汽油辛烷值NIR数据处理与建模仿真

NIR gasoline octane data processing and modeling

随着当今计算机与各类程序软件的开发使用,化学计量学不断发展,人们可以在近红外光谱区内采集大量的数据,并使用各种有效的统计方法,把近红外光谱技术应用于定性与定量。近红外光谱为分子振动光谱的倍频和组合频谱带,主要是对含氢基团的吸收,包含有绝大多数类型有机物的组成与分子结构的丰富信息。原理是基于不同的基团或同一基团在不同化学环境之中吸收波长的差异。该技术具有快速、简便、样品无需处理、适合在线分析等特点。近红外光谱技术主要应用于农产品、医学、制药等行业,研究的热点方向是仪器的相关改进和新的化学计量学算法。对近红外光谱数据进行处理的常用化学计量学方法为多元校正方法,主要有逐步多元线性回归、主成分回归、偏最小二乘法、拓扑学方法以及人工神经网络法等。本文介绍了近红外光谱分析技术以及在相关领域中的应用。通过对汽油近红外光谱数据的考察,建立汽油辛烷值的定量校正模型。分别采用偏最小二乘回归与主成分回归建立校正模型并实现分析。实验结果表明,应用仿真工具可以对近红外光谱数据进行有效并可靠的处理与分析。

With the development of today＇s computer, various types of softwares and chemical metrology, people can collect large amounts of data in the near infrared spectral region, using a variety of effective statistical methods used in the NIR qualitative and quantitative analysis. Near infrared spectroscopy for molecular vibrational spectra of the frequency and combination of spectral bands, mainly on the absorption of hydrogen groups, containing most types of organic matter composition and molecular structure of the wealth of information. Principle is based on the same groups or different groups in different chemical environment differences in absorption wavelength. The technology is fast, simple, sample without treatment, suitable for online analysis and so on. Near infrared spectroscopy is mainly used in agricultural, medical, pharmaceutical and other industries, a hot research direction is related to improved equipment and new chemometric algorithms. The near infrared spectral data processing methods commonly used chemometric multivariate calibration methods, there are stepwise multiple linear regression, principal component regression, partial least squares, topological methods and artificial neural network method. Near infrared spectroscopy technology and applications are introduced in related fields in this article. A quantitative calibration model is established by near-infrared spectral data on gasoline investigation. PLS and PCR are used to establish calibration models and to achieve analysis. Experimental results demonstrate that for the NIR data, MATLAB can be effective and reliable to process and analysis.