鉴于光谱分析仪在使用过程中会发生谱图微缓形变, 造成光谱保真度下降, 本文提出一种结合标气定期校验和最小二乘支持向量机(LS-SVM)的方法。在校验时, 比较相同标准气体的当前光谱和出厂标定时的光谱, 利用LS-SVM算法计算谱图形变, 并根据形变分析结果进行复原。由于形变具有长期累积性和不可控性, 所以用激光器驱动电流的可控性变化引起的谱图变化模拟形变。采用TDLAS技术实验, 检测烯烃裂解炉清焦过程中的CO和CO2, 使用20 m光程的Herriot气室和中心波长为1 580 nm的DFB激光器。共采集了50组不同形变程度的光谱, 根据LS-SVM算法的形变分析结果, 利用插值法还原谱图。对比了该方法与不复原以及传统寻峰复原方法的气体浓度分析精度, 发现当分析仪存在复杂形变时, 利用LS-SVM算法和插值法复原的方法比传统的寻峰方法精度提高2~3倍。实验结果验证了LS-SVM算法的谱图形变分析能力以及插值法复原的有效性。

The output of spectroscopy analyzer undergoes slow and subtle distortion when operational for a long period. This results in degradation of the spectral fidelity and it adversely affects the analyzing accuracy. A design based on combination of periodic validation with standard gas and Least Square Supporting Vector Machine (LS-SVM) algorithm was proposed to overcome this issue. For validation, the current spectra and factory calibration spectra of the same standard validation gas were compared. The LS-SVM method was employed to calculate the spectra deformation and the deformation value was further used to restore the spectra. Because of the chronic accumulation and non-controllable characteristic, the deformation was simulated by changing the diode laser driving current. The simultaneous detection of CO and CO2 in the olefin cracking furnace decoking process was chosen as the application example. Tunable Diode Laser Spectroscopy (TDLAS) platform was employed with a 20 m Herriot cell and a 1 580 nm DFFB. In total, 50 spectra with different deformation levels were collected. The spectra were restored with an interpolation method based on the LS-SVM model output. Accuracy of the gas concentration calculation was compared with no restoration and traditional peak tracking restoration methods; we found that when the analyzer has a complex deformation, accuracy of the LS-SVM algorithm and the interpolation recovery method is 2—3 times higher than that of the traditional peak tracking method. The experiment result proves the effectiveness of the LS-SVM algorithm and the interpolation restoration method.