油中特征气体(H2, CO, CO2, CH4, C2H4, C2H6, C2H2)的快速准确检测是变压器在线监测的重要环节。 激光拉曼光谱技术适用于特征气体的检测, 能克服传统在线监测的诸多不足。 在拉曼光谱图2 900～3 300 cm-1谱段, 甲烷(CH4)和乙烷(C2H6)气体特征谱峰聚集, 研究此谱段中不同含量比的混合气体样本对变压器油中混合气体定量分析具有重要意义。 在单一特征气体拉曼光谱检测的研究基础上, 选取预处理后光谱图中特征峰的谱峰高度、 半高宽以及谱峰面积多个参量作为特征因素, 对变压器油中混合气体进行定量分析。 以二阶微扰理论分析计算得出, 甲烷拉曼光谱中存在四个特征谱峰, 选取的谱段中包含以3 111与3 284 cm-1为拉曼频移中心的两峰, 乙烷拉曼光谱中存在六个特征谱峰, 选取谱段中存在3 111与3 187 cm-1两峰, 理论上通过谱段中携带的特征谱峰信息能够计算两种气体含量; 通过拉曼光谱平台检测, 混合气体光谱图特征谱峰会产生平移以及聚合, 对光谱图中寻峰得到的中心频移为2 902, 2 918, 2 956和3 022 cm-1的四个混合峰建立高斯函数模型, 得到特征谱峰的谱峰高度、 半高宽以及谱峰面积; 建立偏最小二乘回归模型, 以谱峰高度、 半高宽、 谱峰面积为自变量, 两种气体含量为因变量计算分析。 模型潜在因子取到t6时, 调整后的R2为0.993, 表明自变量与因变量具有确切关系, 回归模型可靠。 对回归方程参数分析发现, 谱峰半高宽相比谱峰面积以及谱峰高度有显著贡献, 符合预期目标, 混合气体光谱图中四个特征谱峰对两种气体均有影响。 通过实验可总结得出, 针对甲烷乙烷混合气体, 在室温25 ℃, 积分时间15 s, 积分次数2, 狭缝100 μm条件下, 通过获取谱峰高度、 谱峰面积以及半高宽三个参量, 能够准确测量气体含量, 为变压器油中多种特征气体的同时检测奠定了基础。

The rapid and accurate detection of characteristic gases(H2, CO, CO2, CH4, C2H4, C2H6, C2H2) in oil is an important part of transformer on-line monitoring. Laser Raman spectroscopy is suitable for the detection of characteristic gases and can overcome many shortcomings of traditional on-line monitoring. When the characteristic gas in transformer oil is detected, the characteristic peaks of methane (CH4) and ethane (C2H6) gather in the Raman spectrum from 2 900 to 3 300 cm-1. It is of great significance to study the mixed gases samples with different content ratios in this spectrum for the quantitative analysis of mixed gases in transformer oil. Based on the research of Raman spectroscopy detection of single characteristic gas, the spectral peak height, full-width at half-maximum and spectral peak area parameters of the characteristic peaks in the pretreated spectrum are selected as the characteristic factors to quantitatively analyze the mixed gas in the transformer oil. According to the second-order perturbation theory, there are four characteristic peaks in the methane Raman spectrum, and the selected spectrum contains two peaks with 3 111 and 3 284 cm-1 as the Raman shift center. Six characteristic peaks exist in the Raman spectra of ethane, and there are two peaks of 3 111 and 3 187 cm-1 in the selected spectral bands. In theory, the amount of both gases can be calculated by the information of the characteristic peak carried in the spectrum bands. Through the detection of Raman spectroscopy platform, the characteristic peaks of the mixed gas spectrum will produce translation and polymerization. In practice, four peaks with shifts center of 1 902, 2 918, 2 956 and 3 022 cm-1 were found in the spectrum. A Gaussian function model was established for the four mixed peaks, and the spectral peak height, full-width at half-maximum and spectral peak area of the characteristic peak was obtained. A partial least squares regression model (PLS) was established. The spectral peak height, full-width at half-maximum and spectral peak area were taken as independent variables, and the two gas contents were taken as dependent variables for calculation and analysis. When the potential factor of the model is taken to t6, the adjusted R-square is 0.993, indicating that the independent variable has a definite relationship with the dependent variable, and the regression model is reliable. The analysis of regression equation parameters shows that the full-width at half-maximum of spectral peaks contribute significantly to the area and height of spectral peaks, which is in line with the expected target. Four characteristic spectral peaks in the spectrum of mixed gases have an effect on both gases. It can be concluded from the experiments that for the methane-ethane mixed gas, at room temperature 25 ℃, integration time 15 s, integral number 2, slit 100 μm, by obtaining the peak height, peak area and full-width at half-maximum three parameters, the gas content can be accurately measured, which lays a foundation for the simultaneous detection of various characteristic gases in transformer oil.