基于动态注意力深度迁移网络的高炉铁水硅含量在线预测方法

蒋珂; 蒋朝辉; 谢永芳; 潘冬; 桂卫华

doi:10.16383/j.aas.c210524

基于动态注意力深度迁移网络的高炉铁水硅含量在线预测方法

doi: 10.16383/j.aas.c210524

蒋珂^1,,
蒋朝辉^{1, 2,},
谢永芳^1,,
潘冬^1,,
桂卫华^1,

1.
中南大学自动化学院长沙 410000
2.
鹏城实验室深圳 518000

基金项目: 国家自然科学基金(61773406, 61725306, 61290325), 国家重大科研仪器研制项目(61927803), 中南大学研究生自主探索创新项目(2021zzts0183), 湖南省研究生科研创新项目(CX20210242)资助

详细信息

作者简介:
蒋珂：中南大学博士研究生. 2019年获得中南大学硕士学位. 主要研究方向为数据驱动的工业过程建模与控制, 过程数据分析和机器学习. E-mail: jiangke@csu.edu.cn

蒋朝辉：中南大学自动化学院教授. 2011年获得中南大学博士学位. 主要研究方向为智能传感与检测技术, 图像处理与智能识别, 人工智能和机器学习. 本文通信作者. E-mail: jzh0903@csu.edu.cn

谢永芳：中南大学自动化学院教授. 1993 年获得中南工业大学学士学位. 主要研究方向为分散控制, 鲁棒控制, 过程控制, 工业大数据和知识自动化. E-mail: yfxie@csu.edu.cn

潘冬：中南大学自动化学院讲师. 分别于2015年和2021年获得中南大学学士和博士学位. 2019年至2021年, 在加拿大拉瓦尔大学电子与计算工程系联合培养. 主要研究方向为红外热成像, 视觉检测, 图像处理和深度学习. E-mail: pandong@csu.edu.cn

桂卫华：中国工程院院士, 中南大学自动化学院教授. 1981年获得中南矿冶学院硕士学位. 主要研究方向为复杂工业过程建模, 优化与控制应用, 故障诊断与分布式鲁棒控制. E-mail: gwh@csu.edu.cn

计量
- 文章访问数: 2420
- HTML全文浏览量: 633
- PDF下载量: 518
- 被引次数: 0
出版历程
- 收稿日期: 2021-06-10
- 录用日期: 2021-11-02
- 网络出版日期: 2021-11-21
- 刊出日期: 2023-05-20

Online Prediction Method for Silicon Content of Molten Iron in Blast Furnace Based on Dynamic Attention Deep Transfer Network

JIANG Ke^1
,,
JIANG Zhao-Hui^{1, 2
,},
XIE Yong-Fang^1
,,
PAN Dong^1
,,
GUI Wei-Hua^1
,

1.
School of Automation, Central South University, Changsha 410000
2.
Peng Cheng Laboratory, Shenzhen 518000

Funds: Supported by National Natural Science Foundation of China (61773406, 61725306, 61290325), National Major Scientific Research Equipment of China (61927803), Independent Exploration and Innovation Project for Postgraduate of Central South University (2021zzts0183), and Hunan Provincial Innovation Foundation for Postgraduate (CX20210242)

More Information

Author Bio:
JIANG Ke　Ph.D. candidate at the School of Automation, Central South University. She received her master degree from Central South University in 2019. Her research interest covers data-based modeling and control of industrial process, process data analysis, and machine learning

JINAG Zhao-Hui　Professor at the School of Automation, Central South University. He received his Ph.D. degree from Central South University in 2011. His research interest covers intelligent sensing and detection technology, image processing and intelligent recognition, artificial intelligence, and machine learning. Corresponding author of this paper

XIE Yong-Fang　Professor at the School of Automation, Central South University. He received his bachelor degree from Central South University of Technology in 1993. His research interest covers decentralized control, robust control, process control, industrial big data, and knowledge automation

PAN Dong　Lecturer at the School of Automation, Central South University. He received his bachelor and Ph.D. degrees from Central South University in 2015 and 2021, respectively. He was a joint training Ph.D. candidate in the Department of Electrical and Computing Engineering from Université Laval, Canada, from 2019 to 2021. His research interest covers infr-ared thermal imaging, vision-based measurement, image processing, and deep learning

GUI Wei-Hua　Academician of Chinese Academy of Engineering, professor at the School of Automation, Central South University. He received his master degree from Central South Institute of Mining and Metallurgy in 1981. His research interest covers complex industrial process modeling, optimization and control applications, fault diagnosis, and distributed robust control

