基于优化的改进移动最小二乘代理模型方法  

An Improved Moving Least Square Method Metamodal Based on Optimization

作　　者：于成龙[1] 刘莉[1] 龙腾[1] 邢超[1] 彭磊[1] 

机构地区：[1]北京理工大学宇航学院,北京100081

出　　处：《航空计算技术》2013年第1期85-88,共4页Aeronautical Computing Technique

摘　　要：针对移动最小二乘代理模型精度受影响域半径影响的问题,提出了一种基于优化的改进移动最小二乘法代理模型方法。构造移动最小二乘代理模型时采用遗传算法获取最佳影响域半径,提高近似精度进而达到减少样本点的目的。通过标准数值测试算例和NASA减速器优化设计实例验证,大大提高了移动最小二乘法代理模型的近似精度,与标准的移动最小二乘法相比,仅需较少的样本点即可达到相同的近似精度。Since the accuracy of moving least square method metamodal is influenced by the radius of in- fluence domain, an improvement of moving least square method based on optimization is proposed to over- come such defect above. To improve the approximate accuracy of moving least square method metamodel and reduce the amount of sample points, the best radius of influence domain is got by the genetic algo- rithm before the moving least square method metamodel is structured. Validated by using two benchmark numerical test problems and the NASA speed reducer optimization design, this method improves the ap- proximate accuracy of moving least square method metamodel. Compared to standard Moving Least Square Method, this method needs less sample points when approximate accuracy is the same.

关 键 词：代理模型 移动最小二乘 影响域半径 遗传算法 

分 类 号：O241.5[理学—计算数学]