基于贝叶斯支持向量回归机的稳健参数设计  

Robust Parameter Design Based on Bayesian Support Vector Regression Machine

作　　者：周晓剑 顾翔 Zhou Xiaojian;Gu Xiang(School of Management,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学管理学院,南京210023

出　　处：《统计与决策》2023年第24期23-28,共6页Statistics & Decision

基　　金：国家自然科学基金资助项目(71872088)。

摘　　要：稳健参数设计是一种质量改进的重要技术,能够从产品生产的源头上减少和控制波动的产生。双响应曲面法是其常用的方法,主要是利用低阶多项式模型来拟合均值和方差响应,但当样本较复杂(如为非线性或者高维样本)时,低阶多项式模型的拟合性能往往较差,求解优化问题效果不佳。支持向量回归机对非线性数据有良好的拟合潜力,但其性能依赖于参数的合理设置,文章将贝叶斯优化应用于支持向量回归机的参数选择,并将优化后的模型应用于稳健参数设计中响应曲面模型的构建,提出一种基于贝叶斯支持向量回归机的稳健参数设计方法。试验结果表明,所提方法和其他常见优化方法相比,可以获得更精确的响应曲面,可以在实际应用中近似得到可靠的最优因子搭配水平。Robust parameter design is an important technique for quality improvement,which can be used to reduce and control fluctuations from the source of production.The dual response surface method is a commonly used one.It mainly uses a low-order polynomial model to fit the mean and variance responses.However,when the sample is complex(such as nonlinear or high-dimensional samples),the fitting performance of the low-order polynomial model is often worse,and the solution to the optimization problem is not effective.Support vector regression machine has good fitting potential to nonlinear data,but its performance depends on the reasonable setting of parameters.This paper applies Bayesian optimization to parameter selection of support vector regression machine,then uses the optimized model for the construction of response surface model in robust parameter design,and finally proposes a robust parameter design method based on Bayesian support vector regression machine.The experimental results show that the proposed method can be used to obtain more accurate response surfaces than other common optimization methods,capable of approximating reliable optimal factor collocation levels in practical applications.

关 键 词：稳健参数设计 支持向量回归机 贝叶斯优化 

分 类 号：O212.8[理学—概率论与数理统计]