基于1D-WCWKCNN的痕量甲烷气体浓度检测

doi:10.3964/j.issn.1000-0593(2024)03-0829-07

摘要
参考文献
相关文章 (1)

全文: PDF (4753 KB)
输出: BibTeX | EndNote (RIS)

摘要：可调谐激光吸收光谱（TDLAS）技术在检测甲烷气体浓度过程中，甲烷气体透射光强二次谐波信号幅值与痕量甲烷气体浓度值成正比关系。如何准确和快速筛选目标甲烷透射光强二次谐波信号幅值至关重要。通过光电探测器获取1 000个甲烷气体透射光强信号样本，解调该透射光强信号获得二次谐波信号。人工在获得多种痕量甲烷气体透射光强和通过透射光强解调二次谐波信号时，存在噪声和人为操作对二次谐波信号的幅值产生影响，从而造成人工筛选二次谐波信号的时间增加。针对传统TDLAS技术筛选痕量甲烷气体二次谐波信号过程中存在高额时间成本的问题，提出了一种基于宽卷积和宽卷积核一维卷积神经网络（1D-WCWKCNN）的痕量甲烷气体浓度检测方法。首先，借助甲烷气体数据集训练1D-WCWKCNN模型，根据训练结果不断调整模型参数。其次，利用宽卷积层和宽卷积核一维卷积层对痕量甲烷气体二次谐波信号进行特征提取，使网络进行一次卷积后能够获得甲烷气体浓度信号中更长序列以及该序列边界信息与气体浓度之间的特征关系。甲烷透射光强二次谐波信号通过6层卷积层提取该信号与甲烷气体浓度关系的深层特征，然后通过6层最大池化层保留该信号与甲烷气体浓度关系的主要特征，再经过Flatten层将前一层处理的痕量甲烷气体透射光强二次谐波信号数据进行一维化处理。最后，根据训练好的1D-WCWKCNN模型通过Dense层输出痕量甲烷气体浓度。利用基于1D-WCWKCNN的痕量甲烷气体浓度检测模型代替了TDLAS技术中人工花费高额时间成本筛选二次谐波信号进行拟合直线对痕量甲烷气体浓度检测的过程。在实际实验中验证了该方法的有效性，实验结果表明利用该方法能够对50~1 000 mg·L^-1的痕量甲烷气体浓度进行有效检测，其准确度达到99.85%，与其他方法相比该方法信号特征提取能力强，检测甲烷气体精度高。该方法有助于气体检测领域中待测气体浓度信号的筛选。

关键词：痕量甲烷气体检测；TDLAS技术；宽卷积核；一维宽卷积

Abstract：In detecting methane concentration by tunable laser absorption spectroscopy (TDLAS), the second harmonic signal amplitude of methane transmitted light intensity is directly proportional to the concentration of trace methane gas. How to accurately and quickly screen the amplitude of the second harmonic signal of the target methane transmitted light intensity is crucial. The photodetector obtains the 1 000 methane gas transmitted light intensity signal samples, and it is demodulated to obtain the second harmonic signal. When obtaining a variety of trace methane gas transmitted light intensity and demodulating the second harmonic signal by transmitted light intensity, noise and artificial operation affect the amplitude of the second harmonic signal，resulting in an increase in the time for manual screening of the second harmonic signal. Using traditional TDLAS technology to screen trace methane second harmonic signals had the problem of high time cost. A trace methane concentration detection method based on wide convolution and wide kernel 1D convolutional neural networks (1D-WCWKCNN) was proposed. Firstly, the 1D-WCWKCNN model is trained with the help of the methane gas dataset, and the model parameters are continuously adjusted according to the training results. Secondly, the method used a wide convolution layer and wide convolution kernel 1D convolution layer to extract the features of the trace methane second harmonic signal so that the network obtained the characteristic relationship between a longer sequence and the sequence boundary information in the methane concentration signal and the gas concentration after one convolution. The second harmonic signal of methane transmitted light intensity is extracted through the 6-layer convolutional layer to extract the main characteristics of the relationship between the signal and methane gas concentration. The 6-layer maximal pooling layer retains the main characteristics. The Flatten layer processes the signal data processed by the previous layer in one dimension. Finally, the trained 1D-WCWKCNN model outputs trace methane gas concentration through the Dense layer. The trace methane gas concentration detection model based on 1D-WCWKCNN replaces manually screening second harmonic signals for detecting trace methane gas concentration in a fitted straight line in TDLAS technology. The effectiveness of this method is verified in actual experiments. The results show that it can effectively detect the concentration of trace methane in 50~1 000 mg·L^-1, and its accuracy reaches 99.85%. Compared with other methods, it has strong signal feature extraction ability and high detection accuracy of methane gas.This method facilitates the screening of gas concentration signals to be measured in the field of gas detection.