摘要

摘要: 铁水硅含量是反映高炉冶炼过程中热状态变化的灵敏指示剂, 但无法实时在线检测, 造成铁水质量调控盲目. 为此, 提出一种基于动态注意力深度迁移网络(Attention deep transfer network, ADTNet)的高炉铁水硅含量在线预测方法. 首先, 针对传统深度网络静态建模思路无法准确描述过程变量与铁水硅含量之间的关系, 提出一种基于注意力机制模块的输入过程变量与输出硅含量之间的动态关系描述方法; 其次, 为降低硅含量预测模型训练时对标签数据的依赖, 考虑到铁水温度与硅含量数据之间的正相关性, 利用小时级硅含量标签数据微调基于分钟级铁水温度数据预训练好的深度模型的结构, 进而提高基于动态注意力深度迁移网络的硅含量预测精度; 同时, 为增强预测网络的可解释性, 实时给出了基于动态注意力机制模块计算的每个样本各过程变量对铁水硅含量的贡献度; 最后, 基于某钢铁厂2号高炉的工业实验, 验证了该方法的准确性、有效性和先进性.
- 高炉炼铁 /
- 铁水硅含量 /
- 深度网络 /
- 迁移学习 /
- 动态注意力机制 /
- 预测
Abstract: The molten iron silicon content, which can reflect the thermal state in blast furnace hearth during the ironmaking process, is difficult to detect in real time. This will cause blind adjustment to the quality of the molten iron. Hence, this paper proposes a data-driven model for the online prediction of the silicon content based on dynamic attention deep transfer network (ADTNet). First, considering that the deep network cannot accurately describe the relationship between process variables and the silicon content based on static modeling process, a dynamic attention module is designed to describe the dynamic relationship between the inputs and outputs. Then, to reduce the dependence of labeled silicon content samples during model training process, an online prediction model is established based on minute-level temperature data after considering the positive correlation between molten iron temperature and silicon content data. Subsequently, using the labeled silicon content data to fine tune the parameters of the well-trained deep model and improve the silicon content prediction performance based on dynamic attention deep transfer network. Furthermore, to enhance the interpret-ability of the deep black box model, the contribution of process variables of each sample to the silicon content is presented based on the dynamic attention module. Finally, industrial experiments of No. 2 blast furnace in a steel plant verify the accuracy, effectiveness and advancement of the proposed method.
- Blast furnace ironmaking process /
- silicon content of molten iron /
- deep network /
- transfer learning /
- dynamic attention mechanism /
- prediction

HTML全文

图 1 高炉三维仿真模拟图

Fig. 1 Three-dimensional simulation diagram ofthe blast furnace cast field

下载: 全尺寸图片幻灯片

图 2 去噪自编码机基本结构

Fig. 2 Architecture of a denoising autoencoder

下载: 全尺寸图片幻灯片

图 3 堆叠去噪自编码机训练过程

Fig. 3 The training process of stacking denoising autoencoders

下载: 全尺寸图片幻灯片

图 4 动态注意力机制模块

Fig. 4 The dynamic attention mechanism module

下载: 全尺寸图片幻灯片

图 5 基于动态注意力机制模块的深度去噪自编码机网络

Fig. 5 Deep denoising autoencoders network based on dynamic attention mechanism module

下载: 全尺寸图片幻灯片

图 6 铁水温度与铁水硅含量的散点图

Fig. 6 The scatter plot of temperature and silicon content of molten iron

下载: 全尺寸图片幻灯片

图 7 高炉铁水测温系统

Fig. 7 Molten iron temperature measuring system in a blast furnace

下载: 全尺寸图片幻灯片

图 8 基于深度迁移网络的铁水硅含量在线预报模型

Fig. 8 Online prediction model of silicon content in molten iron based on deep transfer network

下载: 全尺寸图片幻灯片

图 9 基于支持向量回归机的铁水硅含量预测结果

Fig. 9 Prediction results of silicon content based on SVR

下载: 全尺寸图片幻灯片

图 10 基于堆叠去噪自编码机的铁水硅含量预测结果

Fig. 10 Prediction results of silicon content based on S-DAE

下载: 全尺寸图片幻灯片

图 11 基于动态注意力机制深度网络的铁水硅含量预测结果

Fig. 11 Prediction results of silicon content based on ADNet

下载: 全尺寸图片幻灯片

图 12 基于动态注意力深度迁移网络的铁水硅含量预测结果

Fig. 12 Prediction results of silicon content based on ADTNet

下载: 全尺寸图片幻灯片

图 13 基于不同模型的铁水硅含量误差分布图

Fig. 13 Prediction errors of silicon content in molten iron based on different models

下载: 全尺寸图片幻灯片

图 14 基于不同模型的预测和实际铁水硅含量分布散点图

Fig. 14 The scatter plot of predictive and observed silicon content based on different models

