论文部分内容阅读
血糖浓度的近红外光谱分析中,奇异样本的存在会影响多元校正模型的精度。本研究建立了基于蒙特卡洛交互验证法(MCCV)的奇异样本去除方法,通过人体离体血浆实验和人体在体试验,验证了该方法在血糖近红外光谱分析中的应用效果,并与基于改进的无信息变量消除的无信息样本去除方法(MUVE-USE)进行了比较研究。实验结果表明,基于MCCV的奇异样本去除方法,除了与MUVE-USE一样可去除由于粗大误差(如样品损坏)或系统误差(如仪器漂移)产生的奇异样本外,还能同时去除对模型精度有影响的由于不确定原因产生的随机误差等奇异样本。去除多种奇异样本后建立的多元校正模型的精度明显提高。
In near-infrared spectroscopy of glucose concentrations, the presence of singular samples can affect the accuracy of the multivariate calibration model. In this study, we established a method based on Monte Carlo Mutual Verification (MCCV) for singularity sample removal. The results of this method were verified by in vitro plasma experiments and in vivo experiments in vivo. An improved non-information sample removal method without information variable elimination (MUVE-USE) was compared. The experimental results show that the MCCV-based singularity sample removal method can remove the singularity samples generated by coarse errors (such as sample damage) or system errors (such as instrument drift) as well as MUVE-USE, Singularity samples such as random errors due to uncertainties. The accuracy of the multivariate calibration model established after removing many kinds of singular samples is obviously improved.