面向多采样率数据的TTPA-LSTM软测量建模

doi:10.11949/0438-1157.20241121

摘要/Abstract

摘要：

实际工业生产中，过程变量间存在的时滞和采样率差异会降低建模质量，使得许多软测量模型无法适用。因此，提出一种基于时间感知模式注意力（time-aware temporal pattern attention，TTPA）机制和长短时记忆网络的软测量建模方法。首先，将高、低采样率对应的数据分别重构为短期和长期信息，采用时间感知模块将输入信息分解并考虑时间间隔特性，针对质量相关信息占比低的问题，设计非递增启发式衰减函数对短期信息进行加权，组合后获得长短期信息集成特征，降低因多采样率产生的数据缺失影响。其次，引入特征优化模块实现特征二维滤波，跨时间步解析多元时间序列中的时滞信息，获取更有效的质量相关特征。最后，搭建了基于TTPA的长短期记忆网络软测量模型。通过工业青霉素发酵过程和脱丁烷塔过程的应用仿真，验证了所提模型的有效性和优越性。

关键词: 多采样率, 时间感知模式注意力, 长短时记忆网络, 软测量, 神经网络, 过程控制, 动态建模

Abstract:

In practical industrial production, the time lags and sampling rate differences among process variables can deteriorate the modeling quality, rendering many soft sensing models inapplicable. Therefore, a soft sensing modeling approach based on time-aware temporal pattern attention (TTPA) mechanism and long short-term memory network is proposed. In this study, we first reconstruct the data corresponding to high and low sampling rates into short-term and long-term information, respectively. A time-aware module is utilized to decompose the input information while considering the characteristics of time intervals. To address the issue of low proportion of quality-related information, a non-increasing heuristic decay function is designed to weight the short-term information. By combining these weighted components, we derive an integrated feature set that encapsulates both short-term and long-term information, thereby mitigating the impact of data loss resulting from multiple sampling rates. Secondly, a feature optimization module is introduced to achieve two-dimensional filtering of features, and the time lag information in the multivariate time series is analyzed across time steps to obtain more effective quality-related features. Finally, a soft sensing model based on TTPA-based long short-term memory network is established. The effectiveness and superiority of the proposed model were verified through the application simulation of IndPensim process and debutanizer process.

Key words: multi-sampling rate, time-aware temporal pattern attention, long short-term memory network, soft-sensor, neural networks, process control, dynamic modeling

中图分类号:

TP 273

王法正, 隋璘, 熊伟丽. 面向多采样率数据的TTPA-LSTM软测量建模[J]. 化工学报, 2025, 76(4): 1635-1646.

Fazheng WANG, Lin SUI, Weili XIONG. TTPA-LSTM soft sensor modeling for multi-sampling rate data[J]. CIESC Journal, 2025, 76(4): 1635-1646.

图/表 18

图1 TTPA-LSTM软测量模型结构

Fig.1 Structure of TTPA-LSTM soft sensor model

图2 工业过程中的多采样率数据说明

Fig.2 Illustrations of the multirate sampled data in industrial processes

图3 多采样率特征处理流程

Fig.3 Multirate sampling feature processing flow

表1 工业青霉素发酵变量描述

Table 1 Variable description of IndPensim

符号	变量名称	采样间隔/min
R_W	加热/冷却水流速	12
V_V	容器体积	12
V_W	容器质量	12
$E C O 2$	废气二氧化碳浓度	24
$E O 2$	废气氧浓度	24
$R C O 2$	碳吸收速率	12
$R O 2$	氧吸收速率	12
P	青霉素浓度	60

表1 工业青霉素发酵变量描述

Table 1 Variable description of IndPensim

符号	变量名称	采样间隔/min
R_W	加热/冷却水流速	12
V_V	容器体积	12
V_W	容器质量	12
$E C O 2$	废气二氧化碳浓度	24
$E O 2$	废气氧浓度	24
$R C O 2$	碳吸收速率	12
$R O 2$	氧吸收速率	12
P	青霉素浓度	60

表2 各模型青霉素浓度预测评价指标

Table 2 The prediction evaluation metrics for penicillin concentration of various models

所用模型	RMSE	MAE	MAPE	R²
MLP	5.55531	4.48582	4.03364	0.69172
RNN	4.54358	3.77206	5.97488	0.79377
NI^[16]	3.52126	2.63703	1.24222	0.87614
MSDF^[17]	3.32183	2.54239	1.53001	0.88978
SFA-MLP^[19]	2.81958	2.05189	1.21357	0.92058
SA-TLSTM	2.57029	1.87089	0.84318	0.93401
TTPA-LSTM	1.22438	0.92089	0.34144	0.98503

