Robust Dynamic Latent Variable Model for Dynamic Process Monitoring With Missing Data in Cyber-Physical Systems

Although remarkable studies on dynamic process modeling and monitoring in cyber-physical systems have been conducted, there are still several limitations to these approaches, rendering them unsuitable for practical applications. Firstly, the common dynamic models are built based on complete data, which means that their modeling performance usually deteriorates when there are missing measurements. Secondly, practical data often exhibits more complex characteristics, such as non-stationary and non-Gaussian noises. Therefore, a robust dynamic latent variable model is proposed to address the challenges of dynamic process modeling and monitoring in complex measurement environments with missing values. The proposed method effectively handles non-Gaussian noises and recursively estimates missing values, thus enhancing the monitoring performance within the dynamic latent variable modeling framework. Finally, the feasibility and superiority of the proposed method are evaluated through a numerical example and a real wastewater treatment case, demonstrating its effectiveness and practicality in real-world scenarios.