Recursive Autoregressive Dynamic Latent Variable Model for Fault Detection of Dynamic Process with Missing Values

摘要

For the dynamic processes, both the auto-correlations and the cross-correlations need to be extracted. In the previous work, the autoregressive dynamic latent variable (ARDLV) model is able to achieve this goal since an AR process is used for high-order dynamic process modelling. However, the training data set usually contains the missing values, which leads to the normal ARDLV invalid. In this paper, a novel recursive ARDLV model is proposed for fault detection of the dynamic process with missing values. In the proposed model, the missing value and the model parameters are estimated alternatively in the probabilistic framework. Finally, a case study is illustrated to reveal the performance of proposed method, in which an incomplete data set is used for fault detection purpose.