[1]韩哲哲,段德智,倪浩伟,等.基于火焰成像和堆栈降噪自编码的燃烧工况识别[J].东南大学学报(自然科学版),2020,50(3):537-544.[doi:10.3969/j.issn.1001-0505.2020.03.017]
 Han Zhezhe,Duan Dezhi,Ni Haowei,et al.Combustion condition identification through flame imaging and stacked denoising autoencoder[J].Journal of Southeast University (Natural Science Edition),2020,50(3):537-544.[doi:10.3969/j.issn.1001-0505.2020.03.017]
点击复制

基于火焰成像和堆栈降噪自编码的燃烧工况识别()
分享到:

《东南大学学报(自然科学版)》[ISSN:1001-0505/CN:32-1178/N]

卷:
50
期数:
2020年第3期
页码:
537-544
栏目:
能源与动力工程
出版日期:
2020-05-20

文章信息/Info

Title:
Combustion condition identification through flame imaging and stacked denoising autoencoder
作者:
韩哲哲1段德智2倪浩伟1李金健1刘煜东1李健1张彪1许传龙1
1东南大学火电机组振动国家工程研究中心, 南京 210096; 2江西中船航海仪器有限公司, 九江 332000
Author(s):
Han Zhezhe1 Duan Dezhi2 Ni Haowei1 Li Jinjian1 Liu Yudong1 Li Jian1Zhang Biao1 Xu Chuanlong1
1National Engineering Research Center of Turbogenerator Vibration, Southeast University, Nanjing 210096, China
2Jiangxi Zhonghuai Navigation Instruments Co., Ltd., Jiujiang 332000, China
关键词:
燃烧工况识别 火焰图像 堆栈降噪自编码 高斯过程分类器
Keywords:
combustion condition identification flame image stacked denoising autoencoder Gaussian process classifier
分类号:
TK227.1
DOI:
10.3969/j.issn.1001-0505.2020.03.017
摘要:
提出一种基于深度神经网络的燃烧监测方法. 该方法利用具有深层结构的堆栈降噪自编码(SDAE)提取火焰图像特征, 并将其输入到高斯过程分类器(GPC)中, 从而识别燃烧工况. 针对SDAE训练集中未出现的新燃烧工况, 使用少量新工况的标签图像对GPC进行重新训练, 即可扩大监测模型的识别范围. 在重油燃烧试验装置上开展了试验研究, 利用获得的火焰图像对SDAE-GPC网络进行模型训练以及性能测试. 结果表明, 所提出的监测方法对训练集所包含的燃烧工况具有99.3%的识别精度, 对新工况具有98.2%的识别精度, 且对图像噪声具有良好的鲁棒性, 在燃烧工况识别中具有潜在的应用前景.
Abstract:
A combustion monitoring method based on the deep neural network was proposed. The features of flame images were extracted by a stacked denoising autoencoder(SDAE)with deep architectures and then inputted into the Gaussian process classifier(GPC)to identify combustion conditions. For the new combustion conditions without being contained in the SDAE training dataset, the monitoring model could expand the identification range if the GPC was retrained using a small number of labeled images from new conditions. The experiments were carried out on the heavy-oil combustion test facility. The SDAE-GPC network was trained and tested using the flame images. The results show that the identification accuracy of the proposed monitoring method is 99.3% for the combustion conditions contained in the training dataset and 98.2% for the new conditions. Also, the model has a good robustness to the image noise and potential application prospects on the combustion condition identification.

参考文献/References:

[1] 刘伟, 司风琪, 徐治皋. 基于燃烧特征量和模糊C均值聚类的燃烧诊断[J]. 东南大学学报(自然科学版), 2012, 42(S2): 326-330. DOI:10.3969/j.issn.1001-0505.2012.S2.019.
Liu W, Si F Q, Xu Z G. Combustion diagnosis based on combustion feature and fuzzy C-means clustering[J]. Journal of Southeast University(Natural Science Edition), 2012, 42(S2): 326-330. DOI:10.3969/j.issn.1001-0505.2012.S2.019. (in Chinese)
[2] 白卫东, 严建华, 马增益, 等. 基于支持向量机的火焰状态识别方法[J]. 动力工程, 2004, 24(4): 548-551. DOI:10.3969/j.issn.1674-7607.2004.04.021.
Bai W D, Yan J H, Ma Z Y,et al. Method of flame identification based on support vector machine[J]. Power Engineering, 2004, 24(4): 548-551. DOI:10.3969/j.issn.1674-7607.2004.04.021. (in Chinese)
[3] Sun D, Lu G, Zhou H, et al. Quantitativeassessment of flame stability through image processing and spectral analysis[J]. IEEE Transactions on Instrumentation and Measurement, 2015, 64(12): 3323-3333. DOI:10.1109/tim.2015.2444262.
[4] Wang Z Y, Song C F, Chen T. Deep learning based monitoring of furnace combustion state and measurement of heat release rate[J]. Energy, 2017, 131: 106-112. DOI:10.1016/j.energy.2017.05.012.
[5] Abdurakipov S S, Gobyzov O A, Tokarev M P, et al. Combustion regime monitoring by flame imaging and machine learning[J]. Optoelectronics, Instrumentation and Data Processing, 2018, 54(5): 513-519. DOI:10.3103/s875669901805014x.
[6] Vincent P,Larochelle H, Lajoie I, et al. Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion[J]. Journal of Machine Learning Research, 2010, 11(12): 3371-3408.
[7] Hinton G E. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786): 504-507. DOI:10.1126/science.1127647.
[8] Zhou P, Austin J. Learning criteria for training neural network classifiers[J]. Neural Computing & Applications, 1998, 7(4): 334-342. DOI:10.1007/bf01428124.
[9] Zhao K G, Popescu S, Zhang X S. Bayesian learning with Gaussian processes for supervised classification of hyperspectral data[J]. Photogrammetric Engineering & Remote Sensing, 2008, 74(10): 1223-1234. DOI:10.14358/pers.74.10.1223.
[10] Han Z Z, Hossain M M, Wang Y W, et al. Combustion stability monitoring through flame imaging and stacked sparse autoencoder based deep neural network[J]. Applied Energy, 2020, 259: 114159. DOI:10.1016/j.apenergy.2019.114159.
[11] Bernard S,Heutte L, Adam S. Influence of hyperparameters on random forest accuracy[M]//Multiple Classifier Systems. Berlin, Heidelberg: Springer, 2009: 171-180. DOI:10.1007/978-3-642-02326-2_18.

备注/Memo

备注/Memo:
收稿日期: 2019-11-21.
作者简介: 韩哲哲(1987—), 男,博士生; 许传龙(联系人), 男, 博士, 教授, 博士生导师, chuanlongxu@seu.edu.cn.
基金项目: 国家自然科学基金资助项目(51976038)、中央高校基本科研业务费专项资金资助项目(2242019k1G018).
引用本文: 韩哲哲,段德智,倪浩伟,等.基于火焰成像和堆栈降噪自编码的燃烧工况识别[J].东南大学学报(自然科学版),2020,50(3):537-544. DOI:10.3969/j.issn.1001-0505.2020.03.017.
更新日期/Last Update: 2020-05-20