中国药事
中國藥事
중국약사
Chinese Pharmaceutical Affairs
2015年
8期
879-884
,共6页
近红外光谱%快速检验%无损分析%复方益肝灵胶囊%聚类分析%水飞蓟素%定量模型
近紅外光譜%快速檢驗%無損分析%複方益肝靈膠囊%聚類分析%水飛薊素%定量模型
근홍외광보%쾌속검험%무손분석%복방익간령효낭%취류분석%수비계소%정량모형
near infrared spectroscopy (NIRS)%rapid testing%non-destructive analysis%Compound Yiganling Capsules%clustering analysis%silymarin%quantitative model
目的：采用近红外光纤光谱技术对复方益肝灵胶囊进行快速判断厂家归属以及快速检定其中水飞蓟素的含量。方法：以全国6个企业生产的39批复方益肝灵胶囊作为分析对象，建立聚类分析模型。光谱预处理方法为二阶导数加矢量归一化；平滑点：13；光谱范围：9000~4100 cm-1；光谱之间距离采用欧氏距离，光谱和类之间距离采用方差平方和法（Ward’s算法）。同样以全国6个企业生产的39批复方益肝灵胶囊作为分析对象，建立定量分析模型。光谱预处理方法为一阶导数加矢量归一化法；光谱范围：7100~4250 cm-1和9100~7300 cm-1；采用偏最小二乘法回归。结果：聚类模型可以很好的区分训练集中的药品，在对作为测试集的光谱进行验证中全部判断正确。所建立的定量模型交叉验证均方根误差（RMSECV）为0.128，决定系数为96.54%，外部验证均方根误差（RMSEP）为0.144，相关系数为96.87%。结论：该聚类方法和定量模型快速、简便、准确和可靠，可以满足药品现场快速检查的需要。
目的：採用近紅外光纖光譜技術對複方益肝靈膠囊進行快速判斷廠傢歸屬以及快速檢定其中水飛薊素的含量。方法：以全國6箇企業生產的39批複方益肝靈膠囊作為分析對象，建立聚類分析模型。光譜預處理方法為二階導數加矢量歸一化；平滑點：13；光譜範圍：9000~4100 cm-1；光譜之間距離採用歐氏距離，光譜和類之間距離採用方差平方和法（Ward’s算法）。同樣以全國6箇企業生產的39批複方益肝靈膠囊作為分析對象，建立定量分析模型。光譜預處理方法為一階導數加矢量歸一化法；光譜範圍：7100~4250 cm-1和9100~7300 cm-1；採用偏最小二乘法迴歸。結果：聚類模型可以很好的區分訓練集中的藥品，在對作為測試集的光譜進行驗證中全部判斷正確。所建立的定量模型交扠驗證均方根誤差（RMSECV）為0.128，決定繫數為96.54%，外部驗證均方根誤差（RMSEP）為0.144，相關繫數為96.87%。結論：該聚類方法和定量模型快速、簡便、準確和可靠，可以滿足藥品現場快速檢查的需要。
목적：채용근홍외광섬광보기술대복방익간령효낭진행쾌속판단엄가귀속이급쾌속검정기중수비계소적함량。방법：이전국6개기업생산적39비복방익간령효낭작위분석대상，건립취류분석모형。광보예처리방법위이계도수가시량귀일화；평활점：13；광보범위：9000~4100 cm-1；광보지간거리채용구씨거리，광보화류지간거리채용방차평방화법（Ward’s산법）。동양이전국6개기업생산적39비복방익간령효낭작위분석대상，건립정량분석모형。광보예처리방법위일계도수가시량귀일화법；광보범위：7100~4250 cm-1화9100~7300 cm-1；채용편최소이승법회귀。결과：취류모형가이흔호적구분훈련집중적약품，재대작위측시집적광보진행험증중전부판단정학。소건립적정량모형교차험증균방근오차（RMSECV）위0.128，결정계수위96.54%，외부험증균방근오차（RMSEP）위0.144，상관계수위96.87%。결론：해취류방법화정량모형쾌속、간편、준학화가고，가이만족약품현장쾌속검사적수요。
Objective: To rapidly identify the Compound Yiganling Capsules from different manufactures by near-infrared fibre-optical spectroscopy and detect the content of silymarin.Methods:Thirty nine batches of Compound Yiganling Capsules from six manufactures were used as the objects. Clustering analysis model of Compound Yiganling Capsules was established. The spectra were pretreated with the second derivative and vector normalization; the number of smoothing point was thirteen; the wavelength range was 9000~4100 cm-1; the distance between the spectra was Euclidean distance, and the distance between the spectra and the class was the sum of square of the variance method (Ward’s method). The quantitative model was established using the same samples. The spectra were pretreated with the ifrst Derivative and vector normalization; The wavelength ranges were 7100~4250 cm-1 and 9100~7300 cm-1; The regression method was partial least squares (PLS) algorithm. Results:The clustering model can identify the samples of training set, and validate test set of spectra accurately. For the quantitative model of the content of silymarin, the root mean square error of cross validation (RMSECV) was 0.128 with the determination coefifcient R 96.54%. The root mean square error of prediction (RMSEP) was 0.144 with a correlation coefifcient R of 96.87%.Conclusion:The clustering analysis and the quantitative model were rapid, simple, accurate and reliable which can be applied to fast drug analysis.
