为了提高人体血糖近红外光谱定量分析模型的预测精度， 结合净信号预处理（NAP）算法和径向基偏最小二乘（RBFPLS）回归建立了一种适合于人体血糖测量的非线性建模方法NAP-RBFPLS。 本文首先利用NAP对近红外光谱进行预处理来有效地提取原始光谱中仅与葡萄糖信号相关的光谱信息， 从而有效地减弱了人体血液中水、 白蛋白、 血红蛋白、 脂肪等成分的吸收干扰以及人体体温的变化、 测量仪器本身的漂移、 测量环境的变化和测量条件的变化引起的干扰因素与血糖变化的偶然相关问题； 然后把净信号预处理后的近红外光谱数据通过RBFPLS建立了非线性定量分析模型来解决由于人体强散射引起的血糖浓度与近红外光谱之间的非线性关系， 并与偏最小二乘（PLS）、 基于净信号预处理的偏最小二乘（NAP-PLS）和RBFPLS这三种建模方法建立的定量分析模型进行了对比分析。 实验结果表明， 这两种方法相结合建立的非线性校正模型对预测集的预测精度有了很大的提高， 这将对人体血糖浓度无创检测技术的研究具有实际应用价值。

In order to improve the prediction accuracy of quantitative analysis model in the near-infrared spectroscopy of blood glucose, this paper, by combining net analyte preprocessing(NAP) algorithm and radial basis functions partial least squares(RBFPLS) regression, builds a nonlinear model building method which is suitable for glucose measurement of human, named as NAP-RBFPLS. First, NAP is used to pre-process the near-infrared spectroscopy of blood glucose, in order to effectively extract the information which only relates to glucose signal from the original near-infrared spectra, so that it could effectively weaken the occasional correlation problems of the glucose changes and the interference factors which are caused by the absorption of water, albumin, hemoglobin, fat and other components of the blood in human body, the change of temperature of human body, the drift of measuring instruments, the changes of measuring environment, and the changes of measuring conditions; and then a nonlinear quantitative analysis model is built with the near-infrared spectroscopy data after NAP, in order to solve the nonlinear relationship between glucose concentrations and near-infrared spectroscopy which is caused by body strong scattering. In this paper, the new method is compared with other three quantitative analysis models building on partial least squares(PLS)， net analyte preprocessing partial least squares(NAP-PLS) and RBFPLS respectively. At last, the experimental results show that the nonlinear calibration model, developed by combining NAP algorithm and RBFPLS regression, which was put forward in this paper, greatly improves the prediction accuracy of prediction sets， and what has been proved in this paper is that the nonlinear model building method will produce practical applications for the research of non-invasive detection techniques on human glucose concentrations.