文献类型：期刊文献

中文题名：在线自适应波长选择方法及其在汽油调和过程中的应用

英文题名：Online adaptive wavelength selection method and its application in gasoline blending process

作者：汪恺[1,2];杜文莉[1,2,3];隆建[1,2]

机构：[1]华东理工大学信息科学与工程学院,上海200237;[2]华东理工大学化工过程先进控制和优化技术教育部重点实验室,上海200237;[3]化学工程联合国家重点实验室(华东理工大学),上海200237

年份：2021

卷号：72

期号：2

起止页码：1059

中文期刊名：化工学报

外文期刊名：CIESC Journal

收录：CSTPCD;;EI(收录号：20211010022553);Scopus;北大核心:【北大核心2020】；CSCD:【CSCD2021_2022】；

基金：国家杰出青年科学基金项目(61725301);国家自然科学基金项目(61973124,61873093)。

语种：中文

中文关键词：测量;模型;数值分析;近红外光谱分析技术;波长选择;汽油调和;高斯过程回归

外文关键词：measurement;model;numerical analysis;NIR;wavelength selection;gasoline blending process;Gaussian process regression

摘要：近红外光谱分析技术作为一种非侵入性的分析手段在工业上得到了广泛应用。然而,大多数近红外模型的波长选择方法是离线建立的,无法有效跟踪过程特性的变化。提出了一种新的在线自适应波长选择方法——在线自适应区间高斯过程回归波长选择方法(adaptive interval Gaussian process regression,AIGPR),并用于汽油调和过程中的近红外模型的建立。该方法可以根据待测样本的特性对波长结构进行调整。为了降低在线应用的计算成本,该方法分为离线和在线两个部分,离线部分将光谱分割成若干个波长区间,并在每个波长区间上建立局部模型,为在线应用做准备;在线部分中根据划分规则将采样得到待测样本光谱进行分割并代入相应的局部模型中计算波长区间重要性指标,获得最优波长区间。在汽油辛烷值的光谱数据上证明了该方法的有效性。与重要变量投影法和改进的相关系数法相比,该方法具有更好的性能。
The near-infrared spectroscopy analysis technology has been widely used in industry as a non-invasive analysis method.However,most of the wavelength selection methods of NIR models are established offline,which cannot effectively track the change of process characteristics.In this paper,a new online adaptive wavelength selection method,online adaptive interval Gaussian process regression(AIGPR),is proposed and applied in gasoline blending process.The proposed method can adjust the wavelength structure according to the characteristics of the query samples.In order to reduce the computing cost of online applications,the proposed method is divided into two parts:offline and online.In the offline part,the spectrum is divided into several wavelength intervals,and a local model is established in each wavelength interval to prepare for online application;in the online part,the query samples are segmented according to the division rules and used to calculate the wavelength importance index by the corresponding local model to obtain the optimal wavelength interval.The effectiveness of the method is proved on the gasoline data.Compared with the variable importance in the projection(VIP)method and the improved relationvariance(RV)method,AIGPR has better performance.