多参数非线性系统全局敏感性分析与动态代理模型研究

发布时间：2018-04-21 09:04

本文选题：多参数非线性系统 + 代理模型　；参考：《湖南大学》2014年硕士论文

【摘要】：本文的研究来源于国家自然科学基金项目(51275164)“基于全局敏感性分析和拟合基函数代理模型的汽车碰撞人体损伤稳健优化”。多参数非线性系统具有两个显著的特征一是参数众多，二是参数之间存在着强烈的交互效应，一个参数微小变动便可能使系统的输出结果产生很大的变化。多参数非线性系统的优化设计不仅计算工作量大，而且难以获得理论上的最优解。因此具有较高计算效率的代理模型技术和能够筛选重要参数的全局敏感性分析已成为多参数非线性系统的重点研究内容。本文在前人研究工作的基础上，进一步研究了全局敏感性分析和代理模型技术，，并应用于汽车乘员约束系统的概念优化设计中。本文的主要研究内容和成果如下： (1)通过代理模型的精度、效率和特征拟合三个方面，分别比较响应面模型、Kriging代理模型和径向基函数模型对不同设计变量个数和非线性程度高低的拟合能力。 (2)本文通过采用基于代理模型的全局敏感性分析方法，研究每个参数的重要程度，得到每个参数的主效应图。并通过四维非线性函数的测试算例说明了全局敏感性分析的计算原理与分析方法，验证了基于代理模型的全局敏感性分析的有效性。 (3)传统代理模型在拟合设计空间较大和设计变量较多的情况时，整个设计空间的局部精度不能完全得到保证且空间缩减方法容易误删全局最优解，针对这一问题，基于全局敏感性分析的主效应图信息，提出了设计空间自适应调整的动态代理模型优化策略。通过算例函数和一个减速器的优化设计实例，验证了本文优化策略的优越性。 (4)通过MADYMO软件，建立正面碰撞乘员约束系统的仿真模型，确定系统设计参数，通过对设计参数进行全局敏感性分析，得到敏感性强的重要参数。将本文提出的基于全局敏感性分析的设计空间自适应调整的动态代理模型优化策略应用到乘员约束系统重要参数的优化设计中，优化结果表明，乘员伤害值得到了明显的降低，该优化策略对提高汽车乘员约束系统的安全性具有明显的效果，在正面碰撞的汽车乘员约束系统的概念设计具有很好的参考和借鉴作用。
[Abstract]:The research of this paper comes from the National Natural Science Foundation of China 51275164) "robust optimization of human body damage in automobile collision based on global sensitivity analysis and fitting basis function agent model". There are two remarkable characteristics of multiparameter nonlinear system: one is that there are many parameters, the other is that there is a strong interaction effect between parameters. A small change of parameters may cause great changes in the output results of the system. The optimal design of multi-parameter nonlinear systems is difficult to obtain the theoretical optimal solution. Therefore, the agent model technology with high computational efficiency and the global sensitivity analysis, which can screen important parameters, have become the focus of research in multi-parameter nonlinear systems. Based on the previous research work, this paper further studies the global sensitivity analysis and agent model technology, and applies it to the conceptual optimization design of the vehicle occupant constraint system. The main contents and achievements of this paper are as follows: 1) through the accuracy, efficiency and feature fitting of the agent model, the ability of the response surface model and the radial basis function model to fit the number of different design variables and the degree of nonlinearity are compared respectively. In this paper, the importance of each parameter is studied by using a global sensitivity analysis method based on agent model, and the main effect graph of each parameter is obtained. The calculation principle and analysis method of global sensitivity analysis are illustrated by a test example of four-dimensional nonlinear function, and the validity of global sensitivity analysis based on agent model is verified. 3) when the traditional agent model has large design space and more design variables, the local precision of the whole design space can not be completely guaranteed and the space reduction method is prone to delete the global optimal solution by mistake. Based on the main effect graph information of global sensitivity analysis, a dynamic agent model optimization strategy with adaptive spatial adjustment is proposed. The superiority of the optimization strategy is verified by an example function and an example of optimal design of the reducer. 4) through MADYMO software, the simulation model of frontal collision occupant constraint system is established, and the design parameters are determined. By analyzing the global sensitivity of the design parameters, the important parameters with strong sensitivity are obtained. The dynamic agent model optimization strategy based on global sensitivity analysis is applied to the optimization of the important parameters of the occupant constraint system. The optimization results show that the damage value of the occupants is significantly reduced. The optimization strategy has obvious effect on improving the safety of the vehicle occupant restraint system, and it has a good reference and reference function for the conceptual design of the vehicle occupant restraint system in the frontal impact.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U491.61

【参考文献】