CN-121997467-A - Multivariate parameter optimization analysis method based on orthogonal test and proxy model

CN121997467ACN 121997467 ACN121997467 ACN 121997467ACN-121997467-A

Abstract

The application belongs to the technical field of electric data processing, and particularly relates to a multivariate parameter optimization analysis method based on orthogonal tests and agent models, which is used for determining an initial training data set of an aircraft wing to be optimized, constructing two different types of agent models for establishing an approximate mapping relation between design variables and fatigue crack life, evaluating the prediction precision of the fatigue crack life by using a cross verification method, selecting a model with higher precision as a main optimization agent model according to an evaluation result, performing global optimization by adopting a mixed strategy combining two optimization algorithms, outputting an optimal design variable combination, and verifying. According to the application, the number of required initial tests or simulation times is obviously reduced through orthogonal test design, and meanwhile, the proxy model is utilized to replace the original objective function with high calculation cost, so that frequent calling of the original objective function is avoided, and the overall optimization efficiency is greatly improved.

Inventors

BAI JUN
MA YAOHUI
GUO WENJIE
AI SEN
GUO YUCHAO

Assignees

中国飞机强度研究所

Dates

Publication Date: 20260508
Application Date: 20260410

Claims (8)

1. A multivariate parameter optimization analysis method based on orthogonal test and proxy model is characterized by comprising the following steps: determining a plurality of design variables and value ranges thereof of an aircraft wing to be optimized, generating a sample point set covering the value ranges of the design variables based on an orthogonal test method, and acquiring a wing fatigue crack life value corresponding to each sample point to form an initial training data set; Step 2, based on the initial training data set, constructing two different types of agent models in parallel, wherein the agent models are used for establishing an approximate mapping relation between design variables and fatigue crack life; step 3, evaluating the prediction accuracy of the fatigue crack life of the two agent models constructed in the step 2 by using a cross verification method, and selecting a model with higher accuracy as a main optimization agent model according to an evaluation result; step 4, taking the maximized fatigue crack life as an optimization target, taking the main optimization proxy model optimized in the step 3 as a replacement prediction model of the optimization target, and adopting a mixed strategy combining two optimization algorithms to perform global optimization to obtain an optimal design variable combination; And 5, outputting the optimal design variable combination obtained by the optimization in the step 4, and verifying the optimal design variable combination through an actual test or simulation.
2. The orthogonal trial and proxy model based multivariate parameter optimization analysis method of claim 1, wherein the design variables comprise geometric parameters and material property parameters.
3. The multivariate parameter optimization analysis method based on orthogonal test and proxy model of claim 1, wherein the wing fatigue crack life value corresponding to each sample point is obtained by finite element analysis or fatigue test.
4. The method of multivariate parameter optimization analysis based on orthogonal experiments and proxy models of claim 1, wherein the two proxy models comprise a polynomial response surface model and a kriging model.
5. The method of multivariate parameter optimization analysis based on orthogonal test and proxy models of claim 2, wherein the geometric parameters include, but are not limited to, airfoil relative thickness, chord length, sweep angle, aspect ratio, and root tip ratio, and the material property parameters include, but are not limited to, elastic modulus, poisson ratio, material density, and fatigue strength coefficient.
6. The multivariate parameter optimization analysis method based on orthogonal experiment and proxy model of claim 1, wherein a hybrid strategy of a particle swarm optimization algorithm combined with a genetic algorithm is used for global optimization.
7. The multivariate parameter optimization analysis method based on orthogonal experiment and proxy model of claim 6, wherein the specific steps of global optimization by adopting a mixed strategy of a particle swarm optimization algorithm and a genetic algorithm are as follows: firstly, a particle swarm optimization algorithm is operated to perform rapid global exploration, and a high-quality combination of design variables is obtained; And taking the high-quality combination of the design variables obtained by the particle swarm optimization algorithm as an initial combination of the design variables of the genetic algorithm, starting the genetic algorithm to perform subsequent global search, and obtaining an optimal design variable combination.
8. The method of multivariate parameter optimization analysis based on orthogonal test and proxy model of claim 1, wherein when the verification result in step 5 does not reach the life expectancy target, then adding the design variable combination under verification and fatigue crack life values to the initial training dataset, updating the proxy model, and repeating steps 2 to 5 until the verification result reaches the life expectancy target.

