首页 | 本学科首页   官方微博 | 高级检索  
     

基于多变量高斯过程模型的贝叶斯建模与稳健参数设计
引用本文:冯泽彪,汪建均,马义中. 基于多变量高斯过程模型的贝叶斯建模与稳健参数设计[J]. 系统工程理论与实践, 2020, 40(3): 703-713. DOI: 10.12011/1000-6788-2019-0200-11
作者姓名:冯泽彪  汪建均  马义中
作者单位:南京理工大学 经济管理学院, 南京 210094
基金项目:国家自然科学基金(71771121,71931006);中央高校基本科研业务费专项资金(30915011102);江苏省研究生科研与实践创新计划项目(KYCX19_0345)
摘    要:针对模型预测偏差和波动的稳健参数设计问题,在多变量高斯过程(multivariate Gaussian process,MGP)建模的框架下,结合质量损失函数和非线性优化约束方法构建一个新的多响应优化模型.首先,利用成对估计方法获得超参数近似值,构建多变量高斯模型;其次,结合MGP模型特征,构造充分考虑响应波动因素的质量损失函数.利用蒙特卡罗模拟方法,获得响应落入指定区间的期望概率;然后,以期望概率为约束,结合本文所提质量损失函数建立优化模型;最后,利用全局优化算法进行寻优,获得考虑响应期望概率的优化结果.实际案例和软件仿真表明,该方法综合权衡了预测偏差和预测波动引起的不确定性对优化结果的影响.获得了兼顾质量损失和期望概率最优均衡解,从而实现稳健参数设计.

关 键 词:多变量高斯过程模型  质量损失函数  期望概率  多响应  稳健参数设计
收稿时间:2019-01-30

Bayesian modeling and robust parameter design based on multivariate Gaussian process model
FENG Zebiao,WANG Jianjun,MA Yizhong. Bayesian modeling and robust parameter design based on multivariate Gaussian process model[J]. Systems Engineering —Theory & Practice, 2020, 40(3): 703-713. DOI: 10.12011/1000-6788-2019-0200-11
Authors:FENG Zebiao  WANG Jianjun  MA Yizhong
Affiliation:School of Economics and Management, Nanjing University of Science and Technology, Nanjing 210094, China
Abstract:For the robust parameter design problem of prediction deviation and variability, a new optimization model is constructed by combining the quality loss and Bayesian posterior estimation method under the framework of multivariate Gaussian process (MGP) modeling. Firstly, the hyperparameters are obtained by using the pairwise estimation method, and the MGP is constructed. Secondly, Monte Carlo simulation method is used to obtain the expected probability that the responses fall within the specified intervals. Then, the optimization model is established by using the quality loss function proposed in this paper with the expected probability as the constraint. Finally, the global optimization algorithm is used to perform global optimization, and the optimization results considering the expected probability are obtained. The example shows that proposed method comprehensively considers the impact of prediction deviation and variability on the optimization result. The optimal considering quality loss and expected probability is obtained, and realizing robust parameter design.
Keywords:multivariate Gaussian process model  quality loss function  expected probability  multi-response  robust parameter design  
本文献已被 CNKI 维普 等数据库收录!
点击此处可从《系统工程理论与实践》浏览原始摘要信息
点击此处可从《系统工程理论与实践》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号