2026-11-18 合肥物質科学研究院(HFIPS)

Schematic Diagram of the FTIR Gas Sensing System (Image by ZHAO Tianxiang)
<関連情報>
- https://english.hf.cas.cn/nr/rn/202611/t20261118_1219749.html
- https://www.sciencedirect.com/science/article/abs/pii/S1386142526015982
スペクトル重複条件下におけるLSTM-PSO-SVRハイブリッド手法を用いた高精度FTIRガス定量分析(Robust quantitative FTIR gas analysis under spectral overlap using an LSTM-PSO-SVR hybrid framework)
Kaixiang Fan, Cunguang Zhu, Pengpeng Wang, Tianxiang Zhao, Zhanshang Su, Yao Dong, Pengzhen Huang, Yibo Gao, Xiangxian Li
Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy Volume 367, Part 2, Article 129027 (2027)オンライン公開日:2026-10-23
DOI:10.1016/j.saa.2026.129027
This study developed a hybrid analytical framework combining long short-term memory networks (LSTM), particle swarm optimization (PSO), and support vector regression (SVR) to improve quantitative Fourier transform infrared (FTIR) analysis of gas mixtures under severe spectral overlap. Focusing on the 2210–2250 cm⁻¹ absorption region shared by carbon monoxide (CO) and carbon dioxide (CO₂), a three-layer LSTM network transformed 83-dimensional spectral inputs into 128-dimensional features, followed by concentration prediction using PSO-optimized SVR. The proposed framework reduced prediction errors by 92.74% for CO₂ and 79.84% for CO compared with conventional SVR. It also demonstrated accurate prediction under limited calibration sample conditions, with an average inference time of 3.15 milliseconds per spectrum. These findings suggest potential applications in rapid and accurate monitoring of industrial emissions and greenhouse gases.