Description

Multivariate parameter optimization analysis method based on orthogonal test and proxy model Technical Field The application belongs to the technical field of electric data processing, and particularly relates to a multivariate parameter optimization analysis method based on an orthogonal test and a proxy model. Background In many engineering fields such as engineering design, material research and development, process optimization and the like, there is a general need for collaborative optimization of a plurality of influencing factors (i.e., variables) in order to obtain an optimal performance index. Conventional parameter optimization methods, such as a full-factor test method or an experience-dependent trial and error method, generally require a large number of tests, so that the optimization process is high in cost and long in period, and is difficult to adapt to urgent requirements of modern engineering practice on efficiency. To reduce the number of tests, the orthogonal test method (Orthogonal Experimental Design) is widely used as an efficient test design method. The method can balance and cover various horizontal combinations of design variables in fewer test times, so that the main effect of each variable is effectively analyzed. However, the orthogonal test method is mainly a sample point arrangement strategy, does not have the capability of predicting untested points, and is difficult to accurately describe complex nonlinear response relations. In addition, when the objective function (such as high-fidelity simulation calculation) on which the optimization depends is extremely high in calculation cost, even if the number of tests is reduced, the efficiency of directly performing iterative optimization is still low. The proxy model (Surrogate Model) technique provides insight to cope with the high-cost computing problem. The technique is used for replacing an expensive original calculation process by constructing an approximate mathematical model of the objective function based on limited sample data, thereby greatly reducing the calculation overhead in the optimization process. In various proxy models, a polynomial response surface model is widely used because of its compact form and definite physical meaning, and a kriging model is used as a spatial interpolation method to represent higher precision when fitting a response surface with local fluctuation and nonlinear characteristics. However, a single proxy model has the adaptability limitation that a polynomial response surface model is difficult to accurately capture a high-order nonlinear relation, and a kriging model can have the problems of over fitting or extrapolation instability in a sample point sparse area. After the agent model is obtained, the choice of optimization algorithm is also critical. Traditional gradient-based optimization algorithms tend to fall into locally optimal solutions. Intelligent optimization algorithms such as genetic algorithm and particle swarm optimization algorithm have stronger global searching capability, but have defects, such as slower convergence speed or insufficient optimization precision in the later period of optimization. In summary, in the prior art, the test design method, the agent model technology and the optimization algorithm are often used independently or simply in series, and a complete analysis method for systematically fusing efficient test design, multi-class agent model complementary construction and hybrid intelligent optimization strategies is lacking. Such fracturing results in difficulty in achieving the goals of high efficiency, high accuracy and strong robustness simultaneously in the face of complex multivariate parameter optimization problems. Therefore, there is a need in the art for a generalized parametric optimization analysis framework that can integrate these advantages. Disclosure of Invention In order to solve the above problems, the present application provides a multivariate parameter optimization analysis method based on orthogonal test and proxy model, comprising: determining a plurality of design variables and value ranges thereof of an aircraft wing to be optimized, generating a sample point set covering the value ranges of the design variables based on an orthogonal test method, and acquiring a wing fatigue crack life value corresponding to each sample point to form an initial training data set; Step 2, based on the initial training data set, constructing two different types of agent models in parallel, wherein the agent models are used for establishing an approximate mapping relation between design variables and fatigue crack life; step 3, evaluating the prediction accuracy of the fatigue crack life of the two agent models constructed in the step 2 by using a cross verification method, and selecting a model with higher accuracy as a main optimization agent model according to an evaluation result; step 4, taking the maximized fatigue crack life as an optimizatio