Key words：Trace methane gas detection;TDLAS technology;Wide convolution kernel;1D wide convolution

收稿日期: 2022-08-12 修订日期: 2022-11-14

中图分类号:

TE88

基金资助: 国家自然科学基金青年科学基金项目（62103096）资助

通讯作者: 梁洪卫 E-mail: lianghongwei@nepu.edu.cn

作者简介: 阚玲玲，女，1981年生，东北石油大学电气信息工程学院副教授 e-mail: kanlingling2004@126.com

引用本文:

阚玲玲，朱富海，梁洪卫. 基于1D-WCWKCNN的痕量甲烷气体浓度检测[J]. 光谱学与光谱分析, 2024, 44(03): 829-835.
KAN Ling-ling, ZHU Fu-hai, LIANG Hong-wei. Detection of Trace Methane Gas Concentration Based on 1D-WCWKCNN. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2024, 44(03): 829-835.

链接本文:

https://www.gpxygpfx.com/CN/10.3964/j.issn.1000-0593(2024)03-0829-07 或 https://www.gpxygpfx.com/CN/Y2024/V44/I03/829

[1] YANG Dong-shang，ZENG Yi，XI Liang，et al(杨东上，曾议，奚亮，等). Acta Optica Sinica(光学学报)，2020，40(5): 0501002.
[2] REN Hong-mei，LI Ang，HU Zhao-kun，et al(任红梅，李昂，胡肇焜，等). Acta Physica Sinica(物理学报)，2020，69(20)：204204.[3] HUANG Ye-yuan，LI Ang，QIN Min，et al(黄业园，李昂，秦敏，等). Acta Optica Sinica(光学学报)，2021，41(10)：1030002.
[4] CAI Hao-ze，BU Ling-bing，GONG Yu，et al(蔡镐泽，卜令兵，龚宇，等). Acta Photonica Sinica(光子学报)，2019，48(7)：701001.
[5] WANG Yu-hao，LIU Jian-guo，XU Liang，et al(王钰豪，刘建国，徐亮，等). Acta Physica Sinica(物理学报)，2022，71(9)：093201.
[6] DING Wu-wen，SUN Li-qun，YI Lu-ying(丁武文，孙利群，衣路英). Acta Physica Sinica(物理学报)，2017，66(10)：100702.
[7] Shao L G，Fang B，Zheng F，et al. Spectrochimica Acta A: Molecular and Biomolecular Spectroscopy，2019，222：117118.
[8] Li G L，Ma K，Jiao Y，et al. Microwave and Optical Technology Letters，2021，63(4)：1147.
[9] Jiang C L，Liu Y F，Yu B，et al. Review of Scientific Instruments，2018，89(8)：083106.
[10] Jiang J，Wang Z W，Han X，et al. IEEE Transactions on Dielectrics and Electrical Insulation，2019，26(1)：153.
[11] Li B，Xue L，Ji N，et al. International Journal of Optics，2021，2021：8829790.
[12] Jiang J，Zhao M X，Ma G M，et al. IEEE Sensors Journal，2018，18(6)：2318.
[13] JI Wen-hai，LÜ Xiao-cui，HU Wen-ze，et al(季文海，吕晓翠，胡文泽，等). Optics and Precision Engineering (光学精密工程)，2018，26(8)：1837.
[14] ZHOU Fei-yan，JIN Lin-peng，DONG Jun(周飞燕，金林鹏，董军). Chinese Journal of Computers(计算机学报)，2017，40(6)：1229.
[15] Zhao X J，Wen Z H，Pan X F，et al. IEEE Access，2019，7：12630.
[16] Zhang T J. International Journal of Computational Intelligence Systems，2020，13(1)：1198.
[17] Peng P，Zhao X J，Pan X F，et al. Sensors，2018，18(1)：157.
[18] Wei G F，Li G，Zhao J，et al. Sensors，2019，19(1)：217.
[19] HAO Hui-min，LIANG Yong-guo，WU Hai-bin，et al(郝惠敏，梁永国，武海彬，等). Spectroscopy and Spectral Analysis(光谱学与光谱分析)，2021，41(3)：782.
[20] WANG Xiu-fang，SUN Shuang，DING Chun-yang(王秀芳，孙双，丁春阳). Journal of Jilin University (Engineering and Technology Edition)[吉林大学学报(工学版)]，2022，52(2)：310.
[21] Zhang Y Q，Miyamori Y，Mikami S，et al. Computer‐Aided Civil and Infrastructure Engineering，2019，34(9)：822.
[22] Van Der Linde A，Tutz G. Statistics，2008，42(1)：4.