样品选择结合分段直接校正法和偏最小二乘法用于近红外光谱分析模型传递研究

为了实现近红外光谱分析模型在不同仪器间的共享,对分析模型的传递方法进行了研究。以混胺燃料的分析模型为例,采用马氏距离法对转换集样品进行选择,采用分段直接校正(PDS)法结合偏最小二乘(PLS)法,对从仪器光谱进行校正。对转换集样品数、PLS回归的主因子数、PDS的窗口宽度进行选择,当转换集样品数为12、主因子数为1、窗口宽度为39时,转换集样品的光谱校正率达到96.5%.混胺分析模型传递后,预测集样品的标准偏差远远好于校正前的结果,与主仪器的非常接近。研究结果表明:采用马氏距离法选择转换集样品,并结合PDS-PLS算法计算转换矩阵,可有效扣除不同仪器间的光谱差异,提高混胺分析模型的传递精度,实现三乙胺、二甲苯胺、水分、密度的准确分析。

Application of Sample Selection and PDS-PLS Algorithms in Near Infrared Spectra Analysis Model Transfer

A transfer method of near infrared (NIR) spectra analysis model is studied in order to enable one instrument (reference instrument) and the other one (target instrument) to share the model. The transfer set samples are selected with Mahalanobis distances by taking an analysis model of mixed-amine fuel for example, and the algorithm of piecewise direct standardization (PDS) combined partial least squares (PLS) is used to set the spectral correction matrix between the reference and target instruments. The number of transfer samples, the number of prime factors of PLS regression and the window width of PDS are confirmed based on the average of root mean square (ARMS) of corrected spectra of transfer samples. The corrected rate of spectra of transfer samples reachs 96.5% with transfer sample number of 12 and prime factor number of 1 and window width of 39. Standard deviation of prediction set (SEP) of every analysis item on target instrument is near SEP on reference instrument after the analysis model of mixed-amine is transferred, the transferred model can predicate triethylamine content, xylidine content, water content and density accurately. The results indicate that the sample selection based on Mahalanobis distances and PDS-PLS algorithms can deduct the difference of NIR spectra among instruments and improve the analytical accuracy of NIR model.