(382f) Predictive Control with Model Performance Monitoring and Re-Identification

The operation of chemical plants faces numerous challenges such as inherent nonlinearity, complex variable interactions and process constraints. The most common control method that can handle these challenges is model predictive control (MPC) [1, 2]. In several industrial applications of MPC, a linear model is used, in part due to the simplicity of developing linear models and in part due to the computational ease with using linear models. In order to handle the resultant plant-model mismatch, MPC with re-identification is proposed.

There exist some results on MPC with re-identification (IMPC) where model validity is accounted for by requiring excitation constraints to ensure that the model parameters remain identifiable [3]. In this approach, identification is performed at every time step. Furthermore, the approach requires finding the right trade-off between the inevitable performance deterioration (due to excitation conditions) and the possibility of loss of model validity.

But, in these approaches, the original training data is not retained in the new model identification [4], and these methods are designed to address situations where the system is changed significantly and previous data are not at all representative of the plant in question. In situations where plant model mismatch arises due to change in operating condition (with the possibility of reverting back to the nominal plant operation), it becomes useful to merge old and new plant data in the re-identification step.

In this work we address the problem of handling plant-model mismatch by designing a subspace identification based MPC framework that includes model monitoring and closed-loop identification components.

Key Words: Model Predictive Control, Subspace Identification Methods, Re-Identification

[1] Nagy, Z. K., & Braatz, R. D. (2003). Robust nonlinear model predictive control of batch processes. AIChE Journal, 49(7), 1776-1786.

[2] Schäfer, J., & Cinar, A. (2004). Multivariable MPC system performance assessment, monitoring, and diagnosis. Journal of process control, 14(2), 113-129.

[3] Genceli, H., & Nikolaou, M. (1996). New approach to constrained predictive control with simultaneous model identification. AIChE journal, 42(10), 2857-2868.

[4] Kheradmandi, M., & Mhaskar, P. (2018). Model predictive control with closed-loop re-identification. Computers & Chemical Engineering, 109, 249-260.