基于红外光谱与多种预处理组合的柴油纯度检测方法

STUDY ON DIESEL OIL PURITY DETECTION BASED ON INFRARED SPECTROSCOPY AND VARIOUS PRETREATMENT COMBINATIONS

利用红外光谱技术采集51种不同体积分数的柴油光谱数据,采用主成分马氏距离法剔除异常样本,通过71种预处理组合及偏最小二乘法建立了柴油纯度性质模型。结果表明:合理预处理后的建模效果明显优于未经预处理的效果,预测均方根误差(RMSEP)降到了0.0400以下,相关系数(R_(p))达到了0.9900以上,且预处理组合方法的顺序不同,其建模效果即模型评价指标RMSEP、R_(p)也不同;1stderivative+SNV、1^(st) derivative+SNV+center、1^(st) derivative+center+SNV、center+1stderivative+SNV这4种预处理组合方法的模型预测效果最好,RMSEP可达0.0187,R_(p)可达0.9978,可适用于柴油纯度光谱数据的处理,实现柴油纯度的检测。

The 51 kinds of diesel spectral data with different volume fractions were collected by infrared spectroscopy technology,the diesel purity property model was established by 71 kinds of pretreatment combination and partial least square method(PLS)after abnormal samples were eliminated by Mahalanobis distance method.The results show that the modeling effect after reasonable pretreatment is significantly better than that without pretreatment.The root-mean-square error of prediction(RMSEP)decreases to below 0.0400,and the correlation coefficient(R_(p))is higher than 0.9900.Moreover,due to the different order of pretreatment combination methods,the modeling effect(model evaluation index RMSEP and R_(p))also varies.Among them,the models with 1^(st) derivative+SNV,1^(st) derivative+SNV+center,1^(st) derivative+center+SNV,and center+1^(st) derivative+SNV have the best prediction performance.The root-mean-square error of prediction is as high as 0.0187,and the correlation coefficient is up to 0.9978.It can be seen that the above pretreatment combination method is suitable for processing diesel purity spectral data and can achieve detection of diesel purity.