Robust control of large-scale nonlinear constrained systems
Current applications of nonlinear model predictive control algorithms are restricted to small-scale processes, due mainly to the computational difficulties encountered when trying to solve the non-convex nonlinear optimization problem on-line. Also, there is no complete procedure in synthesizing a nonlinear model predictive controller that guarantees stability when there is model uncertainty and when the state is not completely measured. Although there are plenty of results available for the nominal closed-loop stability of various NMPC algorithms, few results are available on the robust stability of constrained nonlinear systems. We have proposed a cascade NMPC algorithm for large-scale systems. This control algorithm consists of two levels where the low-level controller guarantees robust stability while the high-level controller optimizes nominal performance subject to robust stability constraint. The low-level controller is an output feedback controller while the high-level controller is a state feedback controller. There are several characteristics to be noticed about this algorithm: it is computationally efficient and is always feasible to be implemented on-line; it uses a closed-loop control strategy; it guarantees robust asymptotic stability with various kinds of model uncertainties, if such a controller exists.
For model uncertainty description, we consider both parametric and structural uncertainty. We develop an uncertainty description that can handle both parametric and structural uncertainties in a uniform manner. We have applied the proposed algorithm to an industrial system consisting of a co-polymerization reactor with recycle. It is a reasonably complex, highly nonlinear and poorly modeled (i.e., uncertain) process. Simulation results show the feasibility of implementing this algorithm on realistic industrial systems where model uncertainties always exist.