温度干扰下的葡萄糖水溶液近红外光谱修正方法与比较

The Correction Methods for Near Infrared Spectrum of Glucose Aqueous Solution to Reduce the Influence from Temperature

水的近红外吸收光谱受温度影响较大,采用近红外光谱法对水溶液样本进行成分测量时,温度的影响不能忽视。特别对于葡萄糖这种弱吸收成分,温度的微小变化会对其测量精度造成较大影响。实验发现葡萄糖水溶液在30℃附近时,温度每变化1摄氏度,在1 160nm处溶液的吸光度变异系数约0.344 7%,这个变化约相当于500~600mg·dL^(-1)葡萄糖浓度引起的溶液吸光度变化量。将近年来发展的几种化学计量学方法用于温度扰动下葡萄糖水溶液的光谱修正,以提高葡萄糖测量精度,具体包括广义最小二乘加权法(GLSW)和外部参数正交化方法(EPO)。对不同温度的葡萄糖水溶液样品在900~1 350nm的光谱进行采集,测试了两种方法对温度变化后光谱的修正效果,考察了修正后光谱的葡萄糖预测误差,并与传统的考虑温度变量的多变量回归方法—偏最小二乘(PLS)回归法进行了比较。结果表明,GLSW和EPO方法对不同温度下的溶液光谱有好的校正效果,不同温度下溶液光谱的变异系数得到显著改善,同时两种方法在模型复杂性、温度修正效果等方面都优于传统方法。研究可推广到其他水份含量较大的样品测试场合,也对人体组织中的葡萄糖测量有借鉴价值。

In terms of the near-infrared spectroscopy, the concentration measurement for a substance in aqueous solution samples would be greatly influenced by the samples~ temperature since the water＇s absorption in the solution will be obviously changed when the temperature varies. Especially for the measurement of glucose, which exhibits relatively weak absorption to the nearinfrared light, the measurement accuracy would drop even with the slightest change of the solution＇s temperature. It is found by experiment that the absorbance of glucose aqueous solutions would be changed in 0. 344 7% at 1 160 nm when their temperature varies by 1 ℃ from around 30 ℃. Some recently developed methods based on chemometrics are presented, which can calibrate the solution^s spectra affected by temperature, to improve the glucose measurement accuracy under various temperatures. The methods include generalized least squares weighting （GLSW） and external parameter orthogonalisation （EPO）. These methods are tested on the glucose aqueous solutions＇ spectral data of 900-1 350 nm recorded at various temperatures＇ the prediction er- rors are evaluated using the spectra calibrated by the methods. The results show that the spectra of the same solution with vari- ous temperatures can be well calibrated to an equivalent result to the spectrum at the reference temperature. Moreover, the methods are compared with the multivariahle regression of （partial least squares, PLS） regression, which includes temperature as a variable in the prediction model. The result shows that GLSW and EPO can well calibrate the solutions＇ spectra from various temperature values because the coefficient of variation was reduced. And then the calibrated spectra show better performance on the prediction of glucose concentration than PLS because the complexity and the prediction errors of the models were well improved. The research in this paper could be referenced for the other substance measurements in the aqueous samples, and it can also he referredto the blood glucose measurement in human tissues.