基于机器学习的硝酸拉曼光谱定量分析方法
Quantitative Analysis of Nitric Acid Using Raman Spectroscopy Based on Machine Learning
张雅茹 1程碧瑶 1杨博 1王小卓 1任文贞2
作者信息
- 1. 中国兵器工业集团第二一二研究所,陕西西安 710065
- 2. 中国科学研究院西安光学精密机械研究所,陕西 西安 710119
- 折叠
摘要
针对太安产线智能化检测中存在的硝酸拉曼光谱随浓度变化复杂、难以通过传统方法实现高精度定量分析的问题,提出一种基于机器学习的硝酸拉曼光谱高精度定量分析方法.三种机器学习算法包括偏最小二乘回归、支持向量机回归及随机森林回归算法首次被用于构建硝酸拉曼光谱定量分析模型.将经去噪、背景扣除、归一化等预处理后的硝酸拉曼光谱作为模型输入,硝酸浓度作为模型输出,采用网格搜索算法结合五折交叉验证优化模型超参数后,三种模型在测试集数据上的R2均大于0.995,且随机森林回归模型最优,浓度预测均方误差可达0.356.实验结果表明,利用基于机器学习的拉曼光谱定量分析技术,可为硝酸浓度测量提供一种无损、高精度检测方法.
Abstract
Aiming at the problem that the traditional methods in a modern production line of pentaerythrite tet-ranitrate(PETN)with intelligent detection technologies cannot achieve high precision quantitative analysis of the nitric acid Raman spectra due to the complicated changes of concentration,a high precision quantitative anal-ysis method based on machine learning was proposed.Three machine learning algorithms including partial least squares regression,supporting vector regression and random forest regression were firstly applied for construc-ting nitric acid quantitative models,which took the pre-processed spectra after smoothing,background subtrac-ting and normalization as input and nitric acid concentration as output.After optimization of the hyper-parame-ters in the three models by using GridSearchCV method coupled with 5-folds cross verification,the three models had all achieved the value of R2＞0.995 in the test packages and the random forest regression model was con-firmed as optimal,of which the mean square error was as low as 0.356.Experimental results showed that non-destructive,high precision detection for nitric acid concentration could be achieved by adopting Raman spectros-copy coupled with machine learning.
关键词
硝酸浓度/机器学习/拉曼光谱/定量分析Key words
nitric acid concentration/machine learning/Raman spectroscopy/quantitative analysis引用本文复制引用
出版年
2024
NETL
NSTL
万方数据