The first-order and second-order differentiation results in processing near-infrared spectral data of gasoline samples with Grumwald-Letnikov fractional differential algorithm and commonly-used Savitzky-Golay algorithm were compared; the spectral data of 25 gasoline samples and 41 coal samples were smoothed with Savitzky-Golay algorithm, and then the first-order and second-order differential processing and 21-order differentiation processing from 0. 2-order to 2. 2-order were implemented with Grumwald-Letnikov algorithm ; through having processed data of gasoline samples combined with octane number and initial boiling point, and that of coal samples combined with volatile component, hydrogen content, nitrogen content index data, their data models were established by PLS and the optimal number of principal component was identified with whole cross validation leave-one method; having the data and processing method assessed with predictive residual error sum of squares (PRESS) and the correlation coefficient(R) shows that fractional differential can be applied to preprocessing near-infrared spectral data, and for the same data, the fractional differential order for their optimal values of different indexes differs.%比较了Grumwald-Letnikov分数阶微分算法和常用的Savitzky-Golay算法对汽油样品近红外光谱数据1阶微分和2阶微分结果.对25个汽油样品和41个煤炭样品的近红外光谱数据通过Savitzky-Golay算法进行平滑,平滑后l阶微分和平滑后2阶微分处理;通过Grumwald-Letnikov算法进行平滑后的0.2 ~2.2阶的21个阶次的微分处理.汽油样品的处理数据结合汽油的辛烷值、初馏点指标数据和煤炭样品的处理数据结合煤炭挥发分、氢含量和氮含量指标数据分别通过PLS建立数据模型,利用留一法全交互验证选取最优主成分.通过预测残差平方和( PRESS)和相关系数(R)对数据,处理方法进行评估.结果表明:分数阶微分可以应用于近红外光谱的数据预处理,对于相同数据的不同指标取得最优值的分数阶微分的阶次是不同的.
展开▼