为了准确检测油田伴生气中微量H2S气体的含量, 设计一种模拟油田伴生气中微量杂质气体H2S的在线实时分析系统, 能够为油田伴生气回收利用工艺的改进和制定提供依据。该系统基于可调谐激光吸收光谱技术和波长调制技术, 利用可调谐的分布反馈式激光器、锁相放大器, 结合改进新型Herriot气室、InGaAs探测器, 实现了模拟油田伴生气中微量气体H2S的实时在线监测。为消除背景气体CH4以及其他杂质气体的干扰, 开展RBF和经典BP神经网络的对比实验。通过模拟混合气站配备多种不同浓度的H2S标准气体测试系统, 实验结果表明, 在强大背景气体的干扰下, 该系统可达到的H2S检测下限为1.2 ppm; 在抗干扰方面, 与经典BP神经网络相比, RBF神经网络具有很强的优势, 其预测误差小于10-10。另外, 该系统还具有较高的检测精度和强鲁棒性, 在油田伴生气中微量气体的检测领域具有很强的适用价值。

In order to accurately detect the content of H2S trace gas in oil field, an on-line, real-time monitoring and analysis system for trace impurity gases, such as hydrogen sulfide(H2S) in oil field associated gas was designed, which can provide basis for the improvement and formulation of associated gas recovery and utilization technology in oil fields. The system was based on Tunable Laser Diode Spectroscopy(TDLAS) and Wavelength Modulation(WMS), using a tunable distributed feedback laser (DFB), lock-in amplifier, combined with a new Herriot chamber, InGaAs detector, to achieve the simultaneous on-line monitoring of H2S trace gases in oil field associated gas. To eliminate interference of background gases CH4 and other impurities in the gas, comparative experiments of RBF and BP neural network were carried out. A variety of H2S standard gas test systems with different concentrations were provided in the analog gas mixing station. The experimental results show that the interference of strong background gas, the detection limit for H2S is 1.2 ppm, in terms of interference, compared with the classical BP neural network, RBF neural network has a strong advantage, the prediction error is less than 10-10, the system had high detection accuracy and robustness, and strong application value in the field of oil gas detection.