基于深层卷积随机配置网络的电熔镁炉工况识别方法研究

李帷韬; 童倩倩; 王殿辉; 吴高昌

doi:10.16383/j.aas.c230272

基于深层卷积随机配置网络的电熔镁炉工况识别方法研究

doi: 10.16383/j.aas.c230272

李帷韬^1,,
童倩倩^1,,
王殿辉^{2, 3,},
吴高昌^3,

1.
合肥工业大学电气与自动化工程学院合肥 230009
2.
中国矿业大学人工智能研究院徐州 221116
3.
东北大学流程工业综合自动化国家重点实验室沈阳 110819

基金项目: 国家重点研发计划(2018AAA0100304), 国家自然科学基金(62173120, 62103092), 安徽省自然科学基金(2108085UD11), 111引智项目(BP0719039)资助

详细信息

作者简介:
李帷韬：合肥工业大学电气与自动化工程学院副教授. 主要研究方向为深度学习, 图像处理和智能认知. E-mail: wtli@hfut.edu.cn

童倩倩：合肥工业大学电气与自动化工程学院硕士研究生. 主要研究方向为智能认知. E-mail: 2021110400@mail.hfut.edu.cn

王殿辉：中国矿业大学人工智能研究院教授. 主要研究方向为工业大数据建模与分析, 随机配置学习理论及工业应用. 本文通信作者. E-mail: dh.wang@deepscn.com

吴高昌：东北大学流程工业综合自动化国家重点实验室副教授. 主要研究方向为智能计算成像, 深度学习和异常工况智能感知与预测. E-mail: wugc@mail.neu.edu.cn

计量
- 文章访问数: 205
- HTML全文浏览量: 117
- PDF下载量: 96
- 被引次数: 0
出版历程
- 收稿日期: 2023-05-10
- 录用日期: 2023-09-26
- 网络出版日期: 2024-02-27
- 刊出日期: 2024-03-29

Research on Fused Magnesium Furnace Working Condition Recognition Method Based on Deep Convolutional Stochastic Configuration Networks

1.
School of Electrical Engineering and Automation, Hefei University of Technology, Hefei 230009
2.
Institute of Artificial Intelligence, China University of Mining and Technology, Xuzhou 221116
3.
State Key Laboratory of Synthetical Automation for Process Industries, Northeastern University, Shenyang 110819

Funds: Supported by National Key Research and Development Program of China (2018AAA0100304), National Natural Science Foundation of China (62173120, 62103092), Anhui Provincial Natural Science Foundation (2108085UD11), and 111 Project (BP0719039)

More Information

Author Bio:
LI Wei-Tao　Associate professor at the School of Electrical Engineering and Automation, Hefei University of Technology. His research interest covers deep learning, image processing, and intelligent cognition

TONG Qian-Qian　Master student at the School of Electrical Engineering and Automation, Hefei University of Technology. Her main research interest is intelligent cognition

WANG Dian-Hui　Professor at the Institute of Artificial Intelligence, China University of Mining and Technology. His research interest covers industrial big data modeling and analysis, stochastic configuration learning theory and industrial applications. Corresponding author of this paper

WU Gao-Chang　Associate professor at the State Key Laboratory of Synthetical Automation for Process Industries, Northeastern University. His research interest covers intelligent computational imaging, deep learning, and intelligent sensing and prediction of abnormal working conditions

摘要

摘要: 为解决电熔镁炉工况识别模型泛化能力和可解释性弱的缺陷, 提出一种基于深层卷积随机配置网络(Deep convolutional stochastic configuration networks, DCSCN)的可解释性电熔镁炉异常工况识别方法. 首先, 基于监督学习机制生成具有物理含义的高斯差分卷积核, 采用增量式方法构建深层卷积神经网络(Deep convolutional neural network, DCNN), 确保识别误差逐级收敛, 避免反向传播算法迭代寻优卷积核参数的过程. 定义通道特征图独立系数获取电熔镁炉特征类激活映射图的可视化结果, 定义可解释性可信度评测指标, 自适应调节深层卷积随机配置网络层级, 对不可信样本进行再认知以获取最优工况识别结果. 实验结果表明, 所提方法较其他方法具有更优的识别精度和可解释性.
- 电熔镁炉 /
- 深层卷积随机配置网络 /
- 高斯差分卷积核 /
- 类激活映射图 /
- 可解释性
Abstract: In order to solve the defects of generalization ability and weak interpretability of fused magnesium furnace working condition recognition model, an interpretable fused magnesium furnace abnormal working condition recognition method based on deep convolutional stochastic configuration networks (DCSCN) is proposed in this paper. Firstly, based on the supervised learning mechanism to generate Gaussian differential convolution kernel with physical meaning, an incremental method is used to construct a deep convolutional neural network (DCNN) to ensure that the recognition error converges step by step, and to avoid the process that back propagation algorithm iteratively finds the optimal convolutional kernel parameters. This paper defines channel feature map independent coefficients to obtain visualization results of fused magnesium furnace feature class activation mapping map, defines interpretable credibility measure to adaptively adjust deep convolutional stochastic configuration network layers, and recognizes untrustworthy samples to obtain optimal working condition recognition results. The experimental results show that the proposed method in this paper has better recognition accuracy and interpretability than other methods.
- Fused magnesium furnace /
- deep convolutional stochastic configuration networks (DCSCN) /
- Gaussian differential convolution kernel /
- class activation mapping map /
- interpretability

