摘要
构建代理模型所用样本点可以分为高保真(high-fidelity,HF)和低保真(low-fidelity,LF)两类,前者准确率高,后者计算成本低,为了同时发挥两者的优点,提出了一种基于加权最小二乘法的变保真代理模型(MFS-WLS),该模型利用少量HF样本和大量LF样本,计算LF修正因子和差异函数的未知系数以表示高保真响应。将MFS-WLS与工程优化领域常用的两种变保真和一种单保真代理模型进行比较,结果表明MFS-WLS具有较高的预测精度和鲁棒性,最后通过压力机上梁轻量化设计案例进一步验证了模型的实用性。
The sample points used to construct the surrogate model can be divided into high-fidelity(HF)and low-fidelity(LF).The former has high accuracy and the latter has low computational cost.In order to give full play to the advantages of both,a multi-fidelity surrogate model(MFS-WLS)based on weighted least squares method is proposed.MFS-WLS uses a few HF samples and many LF samples to calculate the unknown coefficients of LF scale factor and discrepancy function so as to represent the high-fidelity response.MFS-WLS is compared with two multi-fidelity and one single-fidelity surrogate models that are commonly used in the field of engineering optimization.The results show that MFS-WLS has high prediction accuracy and robustness.Finally,the practicability of the MFS-WLS is further verified by the lightweight design of the upper beam of the press machine.
作者
韩畅阳
王硕
刘富文
宋学官
HAN Changyang;WANG Shuo;LIU Fuwen;SONG Xueguan(School of Mechanical Engineering,Dalian University of Technology,Dalian 116024,China)
出处
《组合机床与自动化加工技术》
北大核心
2023年第12期21-24,29,共5页
Modular Machine Tool & Automatic Manufacturing Technique
基金
国家重点研发计划项目(2018YFB1700704)。
关键词
变保真代理模型
加权最小二乘法
优化设计
轻量化
multi-fidelity surrogate model
weighted least squares
optimization design
lightweight
