基于马氏距离和稀疏矩阵技术的激光诱导击穿光谱(LIBS)煤质灰分分析  

LIBS Coal Ash Analysis Based on Mahalanobis Distance and Baseline Estimation and Denoising Using Sparsity Data Preprocessing

作　　者：李盛冬 倪明辉 许斐 韦祎 李燕[2] LI Shengdong;NI Minghui;XU Fei;WEI Yi;LI Yan(Nanjing Coal Quality Supervision and Inspection Co.Ltd.,China Energy Corporation,Nanjing,Jiangsu 210031,China;School of Chemistry and Chemical Engineering,Nanjing University of Science and Technology,Nanjing,Jiangsu 210094,China)

机构地区：[1]国能南京煤炭质量监督检验有限公司,南京210031 [2]南京理工大学化学与化工学院,南京210094

出　　处：《中国无机分析化学》2022年第4期97-102,共6页Chinese Journal of Inorganic Analytical Chemistry

基　　金：国家自然科学基金资助项目(51676100,21207066)。

摘　　要：为了提高激光诱导击穿光谱(LIBS)定量测定煤质的精度,先对原始数据进行预处理,包括异常值剔除、基线校正、谱线筛选,再将LIBS与偏最小二乘回归法(PLSR)结合建立定量模型以应用于煤质灰分的分析。结果表明,经过预处理后训练样品的拟合度(R^(2))从0.9740提高到0.9841,均方根误差(RMSE)从0.9613降低到了0.7527,预测均方根误差(RMSEP)从2.2731降到2.0017,同时平均绝对误差(MAE)和平均相对误差分别从1.9747、0.1094降低到1.5572、0.0757。研究表明,基于马氏距离(MD)的异常数据剔除算法结合基于稀疏矩阵技术的基线估计与降噪算法(BEADS),在一定程度上能够改善数据的稳定性和光谱信噪比,有利于提高数据建模的预测精度。To improve the accuracy of laser-induced breakdown spectroscopy(LIBS)quantitative measurement of coal quality,the original data were preprocessed,including outlier elimination,baseline correction and spectral line screening,and then LIBS and partial least squares regression(PLSR)were combined to establish a quantitative model for the analysis of coal ash.The results showed that the fitting degree(R^(2))of training samples increased from 0.9740 to 0.9841 after pretreatment,the root mean square error(RMSE)decreased from 0.9613 to 0.7527,and the root mean square error(RMSEP)decreased from 2.2731 to 2.0017.The mean absolute error(MAE)and mean relative error decreased from 1.9747 and 0.1094 to 1.5572 and 0.0757,respectively.The study shows that the abnormal data removal algorithm based on Mahalanobis Distance(MD)combined with the baseline estimation and noise reduction algorithm based on Baseline Estimation And Denoising using Sparsity(BEADS)can improve the data stability and spectral signal-to-noise ratio to a certain extent,which is conducive to improving the prediction accuracy of data modeling.

关 键 词：激光诱导击穿光谱 煤灰分 马氏距离 基于稀疏矩阵技术的基线估计与降噪算法 偏最小二乘法 

分 类 号：O657.38[理学—分析化学] TQ533[理学—化学]