为解决常规Kriging模型在设计空间拟合精度低的问题,提出了序列Kriging模型. 通过预期改善函数,在优化解附近和设计空间的稀疏区域增加样本点,不断更新近似模型,从而提高近似模型在兴趣域的拟合精度和全局预测能力. 通过两个数值算例分析了序列Kriging模型的全局搜索特性,从算例优化的统计结果可以看出,序列Kriging模型比常规Kriging模型的优化精度更高,稳健性更好. 最后,序列Kriging模型被应用到车身轻量化设计中,采用粒子群算法所得到的优化结果表明,近似模型的精度有了很大的提高,并且在满足耐撞性约束的同时所选部件的质量降低了23.35%.%In view of the low fitting accuracy of conventional Kriging model in design space, sequential Kriging model is proposed. New sample points are added in the neighborhood of optimization solution and the sparse region of design space by using expected improvement function, and the surrogate model is constantly updated to en-hance the fitting accuracy in the region of interest and the global prediction ability of the model. Through two numer-ical examples, the global searching characteristics of sequential Kriging model are analyzed. It can be seen from the statistical result of optimization numerical examples that the optimization accuracy and robustness of sequential Krig-ing model are higher than those of conventional Kriging model. Finally, sequential Kriging model is applied to a car-body lightweight design and the results of optimization with particle swarm optimization algorithm show that the accu-racy of surrogate model is greatly improved and the mass of selected components is reduced by 23.35% while satis-fying the constraints of crashworthiness.
展开▼
