invention provides a method for correcting a high-temperature aerodynamic unbalance factor based on microscopic data, comprising: acquiring a microscopic state-state simulation data set, extracting mapping relation data of unbalanced factors in the microscopic state-state simulation data set about translation temperature and vibration temperature, and constructing a basic feature space; The condition variation self-encoder is constructed, the distribution of the data in the basic feature space is learned, the directional sampling is carried out in the target temperature interval, and a reconstruction data set conforming to the physical distribution rule is generated through the trained condition variation self-encoder; Constructing a nonlinear constraint equation about pseudo temperature according to a physical definition type and physical quantity coupling relation of an unbalanced factor, carrying out iterative calculation on the unbalanced factor value in the reconstruction data set by using a numerical value optimization method, and reversely solving data of the corresponding pseudo temperature; Based on the function skeleton, global fitting is carried out on a reconstruction data set in a target temperature interval through a constraint optimization algorithm, a group of global optimal reference parameters are obtained, a global reference model is obtained, the relative error of the global reference model in the target temperature interval is calculated, and an error field is constructed; Based on a set error threshold value, identifying an error region in an error field through a connected region analysis algorithm, and dividing a target temperature interval into a plurality of physical subregions; extracting a closed boundary point set between each physical subarea, and obtaining a partition boundary parameter equation between each physical subarea by using a parameterized curve fitting method; And aiming at each physical subarea, independently optimizing the reference parameters in the function skeleton to obtain a plurality of groups of reference parameters, and constructing a mixed expert model integrated by the local optimal model of each subarea according to the function skel