依据IR和GC数据利用PLS建模预测粗苯原料中多组分的含量

PLS Model Predicted the Multicomponent Content in the Crude Benzol Based on IR and GC Data

基于粗苯原料中有效组分不能快速检测的现状,利用化学计量学和偏最小二乘法原理,进行了粗苯中苯、甲苯、邻二甲苯、间二甲苯、对二甲苯含量预测的研究。依据粗苯的红外光谱和气相色谱数据,采用交互验证方法,选取合适的维数、波段及图谱处理方式,建立了数学模型。校正集中苯、甲苯、邻二甲苯、间二甲苯、对二甲苯的相关系数分别为0.950 2,0.969 8,0.952 9,0.970 6和0.970 5,预测均方根误差分别为1.50,0.391,0.060 7,0.086和0.029 8。预测精度能满足企业要求,检测时间从0.5h降到了几分钟。

Because the effective component in crude benzol raw materialcan't be determined fast,according to the principle of chemical metrology,partial least-squares regression analysis was used to predict crude benzol samples which contain high benzene,toluene,and xylene(BTEX)constituents.The concentration data on GC were served as truth-value.Using cross-validation method,IR mathematical models were built and then were optimized by selecting suitable parameters of dimension,wave band and spectra?processing.In calibration set,correlation coefficients(R)of benzene,toluene,o-xylene,m-xylene,and pxylene were 0.950 2,0.969 8,0.952 9,0.970 6and 0.970 5;the root mean square prediction errors(RMSECV)were 1.50,0.391,0.060 7,0.086 and 0.029 8,respectively.The prediction accuracy could satisfy the requirement of enterprises,and testing time reduced from 0.5hours to a few minutes.