大规模混合设计变量结构优化设计研究

项目名称： 大规模混合设计变量结构优化设计研究

项目编号： No.11302173

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 数理科学和化学

项目作者： 肖曼玉

作者单位： 西北工业大学

项目金额： 26万元

中文摘要： 现代复杂结构系统苛刻的性能要求不仅需要对形状、尺寸等连续变量进行优化设计，也对材料、型材选择等分类变量优化问题提出了新的挑战，以上两类混合变量相互耦合，不可分割，成为当今结构系统性能设计丞待解决的关键问题。为此，本项目提炼出对含连续-分类混合设计变量大规模结构优化这一基础问题,探讨基于多核基-支持向量机优化设计新方法，为实现对分类变量属性值的数值描述，借助聚类方法及单纯形法将分类变量属性值空间映射到高维数值空间，消除传统整型和离散描述方式的缺点，提高不同属性值的分辨率，进一步为减少整体优化设计时间，提高效率，构建借助缩减模型遗传算法多级近似优化模型，为高效实现大规模复杂结构连续-分类混合变量优化问题提供可行的优化设计方案，具有重要的理论意义和应用价值。

中文关键词： 混合设计变量；单纯形法；多核基函数；支持向量机；遗传算法

英文摘要： In the context of complex structural system, more and more rigorous performance requirements put forward a new challenge because of not only considering continuous design variables on shape, size of structure, but also needing categorical design variables on material selection, shape structure selection simultaneously. Since the above two kinds of mixed design variables are mutual coupling and inseparable, it has become a handling problem of structural design optimization. Therefore, in this project, a systematic methodology combining multiple kernel learning method and support vector machine is developed. The proposed method should not only remove the influence of the definition of the mixed design spaces, but also increase the weight factor of each attribute with aid of clustering for all attribute values of mixed variables, in which it is possible to map more physically the initial design space into a numerical high dimensional space based on a simplex representation. Furthermore, the evolutionary optimization assisted by model reductions will be performed to reduce the global computational time and increase the design performance. A successful achievement of the project goals will definitely lead to a significant breakthrough in the scientific field as well as in multiple industrial applications.

英文关键词： mixed variables；regular simplex method；multiple kernel functions；support vector machine；evolutionary algorithm