图4 各模型青霉素浓度预测结果对比

Fig.4 Comparison of penicillin concentration results among various models

图5 各模型青霉素浓度预测误差

Fig.5 Penicillin concentration prediction error among various models

表3 消融实验青霉素浓度预测评价指标

Table 3 The prediction evaluation metrics for penicillin concentration of ablation experiments

模型	RMSE	MAE	MAPE	R²
LSTM	3.80403	3.07299	1.71857	0.85545
TLSTM	2.86516	2.46471	2.02471	0.91799
TPA-LSTM	2.74571	2.02671	0.68686	0.92469
TTPA-LSTM	1.22438	0.92089	0.34144	0.98503

图6 消融实验青霉素浓度预测结果对比

Fig.6 Comparison of penicillin concentration results of ablation experiments

表4 K折交叉验证青霉素浓度预测评价指标

Table 4 The prediction evaluation metrics for penicillin concentration of K-fold cross-validation

子集序号	RMSE	MAE	MAPE	R²
1	1.38802	1.07046	0.82576	0.98066
2	1.26901	0.99769	0.85675	0.98213
3	1.36301	1.04380	0.43557	0.98136
4	1.43175	1.07177	0.55785	0.97943
5	1.22438	0.92089	0.34144	0.98503

图7 脱丁烷塔工艺流程

Fig.7 Process flow of debutanizer

表5 脱丁烷塔变量描述

Table 5 Variable description of debutanizer

符号	变量名称	采样间隔/min
x₁	塔顶端温度	10
x₂	塔顶端压力	10
x₃	塔顶端回流量	20
x₄	塔顶端出料量	20
x₅	塔板6温度	10
x₆	塔底温度1	10
x₇	塔底温度2	10
y	塔底丁烷含量	40

表6 多采样率模型在脱丁烷塔数据集的预测评价指标

Table 6 Prediction and evaluation indicators of multirate sampling models in the debutanizer dataset

模型	RMSE	MAE	MAPE	R²
NI^[16]	0.04298	0.03221	0.12219	0.93912
MSDF^[17]	0.03623	0.02836	0.11641	0.95685
SFA-MLP^[19]	0.03494	0.02507	0.09652	0.95978
TTPA-LSTM	0.02424	0.01822	0.08607	0.98089

图8 多采样率模型塔底丁烷含量预测误差直方图

Fig.8 Histogram of prediction error for butane concentration at the bottom of a multirate sampling model tower

图9 各多采样率模型对丁烷含量预测误差箱线图

Fig.9 Box plots of prediction errors for butane concentration using various multirate sampling models

图10 多采样率模型塔底丁烷含量预测结果对比

Fig.10 Comparison of numerical simulation results for multirate models

表7 脱丁烷塔数据集上的消融实验预测评价指标

Table 7 Predictive evaluation indicators for ablation experiments on the debutanizer dataset

模型	RMSE	MAE	MAPE	R²
LSTM	0.04665	0.03765	0.14688	0.92874
TLSTM	0.03296	0.02649	0.10634	0.96469
TPA-LSTM	0.03698	0.02693	0.09838	0.95512
TTPA-LSTM	0.02424	0.01822	0.08607	0.98089

图11 各参数下TTPA-LSTM模型丁烷含量预测RMSE

Fig.11 RMSE prediction of butane concentration using TTPA-LSTM model under various parameters

