基于时序迁移与双流加权的ONLSTM软测量建模

doi:10.11949/0438-1157.20230893

摘要/Abstract

摘要：

实际化工过程建模具有多变量、非线性和动态性等特点，会导致模型复杂度提高且提取特征时产生冗余信息和时序分布漂移问题，因此提出一种基于时序迁移和双流加权的有序神经元长短时记忆网络（ONLSTM）模型。首先，利用时序迁移对特征分布进行匹配以自适应表征特征分布信息，采用划分特征分布差异最大时间域进行训练，减小时序分布失配，从而解决时序分布漂移问题；其次，在时序迁移框架内嵌入双流加权ONLSTM模型，通过对ONLSTM主遗忘门和主输入门分别加权，更精确控制传递信息；进一步结合双流结构设计双信息流控制相应门控单元，减小参数调节过程中的耦合影响，降低模型复杂度，提高其预测性能；最后，将所提模型应用于硫回收过程以及某火电厂脱硫过程排放烟气SO₂浓度软测量建模，并与其他深度学习网络进行对比，验证了模型有效性。

关键词: 时间序列迁移, 加权有序神经元长短时记忆网络, 双流结构, 软测量, 神经网络, 过程控制, 动态建模

Abstract:

The modeling of actual chemical processes has characteristics such as multi-variability, nonlinearity, and dynamism, which can lead to increased model complexity and the generation of redundant information and temporal distribution shift when extracting features. Therefore, an ordered neurons long short-term memory network (ONLSTM) model based on time series transfer and dual stream weighting is proposed. First, temporal transfer is used to match feature distributions to adaptively represent feature distribution information and training is performed by dividing the time domain with the largest difference in feature distribution to reduce timing distribution mismatch, thereby solving the problem of timing distribution drift. Secondly, a dual stream weighted ONLSTM model is embedded within the time series transfer framework, and the ONLSTM main forgetting gate and main input gate are weighted separately to more accurately control the transmission of information. Further combining the dual flow structure to design the corresponding gating unit for dual information flow control, reducing the coupling effect during parameter adjustment, reducing model complexity, and improving its predictive performance. Finally, the proposed model was applied to soft sensing modeling of SO₂ concentration in the sulfur recovery process and the flue gas emissions from a certain thermal power plant desulfurization process, and compared with other deep learning networks to verify the effectiveness of the model.

Key words: time series transfer, weighted ordered neurons long short-term memory, dual stream, soft sensor, neural networks, process control, dynamic modeling

中图分类号:

TP 273

李祥宇, 隋璘, 马君霞, 熊伟丽. 基于时序迁移与双流加权的ONLSTM软测量建模[J]. 化工学报, 2023, 74(11): 4622-4633.

Xiangyu LI, Lin SUI, Junxia MA, Weili XIONG. ONLSTM soft sensor modeling based on time series transfer and dual stream weighting[J]. CIESC Journal, 2023, 74(11): 4622-4633.

图/表 15

图1 ONLSTM结构图

Fig.1 ONLSTM structure diagram

图2 加权ONLSTM结构图

Fig.2 Weighted ONLSTM structure diagram

图3 双流加权ONLSTM结构图

Fig.3 Dual stream weighted ONLSTM structure diagram

图4 TT-DSW-ONLSTM结构图

Fig.4 TT-DSW-ONLSTM structure diagram

图5 硫回收过程流程图

Fig.5 Process flow diagram of sulfur recovery

表1 SRU过程采样变量

Table 1 SRU process sampling variables

变量	变量描述
u₁	GAS气流
u₂	AIR空气流
u₃	AIR二次空气流
u₄	SWS区域GAS气流
u₅	SWS区域AIR空气流
y₁	SO₂浓度

图6 硫回收过程不同K值下TT-DSW-ONLSTM模型的RMSE

Fig.6 RMSE of TT-DSW-ONLSTM model under different K values in sulfur recovery process

图7 不同模型下SRU过程预测曲线

Fig.7 Prediction curves of SRU process under different models

图8 SRU过程各网络模型对SO2浓度预测误差

Fig.8 Prediction error of SO2 concentration by various network models in SRU process

