Please wait a minute...
浙江大学学报(工学版)
Journal of ZheJiang University (Engineering Science)  2021, Vol. 55 Issue (3): 563-570    DOI: 10.3785/j.issn.1008-973X.2021.03.017
    
Transition mode identification method based on maximum mean discrepancy for multimode process
Chao REN(),Gao-wei YAN*(),Lan CHENG,Fang WANG
College of Electrical and Power Engineering, Taiyuan University of Technology, Taiyuan 030024, China
Download: HTML     PDF(1350KB) HTML
Export: BibTeX | EndNote (RIS)      

Abstract  

A transition mode identification method based on maximum mean discrepancy (MMD) was proposed in order to better reveal the law of the operating state and the process data distributional variation of a multi-mode process, and further improve the accuracy of subsequent modeling. Firstly, the sliding window technique was introduced to segment process data, and the MMD was used to measure the distribution difference of the process data. The stable mode and transition mode of the process data were distinguished by comparing the MMD with the stable modal threshold α. Secondly, the width of the sliding window was reduced in transition mode segments, and a transition mode threshold β was used to identify the transition modes. The modal identification results of numerical simulation experiments show that the proposed method can achieve the goal of detecting the step change of the expected value of the input variables and identifying the transient modes. Tennessee Eastman (TE) process simulation data experiments show that the proposed method can effectively divide reasonable modes, select the historical mode modeling with the closest distribution, and improve the soft sensor modeling accuracy of multi-modal processes.

Key words： mode identification      multi-mode process      maximum mean discrepancy      data distribution      sliding window     
Received: 15 October 2019      Published: 25 April 2021
CLC:  TP 274  
Fund:  国家自然科学基金资助项目（61973226，61603267）；山西省科技重大专项资助项目（20181102017）; 山西省重点研发计划项目（201903D121143）
Corresponding Authors: Gao-wei YAN     E-mail: m15988842972@163.com;yangaowei@tyut.edu.cn
Service
E-mail this article
Add to my bookshelf
Add to citation manager
E-mail Alert
RSS
Articles by authors
Chao REN
Gao-wei YAN
Lan CHENG
Fang WANG
Cite this article:

Chao REN,Gao-wei YAN,Lan CHENG,Fang WANG. Transition mode identification method based on maximum mean discrepancy for multimode process. Journal of ZheJiang University (Engineering Science), 2021, 55(3): 563-570.

URL:

http://www.zjujournals.com/eng/10.3785/j.issn.1008-973X.2021.03.017     OR     http://www.zjujournals.com/eng/Y2021/V55/I3/563


基于最大均值差异的多模态过程过渡模态识别方法

为了更好地揭示多模态过程的运行状态和数据分布变化规律，提高后续建模精度，提出基于最大均值差异（MMD）的多模态过程的过渡模态识别方法. 引入滑动窗口对数据进行切割，使用最大均值差异对多模态过程数据的分布差异进行度量，通过与稳定模态阈值α比较区分过程数据的稳定模态和过渡模态. 在过渡模态段内减小滑动窗口窗宽，利用过渡模态阈值β识别出过渡子模态. 数值仿真实验的模态识别结果表明，所提方法可以准确检测出输入变量期望值的阶跃变化，实现对模态的准确识别. 田纳西伊斯曼（TE）过程仿真数据实验表明，所提方法可以有效地划分出合理的模态，进而选择出分布最相近的历史模态建模，提高多模态过程的软测量建模精度.


