紫外可见近红外光谱结合偏最小二乘法测定煤中挥发分  

Determination of volatile content in coal by ultraviolet-visible-near infrared spectroscopy combined with partial least squares method

作　　者：张淑利 翟璐璐 楚琰 张天龙[2] ZHANG Shuli;ZHAI Lulu;CHU Yan;ZHANG Tianlong(Xi'an Siyuan University,Xi'an 710038,China;College of Chemistry and Material Science,Northwest University,Xi'an 710127,China)

机构地区：[1]西安思源学院医学院,陕西西安710038 [2]西北大学化学与材料科学学院,陕西西安710127

出　　处：《冶金分析》2024年第6期18-24,共7页Metallurgical Analysis

摘　　要：以重量法为主的传统的挥发分测定方法具有分析精度差、检测耗时长以及样品前处理复杂等缺点。紫外可见近红外(UV-Vis-NIR)光谱技术分析速度快,操作简便,同时结合化学计量学方法能够有效降低煤样体系的复杂性对挥发分定量分析精确度的干扰,可实现对煤样品挥发分含量的准确定量分析。本文通过UV-Vis-NIR光谱技术结合化学计量学方法对煤炭中的挥发分进行快速定量分析。首先采用UV-Vis-NIR光谱技术得到不同煤样挥发分的光谱数据,其次优化偏最小二乘(PLS)模型的潜变量数(LVs)、一阶导数(1^(st)Der)的平滑点数、Savitzky-Golay平滑(SG平滑)的平滑点数和多项式阶数,最后考察标准正态变换(SNV)、1^(st)Der和SG平滑3种预处理方法分别建模以及集成后建模对模型预测结果的影响。结果表明,基于SNV-1^(st)Der-SG平滑集成预处理方法建立的PLS校正模型的预测性能最佳。当LVs为15,留一交叉验证(LOOCV)的决定系数(R_(cv)^(2))提升至0.9747,均方根误差(RMSE_(cv))降低至1.193%;外部验证得到决定系数(R_(p)^(2))为0.8625,均方根误差(RMSE_(p))为2.767%。该工作为煤炭性质的快速准确分析以及化石能源的高效利用提供实验基础和理论依据。The traditional determination methods of volatile content mainly by gravimetry have some shortcomings,such as poor analysis accuracy,long detection time and complicated sample pretreatment.Ultraviolet-visible-near infrared(UV-Vis-NIR)spectroscopy technology has the advantages of fast analysis speed and simple operation.It combined with chemometrics method could effectively reduce the interference of quantitative analysis accuracy of volatile content caused by the complexity of coal sample system,and achieve accurate quantitative analysis of volatile content in coal samples.The volatile contents in coal were rapidly and quantitatively analyzed by UV-Vis-NIR spectroscopy technology combined with chemometrics.Firstly,the spectral data of coal samples with different volatile contents were obtained by UV-Vis-NIR spectroscopy technology.Secondly,the latent variables(LVs)of partial least squares(PLS)model,the smoothing points of first derivative(1^(st) Der),the smoothing points and polynomial order of Savitzky Golay smoothing(SG smoothing)were optimized.Finally,the modeling was conducted using three preprocessing methods and their ensemble methods,namely standard normal transformation(SNV),1^(st) Der,and SG smoothing.Meanwhile,their impacts on the model prediction results were investigated.The results showed that the PLS correction model based on ensemble preprocessing method of SNV-1^(st) Der-SG smoothing had the best predicting performance.When the number of LVs was 15,the determinant coefficient(R_(cv)^(2))for the leave-one-out cross validation(LOOCV)increased to 0.9747,and the root-mean-square error(RMSE_(cv))was reduced to 1.193%.The determinant coefficient(R_(p)^(2))for external verification was 0.8625,and the root-mean-square error(RMSE_(p))was 2.767%.This study provided experimental and theoretical basis for the rapid and accurate analysis of coal properties and the efficient utilization of fossil fuels.

关 键 词：紫外可见近红外(UV-Vis-NIR)光谱 偏最小二乘法 挥发分 定量分析 煤 

分 类 号：O433.54[机械工程—光学工程] O657.38[理学—光学]