表2 各网络模型对SO2浓度预测结果

Table 2 Prediction results of SO2 concentration by various network models

算法模型	MSE/10^-4	MAE	R²
LSTM	6.9696	0.0257	0.9933
ONLSTM	3.4971	0.0143	0.8925
WONLSTM	2.1054	0.0114	0.9353
DSW-ONLSTM	1.5775	0.0104	0.9589
TT-WONLSTM	0.6773	0.0072	0.9791
TT-DSW-ONLSTM	0.3158	0.0042	0.9919

图9 烟气吸收脱硫工艺流程图

Fig.9 Process flow diagram of flue gas absorption desulfurization

图10 烟气脱硫过程双流结构中不同α和β值下TT-DSW-ONLSTM模型的RMSE

Fig.10 RMSE of TT-DSW-ONLSTM model under different α and β values in dual flow structure of flue gas desulfurization process

图11 不同模型下烟气脱硫过程预测曲线

Fig.11 Prediction curves of flue gas desulfurization process under different models

图12 烟气脱硫过程各网络模型对SO2浓度预测误差

Fig.12 Prediction error of SO2 concentration by various network models in flue gas desulfurization process

表3 烟气脱硫过程各网络模型对SO2浓度预测结果

Table 3 Prediction results of SO2 concentration by various network models in flue gas desulfurization process

算法模型	MSE	MAE	R²
LSTM	7.7574	2.2530	0.6961
ONLSTM	5.8056	1.9712	0.7630
WONLSTM	4.4563	1.5972	0.8254
DSW-ONLSTM	3.9376	1.5350	0.8457
TT-WONLSTM	3.5981	1.4902	0.8651
TT-DSW-ONLSTM	3.1725	1.5196	0.8757

