双模型结合进一步降低预测均方根误差和均方根相对误差的方法

前期研究工作提出了以预测均方根相对误差最小为回归目标的方法(Minimization of prediction relative error,MPRE),它能使得预测结果的均方根相对误差更小。偏最小二乘法(Partial least squares,PLS)是以预测均方根误差为回归目标,能使得预测结果的均方根误差更小。基于多模型结合的思想,提出将MPRE与PLS相结合的双模型结合多元校正方法。本方法步骤为:(1)分别采用MPRE与PLS法对校正集建模;(2)计算阈值;(3)分别采用已建立好的MPRE与PLS模型进行预测;(4)将预测结果与阈值进行比较,得到预测结果。通过对酒精的近红外光谱与汽...

Full description

Saved in:
Bibliographic Details
Published in分析化学 Vol. 43; no. 5; pp. 754 - 758
Main Author 吴雪梅 刘志强 张天龙 李华
Format Journal Article
LanguageChinese
Published 西北大学分析科学研究所,西安710069 2015
西安文理学院化学与化学工程学院,西安710065%第二炮兵工程大学,西安,710025%西北大学分析科学研究所,西安,710069
Subjects
Online AccessGet full text
ISSN0253-3820
DOI10.11895/j.issn.0253-3820.140915

Cover

More Information
Summary:前期研究工作提出了以预测均方根相对误差最小为回归目标的方法(Minimization of prediction relative error,MPRE),它能使得预测结果的均方根相对误差更小。偏最小二乘法(Partial least squares,PLS)是以预测均方根误差为回归目标,能使得预测结果的均方根误差更小。基于多模型结合的思想,提出将MPRE与PLS相结合的双模型结合多元校正方法。本方法步骤为:(1)分别采用MPRE与PLS法对校正集建模;(2)计算阈值;(3)分别采用已建立好的MPRE与PLS模型进行预测;(4)将预测结果与阈值进行比较,得到预测结果。通过对酒精的近红外光谱与汽油紫外光谱进行定量分析结果表明,本方法可进一步减小预测均方根误差与相对误差。
Bibliography:Double models; Multivariate calibration; The root-mean-square relative error; Root-mean-squareerror
22-1125/O6
WU Xue-Mei, LIU Zhi-Qiang, ZHANG Tian-Long , LI Hua 1 ( Institute of Analytical Science, Northwest University, Xi'an 710069, China) 2(Department of Chemistry and Chemical Engineering, Xi'an University, Xi'an 710065, China) 3( The Second Artillery Engineering University, Xi'an 710025, China)
A method for multivariate calibration with minimization of root-mean-square relative error of prediction (RMSREP) has been proposed in previous work, and the method in this paper is named MPRE method. MPRE is based on the use of back-propagation artificial neural network (BP-ANN). The regression objective of MPRE method is to minimize RMSREP by changing the output values of BP-ANN. Partial least squares (PLS) model was widely used in analytical field as it can minimize the root-mean-square error of prediction (RMSEP). A method based on double models combination that employed the idea of ensemble MPRE and PLS models i
ISSN:0253-3820
DOI:10.11895/j.issn.0253-3820.140915