Abstract: The existing calibration algorithms for Raman spectral quantitative analysis, such as principal component regression, partial least squares and support vector machine need a large number of training samples and the regression model may have unsatisfactory generalization ability. Indirect hard modeling (IHM)is a recently introduced method for spectral quantitative analysis, which is used to analyze spectra with highly overlapping band and nonlinear variation. This method can be used to build a calibration model with high generalization ability only by a few training samples. However, all the spectra of pure-components in the mixture is required in IHM method. Obviously, this condition is difficult to achieve in practical applications. To overcome the limitation of IHM, a novel method based on Voigt function for spectral quantitative analysis, i.e., direct hard modeling (DHM), was presented. The novel method did not require the pure-component spectra. By DHM algorithm, the characteristic peaks of every pure component in the spectra of mixture could be fixed, and then a linear calibration model could be built based on the areas of unique peaks and the concentration of pure-components. The DHM algorithm was successfully demonstrated while testing on the quantitative analysis of xylene mixture in a PX unit. Experimental results showed that the DHM algorithm had the advantages of robustness and fewer training samples.

Key words: spectral analysis, quantitative analysis, regression model, PX unit

摘要： 目前用于拉曼光谱定量分析的方法，如PCA、PLS及SVM等算法需要较多的训练样本，且所建回归模型的外推性较差。间接硬建模（indirect hard modeling，IHM）是一种新型的光谱定量分析技术，适用于光谱的叠加及非线性变化情况，只需少量训练样本即可得到外推性较高的回归模型。但IHM方法需要已知混合物中所有常成分的光谱，这一条件在实际应用中较难达到。为此，提出了一种新的定量分析方法——直接硬建模算法（direct hard modeling，DHM）。新算法不需已知待测成分光谱，而是直接在混合物光谱中确定待测成分所对应的特征峰，然后利用特征峰面积与待测成分浓度之间建立线性模型。通过对PX装置中二甲苯成分的定量分析实验证明DHM具有训练样本数量少、回归模型稳健性强等优点。

关键词: 光谱分析, 定量分析, 回归模型, PX装置