Abstract
This paper proposes a novel dynamic latent structure with time-varying parameters for virtual sensing of industrial process with irregular missing data. The proposed latent structure is based on the linear dynamic system (LDS) model. In order to capture the time-varying process characteristics, a Karman filter based parameter updating method is developed and a virtual sensor is constructed to predict hard-to-measure quality variables. The latent variable structure of the improved model enables the virtual sensor to capture the variable cross-correlation and autocorrelation in the missing data by considering both spatial and temporal information, so that the information in the missing data can be learned from those not missing in the samples as well as the Markov process in the temporal stream. Incorporation of both spatial and temporal information renders more flexibility, resulting in a virtual sensor with higher prediction accuracy even with irregular missing data. The better performance of the proposed method is verified by two industrial applications with different ratios of irregular missing data.
Highlights
With the development of distributed control and proliferation networks in modern industries, a large amount of data is collected and stored, making data-driven modeling, monitoring and control an attractive research direction [1]–[3]
In order to overcome these difficulties, this paper develops a virtual sensor for time-varying dynamic system based on the linear dynamic system (LDS) model, which can adapt to data with unknown missing patterns
Unlike traditional data deletion and filling methods, LINEAR DYNAMIC TIME-VARYING PARAMETER STRUCTURE (LDVPS) can construct a probabilistic dynamic model that meets the characteristics of missing data without discarding any valuable data information or introducing additional noise
Summary
With the development of distributed control and proliferation networks in modern industries, a large amount of data is collected and stored, making data-driven modeling, monitoring and control an attractive research direction [1]–[3]. In order to overcome this difficulty, data-driven virtual sensors have been widely used as a popular alternative to estimate hard-to-measure key quality variables from easyto-measure process variables [4]–[6]. Z. Ying et al.: Dynamic Latent Structure With Time-Varying Parameters for Virtual Sensing of Industrial Process required data features from the existing observations in the original data matrix, so as to achieve an accurate prediction for missing serve-quality data. The above mentioned work showed mixed success, as inappropriate imputation brings additional information and may deteriorate the performance of modeling and monitoring These deterministic methods do not perform well when faced with uncertainty and disturbances, which is common in industrial processes. In order to overcome these difficulties, this paper develops a virtual sensor for time-varying dynamic system based on the LDS model, which can adapt to data with unknown missing patterns.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