参考文献 33

1	熊伟丽, 孙文心, 马君霞. 带自相关约束的NARX动态软测量模型[J]. 控制与决策, 2020, 35(4): 816-822.
	Xiong W L, Sun W X, Ma J X. Autocorrelation constrained NARX dynamic soft sensing model[J]. Control and Decision, 2020, 35(4): 816-822.
2	Jiang Y C, Yin S, Dong J W, et al. A review on soft sensors for monitoring, control, and optimization of industrial processes[J]. IEEE Sensors Journal, 2021, 21(11): 12868-12881.
3	黎宏陶, 王振雷, 王昕. 基于即时学习的改进条件高斯回归软测量[J]. 化工学报, 2024, 75(6): 2299-2312.
	Li H T, Wang Z L, Wang X. Improved conditional Gaussian regression soft sensor based on just-in-time learning[J]. CIESC Journal, 2024, 75(6): 2299-2312.
4	李祥宇, 隋璘, 马君霞, 等. 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.
	Li X Y, Sui L, Ma J X, et al. ONLSTM soft sensor modeling based on time series transfer and dual stream weighting[J]. CIESC Journal, 2023, 74(11): 4622-4633.
5	Narayanan H, Luna M F, von Stosch M, et al. Bioprocessing in the digital age: the role of process models[J]. Biotechnology Journal, 2020, 15(1): e1900172.
6	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.
7	Yin S L, Li Y G, Sun B, et al. Mixed kernel principal component weighted regression based on just-in-time learning for soft sensor modeling[J]. Measurement Science and Technology, 2022, 33(1): 015102.
8	程康明, 熊伟丽. 一种自训练框架下的三优选半监督回归算法[J]. 智能系统学报, 2020, 15(3): 568-577.
	Cheng K M, Xiong W L. Three-optimal semi-supervised regression algorithm under self-training framework[J]. CAAI Transactions on Intelligent Systems, 2020, 15(3): 568-577.
9	Li X H, Ma J W. Domain adaptation based on semi-supervised cross-domain mean discriminative analysis and kernel transfer extreme learning machine[J]. Sensors, 2023, 23(13): 6102.
10	Ye Z, Cheng X, Ding X, et al. A new unconditionally stable FDTD method based on artificial neural network[J]. IEEE Transactions on Microwave Theory and Techniques, 2024, 72(10): 5759-5767.
11	Wang J Y, Yao L, Xiong W L. Novel semi-supervised deep probabilistic slow feature extraction for online chemical process soft sensing application[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 2517711.
12	Chai Z, Zhao C H, Huang B. Variational progressive-transfer network for soft sensing of multirate industrial processes[J]. IEEE Transactions on Cybernetics, 2022, 52(12): 12882-12892.
13	Liu D J, Wang Y L, Liu C L, et al. Multirate-former: an efficient transformer-based hierarchical network for multistep prediction of multirate industrial processes[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 73: 2502313.
14	Lyu Y T, Zhou L, Cong Y, et al. Multirate mixture probability principal component analysis for process monitoring in multimode processes[J]. IEEE Transactions on Automation Science and Engineering, 2024, 21(2): 2027-2038.
15	Lin B, Recke B, Knudsen J K H, et al. Data-driven soft sensor design with multiple-rate sampled data: a comparative study[C]//2007 European Control Conference (ECC). Kos, Greece: IEEE, 2007: 4740-4745.
16	Grujić D N. Numerical Hilbert transform algorithm for causal interpolation of functions represented by cubic and exponential splines[J]. IEEE Access, 2021, 9: 136702-136709.
17	Zhang S M, Zhao C H. Slow-feature-analysis-based batch process monitoring with comprehensive interpretation of operation condition deviation and dynamic anomaly[J]. IEEE Transactions on Industrial Electronics, 2019, 66(5): 3773-3783.
18	He K J, Yang Q, Ji L, et al. Financial time series forecasting with the deep learning ensemble model[J]. Mathematics, 2023, 11(4): 1054.
19	Zheng J Q, Ma L W, Wu Y, et al. Nonlinear dynamic soft sensor development with a supervised hybrid CNN-LSTM network for industrial processes[J]. ACS Omega, 2022, 7(19): 16653-16664.
20	Baytas I M, Xiao C, Zhang X, et al. Patient subtyping via time-aware LSTM networks[C]//Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Halifax, NS, Canada: ACM, 2017: 65-74.
21	曹鲁慧, 秦丰林, 闫中敏. 基于TLSTM的医疗保险欺诈检测[J]. 计算机工程与应用, 2020, 56(21): 237-241.
	Cao L H, Qin F L, Yan Z M. TLSTM-based medical insurance fraud detection[J]. Computer Engineering and Applications, 2020, 56(21): 237-241.
22	李妍君, 熊伟丽, 徐保国. 一种带过程变量时滞估计的在线软测量建模方法[J]. 信息与控制, 2016, 45(6): 641-646.
	Li Y J, Xiong W L, Xu B G. An online soft-sensor modeling method including process variable time-delay estimation[J]. Information and Control, 2016, 45(6): 641-646.
23	陈为胜, 李俊民. 非线性时滞大系统自适应神经网络分散控制[J]. 控制与决策, 2006, 21(8): 873-878.
	Chen W S, Li J M. Adaptive neural network decentralized control for nonlinear time-delay large-scale systems[J]. Control and Decision, 2006, 21(8): 873-878.
24	张晓迪, 蔡云泽, 何星, 等. 基于神经网络的不稳定时滞对象控制[J]. 上海交通大学学报, 2014, 48(7): 1033-1038.
	Zhang X D, Cai Y Z, He X, et al. Neural network control for unstable processes with time delay[J]. Journal of Shanghai Jiao Tong University, 2014, 48(7): 1033-1038.
25	孙凯, 隋璘, 张芳芳, 等. 基于非负绞杀与长短期记忆神经网络的动态软测量算法[J]. 控制理论与应用, 2023, 40(1): 83-93.
	Sun K, Sui L, Zhang F F, et al. Dynamic soft sensor algorithm based on nonnegative garrote and long short-term memory neural network[J]. Control Theory & Applications, 2023, 40(1): 83-93.
26	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.
	Yan L Q, Wang Z L. Multi-step predictive soft sensor modeling based on STA-BiLSTM-LightGBM combined model[J]. CIESC Journal, 2023, 74(8): 3407-3418.
27	Xia J, Feng Y W, Lu C, et al. LSTM-based multi-layer self-attention method for remaining useful life estimation of mechanical systems[J]. Engineering Failure Analysis, 2021, 125: 105385.
28	Shih S Y, Sun F K, Lee H Y. Temporal pattern attention for multivariate time series forecasting[J]. Machine Learning, 2019, 108(8): 1421-1441.
29	Duan R, Liu J, Zhou J Z, et al. Study on performance evaluation and prediction of francis turbine units considering low-quality data and variable operating conditions[J]. Applied Sciences, 2022, 12(10): 4866.
30	Ren L, Wang T, Yuanjun L L, et al. A data-driven self-supervised LSTM-DeepFM model for industrial soft sensor[J]. IEEE Transactions on Industrial Informatics, 2022, 18(9): 5859-5869.
31	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.
32	Jia M W, Xu D Y, Yang T, et al. Graph convolutional network soft sensor for process quality prediction[J]. Journal of Process Control, 2023, 123: 12-25.
33	Sun W X, Xiong W L, Chen H T, et al. A novel CVAE-based sequential Monte Carlo framework for dynamic soft sensor applications[J]. IEEE Transactions on Industrial Informatics, 2024, 20(3): 3789-3800.