HTML全文

图 1 基于深层卷积随机配置网络的可解释电熔镁炉工况识别模型结构图

Fig. 1 Structure of interpretable fused magnesium furnace working condition recognition model based on deep convolutional stochastic configuration networks

下载: 全尺寸图片幻灯片

图 2 深层卷积随机配置网络结构图

Fig. 2 Deep convolutional stochastic configuration networks structure diagram

下载: 全尺寸图片幻灯片

图 3 基于特征图独立性得分的类激活映射示意图

Fig. 3 Schematic diagram of the class activation mapping based on feature map independence scores

下载: 全尺寸图片幻灯片

图 4 正常工况图像数据增强后的结果

Fig. 4 Results of normal conditions image data enhancement

下载: 全尺寸图片幻灯片

图 5 欠烧工况图像数据增强后的结果

Fig. 5 Results after image data enhancement for underburning conditions

下载: 全尺寸图片幻灯片

图 6 过热工况图像数据增强后的结果

Fig. 6 Results after image data enhancement for superheated operating conditions

下载: 全尺寸图片幻灯片

图 7 异常排气工况图像数据增强后的结果

Fig. 7 Results after image data enhancement for abnormal exhaust conditions

下载: 全尺寸图片幻灯片

图 8 不同卷积核大小条件下的识别精度曲线

Fig. 8 Recognition accuracy curves under different convolutional kernel sizes

下载: 全尺寸图片幻灯片

图 9 强化学习训练过程的平均奖励曲线

Fig. 9 Average reward curves for training process of reinforcement learning methods

下载: 全尺寸图片幻灯片

图 10 不同卷积层类激活映射图

Fig. 10 Different convolutional layer class activation mapping maps

下载: 全尺寸图片幻灯片

图 11 本文方法与基于强化学习的类激活映射图对比

Fig. 11 Comparison of the method proposed in this paper with the class activation mapping maps based on reinforcement learning

下载: 全尺寸图片幻灯片

图 12 本文方法与基于强化学习的可信识别样本比例变化曲线

Fig. 12 The proportion change curves of trusted recognition samples based on reinforcement learning and the method proposed in this paper

下载: 全尺寸图片幻灯片

图 13 不同网络模型的训练样本识别精度曲线

Fig. 13 Recognition accuracy curves of training samples for different network models

下载: 全尺寸图片幻灯片

表 1 基于强化学习的漏诊率、误诊率和精度对比 (%)

Table 1 Comparison of missed diagnosis rate, misdiagnosis rate and accuracy based on reinforcement learning (%)

模型		训练集			测试集
模型		漏诊率	误诊率	精度	漏诊率	误诊率	精度
单层	本文方法	7.61 ± 0.189	9.15 ± 0.331	83.24 ± 0.195	9.95 ± 0.216	10.30 ± 0.231	79.75 ± 0.108
单层	强化学习	9.08 ± 0.082	10.14 ± 0.354	80.76 ± 0.228	10.51 ± 0.172	12.81 ± 0.390	76.68 ± 0.305
三层	本文方法	5.31 ± 0.239	1.96 ± 0.165	92.73 ± 0.166	5.24 ± 0.245	2.45 ± 0.203	92.31 ± 0.283
三层	强化学习	7.36 ± 0.361	2.58 ± 0.313	90.06 ± 0.313	6.57 ± 0.361	3.61 ± 0.313	89.82 ± 0.329

下载: 导出CSV

表 2 消融实验结果 (%)

Table 2 Results of ablation experiments (%)

