(325b) On Monitoring of Economic Model Predictive Control Systems

Economic model predictive control (EMPC), which replaces the quadratic cost of a conventional model predictive control (MPC) system with an economics-based, typically non-quadratic, cost function, has recently been proposed to dynamically optimize process economic performance. Specifically, EMPC schemes typically operate process systems in a time-varying fashion and have been demonstrated to yield improved closed-loop economic performance over steady-state operation [1]-[3]. An open fundamental challenge to time-varying operation is introducing methods that can assess and monitor the performance of EMPC schemes [4]. Existing results on monitoring of MPC deal with operation at steady-state and are based on the use of historical fault-free operation data to construct state-space regions of acceptable operation around the desired operating steady-state. These regions account for common cause variance present in the process and the control system due to model inaccuracy, sensor noise and actuator imperfect operation. However, the regions of acceptable process operation are computed in the context of steady-state operation and may not be suitable to be used to infer acceptable control system performance when the control system operates the process in an inherently time-varying fashion to optimize the economic performance as EMPC does.

Motivated by this issue, we will present a framework for the computation of acceptable operating regions for EMPC systems that operate the process in a time-varying fashion within a well-defined region of the state-space. To this end, it is critical to capture the interplay between sources of common cause variance in the process and dynamic process operation, for fault-free operation. Therefore, given a well-defined region in the state-space of the process where time-varying operation is allowed to take place to optimize economics, meet state/input constraints and stabilizability requirements, open-loop as well as closed-loop runs are carried out to collect process dynamic operation data which are normalized with respect to expected (noise-free) process operation profiles and suitable, dynamic process operation regions are computed for time-varying operation. These regions are utilized to monitor the performance of EMPC by comparing real-time process operation data under EMPC and the corresponding regions of acceptable EMPC operation computed in the fault-free, dynamic data generation step. The proposed framework is evaluated using several chemical process examples.

References:

1. Heidarinejad M, Liu J, Christofides PD. Economic model predictive control of nonlinear process systems using Lyapunov techniques. AIChE Journal. 2012;58:855-870.
2. Ellis M, Christofides PD. Optimal time-varying operation of nonlinear process systems with economic model predictive control. Industrial & Engineering Chemistry Research, in press.
3. Huang R, Harinath E, Biegler LT. Lyapunov stability of economically oriented NMPC for cyclic processes. Journal of Process Control. 2011;21:501-509.
4. Daoutidis P, Bartusiak D. Prespective on CPC VIII. Computers & Engineering. 2013;51:1-3.