基于增广SVM的结构动力学模型修正方法研究

发布时间：2018-02-11 14:57

本文关键词： 代理模型多目标优化增广SVM 模型修正　出处：《振动与冲击》2017年15期 　论文类型：期刊论文

【摘要】：研究了基于代理模型的有限元模型修正方法,针对支持向量机(Support Vector Machine,SVM)在处理非线性程度不高函数时容易出现过拟合,提出了一种采用混合基函数形式的增广SVM方法。该方法首先是在结构动力学试验结果和结构有限元模型计算分析结果的基础上,根据设计要求、灵敏度计算或工程经验选择适合的待修正参数、修正范围来确定修正样本空间,并给出样本点,其次采用增广SVM方法构造每组样本点和与之对应的目标函数之间的代理模型,采用基于Pareto最优解的多目标优化方法,以代理模型输出为目标,样本空间为变量,寻找待修正参数在修正区间内的全局最优解。用代理模型代替原有的有限元模型进行相关的计算分析,避免在模型修正过程中反复调用原有限元模型进行计算带来的高昂计算成本。通过算例一表明,增广SVM的预测结果较传统SVM方法精度更高,而算例二、三则说明所提出的基于增广SVM方法的结构动力学模型修正方法具有实际应用价值,同时计算结果具有很高的精度。
[Abstract]:In this paper, the finite element model modification method based on agent model is studied. For support vector machine support Vector machine, it is easy to overfit when dealing with the function of low nonlinearity. An augmented SVM method in the form of mixed basis function is proposed. The method is based on the experimental results of structural dynamics and the calculation and analysis of the finite element model of the structure, according to the design requirements. Sensitivity calculation or engineering experience selects suitable parameters to be corrected, modifies the range to determine the modified sample space, and gives the sample points. Secondly, an augmented SVM method is used to construct a proxy model between each set of sample points and the corresponding objective functions. The multi-objective optimization method based on Pareto optimal solution is adopted. The output of the agent model is taken as the target and the sample space is taken as the variable. The global optimal solution of the parameters to be modified in the modified interval is found. The agent model is used instead of the original finite element model to carry out the relevant calculation and analysis. This paper avoids the high calculation cost caused by repeated calling the original finite element model in the process of model modification. The calculation example shows that the prediction result of the augmented SVM is more accurate than that of the traditional SVM method, and the second example shows that the proposed method is more accurate than the traditional SVM method. Three points show that the proposed modified structural dynamic model based on augmented SVM method has practical application value, and the calculation results have high accuracy at the same time.
【作者单位】：南京航空航天大学机械结构力学及控制国家重点实验室;
【基金】：国家自然科学基金(11472132) 江苏省研究生培养创新工程(KYLX_0223) 中央高校基本科研业务费资助(NS2014002) 江苏高校优势学科建设工程
【分类号】：O342

【相似文献】