Abstract—Traditional Multiway Partial Least Squares (MPLS) methods focusing on the monitoring and quality prediction of batch processes have the model mismatch problem when encountered with different conditions such as variations in operating conditions, weather, environment, and raw materials, however local Just In Time Learning (JITL) MPLS methods forced on the aforementioned problems have the problem that the online selection of historical modelling data suffers unneglectable computing loading costs, which is heavier if dealing with large amount of historical data or using insufficient hardware capability. Focused on addressing these problems, this paper proposes a novel local modeling MPLS method. First, a pre-clustering procedure is applied during offline stage on historical data, giving the corresponding fuzzy c-means memberships and cluster centers. Second, during online stage the collected local online sample’s fuzzy membership is calculated. Then rough historical samples are selected under the threshold derived from offline fuzzy memberships and then detailed selected samples are selected under the guidance of JITL. Finally, a local MPLS model will be built for online monitoring and soft sensing. The proposed Pre-Clustered Fuzzy-C-JITL MPLS algorithm reduces and transfers main part of the computational pressure from online historical sample selection stage to offline stage, improving the efficiency of online monitoring and soft sensing, while keeping the level of prediction accuracy in the same extent. This improvement can be obvious when under the circumstance like high timeliness requirement, low processing capability, huge and complex historical data, or long cycle length. The proposed approach is demonstrated through applications on the penicillin benchmark and an E. coli fermentation process and has its effectiveness examined.
Index Terms—Batch processes, machine learning, multiway partial least squares, soft sensing.
X. Wang, P. Wang, X. Gao, P. Chang, and Z. Li are with the Faculty of Information Technology, Engineering Research Center of Digital Community (Ministry of Education), Beijing Laboratory of Urban Rail Transit, and Beijing Laboratory of Computational Intelligence System, Beijing University of Technology, Beijing, 100124 China (e-mail: wind-k@hotmail.com).
J. Zhang is with the School of Chemical Engineering and Advance Materials, Newcastle University, Newcastle upon Tyne, NE1 7RU UK.
[PDF]
Cite: Xichang Wang, Pu Wang, Jie Zhang, Xuejin Gao, Peng Chang, and Zheng Li, "Accelerated Online Batch Process Local Monitoring and Soft Sensing Based on Pre-Clustered Fuzzy-C-JITL Multiway Partial Least Squares," International Journal of Modeling and Optimization vol. 7, no. 4, pp. 194-201, 2017.