船舶结构多学科设计优化近似方法的比较

随着船舶结构设计日益复杂,使用统计学方法为复杂的计算机分析建立近似模型已经逐渐成为一种趋势.文章简要地介绍了三种多学科设计优化中常用的建立近似模型的方法;响应面、Kriging和径向基神经网络,并通过两个算例对其性能进行了比较.建立近似模型所需的样本点由一系列随机正交表提供,三种近似方法的性能则通过误差分析来进行比较.除了常用的二阶响应面模型,文中也构造了三阶响应面模型来进行比较.研究结果表明:Kriging方法以其准确性和稳定性,比其它方法更加适合船舶结构的多学科设计优化.

A Comparison of Approximation Methods for Multidisciplinary Design Optimization of Ship Structures

As the complexity of ship structural designs keeps on increasing, the use of statistical techniques to build approximation models for expensive computer analysis is getting popular. In this paper, three commonly used approximation methods for multidisciplinary design optimization (MDO) of complex engineering systems, including response surface modeling, Kriging and radial basis function network,are reviewed briefly and compared through two case studies. The sample points are selected by a series of randomized orthogonal arrays.Error analysis is then performed to measure the accuracy of the approximation models.Besides the traditional quadratic response surface,fully three-order polynomial response surface models are also constructed for the comparison.It is found that Kriging model is much more suitable for MDO of ship structures than other models for its high accuracy and stability.