A modified observer-based prediction approach for industrial applications

The prediction of key variables has great significance to monitor the running status of industrial systems. In this paper, a novel data-driven design of variable predictor is proposed. The basic idea is the realization of prediction observer, which is modified from the observer-based fault diagnose method. Different from the standard data-driven approaches, the proposed scheme is adopted for the dynamic systems due to the superior tracking ability of output observer. Additionally, by introducing an extra design freedom and the estimation of measured value, it can also be used for the case that the key variable is not on-line measurable. Finally, the proposed prediction scheme is applied to the Tennessee-Eastman plant to demonstrate the effectiveness.