基于近红外光谱的辛烷值模型维护判断方法

A Judging Method on Model Maintenance of Octane Number Based on Near Infrared Spectroscopy

近红外光谱定量分析技术用于汽油性质检测具有快速、高效的特点,但该方法对模型的质量要求很高,需要及时进行模型维护。提出一种汽油辛烷值模型维护的判断方法,首先采用主成分分析观察校正集样本与待测样本的分布;再计算得到待测样本与校正集样本间的光谱距离,以及光谱距离阈值;最后通过比较待测样本与校正集样本的光谱距离及阈值来判断待测样本是否为模型界外样本。实验表明,该方法在生产工艺变化,待测样本超出模型范围时,能够及时给出报警,提示进行模型维护,进而确保预测精度。

A judging method on model maintenance of gasoline octane number is proposed in this paper.First,calibration set and the validation sample were analyzed by principal component analysis,then the spectroscopy distance and threshold between the validation sample and calibration set are worked out after calculation.Finally,whether or not the validation sample is outlier sample is judged by comparing the spectroscopy distance with the threshold.