Dynamic Processes Modeling and Monitoring Based on a Novel Dynamic Latent Variable Model.

Le Zhou,Zhiqiang Ge,Zhihuan Song,Si-Zhao Joe Qin

DOI: https://doi.org/10.1109/safeprocess45799.2019.9213443

2019-01-01

Abstract:For dynamic process modeling and monitoring purpose, it is desirable to extract both the auto-correlations and the cross-correlations in measurements. Besides, the proposed dynamic model is also expected to provide an accurate prediction, visualization and an explicit interpretation of the data structures. From this perspective, the dynamic latent variable model is more suitable compared to the traditional dynamic principal component analysis (DPCA). In this paper, a novel independent dynamic latent variable model is proposed to explicitly extract several independent dynamic latent variables with which to capture process dynamics in the measurements. The proposed model is derived in the probabilistic framework and the model parameters are estimated via the expectation-maximum algorithm. Finally, a case study is illustrated to evaluate the performance of the proposed method for dynamic modeling and process monitoring.