论文部分内容阅读
采用新型微机电近红外(MEMS-NIR)光谱技术,在线监测金银花提取过程中绿原酸成分的含量。以高效液相色谱法为参考方法,采用Kennard-Stone法(KS)划分样本集,运用偏最小二乘(PLS)法建立其含量与NIR光谱之间的多元分析模型。通过组合间隔偏最小二乘法(Si PLS)对建模波段进行筛选,建立PLS模型。运用相对预测偏差(RPD)来评价模型的预测能力,并基于两类误差检测理论计算模型的多变量检测限(MDL),以MDL值进一步科学表达金银花提取过程的在线分析建模方法。结果表明应用MSC预处理方法所建模型最好,其交叉验证均方根误差(RMSECV)、校正均方根误差(RMSEC)和预测均方根误差(RMSEP)分别为1.707,1.487,2.362,校正集决定系数Rcal2为0.998 5,预测集决定系数Rpre2为0.988 1,且其RPD值为9.468,表明模型具有良好的预测性能。经Si PLS法筛选的多变量检测限MDL(0.042 15 g·L-1)小于未经变量筛选的MDL值,表明经Si PLS筛选有利于提高模型的预测性能。该研究从两类误差检测理论更加准确表达模型的预测能力,进一步说明MEMS-NIR光谱技术可用于金银花提取过程的在线监测。
A new micro-electromechanical near-infrared (MEMS-NIR) spectroscopy was used to monitor the content of chlorogenic acid in the process of honeysuckle extraction. Using high performance liquid chromatography as a reference method, Kennard-Stone method (KS) was used to divide the sample set and the PLS method was used to establish a multivariate analysis model between the content and NIR spectra. The PLS model was established by combining the interval partial least squares (Si PLS) to filter the modeling band. The relative predictive bias (RPD) was used to evaluate the predictive ability of the model. Based on two types of error detection theory to calculate the multivariate detection limit (MDL) of the model, an on-line analytical modeling method of further expressing the honeysuckle extraction process with MDL value was established. The results showed that the best model was established by MSC preprocessing. The root mean square error of validation (RMSECV), root mean square error of correction (RMSEC) and root mean square error of prediction (RMSEP) were 1.707, 1.487 and 2.362 respectively. Set decision coefficient Rcal2 is 0.998 5, prediction set decision coefficient Rpre2 is 0.988 1, and its RPD value is 9.468, which shows that the model has a good prediction performance. The multivariate detection limit (MDL) of 0.042 15 g · L-1, which was screened by Si PLS method, was smaller than that of the non-variable MDL, which indicated that screening by Si PLS was helpful to improve the predictive performance of the model. The research from two types of error detection theory to more accurately predict the model’s ability to predict that the MEMS-NIR spectroscopy technology can be used for on-line monitoring of honeysuckle extraction process.