藏药五脉绿绒蒿提取过程的在线近红外光谱质量控制研究  

Online Near Infrared Quality Control on Extraction Process of Tibetan Medicine Meconopsis Quintuplinervia Regel.

作　　者：龙若兰 冯丹 罗西 孙菁[1,2] LONG Ruo-lan;FENG Dan;LUO Xi;SUN Jing(Qinghai Provincial Key Laboratory of Qinghai-Tibet Plateau Biological Resource Research,Northwest Institute of Plateau Biology,Chinese Academy of Sciences,Xining 810008,China;Key Laboratory of Tibetan Medicine Research,Northwest Institute of Plateau Biology,Chinese Academy of Sciences,Xining 810008,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院西北高原生物研究所,青海省青藏高原特色生物资源研究重点实验室,青海西宁810008 [2]中国科学院西北高原生物研究所中国科学院藏药研究重点实验室,青海西宁810008 [3]中国科学院大学,北京100049

出　　处：《分析测试学报》2023年第8期920-929,共10页Journal of Instrumental Analysis

基　　金：国家自然科学基金面上项目(32270402);青海省自然科学基金面上项目(2023-ZJ-919M)。

摘　　要：利用近红外光谱技术和自建的在线检测系统,实现了藏药五脉绿绒蒿提取过程中总黄酮含量的在线近红外光谱监测和提取终点的判定。以403个样品为建模集,分别获得了主成分回归(PCR)、偏最小二乘(PLS)、决策树(DT)、随机森林(RF)算法下的最佳光谱预处理方法和建模区间,以残差预测偏差(RPD)值为指标选择最佳建模方法。以62个样品为外部验证集,考察模型应用于总黄酮含量实时监测的可行性。此外,还探讨了利用模型预测值进行相对浓度变化率(RCCR)分析直接判定提取终点的可行性,并比较了标准偏差绝对距离法(ADSD)和移动窗口标准偏差法(MBSD)对提取终点判定的适用性。结果表明,在预处理方法为Constant+一阶导数+SG平滑、建模区间5300~9000 cm^(-1)条件下所建的总黄酮含量的PLS模型效果最好,其校正集和验证集的误差均方根均小于0.14、相关系数均大于0.97,RPD值为4.68。所建PLS模型对未知样品的平均预测率为79%,实际值与预测值的相关系数大于0.98,表明模型有较好的预测效果。外部验证集中RCCR法判定的预测提取终点和ADSD法判定的提取终点均与实际提取终点一致。所建模型性能较好,通过对未知样品进行准确快速的定量分析,实现了五脉绿绒蒿提取过程中总黄酮含量的实时监测,同时,以RCCR和ADSD作为提取终点的判定方法较为准确,可为藏药材提取过程在线近红外光谱分析技术的研究提供有益借鉴。A near infrared spectroscopy with self-built online detection system was developed for the online detection of total flavonoids and the end-point determination in the extraction process of Meconopsis quintuplinervia Regel.in this paper.Total 403 samples were used as the modeling set to obtain the best pretreatment methods and modeling bands for principal component regression(PCR),partial least squares(PLS),decision tree(DT),and random forest(RF)algorithms,respectively.And the best modeling method was selected with the ration of prediction to deviation(RPD)value as the index.The feasibility for the assay model applied to real-time monitoring of total flavonoids con⁃tent was investigated with 62 samples as an external validation set.In addition,the feasibility for di⁃rect determination of the extraction end-point by relative concentration changing rate(RCCR)analysis was also investigated using the model prediction values.Futhermore,the suitabilities for the deter⁃mination of extraction endpoints by the absolute distance of standard deviation(ADSD)and moving block standard deviation(MBSD)method were compared.The results showed that the PLS model constructed under the pretreatment method Constant+first derivative+Savitzky-Golay smoothing and the modeling bands 5300-9000 cm^(-1) had the best results,which had the root mean squared er⁃rors for calibration and validation both less than 0.14,correlation coefficients for calibration and vali⁃dation both greater than 0.97,and a RPD value of 4.68.The average prediction rate of the con⁃structed PLS model for unknown samples was 79%,the correlation coefficient between the actual and predicted values was greater than 0.98,which meant that the model had a good prediction effect.The prediction extraction end-points determined by both RCCR and ADSD methods in the external validation sets were consistent with the actual end-point of 84 min.It can be seen that the performance of the proposed model was good enough.The real-time monitoring of the total flavonoids content

关 键 词：近红外光谱技术 质量控制 在线检测 五脉绿绒蒿 总黄酮 

分 类 号：O657.3[理学—分析化学] R284[理学—化学]