数据预处理方法对油页岩含油率近红外光谱分析的影响

Influence of data preprocessing on NIRS analysis for oil yield of oil shale

为提高利用近红外光谱(NIRS)分析技术进行油页岩含油率的原位检测时的建模精度,需要采用适当的方法进行数据预处理。本研究利用实际和合成油页岩样品,结合光谱和矩阵2类数据预处理方法,研究不同方法及其组合对油页岩光谱数据一致性、样品含油率偏最小二乘法(PLS)的模型精度的影响。结果表明:在11种光谱数据预处理方法中,一阶导数、正则化、中心化以及适当的组合(如平滑、消噪和消基线后分别加中心化)等光谱预处理方法,可提高相同样品光谱数据间的一致性;在11种光谱预处理和3种矩阵处理的组合方法中,3种数据预处理组合方法(中心化、一阶导数2种光谱预处理+中心化矩阵预处理、一阶导数光谱预处理+正则化矩阵预处理)可提高合成样品含油率PLS模型的精度。

An appropriate data pre-processing method is needed to improve the modeling precision for oil yield analysis of oil shale in situ by the near infrared spectroscopy（NIRS）analysis technique. With nature and proportioning samples, the spectra preprocessing and Partial Least Square （PLS） modeling experiments were carried out by two ways （spectra and matrix） of data preprocessing. Results show that the appropriate spectra preprocessing （ including first derivative, auto-scaling, mean centralization and some combined methods, such as adding mean centralization after smoothing, denoising and de-baseline, respectively） can enhance the spectrum identity of the same sample. Three combined data preprocessing methods （including combined mean centralization or first derivative spectra preprocessing with auto-scaling matrix preprocessing） can improve the precision of NIRS analysis of PLS model for oil yield from proportioning samples.