支持向量回归建立测定醇烯比的近红外光谱校正模型

Developing Near Infrared spectroscopy Calibration Model of Molar Ratio between Methanol and Isobutylene by Support Vector Regression

在石化工业中,甲基叔丁基醚(MTBE)是20多年来发展最快的石化产品之一。在生产过程中,反应器混合进料的醇烯比(MRMI)都是一个至关重要的操作参数,但传统的色谱方法由于速度慢、硬件维护量大等原因,应用效果大都不理想。近红外光谱方法因其快速、测量方便、维护少等特点,特别适合这一参数的在线测量。但是,由于醇烯比与近红外光谱之间呈严重的非线性响应,基于线性校正的偏最小二乘方法将不能得到准确的测定结果。因此,文章采用近几年新兴的支持向量回归方法建立近红外光谱测定醇烯比的校正模型,结果表明,该方法的预测能力优于偏最小二乘和人工神经网络方法,可以推广应用于实际的醇烯比在线近红外测量中。

In petrochemical industries, the molar ratio between methanol and isobutylene is one of the most important control parameters in methyl tertiary butyl ether （MTBE） production plant. However, traditional on-line gas chromatography method is difficult to use in practice because of its high maintenance and low speed. On-line near infrared spectroscopy is hopeful to become an excellent alternative method for determining the parameter due to its rapidness, convenience, and less maintenance. Because of the nonlinearity of the measured parameter and near infrared spectra, support vector regression, a novel powerful nonlinear calibration method, was used to build calibration model in the present paper. Compared with the results of partial least squares （PLS） and artificial neural network （ANN） method, the prediction accuracy of support vector regression model is high enough to meet the demand for protests control of MTBE unit. This calibration method can be applied to real online analysis of the molar ratio between methanol and isobutylene by near infrared spectroscopy.