基于参数优化卷积神经网络的预混火焰温度与H<sub>2</sub>O浓度重建方法研究

单彦博; 阮仕双; 侯艺杰; 张立芳

doi:10.21656/1000-0887.460171

基于参数优化卷积神经网络的预混火焰温度与H₂O浓度重建方法研究

doi: 10.21656/1000-0887.460171

1. 许继电气股份有限公司，河南许昌 461000;^{2. 太原理工大学热能工程系，太原 030024}

基金项目:

国网山西省电力公司科技项目（520530240037）

详细信息

作者简介:
单彦博(1998—)，男，硕士（E-mail: 2623908774@qq.com）；张立芳（1988—），女，讲师，博士（通信作者. E-mail: 21227023@zju.edu.cn）.

通讯作者:
张立芳（1988—），女，讲师，博士(通信作者. E-mail: 21227023@zju.edu.cn).

中图分类号: O242
计量
- 文章访问数: 29
- HTML全文浏览量: 8
- PDF下载量: 1
- 被引次数: 0
出版历程
- 收稿日期: 2025-09-22
- 修回日期: 2025-11-22
- 网络出版日期: 2026-10-31

Research on the Reconstruction Method for Premixed Flame Temperature and H₂O Concentration Based on Parameter Optimization of Convolutional Neural Networks

1. XJ Electric Co., Ltd., Xuchang, Henan 461000, P.R. China;^{2. Department of Thermal Engineering, Taiyuan University of Technology, Taiyuan 030024, P.R. China}

摘要

摘要: 可调谐激光吸收光谱技术（tunable diode laser absorption spectroscopy，TDLAS）因其高灵敏度、高分辨率、响应快速、非侵入性、抗干扰能力强等特点，被广泛应用于燃烧诊断测量.本研究结合TDLAS，采用卷积神经网络（convolutional neural network,CNN）算法对预混火焰温度与H₂O浓度进行重建.首先，分析了影响CNN重建精度的关键参数，包括激活函数、损失函数、优化函数、学习率等，根据重建结果、损失值及收敛速度优选出最佳参数组合.研究表明，优化后的CNN重建精度显著提升，温度与H₂O浓度重建误差较优化前分别降低2.08% 和 0.75%.然后，利用CNN与自适应数值迭代算法对McKenna燃烧器平面预混火焰进行燃烧重建对比，结果表明基于CNN重建的温度与H₂O浓度分布更平滑，计算耗时更短，重建结果更符合实际燃烧状态.
- 卷积神经网络 /
- 温度和H₂O浓度重建 /
- 激光吸收光谱 /
- 算法参数优化
Abstract: The tunable diode laser absorption spectroscopy (TDLAS) technology has been widely applied in combustion diagnosis due to their high sensitivity, high spectral resolution, fast time response, robustness and nonintrusive character. Based on the TDLAS and the tomographic imaging, the convolutional neural network was used to reconstruct the temperature and the H₂O concentration of the premixed flame. First, the key parameters influencing the reconstruction accuracy of the convolutional neural network, including the activation function, the loss function, the optimization function, and the learning rate, were analyzed. The optimal parameters were selected according to the reconstruction results, the loss value and the convergence speed. The research show that, the optimized convolutional neural network significantly improved the reconstruction accuracy. Furthermore, the reconstructed errors of 2D temperature and water vapor concentration were reduced by 2.08% and 0.75% compared to the previous values, respectively. Then, the simulation and verification of the plane premixed flame on a McKenna burner were carried out with the convolutional neural network and the adaptive algebraic reconstruction technique. The results indicate that, the temperature and H₂O concentration distributions reconstructed with the convolutional neural network are smoother and computation timesaving, and are more consistent with the actual combustion situation.
- convolutional neural network /
- temperature and water vapor concentration reconstruction /
- laser absorption spectroscopy /
- algorithm parameter optimization

HTML全文

参考文献(23)

[2]Zhai Y C, Wang F, Chi Y, et al. Optical absorption tomography for temperature and H2O concentration imaging by a compact photoelectric detecting system combined with twice adaptive algebraic algorithm[J].Optical Engineering,2022,61(3): 034103.

Demirbas A. Potential applications of renewable energy sources, biomass combustion problems in boiler power systems and combustion related environmental issues[J].Progress in Energy and Combustion Science,2005,31(2): 171-192.

[3]Liu C, Xu L J, Chen J L, et al. Development of a fan-beam TDLAS-based tomographic sensor for rapid imaging of temperature and gas concentration[J].Optics Express,2015,23(17): 22494-22511.

[4]Ma L, Li X S, Sanders S T, et al. 50-kHz-rate 2D imaging of temperature and H2O concentration at the exhaust plane of a J85 engine using hyperspectral tomography[J].Optics Express,2013,21(1): 1152-1162.

[5]Zhang L F, Wang F, Zhang H D, et al. Simultaneous measurement of gas distribution in a premixed flame using adaptive algebraic reconstruction technique based on the absorption spectrum[J].Chinese Optics Letters,2016,14(11): 111201.

[6]Cai W, Kaminski C F. Multiplexed absorption tomography with calibration-free wavelength modulation spectroscopy[J].Applied Physics Letters,2014,104(15): 154106.

[7]An X L, Kraetschmer T, Takami K, et al. Validation of temperature imaging by H2O absorption spectroscopy using hyperspectral tomography in controlled experiments[J].Applied Optics,2011,50(4): A29-A37.

