摘要
在近红外无创伤血糖浓度检测的基础研究中 ,对于多组分的混合物的分析 ,常因光谱与样品浓度之间呈现非线性响应 ,使得基于线性模型的校正方法失效。本文讨论了非线性校正方法径向基函数神经网络(RBFN)的有效性 ,并与线性校正方法中的主成分分析和偏最小二乘法作了对比研究。验证实验所用样品为①葡萄糖水溶液②包含牛血红蛋白和白蛋白的葡萄糖水溶液 ,结果表明 :在①实验中PLS模型和RBFN预测标准偏差分别为 8.2、8.9;在②实验中分别为 1 5 .6、8.8。可见在样品组分增多时 。
In fundamental study of non-invasive measurement of human blood glucose concentration with near-infrared spectroscopy, the linear calibration algorithms are not effective due to the nonlinear relationship between the infrared spectral absorbance and the concentration in multicomponent mixtures. The effectiveness of the radial basis function networks of nonlinear calibration model was presented and compared with the linear algorithms of the principal component regression and partial least-squares calibration models. The testing samples of this method are (1) glucose in aqueous matrix and (2) glucose in aqueous matrix containing bovine serum albumin and heucoglobin. The result shows that the root mean square error of prediction obtained with linear partial least-squares and the nonlinear radial basis function networks are 8.2, 8.9 in the first experiment respectively, the root mean square. error of prediction are 15.6, 8.8 in the second experiment respectively. Our results revealed that the calibration model of radial basis function networks produced significantly better prediction of glucose than the model of partial least-squares when the components of the samples increase.
出处
《分析化学》
SCIE
EI
CAS
CSCD
北大核心
2003年第4期440-443,共4页
Chinese Journal of Analytical Chemistry
基金
教育部科学技术研究重点项目
关键词
血糖浓度检测
基础研究
近红外光谱
糖尿病
光谱测量
near-infrared spectroscopy
non-invasive measurement
human blood glucose
radial basis function
neural network
