A variable selection aided residual generator design approach for process control and monitoring

Abstract

This paper investigates the design of residual generator based on data-driven techniques, which is optimized by variable selection and can be applied to control and monitoring in industrial process to satisfy the high performance requirements of systems. The basic idea is designing the residual generator based on the foundation of the Luenberger equations as well as the relationship between diagnosis observer (DO) and the parity vector. The crucial and innovative part of the scheme is to acquire the parity space by the Subspace Identification Method (SIM) and obtain the simplified and optimized data from a new variable selection system according to partial least squares (PLS) method during the procedure of SIM. The design and optimization of the residual generator make it independent of model and effective to deal with big data, and avoid the challenges of handling bulk data. After the realization of the Youla parameterization, the proposed approach of residual generator can not only be applied to control purposes, but also monitoring objective of practical industrial systems. A numerical example is used to demonstrate the performance and effectiveness of the proposed scheme.