汽轮机油中含水量的近红外光谱研究

汽轮机油在使用过程中会因为种种原因混入水分，使油液理化性能发生变化，影响系统的正常工作，采取合理的措施有效地对汽轮机油中的含水量进行分析一直是汽轮机油的研究热点之一。应用近红外光谱技术结合连续投影算法(SPA)实现了汽轮机油中含水量的分析，通过对汽轮机油中含水量为0～0.156%的57个油样进行光谱扫描，运用原始光谱、一阶导数光谱和多项式最小二乘拟合求导算法Savitzky-Golay(SG)的不同预处理方法，并应用SPA提取近红外光谱的有效波长，以相关系数(r)和均方根误差(RMSE)作为模型评价指标对汽轮机油中含水量进行了研究。结果表明原始光谱经过一阶导数与SG预处理后，应用SPA提取近红外光谱的16个特征波长，作为最小二乘支持向量机(LS-SVM)模型的输入变量，建立了SPA- LS-SVM模型。运用此模型对验证集样本进行预测的相关系数和均方根误差分别为0.975 9和2.656 8×10-3，说明应用SPA结合近红外光谱技术来检测汽轮机油中含水量是可行的，并能获得满意的预测精度，为进一步应用光谱技术进行汽轮机油中其他污染物的检测提供了新的方法。

Near Infrared Spectroscopy Study on Water Content in Turbine Oil

Near infrared (NIR) spectroscopy combined with successive projections algorithm (SPA) was investigated for determination of water content in turbine oil. Through the 57 samples of different water content in turbine oil scanned applying near infrared (NIR) spectroscopy, with the water content in the turbine oil of 0～0.156%, different pretreatment methods such as the original spectra, first derivative spectra and differential polynomial least squares fitting algorithm Savitzky-Golay (SG), and successive projections algorithm (SPA) were applied for the extraction of effective wavelengths, the correlation coefficient (R) and root mean square error (RMSE) were used as the model evaluation indices, accordingly water content in turbine oil was investigated. The results indicated that the original spectra with different water content in turbine oil were pretreated by the performance of first derivative +SG pretreatments, then the selected effective wavelengths were used as the inputs of least square support vector machine (LS-SVM). A total of 16 variables selected by SPA were employed to construct the model of SPA and least square support vector machine (SPA-LS-SVM). There is 9 as The correlation coefficient was 0.975 9 and the root of mean square error of validation set was 2.655 8×10-3 using the model，and it is feasible to determine the water content in oil using near infrared spectroscopy and SPA-LS-SVM, and an excellent prediction precision was obtained. This study supplied a new and alternative approach to the further application of near infrared spectroscopy in on-line monitoring of contamination such as water content in oil.