下载: 全尺寸图片幻灯片

图 15 过程变量注意力得分热力图

Fig. 15 The heat map of process variables attention scores

下载: 全尺寸图片幻灯片

表 1 过程变量最大互信息系数

Table 1 Maximal information coefficient of process variables

过程变量	MIC_MIT	MIC_Si	过程变量	MIC_MIT	MIC_Si
富氧率	0.104	0.115	冷风压力	0.104	0.094
透气性指数	0.104	0.111	全压差	0.104	0.130
CO	0.103	0.104	热风压力	0.103	0.116
CO₂	0.111	0.145	实际风速	0.100	0.113
标准风速	0.117	0.111	冷风温度	0.102	0.109
富氧流量	0.120	0.129	热风温度	0.101	0.115
冷风流量	0.117	0.111	顶温	0.120	0.162
鼓风动能	0.101	0.107	顶温下降管	0.111	0.155
炉腹煤气量	0.108	0.127	阻力系数	0.103	0.110
炉腹煤气指数	0.109	0.128	鼓风湿度	0.135	0.140
顶压	0.128	0.156	富氧压力	0.103	0.096
本小时实际喷煤量	0.100	0.136	上一小时实际喷煤量	0.110	0.165

下载: 导出CSV

表 2 基于不同模型的预测性能

Table 2 Prediction performance based on different models

模型	RMSE	MAE	HR (%)
SVR	0.0832	0.0635	77.5
S-DAE	0.0794	0.0616	84.6
ADNet	0.0772	0.0583	86.4
ADTNet	0.0649	0.0509	90.0

下载: 导出CSV

参考文献(28)