模型	训练集			测试集
模型	漏诊率	误诊率	精度	漏诊率	误诊率	精度
本文方法	5.31 ± 0.239	1.96 ± 0.165	92.73 ± 0.166	5.24 ± 0.245	2.45 ± 0.203	92.31 ± 0.283
未加入可解释性模块	5.57 ± 0.232	2.51 ± 0.223	91.92 ± 0.278	7.29 ± 0.173	1.59 ± 0.181	91.12 ± 0.347
未加入高斯卷积核	4.29 ± 0.274	4.51 ± 0.391	91.20 ± 0.264	3.45 ± 0.255	2.50 ± 0.329	90.54 ± 0.231
未加入可解释性模块以及高斯卷积核	6.02 ± 0.183	4.25 ± 0.231	89.73 ± 0.325	4.13 ± 0.242	6.73 ± 0.228	89.14 ± 0.179

下载: 导出CSV

表 3 不同高斯噪声的实验结果 (%)

Table 3 Experimental results with different Gaussian noises (%)

模型	训练集			测试集
模型	漏诊率	误诊率	精度	漏诊率	误诊率	精度
本文方法($\eta=0.3$)	5.31 ± 0.239	1.96 ± 0.165	92.73 ± 0.166	5.24 ± 0.245	2.45 ± 0.203	92.31 ± 0.283
$\eta=0.6$模型	6.92 ± 0.232	2.21 ± 0.223	90.87 ± 0.206	7.19 ± 0.173	2.52 ± 0.181	90.29 ± 0.347
$\eta=0.9$模型	8.31 ± 0.423	2.29 ± 0.248	89.40 ± 0.297	7.45 ± 0.382	7.01 ± 0.274	85.54 ± 0.288

下载: 导出CSV

表 4 不同模型的测试样本漏诊率、误诊率和精度对比 (%)

Table 4 Comparison of missed diagnosis rate, misdiagnosis rate and accuracy of test samples with different models (%)

模型	漏诊率	误诊率	精度
SCN	14.21 ± 0.228	14.21 ± 0.228	76.14 ± 0.215
块增量BSC	12.58 ± 0.285	10.57 ± 0.153	76.85 ± 0.233
2DSCN	6.49 ± 0.263	15.52 ± 0.303	77.99 ± 0.353
DeepSCN	9.04 ± 0.285	7.32 ± 0.075	83.64 ± 0.209
CNN	6.82 ± 0.376	5.46 ± 0.167	87.72 ± 0.231
贝叶斯网络^[6]	5.36 ± 0.268	4.72 ± 0.252	89.92 ± 0.256
CNN+LSTM^[8]	6.91 ± 0.201	3.52 ± 0.184	89.57 ± 0.337
本文方法	5.24 ± 0.245	2.45 ± 0.203	92.31 ± 0.283

下载: 导出CSV

表 5 不同识别模型的综合性能对比

Table 5 Comprehensive performance comparison of different recognition models

模型	参数量(MB)	训练时间(s)	测试时间(s)
SCN	500.038	10278.834	0.011
块增量BSC	500.038	8341.094	0.011
2DSCN	1000.038	12352.771	0.013
DeepSCN	127.899	15411.081	0.013
CNN	0.664	20714.322	0.014
贝叶斯网络^[6]	0.046	26.258	0.022
CNN+LSTM^[8]	4.127	20159.642	0.015
本文方法	12.854	18218.021	0.014

下载: 导出CSV

表 6 太阳能电池板数据集实验结果对比 (%)

Table 6 Comparison of experimental results for solar panel dataset (%)

模型		漏诊率	误诊率	精度
单层	本文方法	7.31$\pm$0.187	7.86$\pm$0.259	84.83$\pm$0.245
单层	未加入可解释性模块	9.87$\pm$0.252	6.94$\pm$0.243	83.19$\pm$0.279
三层	本文方法	3.45$\pm$0.213	3.51$\pm$0.169	93.04$\pm$0.323
三层	未加入可解释性模块	4.13$\pm$0.192	4.22$\pm$0.257	91.65$\pm$0.236

下载: 导出CSV

参考文献(35)

