摘要
在小子样结构响应试验数据样本的基础上,利用支持向量机回归的方法模拟了圆柱壳体动态极限应变峰值同壳体几何尺寸和外加脉冲载荷大小的非线性函数关系,同时通过改进的模拟退火单纯形混合算法优化了支持向量机的性能参数,并将支持向量机回归分析的预测性能同BP人工神经网络方法做了比较,验证了具有优化性能参数组合的支持向量机在小样本条件下更好的预测和推广能力.最后,从支持向量机回归模型导出了大尺寸圆柱壳体抗脉冲载荷的强度极限同自身几何尺寸的多元函数关系,从而为该类型壳体设备抗脉冲载荷的强度分析提供了一个可借鉴的预估模型.研究结果表明了支持向量机在机械结构的强度预估和可靠性分析等力学领域具有广泛的应用前景.
Based on experimental data of structural response of small samples, support vector machine (SVM) regression method is employed to simulate the nonlinear functional relationship among the peak value of dynamic strain in the cylinder shell, its size and external pulse loading. Meanwhile, an improved simplexsimulated annealing hybrid algorithm is developed to accomplish the optimization of SVM parameters. In addition, the comparative analysis of forecasting capacity between SVM and back propagation artificial neural network method is conducted. The theoretical results verify that SVM with optimal performance parameters has a better forecasting capacity under the small sampling condition. Finally, the multi-variable functional relationship between the ultimate strength of the large-sized cylinder shell and its size against pulse loading is inferred from the SVM regression model. This functional relationship can be served as a referable predication model for the strength analysis of this kind of cylinder shell devices. Theorefore, the above research shows that SVM will have wide applications in the mechanical structure analysis, such as strength predication and reliability analysis.
出处
《力学学报》
EI
CSCD
北大核心
2009年第3期383-388,共6页
Chinese Journal of Theoretical and Applied Mechanics
基金
国防预研基金资助项目(51411020603HTl402).~~
关键词
支持向量机
圆柱壳体
脉冲载荷
人工神经网络
强度预估模型
support vector machine, cylinder shell, pulse loading, artificial neural network, strength predication model