论文部分内容阅读
采用近红外光谱进行无创血糖检测时,样品背景变动造成的预测集样本与校正集样本量测体系不一致的问题是导致预测精度低的原因之一.提出一种将母体背景作为变量引入回归建模中,结合各个母体背景下的样本光谱信息构建三维光谱矩阵以提高校正模型稳健性的分析方法.将平行因子分析(PARAFAC)与多元线性回归(MLR)相结合,对人体三层皮肤模型的蒙特卡罗模拟实验和葡萄糖水溶液及其混合物的离体实验进行了验证.实验结果表明,与传统的单一母体背景所建立的偏最小二乘模型相比,将母体背景作为建模元素采用PARAFAC-MLR法所建立的校正模型具有更好的预测能力和稳健性.
When using near infrared spectroscopy to detect noninvasive blood glucose, the inconsistency between the sample set and the calibration sample set caused by the sample background change is one of the reasons leading to the low prediction accuracy.This paper presents a method to introduce the maternal background into regression modeling as a variable , Combined with the matrix information of each mother background to build a three-dimensional spectral matrix to improve the robustness of the calibration model analysis method.By combining parallel factor analysis (PARAFAC) and multiple linear regression (MLR), the three-layer human skin model of Monte Carlo simulation experiments and in vitro experiments with aqueous dextrose solution and their mixtures.The experimental results show that using the parental background as a modeling element compared with the traditional partial least squares model with single parental background, PARAFAC-MLR The calibration model established by the law has better predictive ability and robustness.