[8]Cai W W, Ewing D J, Ma L. Investigation of temperature parallel simulated annealing for optimizing continuous functions with application to hyperspectral tomography[J].Applied Mathematics and Computation,2011,217(12): 5754-5767.

[9]Ma L, Cai W W, Caswell A W, et al. Tomographic imaging of temperature and chemical species based on hyperspectral absorption spectroscopy[J].Optics Express,2009,17(10): 8602-8613.

[10]Gu J X, Wang Z H, Kuen J, et al. Recent advances in convolutional neural networks[J].Pattern Recognition,2018,77: 354-377.

[11]Schmidhuber J. Deep learning in neural networks: an overview[J].Neural Networks,2015,61: 85-117.

[12]McCann M T, Jin K H, Unser M. Convolutional neural networks for inverse problems in imaging: a review[J].IEEE Signal Processing Magazine,2017,34(6): 85-95.

[13]Zhang H, Li L,Qiao K, et al. Image prediction for limited-angle tomography via deep learning with convolutional neural network[PP/OL]. v1. arXiv (2016-07-29)[2025-09-22]. https://doi.org/10.48550/arXiv.1607.08707.

[14]Chen H, Zhang Y, Zhang W H, et al. Low-dose CT via convolutional neural network[J].Biomedical Optics Express,2017,8(2): 679-694.

[15]Huang J Q, Liu H C, Dai J H, et al. Reconstruction for limited-data nonlinear tomographic absorption spectroscopy via deep learning[J].Journal of Quantitative Spectroscopy and Radiative Transfer,2018,218: 187-193.

[16]Fu Y L, Zhang R, Enemali G, et al. Convolutional neural network aided chemical species tomography for dynamic temperature imaging[C]//2022 IEEE International Instrumentation and Measurement Technology Conference (I2MTC). Ottawa, ON, Canada. IEEE, 2022: 1-5.

[17]Cai W W, Ewing D J, Ma L. Application of simulated annealing for multispectral tomography[J].Computer Physics Communications,2008,179(4): 250-255.

[18]张立芳. 激光吸收光谱技术测量低浓度多组分气体和二维温度浓度分布的研究[D]. 杭州: 浙江大学, 2017.(Zhang Lifang. Investigations on measurement of low concentration and multi-component gas and reconstruction of 2D temperature and concentration distribution using laser absorption spectroscopy[D]. Hangzhou: Zhejiang University, 2017. (in Chinese))

[19]Verhoeven D. Limited-data computed tomography algorithms for the physical sciences[J].Applied Optics,1993,32(20): 3736-3754.

[20]Zhou Xin. Diode-laser absorption sensors for combustion control[D]. California, United States: Stanford University, 2005.

[21]单彦博, 张立芳, 赵贯甲, 等. 基于可调谐激光吸收光谱技术的预混火焰重建算法比较[J]. 光学技术, 2023,49(5): 600-608.(Shan Yanbo, Zhang Lifang, Zhao Guanjia, et al. Comparison of premixed flame reconstruction algorithms based on tunable laser absorption spectroscopy[J].Optical Technique,2023,49(5): 600-608. (in Chinese))

[22]Hyvrinen A, Kster U. Complex cell pooling and the statistics of natural images[J].Network,2007,18(2): 81-100.

[23]runa J, Szlam A, LeCun Y. Signal recovery from pooling representations[C]//Proceedings of the 31st International Conference on Machine Learning. PMLR, 2014,32(2): 307-315.

施引文献

资源附件(0)

访问统计

计量

文章访问数: 29
HTML全文浏览量: 8
PDF下载量: 1
被引次数: 0

姓名
邮箱
手机号码
标题
留言内容
验证码

留言板

基于参数优化卷积神经网络的预混火焰温度与H₂O浓度重建方法研究

doi: 10.21656/1000-0887.460171

作者简介:
单彦博(1998—)，男，硕士（E-mail: 2623908774@qq.com）；张立芳（1988—），女，讲师，博士（通信作者. E-mail: 21227023@zju.edu.cn）.

通讯作者:
张立芳（1988—），女，讲师，博士(通信作者. E-mail: 21227023@zju.edu.cn).

计量

Research on the Reconstruction Method for Premixed Flame Temperature and H₂O Concentration Based on Parameter Optimization of Convolutional Neural Networks

计量

目录

留言板

基于参数优化卷积神经网络的预混火焰温度与H2O浓度重建方法研究

doi: 10.21656/1000-0887.460171

作者简介: 单彦博(1998—)，男，硕士（E-mail: 2623908774@qq.com）；张立芳（1988—），女，讲师，博士（通信作者. E-mail: 21227023@zju.edu.cn）.

通讯作者: 张立芳（1988—），女，讲师，博士(通信作者. E-mail: 21227023@zju.edu.cn).

计量

出版历程

Research on the Reconstruction Method for Premixed Flame Temperature and H2O Concentration Based on Parameter Optimization of Convolutional Neural Networks

计量

出版历程

目录

基于参数优化卷积神经网络的预混火焰温度与H₂O浓度重建方法研究

作者简介:
单彦博(1998—)，男，硕士（E-mail: 2623908774@qq.com）；张立芳（1988—），女，讲师，博士（通信作者. E-mail: 21227023@zju.edu.cn）.

通讯作者:
张立芳（1988—），女，讲师，博士(通信作者. E-mail: 21227023@zju.edu.cn).

Research on the Reconstruction Method for Premixed Flame Temperature and H₂O Concentration Based on Parameter Optimization of Convolutional Neural Networks