[1]	周平, 张丽, 李温鹏, 戴鹏, 柴天佑. 集成自编码与PCA的高炉多元铁水质量随机权神经网络建模. 自动化学报, 2018, 44(10): 1799-1811 Zhou Ping, Zhang Li, Li Wen-Peng, Dai Peng, Chai Tian-You. Modeling of blast furnace multi-element molten iron quality with random weight neural network based on self-encoding and PCA. Acta Automatica Sinica, 2018, 44(10): 1799-1811
[2]	Zhou H, Zhang H F, and Yang C J. Hybrid model based intelligent optimization of ironmaking process. IEEE Transaction on Industrial Electronics, 2020, 67(3): 2469-247 doi: 10.1109/TIE.2019.2903770
[3]	Jiang K, Jiang Z H, Xie Y F, Pan D, Gui W H. Abnormality monitoring in the blast furnace ironmaking process based on stacked dynamic target-driven denoising autoencoders. IEEE Transactions on Industrial Informatics, 2022, 18(3): 1854−1863
[4]	郜传厚, 渐令, 陈积明, 孙优贤. 复杂高炉炼铁过程的数据驱动建模及预测算法. 自动化学报, 2009, 35(06): 725-730 doi: 10.3724/SP.J.1004.2009.00725 Gao Chuan-Hou, Jian Ling, Chen Jia-Ming, Sun You-Xian. Data-driven modeling and prediction algorithm for complex blast furnace ironmaking process. Acta Automatica Sinica, 2009, 35(6): 725-730 doi: 10.3724/SP.J.1004.2009.00725
[5]	Chen S H, Gao C H. Linear priors mined and integrated for transparency of blast furnace black-Box SVM model. IEEE Transactions on Industrial Informatics, 2020, 16(6): 3862-3870 doi: 10.1109/TII.2019.2940475
[6]	Zhou P, Lv Y B, Wang H, and Chai T Y. Data-driven robust RVFLNs modeling of a blast furnace iron-making process using Cauchy distribution weighted M-Estimation. IEEE Transaction on Industrial Electronics, 2017, 64(9): 7141–7151 doi: 10.1109/TIE.2017.2686369
[7]	宋贺达, 周平, 王宏, 柴天佑. 高炉炼铁过程多元铁水质量非线性子空间建模及应用. 自动化学报, 2016, 42(11): 1664-1679 Song He-Da, Zhou Ping, Wang Hong, Chai Tian-You. Nonlinear subspace modeling of multivariate molten iron quality in blast furnace ironmaking and its application. Acta Automatica Sinica, 2016, 42(11): 1664-1679
[8]	Spirin N A, Onorin O P, Istomin A S. Study of transition processes of blast-furnace smelting by the mathematical model me-thod. In: Proceedings of the IOP Conference Series, Materials Science and Engineering. Suzhou, China: Institute of Physics Pu-blishing, 2018. 12−73
[9]	Spirin N, Onorin O, Alexander I. Prediction of blast furnace thermal state in real-time operation. Solid State Phenomena, 2020, 299: 518-523 doi: 10.4028/www.scientific.net/SSP.299.518
[10]	Spirin N. A, Polinov A A, Gurin I A, Pishnograev SN. Information system for real-time prediction of the silicon content of iron in a blast furnace. Metallurgist, 2020, 63(9): 898-905
[11]	Saxen H, Gao C H, and Gao Z W. Data-driven time discrete models for dynamic prediction of the hot metal silicon content in the blast furnace—A review. IEEE Transactions on Industrial Informatics, 2013, 9(4): 2213-2225 doi: 10.1109/TII.2012.2226897
[12]	李温鹏, 周平. 高炉铁水质量鲁棒正则化随机权神经网络建模. 自动化学报, 2020, 46(04): 721-733 Li Wen-Peng, Zhou Ping. Blast furnace hot metal quality robust regularization random weight neural network modeling. Acta Automatica Sinica, 2020, 46(04): 721-733
[13]	蒋朝辉, 许川, 桂卫华, 蒋珂. 基于最优工况迁移的高炉铁水硅含量预测方法. 自动化学报 Jiang Zhao-Hui, Xu Chuang, Gui Wei-Hua, Jiang Ke. Prediction method of hot metal silicon content in blast furnace based on optimal smelting condition migration. Acta Automatica Sinica, to be published.
[14]	Zhou P, Guo D W, Wang H, and Chai T Y. Data-driven robust M-LS-SVR-based NARX modeling for estimation and control of molten iron quality indices in blast furnace ironmaking. IEEE Transactions on Neural Networks and Learning Systems, 2018, 29(9): 4007-4021 doi: 10.1109/TNNLS.2017.2749412
[15]	蒋朝辉, 董梦林, 桂卫华, 阳春华, 谢永芳. 基于Bootstrap的高炉铁水硅含量二维预报. 自动化学报, 2016, 42(05): 715-723 Jiang Zhao-Hui, Dong Meng-Lin, Gui Wei-Hua, Yang Chun-Hua, Xie Yong-Fang. Two-dimensional prediction for silicon content of hot metal of blast furnace based on bootstrap. Acta Automatica Sinica, 2016, 42(5): 715-723
[16]	Li J P, Hua C C, Yang Y N, Guan X P. Bayesian block structure sparse based T–S fuzzy modeling for dynamic prediction of hot metal silicon content in the blast furnace. IEEE Transactions on Industrial Electronics, 2017, 65(6): 4933-4942
[17]	Hinton G E, Osindero S, and Teh Y W. A fast learning algorithm for deep belief nets. Neural Computing, 2006, 18(7): 1527-1554 doi: 10.1162/neco.2006.18.7.1527
[18]	Hinton G E, Deng L, Yu D, et al. Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups. IEEE Signal Process, 2012, 29(6): 82-97 doi: 10.1109/MSP.2012.2205597
[19]	Ma J, Wu F, Zhu J, Xu D, and Kong D. A pre-trained convolutional neural network based method for thyroid nodule diagnosis. Ultrasonics, 2017, 73: 221 doi: 10.1016/j.ultras.2016.09.011
[20]	Krizhevsky A, Sutskever I, and Hinton G E. Imagenet classification with deep convolutional neural networks. in Process Advance Neural Information Process System, 2012, 1097-1105
[21]	Jiang K, Jiang Z H, Xie Y F, Chen Z P, Pan D, Gui W H. Classification of silicon content variation trend based on fusion of multilevel features in blast furnace ironmaking. Information Sciences, 2020, 521: 32-45 doi: 10.1016/j.ins.2020.02.039
[22]	Wang Y L, Pan Z F, Yuan X F, Yang C H, and Gui W H. A novel deep learning based fault diagnosis approach for chemical process with extended deep belief network. ISA Transactions, 2020, 96: 457-467 doi: 10.1016/j.isatra.2019.07.001
[23]	Pan D, Jiang Z H, Chen Z P, Jiang K, Gui W H. Compensation method for molten iron temperature measurement based on heterogeneous features of infrared thermal images. IEEE Transactions on Industrial Informatics, 2020, 16(11): 7056-7066. doi: 10.1109/TII.2020.2972332
[24]	Pan D, Jiang Z H, Chen Z P, Gui W H, Xie Y F, Yang C H. Temperature measurement and compensation method of blast furnace molten iron based on infrared computer vision. IEEE Transactions on Instrumentation and Measurement, 2018, 68 (10): 3576-3588.
[25]	Vincent P, Larochelle H, Lajoie I, Bengio Y, and Manzagol P. Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion. Journal of Machine Learning Research, 2010, 11: 3371–3408
[26]	Vincent P, Larochelle H, Bengio Y, Manzagol P A. Extracting and composing robust features with denoising autoencoders. In: Proceedings of the 25th International Conference Machine Lea-rning. Helsinki, Finland: 2008. 1096−1103
[27]	Reshef D N, Reshef Y A. Detecting novel associations in large data sets. Science, 2011, 334(6062): 1518-1524 doi: 10.1126/science.1205438
[28]	Agarap A F. Deep learning using rectified linear units (ReLU). arXiv preprint arXiv: 1803.08375, 2018.