近红外光谱分辨率对液体样品近红外定量模型影响的研究

Comparison of the Models of Mixed Liquid Samples under Different Near-Infrared Spectral Resolutions

以四种组分的液体混合样品为实验对象,研究了不同分辨率下(1,4,16,32和64 cm-1)的近红外光谱模型,采用偏最小二乘法建立校正模型,全交互验证法进行检验。用目标函数值来评价定量模型性能。对于原光谱所建立的模型,苯和苯甲醛在分辨率1 cm-1时的目标函数值最大,甲苯在4 cm-1时的目标函数值最大,而氯苯在16 cm-1时的目标函数值最大;对于一阶导数光谱所建立的模型,四种组分的目标函数值均在1 cm-1时最大。结果表明,首先,仪器分辨率对近红外光谱定量分析结果有影响,对于光谱重叠严重的组分,提高分辨率对定量分析有利;但在分析对象的真实带宽较大的定量分析中,可采用较低分辨率以保证信噪比。其次,仪器分辨率对不同的组分的影响是不同的。此外,原光谱的信噪比及样品中不同组分光谱的带宽共同影响定量结果,在保证光谱信噪比的前提下,高分辨率对定量模型有利。

Models of near-infrared spectra under different resolutions, 1, 4, 16, 32 and 64 cm^-1, were studied with the mixed liquid samples of 4 components. The calibration models were developed by the method of partial least square and the validations of the models were carried out by the method of full cross. The value of target function was used to estimate the models perform- ance. For the calibration models developed by the raw spectra, the target function values of benzene and benzaldehyde reached the max value with the resolution of 1 cm^-1 , the target function value of toluene reached the max value with the resolution of 4 cm^-1 , and the target function value of chlorobenzene reached the max value with the resolution of 16 cm^-1 ; for calibration mod- els developed by the 1st derivative spectra, the target function values of the four components all reached the max value with the resolution of 1 cm^-1. The result suggested that, first of all, the resolution of the instrument will influence the quantitative analysis result. For the component with spectrum overlapped seriously, a higher resolution is good for the quantitative analysis, while for the analyte with a broad real band width, a lower resolution can be adopted in order to assure the signal to noise ratio. In addition, the influence of the resolution of the instrument is different for different components. Furthermore, the quantitative analysis result can be affected by both the SNR of the raw spectra and the band width of different components in the analyte, and a higher resolution is good for the quantitative model when the SNR of the spectra is assured.