关键词： 模态识别,  多模态过程,  最大均值差异,  数据分布,  滑动窗口 
Fig.1 Schematic diagram of transition mode identification based on MMD for multimode process
定义模态 ${{{x}}_1}$服从分布 ${{{x}}_2}$服从分布
P $N(5,0.36)$ $N(20,0.49)$
PQ_1 $N(6,0.36)$ $N(20,0.49)$
PQ_2 $N(7,0.36)$ $N(20,0.49)$
PQ_3 $N(8,0.36)$ $N(20,0.49)$
PQ_4 $N(9,0.36)$ $N(20,0.49)$
Q $N(10,0.36)$ $N(20,0.49)$
Tab.1 Mode definition in numerical simulation
Fig.2 Schematic diagram of changes in five output variables in numerical simulation
Fig.3 Mode identification results under different clustering methods
Fig.4 Analysis of mode identification results based on load matrix similarity and maximum mean discrepancy
Fig.5 Schematic diagram of 15 process variables during TE transition proces
工况 方法 RMSE
浓度A 浓度B 浓度C
一 MMD-PLSR 0.221 2 0.118 2 0.251 1
GMM-PLSR 0.227 3 0.132 1 0.256 5
FCM-PLSR 0.246 8 0.198 7 0.267 9
Kmeans-PLSR 0.246 9 0.194 7 0.273 6
LMS-PLSR 0.247 4 0.282 5 0.288 6
二 MMD-PLSR 0.221 2 0.113 1 0.249 7
GMM-PLSR 0.228 8 0.113 8 0.268 4
FCM-PLSR 0.248 2 0.211 9 0.272 5
Kmeans-PLSR 0.247 4 0.205 5 0.274 4
LMS-PLSR 0.265 5 0.293 3 0.285 4
Tab.2 Comparison of RMSE prediction results of multimode soft sensing methods
[1]   杜文莉, 王坤, 钱锋 基于特征空间降维的溶剂脱水分离过程监控[J]. 浙江大学学报: 工学版, 2010, 44 (7): 1255- 1259
DU Wen-li, WANG Kun, QIAN Feng Feature space dimension-reduction based process monitoring of solvent dehydration separation process[J]. Journal of Zhejiang University: Engineering Science, 2010, 44 (7): 1255- 1259
[2]   赵立杰, 汤健, 柴天佑 基于选择性极限学习机集成的磨机负荷软测量[J]. 浙江大学学报: 工学版, 2011, 45 (12): 2088- 2092
ZHAO Li-jie, TANG Jian, CHAI Tian-you Soft sensor of mill load based on selective extreme learning machine ensemble[J]. Journal of Zhejiang University: Engineering Science, 2011, 45 (12): 2088- 2092
[3]   尤海辉, 马增益, 唐义军, 等 循环流化床入炉垃圾热值软测量[J]. 浙江大学学报: 工学版, 2017, 51 (6): 1163- 1171
YOU Hai-hui, MA Zeng-yi, TANG Yi-jun, et al Soft measurement of heating value of burning municipal solid waste for circulating fluidized bed[J]. Journal of Zhejiang University: Engineering Science, 2017, 51 (6): 1163- 1171
[4]   谭帅, 王福利, 彭俊, 等 多模态过程中新模态过程建模方法[J]. 控制与决策, 2012, 27 (8): 1241- 1245
TAN Shuai, WANG Fu-li, PENG Jun, et al Modeling algorithm for new mode in multi-mode process[J]. Control and Decision, 2012, 27 (8): 1241- 1245
[5]   谭帅, 王福利, 常玉清, 等 基于差分分段PCA的多模态过程故障监测[J]. 自动化学报, 2010, 36 (11): 1626- 1636
TAN Shuai, WANG Fu-li, CHANG Yu-qing, et al Fault detection of multi-mode process using segmented PCA based on differential transform[J]. Acta Automatica Sinica, 2010, 36 (11): 1626- 1636
doi: 10.3724/SP.J.1004.2010.01626
[6]   YANG C, ZHOU L, HUANG K, et al Multimode process monitoring based on robust dictionary learning with application to aluminium electrolysis process[J]. Neurocomputing, 2019, 332: 305- 319
doi: 10.1016/j.neucom.2018.12.024
[7]   张雷, 张小刚, 陈华 基于Gath-Geva算法和核极限学习机的多阶段间歇过程软测量[J]. 化工学报, 2018, 69 (6): 2576- 2585
ZHANG Lei, ZHANG Xiao-gang, CHEN Hua Soft sensors for multi-stage batch processes based on Gath-Geva algorithm and kernel extreme learing machine[J]. CIESC Journal, 2018, 69 (6): 2576- 2585
[8]   DU W, FAN Y, ZHANG Y Multimode process monitoring based on data-driven method[J]. Journal of the Franklin Institute, 2017, 354 (6): 2613- 2627
doi: 10.1016/j.jfranklin.2016.11.002
[9]   HE Y, GE Z, SONG Z. Adaptive monitoring for transition process using dynamic mutual information similarity analysis [C]// 2016 Chinese Control and Decision Conference. Yinchuan: IEEE. 2016: 5832-5837.