[1]	卢绍文, 温乙鑫. 基于图像与电流特征的电熔镁炉欠烧工况半监督分类方法. 自动化学报, 2021, 47(4): 891-902 Lu Shao-Wen, Wen Yi-Xin. Semi-supervised classification of semi-molten working condition of fused magnesium furnace based on image and current features. Acta Automatica Sinica, 2021, 47(4): 891-902
[2]	刘强, 孔德志, 郎自强. 基于多级动态主元分析的电熔镁炉异常工况诊断. 自动化学报, 2021, 47(11): 2570-2577 Liu Qiang, Kong De-Zhi, Lang Zi-Qiang. Multi-level dynamic principal component analysis for abnormality diagnosis of fused magnesia furnaces. Acta Automatica Sinica, 2021, 47(11): 2570-2577
[3]	Wu Z W, Wu Y J, Chai T Y, Sun J. Data-driven abnormal condition identification and self-healing control system for fused magnesium furnace. IEEE Transactions on Industrial Electronics, 2015, 62(3): 1703-1715 doi: 10.1109/TIE.2014.2349479
[4]	吴志伟. 嵌入式电熔镁炉智能控制系统研究 [博士学位论文], 东北大学, 中国, 2015. Wu Zhi-Wei. Embedded Intelligent Control System for Fused Magnesium Furnace [Ph.D. dissertation], Northeastern University, China, 2015.
[5]	李荟, 王福利, 李鸿儒. 电熔镁炉熔炼过程异常工况识别及自愈控制方法. 自动化学报, 2020, 46(7): 1411-1419 Li Hui, Wang Fu-Li, Li Hong-Ru. Abnormal condition identification and self-Healing control scheme for the electro-fused magnesia smelting process. Acta Automatica Sinica, 2020, 46(7): 1411-1419
[6]	闫浩, 王福利, 孙钰沣, 何大阔. 基于贝叶斯网络参数迁移学习的电熔镁炉异常工况识别. 自动化学报, 2021, 47(1): 197-208 Yan Hao, Wang Fu-Li, Sun Yu-Feng, HE Da-Kuo. Abnormal condition identification based on bayesian network parameter transfer learning for the electro-fused magnesia. Acta Automatica Sinica, 2021, 47(1): 197-208
[7]	Lu S W, Wen Y X. Semi-supervised condition monitoring and visualization of fused magnesium furnace. IEEE Transactions on Automation Science and Engineering, 2022, 19(4): 3471-3482 doi: 10.1109/TASE.2021.3124015
[8]	吴高昌, 刘强, 柴天佑, 秦泗钊. 基于时序图像深度学习的电熔镁炉异常工况诊断. 自动化学报, 2019, 45(8): 1475-1485 Wu Gao-Chang, Liu Qiang, Chai Tian-You, Qin Si-Zhao. Abnormal condition diagnosis through deep learning of image sequences for fused magnesium furnaces. Acta Automatica Sinica, 2019, 45(8): 1475-1485
[9]	Lu S W, Gao H R. Deep learning based fusion of RGB and infrared images for the detection of abnormal condition of fused magnesium furnace. In: Proceedings of the IEEE 15th International Conference on Control and Automation (ICCA). Edinburgh, UK: IEEE, 2019. 987−993
[10]	Bu K Q, Liu Y, Wang F L. Operating performance assessment based on multi-source heterogeneous information with deep learning for smelting process of electro-fused magnesium furnace. ISA Transactions, 2022, 128: 357-371 doi: 10.1016/j.isatra.2021.10.024
[11]	卢绍文, 王克栋, 吴志伟, 李鹏琦, 郭章. 基于深度卷积网络的电熔镁炉欠烧工况在线识别. 控制与决策, 2019, 34(7): 1537-1544 Lu Shao-Wen, Wang Ke-Dong, Wu Zhi-Wei, Li Peng-Qi, Guo Zhang. Online detection of semi-molten of fused magnesium furnace based ondeep convolutional neural network. Control and Decision, 2019, 34(7): 1537-1544
[12]	Zeiler M D, Fergus R. Visualizing and understanding convolutional networks. In: Proceedings of the 13th European Conference on Computer Vision (ECCV). Zurich, Switzerland: Springer, 2014. 818−833
[13]	Samek W, Binder A, Montavon G, Lapuschkin S, Müller K R. Evaluating the visualization of what a deep neural network has learned. IEEE Transactions on Neural Networks and Learning Systems, 2017, 28(11): 2660-2673 doi: 10.1109/TNNLS.2016.2599820
[14]	Larsen A B L, Sonderby S K, Larochelle H, Winther O. Autoencoding beyond pixels using a learned similarity metric. In: Proceedings of the 33rd International Conference on Machine Learning. New York, USA: PMLR, 2016. 1558−1566
[15]	Zhang Q S, Wu Y N, Zhu S C. Interpretable convolutional neural networks. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018. 8827−8836
[16]	Pham H, Guan M, Zoph B, Le Q, Dean J. Efficient neural architecture search via parameters sharing. In: Proceedings of the 35th International Conference on Machine Learning. Stockholm, Sweden: PMLR, 2018. 4095−4104
