Abstract: A quantitative analysis model for the determination of the oxide concentration in a doublecomponent containing model oil is established based on the infrared spectroscopy data for cyclohexanone and cyclohexanol components in model oil, the regression of prominent component, and the regression of back propagation neural algorithm and support vector machine. The infrared spectroscopy data were compressed by principal component analysis and used as input information to develop models. The three quantitative models can predict the respective oxide contents, with the predictive ability of support vector machine model superior to the other two models, showing a good stability of support vector machine model. Comparing with traditional methods, the established model provides a simple, fast, nodamage, costeffective and green determination method.

Key words: Infrared spectroscopy, Oxygen compound, Prominent component regression, Back propagation network, Support vector machine for regression

摘要： 摘 要:  采用主成分回归方法、 反向传播人工神经网络方法和支持向量机回归方法, 结合含有环己酮和环己
醇双组分的模拟油的红外光谱数据, 建立预测模拟油中双组分含氧化合物浓度的定量校正模型。通过主成分分析
法压缩红外光谱数据, 将主成分作为模型的输入信息。结果表明, 三种方法建立的定量校正模型均可用于预测双组
分含氧化合物的浓度。支持向量机回归模型的预测结果好于其他两种模型, 具有较强的稳定性和良好的预测能力。
对于含氧化合物浓度的测定, 红外光谱法具有操作简单、 测量快速、 无损价廉且绿色的优点。

关键词: 红外光谱, 含氧化合物, 主成分回归, B P网络, 支持向量机回归