[10]   ZHANG S, ZHAO C Stationarity test and Bayesian monitoring strategy for fault detection in nonlinear multimode processes[J]. Chemometrics and Intelligent Laboratory Systems, 2017, 168: 45- 61
doi: 10.1016/j.chemolab.2017.07.010
[11]   JIANG Q, HUANG B, YAN X GMM and optimal principal components-based Bayesian method for multimode fault diagnosis[J]. Computers and Chemical Engineering, 2016, 84: 338- 349
doi: 10.1016/j.compchemeng.2015.09.013
[12]   LIU J, CHEN D S Nonstationary fault detection and diagnosis for multimode processes[J]. AIChE Journal, 2010, 56 (1): 207- 219
[13]   MA Y, SONG B, SHI H, et al Neighborhood based global coordination for multimode process monitoring[J]. Chemometrics and Intelligent Laboratory Systems, 2014, 139: 84- 96
doi: 10.1016/j.chemolab.2014.09.019
[14]   QUIN?ONES-GRUEIRO M, PRIETO-MORENO A, LLANES-SANTIAGO O Modeling and monitoring for transitions based on local kernel density estimation and process pattern construction[J]. Industrial and Engineering Chemistry Research, 2016, 55 (3): 692- 702
[15]   ZHENG Y, QIN S J, WANG F PLS-based similarity analysis for mode identification in multimode manufacturing processes[J]. IFAC-PapersOnLine, 2015, 48 (8): 777- 782
doi: 10.1016/j.ifacol.2015.09.063
[16]   张淑美, 王福利, 谭帅, 等 多模态过程的全自动离线模态识别方法[J]. 自动化学报, 2016, 42 (1): 60- 80
ZHANG Shu-mei, WANG Fu-li, TAN Shuai, et al A fully automatic offline mode identification method for multi-mode processes[J]. Acta Automatica Sinica, 2016, 42 (1): 60- 80
[17]   SONG B, TAN S, SHI H Key principal components with recursive local outlier factor for multimode chemical process monitoring[J]. Journal of Process Control, 2016, 47: 136- 149
doi: 10.1016/j.jprocont.2016.09.006
[18]   HE Y, ZHOU L, GE Z, et al Dynamic mutual information similarity based transient process identification and fault detection[J]. The Canadian Journal of Chemical Engineering, 2018, 96 (7): 1541- 1558
doi: 10.1002/cjce.23102
[19]   谭帅. 多模态过程统计建模及在线监测方法研究[D]. 沈阳: 东北大学, 2012: 17-23.
TAN Shuai. Statical modeling and online monitoring for multiple mode processes [D]. Shenyang: Northeastern University, 2012.
[20]   TAN S, WANG F, PENG J, et al Multimode process monitoring based on mode identification[J]. Industrial and Engineering Chemistry Research, 2012, 51 (1): 374- 388
[21]   LIU T, SHI Z, LIU Y Supervised dimensionality reduction on grassmannian for image set recognition[J]. Neural Computation, 2019, 31 (1): 156- 175
doi: 10.1162/neco_a_01148
[22]   GRETTON A, BORGWARDT K, RASCH M, et al. A kernel method for the two-sample-problem [C]// Advances in Neural Information Processing Systems. Vancouver, British Columbia, Canada: Curran Associates, 2007: 513-520.
[23]   迟学斌 特征值问题的算法、复杂性及并行实现[J]. 数值计算与计算机应用, 1998, 19 (1): 28- 34
CHI Xue-bin Algorithm, complexity and parallel implementation of eigen-problems[J]. Numerical Calculation and Computer Applications, 1998, 19 (1): 28- 34
[1] LIU Lei, YANG Peng,LIU Zuo-jun1. Lower limb locomotion-mode identification based on multi-source information and particle swarm optimization algorithm[J]. Journal of ZheJiang University (Engineering Science), 2015, 49(3): 439-447.
[2] CHEN Jia-Gan, HE Yan, JIANG Jing-Ping. Improved FastSLAM algorithm based on importance weight smoothing[J]. Journal of ZheJiang University (Engineering Science), 2010, 44(8): 1454-1459.