论文部分内容阅读
目的:为提高小柴胡颗粒中黄芩苷近红外校正模型的准确性和预测精度。方法:基于模型的校正均方根误差(RMSEC)、预测均方根误差(RMSEP)、预测平均相对误差(PMRE)和模型对预测集的解释能力(Q2)参数,对比评价竞争自适应重加权法(CARS),蒙特卡洛-无信息变量消除法(MC-UVE),遗传算法(GA),子窗口重排(SPA)算法筛选和全波长变量,采用模群集群分析(MPA)+偏最小二乘法(PLS)方法的建模效果。结果:校正模型准确性和预测精度:CARS>MC-UVE>GA>全波长变量>SPA;CARS算法所建立校正模型预测均方根误差为1.700 4,决定系数R2为0.908 7,在α=0.05水平经配对t检验,50个外部验证样品实测值与预测值间无显著差异。结论:CARS算法筛选波长变量有效简化模型,提高模型预测的准确性和精度,适于小柴胡颗粒中黄芩苷含量的快速、无损检测。
Objective: To improve the accuracy and accuracy of baicalin near-infrared calibration model in Xiaochaihu granules. Methods: Based on model-based RMSEC, RMSEP, PMRE and model’s ability to interpret the predictive set (Q2), comparative assessment of competitive self-adaptive weighting (CARS), Monte-Carlo-no-information variable elimination method (MC-UVE), genetic algorithm (GA), sub-window rearrangement (SPA) Modeling effect of least square method (PLS) method. Results: Correction model accuracy and prediction accuracy: CARS> MC-UVE> GA> full-wavelength variable> SPA; the root mean square error of prediction model established by CARS algorithm was 1.700 4, R2 of determination coefficient was 0.908 7, The level of paired t test, 50 external validation samples measured values and predicted no significant difference between. Conclusion: The CARS algorithm can effectively simplify the model by screening wavelength variables and improve the accuracy and accuracy of the model prediction. It is suitable for rapid and non-destructive detection of baicalin in Xiaochaihu granules.