[1]	徐芳萍, 杨辉, 陈俊, 朱建勇, 陆荣秀. 基于迁移学习与残差注意力卷积网络的稀土元素组分含量软测量[J]. 化工学报, 2025, 76(4): 1647-1660.
[2]	张晗筱, 王瑞琪, 张亚婷. 基于CNN-LSTM的换热器污垢因子预测研究[J]. 化工学报, 2025, 76(4): 1671-1679.
[3]	李征, 庄铠泽, 赵东杰, 宋燕星, 王功明. 事件驱动的深度信念网络软测量模型设计方法[J]. 化工学报, 2025, 76(4): 1693-1701.
[4]	张新梅, 张傲, 邱德华, 刘晓爽, 陈晨. 基于热响应机理的罐区池火灾动态多米诺效应评估方法[J]. 化工学报, 2025, 76(4): 1885-1897.
[5]	周洋, 梁聃, 汪恺, 张艺蓝, 贾立. 基于Alopex进化算法和集成学习的多模态神经网络的污水反渗透过程建模[J]. 化工学报, 2025, 76(4): 1661-1670.
[6]	郑欣雨, 任泽华, 周利, 柴士阳, 吉旭. 基于晶体图卷积神经网络的晶格能回归模型[J]. 化工学报, 2025, 76(3): 1084-1092.
[7]	王大芬, 唐莉丽, 张鑫焱, 聂春雨, 李明珠, 吴菁. 基于时差的多输出tri-training异构软测量建模[J]. 化工学报, 2025, 76(3): 1143-1155.
[8]	王绍吉, 郝矿荣, 陈磊. 基于联邦学习的聚酯纤维酯化过程温度预测研究[J]. 化工学报, 2025, 76(1): 283-295.
[9]	杨晔, 卢建刚. 基于融合Transformer的门尼系数预测建模研究[J]. 化工学报, 2025, 76(1): 266-282.
[10]	冯海军, 章冰璇, 周健. 图神经网络模型预测和解释离子液体毒性的研究[J]. 化工学报, 2025, 76(1): 93-106.
[11]	赵武灵, 满奕. 基于变分编码器的纳米纤维素分子结构预测模型框架研究[J]. 化工学报, 2024, 75(9): 3221-3230.
[12]	郭鑫, 李文静, 乔俊飞. 基于自组织模块化神经网络的污水处理过程出水参数预测[J]. 化工学报, 2024, 75(9): 3242-3254.
[13]	王倩倩, 李冰, 郑伟波, 崔国民, 赵兵涛, 明平文. 氢燃料电池局部动态特征三维模型[J]. 化工学报, 2024, 75(8): 2812-2820.
[14]	李倩, 张蓉民, 林子杰, 战琪, 蔡伟华. 基于机器学习的印刷电路板式换热器流动换热预测与仿真[J]. 化工学报, 2024, 75(8): 2852-2864.
[15]	马君霞, 李林涛, 熊伟丽. 基于Tri-training GPR的半监督软测量建模方法[J]. 化工学报, 2024, 75(7): 2613-2623.