参考文献 31

1	周乐, 沈程凯, 吴超, 等. 深度融合特征提取网络及其在化工过程软测量中的应用[J]. 化工学报, 2022, 73(7): 3156-3165.
	Zhou L, Shen C K, Wu C, et al. Deep fusion feature extraction network and its application in chemical process soft sensing[J]. CIESC Journal, 2022, 73(7): 3156-3165.
2	Yuan X F, Li L, Wang Y L. Nonlinear dynamic soft sensor modeling with supervised long short-term memory network[J]. IEEE Transactions on Industrial Informatics, 2020, 16(5): 3168-3176.
3	Ren J Y, Chen X, Zhao C H. Spatio-temporal view GAIN for data imputation and dynamic soft sensor[C]//2022 IEEE 11th Data Driven Control and Learning Systems Conference (DDCLS). New York: IEEE, 2022: 1446-1451.
4	杜宇浩, 阎高伟, 李荣, 等. 基于局部线性嵌入的测地线流式核多工况软测量建模方法[J]. 化工学报, 2020, 71(3): 1278-1287.
	Du Y H, Yan G W, Li R, et al. Multiple working conditions soft sensor modeling method of geodesic flow kernel based on locally linear embedding[J]. CIESC Journal, 2020, 71(3): 1278-1287.
5	潘贝, 金怀平, 杨彪, 等. 基于多样性加权相似度的集成局部加权偏最小二乘软测量建模[J]. 信息与控制, 2019, 48(2): 217-223, 231.
	Pan B, Jin H P, Yang B, et al. Soft sensor development based on ensemble locally weighted partial least squares using diverse weighted similarity measures[J]. Information and Control, 2019, 48(2): 217-223, 231.
6	Liu J L, Wang Y K, Zhang Y. A novel isomap-SVR soft sensor model and its application in rotary kiln calcination zone temperature prediction[J]. Symmetry, 2020, 12(1): 167.
7	Shi X D, Kang Q, Zhou M C, et al. Soft sensing of nonlinear and multimode processes based on semi-supervised weighted Gaussian regression[J]. IEEE Sensors Journal, 2020, 20(21): 12950-12960.
8	Shang C, Yang F, Huang D X, et al. Data-driven soft sensor development based on deep learning technique[J]. Journal of Process Control, 2014, 24(3): 223-233.
9	Yao L, Ge Z Q. Deep learning of semisupervised process data with hierarchical extreme learning machine and soft sensor application[J]. IEEE Transactions on Industrial Electronics, 2018, 65(2): 1490-1498.
10	Sui L, Sun K, Ma J X, et al. Input variable selection and structure optimization for LSTM-based soft sensor with a dual nonnegative garrote approach[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1-11.
11	Wang K C, Shang C, Liu L, et al. Dynamic soft sensor development based on convolutional neural networks[J]. Industrial & Engineering Chemistry Research, 2019, 58(26): 11521-11531.
12	Wang Y L, Pan Z F, Yuan X F, et al. A novel deep learning based fault diagnosis approach for chemical process with extended deep belief network[J]. ISA Transactions, 2020, 96: 457-467.
13	Guo R Y, Liu H. A hybrid mechanism- and data-driven soft sensor based on the generative adversarial network and gated recurrent unit[J]. IEEE Sensors Journal, 2021, 21(22): 25901-25911.
14	邵伟明, 葛志强, 李浩, 等. 基于循环神经网络的半监督动态软测量建模方法[J]. 电子测量与仪器学报, 2019, 33(11): 7-13.
	Shao W M, Ge Z Q, Li H, et al. Semisupervised dynamic soft sensing approaches based on recurrent neural network[J]. Journal of Electronic Measurement and Instrumentation, 2019, 33(11): 7-13.
15	Lippi M, Montemurro M A, Degli Esposti M, et al. Natural language statistical features of LSTM-generated texts[J]. IEEE Transactions on Neural Networks and Learning Systems, 2019, 30(11): 3326-3337.
16	Zhi Z, Liu L S, Liu D T, et al. Fault detection of the harmonic reducer based on CNN-LSTM with a novel denoising algorithm[J]. IEEE Sensors Journal, 2022, 22(3): 2572-2581.
17	Lui C F, Liu Y Q, Xie M. A supervised bidirectional long short-term memory network for data-driven dynamic soft sensor modeling[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-13.
18	Ren L, Wang T, Laili Y J, et al. A data-driven self-supervised LSTM-DeepFM model for industrial soft sensor[J]. IEEE Transactions on Industrial Informatics, 2021, 18(9): 5859-5869.
19	Xie R M, Hao K R, Huang B, et al. Data-driven modeling based on two-stream λ gated recurrent unit network with soft sensor application[J]. IEEE Transactions on Industrial Electronics, 2020, 67(8): 7034-7043.
20	Shen Y K, Tan S, Sordoni A, et al. Ordered neurons: integrating tree structures into recurrent neural networks[EB/OL]. 2018. .
21	Zhang Q, Chai B, Song B C, et al. A hierarchical fine-tuning based approach for multi-label text classification[C]//2020 IEEE 5th International Conference on Cloud Computing and Big Data Analytics (ICCCBDA). Chengdu, China: IEEE, 2020: 51-54.
22	Dai S Z, Li Z H, Li L, et al. Investigating the dynamic memory effect of human drivers via ON-LSTM[J]. Science China Information Sciences, 2020, 63(9): 1-11.
23	Shi F, Cao H R, Wang Y K, et al. Chatter detection in high-speed milling processes based on ON-LSTM and PBT[J]. The International Journal of Advanced Manufacturing Technology, 2020, 111(11): 3361-3378.
24	Shi X D, Xu J, Bao H Q, et al. Domain adaptation soft sensing with parameter transferring[C]//2022 IEEE International Conference on Networking, Sensing and Control (ICNSC).New York: IEEE, 2023: 1-5.
25	Zhang X R, Song C Y, Zhao J, et al. Domain adaptation mixture of Gaussian processes for online soft sensor modeling of multimode processes when sensor degradation occurs[J]. IEEE Transactions on Industrial Informatics, 2022, 18(7): 4654-4664.
26	Chai Z, Zhao C H, Huang B, et al. A deep probabilistic transfer learning framework for soft sensor modeling with missing data[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(12): 7598-7609.
27	Du Y T, Wang J D, Feng W J, et al. AdaRNN: adaptive learning and forecasting of time series[EB/OL]. 2021. .
28	Schäfer P. Scalable time series classification[J]. Data Mining and Knowledge Discovery, 2016, 30(5): 1273-1298.
29	Luo Y, Wang Y L. A statistical time-frequency model for non-stationary time series analysis[J]. IEEE Transactions on Signal Processing, 2020, 68: 4757-4772.
30	di Bella A, Fortuna L, Graziani S, et al. Soft sensor design for a sulfur recovery unit using genetic algorithms[C]//2007 IEEE International Symposium on Intelligent Signal Processing. New York: IEEE, 2008: 1-6.
31	Sun K, Wu X L, Xue J Y, et al. Development of a new multi-layer perceptron based soft sensor for SO₂ emissions in power plant[J]. Journal of Process Control, 2019, 84: 182-191.

[1]	曹跃, 余冲, 李智, 杨明磊. 工业数据驱动的加氢裂化装置多工况切换过渡状态检测[J]. 化工学报, 2023, 74(9): 3841-3854.
[2]	康飞, 吕伟光, 巨锋, 孙峙. 废锂离子电池放电路径与评价研究[J]. 化工学报, 2023, 74(9): 3903-3911.
[3]	温凯杰, 郭力, 夏诏杰, 陈建华. 一种耦合CFD与深度学习的气固快速模拟方法[J]. 化工学报, 2023, 74(9): 3775-3785.
[4]	尹刚, 李伊惠, 何飞, 曹文琦, 王民, 颜非亚, 向禹, 卢剑, 罗斌, 卢润廷. 基于KPCA和SVM的铝电解槽漏槽事故预警方法[J]. 化工学报, 2023, 74(8): 3419-3428.
[5]	诸程瑛, 王振雷. 基于改进深度强化学习的乙烯裂解炉操作优化[J]. 化工学报, 2023, 74(8): 3429-3437.
[6]	闫琳琦, 王振雷. 基于STA-BiLSTM-LightGBM组合模型的多步预测软测量建模[J]. 化工学报, 2023, 74(8): 3407-3418.
[7]	徐野, 黄文君, 米俊芃, 申川川, 金建祥. 多源信息融合的离心式压缩机喘振诊断方法[J]. 化工学报, 2023, 74(7): 2979-2987.
[8]	高学金, 姚玉卓, 韩华云, 齐咏生. 基于注意力动态卷积自编码器的发酵过程故障监测[J]. 化工学报, 2023, 74(6): 2503-2521.
[9]	邵伟明, 韩文学, 宋伟, 杨勇, 陈灿, 赵东亚. 基于分布式贝叶斯隐马尔可夫回归的动态软测量建模方法[J]. 化工学报, 2023, 74(6): 2495-2502.
[10]	黄磊, 孔令学, 白进, 李怀柱, 郭振兴, 白宗庆, 李平, 李文. 油页岩添加对准东高钠煤灰熔融行为影响的研究[J]. 化工学报, 2023, 74(5): 2123-2135.
[11]	苏晓丹, 朱干宇, 李会泉, 郑光明, 孟子衡, 李防, 杨云瑞, 习本军, 崔玉. 湿法磷酸半水工艺考察与石膏结晶过程研究[J]. 化工学报, 2023, 74(4): 1805-1817.
[12]	贠程, 王倩琳, 陈锋, 张鑫, 窦站, 颜廷俊. 基于社团结构的化工过程风险演化路径深度挖掘[J]. 化工学报, 2023, 74(4): 1639-1650.
[13]	吴心远, 刘奇磊, 曹博渊, 张磊, 都健. Group2vec：基于无监督机器学习的基团向量表示及其物性预测应用[J]. 化工学报, 2023, 74(3): 1187-1194.
[14]	张中秋, 李宏光, 石逸林. 基于人工预测调控策略的复杂化工过程多任务学习方法[J]. 化工学报, 2023, 74(3): 1195-1204.
[15]	张江淮, 赵众. 碳三加氢装置鲁棒最小协方差约束控制及应用[J]. 化工学报, 2023, 74(3): 1216-1227.