光谱预处理和波长选择对混胺燃料各指标近外光谱定量分析结果的影响比较

Effects of spectral preprocessing and wavelength selection on content determination in mixed-amine by NIR spectroscopy

对混胺燃料近红外光谱进行光谱预处理和波长选择,采用偏最小二乘法建立各分析指标的定量校正模型,利用校正集标准偏差(SEC)和预测集标准偏差(SEP)比较光谱预处理和波段选择对各项指标分析模型的影响。平滑处理对各项指标的影响不大,一般采用5点平滑。由于混胺颜色的影响,二阶微分比一阶微分预处理光谱的效果要好。波段选择对各分析指标的影响差别较大,应综合考虑各组成含氢基团的贡献和其它成分共线性的影响,选择合适的波长范围。提出了建立高精度、高稳健性分析模型应注意的问题。

After spectral preprocessing and wavelength selection, quantitative calibration models for different contents of mixed-amine were established. Effects of different spectral preprocessing and wavelength selection on various elements were compared with squared error of calibration（SEC） and squared error of prediction（SEP）. Different number of smoothing points had few influence on the results, so 5 point-smoothing is normally selected. The results of preprocessing in second derivative were better than that in first derivative because of the deep color of mixed-amine. There were distinct results for different elements with spectral wavelength ranges changed, so it is important to select appropriate wavelength ranges according to the contribution of bydric radicals in one component and interference with the others. Some advices were presented to improve the veracity and robustness of NIR models.