1. Simultaneous measurement of NH3 and NO by mid-infrared tunable diode laser absorption spectroscopy based on machine-learning algorithms.
- Author
-
Guo, Songjie, Li, Zhenghui, Liu, Zeming, Wang, Zhu, Liu, Weibin, Lu, Zhimin, Xing, Xiwen, Ren, Wei, and Yao, Shunchun
- Subjects
- *
LASER spectroscopy , *TUNABLE lasers , *MID-infrared lasers , *MACHINE learning , *SEMICONDUCTOR lasers , *PARTICLE swarm optimization , *BEES algorithm - Abstract
• To eliminate the influence of CO 2 and H 2 O on NH 3 and NO concentration measurements, this paper proposes a machine-learning modeling and concentration measurement method using combined spectral signals from reference gases. • Before the concentration measurements, the direct absorption spectroscopy (DAS) signals of NH 3 and NO underwent noise reduction using the variational mode decomposition algorithm optimized by the artificial bee colony algorithm and wavelet threshold denoising (ABC-VMD-WTD) method. The signal-to-noise ratio (SNR) of NH 3 and NO increased from 17.2 and 46.9 before denoising to 167.8 and 532.4 after denoising. • After testing, the average relative errors of NH 3 and NO concentrations measured by the particle swarm optimization-support vector machine (PSO-SVM) model are 4.0 % and 0.8 %, respectively, with measurement precision of 0.05 ppm and 0.42 ppm, which is better than the conventional integral absorbance method. • By analyzing the Allan deviation, the minimum detection limit (MDL) for NH 3 is 16.1 ppb at an average time of 50 s, while for NO, it is 38.4 ppb at an average time of 71 s. Accurate measurements of NH 3 and NO are essential for controlling NO x emissions from coal-fired power plants, reducing ammonia slip, and improving the efficiency of selective catalytic reduction (SCR) operation. We propose a method for measuring NH 3 and NO concentrations based on machine learning algorithms to address the challenges posed by overlapping interference of CO 2 and H 2 O spectral lines in flue gas. Our approach introduces a novel method for acquiring mixture spectra for the model. Initially, we measure the spectra of individual components under different concentrations and temperatures. Subsequently, mixed spectral samples are generated by combining the measured spectra of the individual components. This approach simplifies the spectral measurement process while preserving accuracy. The particle swarm optimization support vector machine (PSO-SVM) algorithm is leveraged, providing a reliable foundation for the continuous and synchronous measurement of NH 3 and NO. Upon testing, the PSO-SVM demonstrates average relative errors of 4.0 % and 0.8 % for NH 3 and NO concentrations, respectively. The corresponding measurement precision is 0.05 ppm for NH 3 and 0.42 ppm for NO, better than the conventional integral absorbance method. The minimum detection limit (MDL) for NH 3 is 16.1 ppb at an average time of 50 s, while for NO, it is 38.4 ppb at an average time of 71 s. The methodology of this paper is expected to play an important role in reducing the influence of interfering components and improving the accuracy of field measurements. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF