摘要
目的建立近红外快速预测片剂硬度的方法。方法采用硬度仪获得片剂真实硬度,运用偏最小二乘回归法(PLSR)和反向人工神经网络(BP-ANN)法建立近红外光谱与硬度之间的校正模型。结果偏最小二乘回归模型的相关系数r=0.9778,内部交叉验证均方根误差(RMSECV)为0.742 kg,预测均方根误差(RMSEP)为0.815 kg;反向人工神经网络训练集、监控集和测试集的相关系数r分别为0.987 3、0.985 6、0.986 8,各数据集的相对标准偏差(RSE%)分别为6.83%、8.77%、6.69%。结论反向人工神经网络非线性模型预测准确度要优于偏最小二乘回归线性模型。
OBJECTIVE To establish a method for predicting tablet hardness by near infrared diffuse reflection spectroscopy. METHODS Tablet hardness value was obtained by hardness meter. Calibration model between NIR spectra and the hardness was es- tablish by partial least squares regression （PLSR） method and BP-ANN method. RESULTS Correlation coefficients （r）, root mean squares error of cross-validation （ RMSECV ）, and root mean square error of prediction （RMSEP） obtained by PLSR model were 0. 977 8, 0. 742 and 0. 815 kg respectively. And the correlation coefficients of training set, monitor set and testing set by BP-ANN were 0. 987 3, 0. 985 6, and 0. 986 8, with RSE% of 6. 83%, 8.77%, and 6. 69%, respectively. CONCLUSION The prediction accuracy of BP-ANN nonlinear model is superior to the PLSR model
出处
《中国药学杂志》
CAS
CSCD
北大核心
2016年第11期904-909,共6页
Chinese Pharmaceutical Journal
基金
“十二五”重大新药创制专项(2012ZX09102201011)
江西省科技厅科技计划专项(20151BBG70031)
江西中医药大学科技计划(2013ZR0076)
关键词
近红外漫反射光谱
硬度
反向人工神经网络
偏最小二乘回归法
near infrared diffuse reflection spectrum
hardness
back-propagation artificial neural networks
PLSR
