在测量人体皮肤中红外光谱的过程中， 皮肤与ATR晶体之间的接触面积很难保持一致， 导致倏逝波与人体皮肤的作用光程会产生差异。 以光程变化信息与葡萄糖浓度信息之间的相关性为基础， 建立了两个分析葡萄糖含量的偏最小二乘回归（PLSR）模型， RMSECV分别为31.3和4.52 mg·dL-1； RMSEP分别为30.3和98.7 mg·dL-1。 结果表明， 当光程变化信息与葡萄糖浓度信息之间偶然相关时， 所建PLSR模型的预测精度与稳健性会受到不同程度的影响， 随着二者之间相关性的增强， 最优模型对葡萄糖信息的识别能力越来越弱。 为提高中红外无创血糖测量分析结果的可靠性， 避免获得伪优定标模型提供了实验依据。

In noninvasive blood glucose measurement, it is difficult to keep the contact area between skin and internal reflectance element uniform while the mid-infrared spectra of human skin are taken, and this would lead to path-length variations. To study the effect of path-length variations on PLSR calibration model, in the present paper, according to the correlation coefficients between path-lengths and glucose concentrations, two PLSR models were achieved respectively and RMSECV were 31.3 and 4.52 mg·dL-1, RMSEP were 30.3 and 98.7 mg·dL-1 for the validation set. The results show that the chance correlations between path-lengths and glucose concentrations will lead to calibration models with different accuracy and robustness. This is useful to improving the reliability of noninvasive measurement of blood glucose by mid-infrared spectroscopy.