一种基于代理模型引导采样的多目标优化方法  

Multi-objective Optimization Method Based on Surrogate Model Guiding Sample

作　　者：陈国栋[1] 韩旭[1] 刘桂萍[1] 宁慧铭[1] 张正[1] 

机构地区：[1]湖南大学汽车车身先进设计制造国家重点实验室,长沙410082

出　　处：《系统仿真学报》2011年第10期2103-2107,共5页Journal of System Simulation

基　　金：国家重点基础研究发展计划(973)(2010CB832705);湖南大学汽车车身先进设计制造国家重点实验室自主课题(61075003)

摘　　要：针对复杂的实际工程多目标优化问题,提出代理模型引导采样的多目标优化方法。通过自适应的加强径向基函数构造的代理模型寻找Pareto优化解集,从找到的解集中通过一定策略筛选出部分作为样本点加入到下代样本空间中,样本点随着迭代的进行越来越靠近全局Pareto最优解集。从当前所有样本点中获得Pareto解,并根据其分布情况作为收敛条件。该方法中代理模型仅仅用来引导采样,也不需要反复计算大量样本点验证代理模型的精度,得到的解都被实际模型验证过。在典型多目标测试函数中体现了精度和效率。最后成功应用于薄板冲压成形变压边力优化中,表明了具有解决多目标实际工程优化问题的能力。In order to deal with engineering design problem with computation-intensive black-box functions, a multiobjective optimization method based on the surrogate model guiding samples was proposed. Some Pareto solutions from the surrogate models built by adaptive extended radial basis functions were selected and added to the samples space for the next iteration. In that trend, the samples were more and more close to the Pareto solutions of actual problem. According to the distribution of the Pareto solutions obtained from the samples, it was considered as convergence condition. This method only uses the surrogate models to guide samples without validating the accuracy of surrogate models, and the obtained Pareto solutions are validated by the actual model. The current method shows the accuracy and effectiveness in the multi-objective testing function and is successfully applied in the optimization of variable binder force for sheet metal forming. Finally, it is believed that it has a great potential to be a practical tool for multi-objective optimization problem.

关 键 词：多目标优化 代理模型 引导采样 薄板冲压成形 变压边力 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TG386[自动化与计算机技术—控制科学与工程]