[17]	Ding Z X, Chen Y R, Li N N, Zhao D B, Sun Z Q, Chen C L P. BNAS: Efficient neural architecture search using broad scalable architecture. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(9): 5004-5018 doi: 10.1109/TNNLS.2021.3067028
[18]	Alvarez J M, Salzmann M. Learning the number of neurons in deep networks. In: Proceedings of the 30th International Conference on Neural Information Processing Systems. Barcelona, Spain: Curran Associates Inc., 2016. 2270−2278
[19]	Kawaguchi K. Deep learning without poor local minima. In: Proceedings of the 30th International Conference on Neural Information Processing Systems. Barcelona, Spain: Curran Associates Inc., 2016. 586−594
[20]	Pao Y H, Park G H, Sobajic D J. Learning and generalization characteristics of the random vector functional-link net. Neurocomputing, 1994, 6(2): 163-180 doi: 10.1016/0925-2312(94)90053-1
[21]	Wang D H, Li M. Stochastic configuration networks: Fundamentals and algorithms. IEEE Transactions on Cybernetics, 2017, 47(10): 3466-3479 doi: 10.1109/TCYB.2017.2734043
[22]	Li M, Wang D H. 2-D stochastic configuration networks for image data analytics. IEEE Transactions on Cybernetics, 2021, 51(1): 359-372 doi: 10.1109/TCYB.2019.2925883
[23]	Wang D H, Li M. Deep stochastic configuration networks with universal approximation property. In: Proceedings of the International Joint Conference on Neural Networks (IJCNN). Rio de Janeiro, Brazil: IEEE, 2018. 1−8
[24]	Pratama M, Wang D H. Deep stacked stochastic configuration networks for lifelong learning of non-stationary data streams. Information Sciences, 2019, 495: 150-174 doi: 10.1016/j.ins.2019.04.055
[25]	Li W T, Zhang Q, Wang D H, Sun W, Li Q Y. Stochastic configuration networks for self-blast state recognition of glass insulators with adaptive depth and multi-scale representation. Information Sciences, 2022, 604: 61-79 doi: 10.1016/j.ins.2022.04.061
[26]	Li W T, Deng Y L, Ding M S, Wang D H, Sun W, Li Q Y. Industrial data classification using stochastic configuration networks with self-attention learning features. Neural Computing and Applications, 2022, 34: 22047-22069 doi: 10.1007/s00521-022-07657-9
[27]	Peng S Y, Ding L J, Li W T, Sun W, Li Q Y. Research on intelligent recognition method for self-blast state of glass insulator based on mixed data augmentation. High Voltage, 2023, 8(4): 668-681 doi: 10.1049/hve2.12296
[28]	Zhang Q, Li W T, Li H, Wang J P. Self-blast state detection of glass insulators based on stochastic configuration networks and a feedback transfer learning mechanism. Information Sciences, 2020, 522: 259-274 doi: 10.1016/j.ins.2020.02.058
[29]	Li W T, Tao H, Li H, Chen K Q, Wang J P. Greengage grading using stochastic configuration networks and a semi-supervised feedback mechanism. Information Sciences, 2019, 488: 1-12 doi: 10.1016/j.ins.2019.02.041
[30]	Li W T, Chen K Q, Wang D H. Industrial image classification using a randomized neural-net ensemble and feedback mechanism. Neurocomputing, 2016, 173: 708-714 doi: 10.1016/j.neucom.2015.08.019
[31]	He Y, Ding Y H, Liu P, Zhu L C, Zhang H W, Yang Y. Learning filter pruning criteria for deep convolutional neural networks acceleration. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE, 2020. 2006−2015
[32]	Sutton R S, Barto A G. Reinforcement Learning: An Introduction. Cambridge: MIT Press, 1998.
[33]	Dai W, Li D P, Zhou P, Chai T Y. Stochastic configuration networks with block increments for data modeling in process industries. Information Sciences, 2019, 484: 367-386 doi: 10.1016/j.ins.2019.01.062
[34]	LeCun Y. LeNet-5, convolutional neural networks [Online], available: http://yann.lecun.com/exdb/lenet, January 11, 2024
[35]	Deitsch S, Christlein V, Berger S, Buerhop-Lutz C, Maier A, Gallwitz F, et al. Automatic classification of defective photovoltaic module cells in electroluminescence images. Solar Energy, 2019, 185: 455-468 doi: 10.1016/j.solener.2019.02.067