Nonlinear model predictive control under uncertainty: Enhancing efficiency, stability and robustness

Abstract

Model predictive control (MPC) is an advanced control technique that is used in process industries and has recently gained attention for application in various fields. It is an optimization-based control that uses predictions from a model over a future control horizon to determine control inputs. MPC optimizes a cost function subjected to process constraints, that makes it suitable for constrained multiple-input multiple-output processes. Including nonlinear models and the explicit consideration of model uncertainty significantly increases the MPC problem complexity. As a result, nonlinear MPC under uncertainty requires a large computational effort and a high degree of conservativeness in the control performance.

This thesis proposes new methods to efficiently solve nonlinear MPC under uncertainty with a reduced degree of conservativeness. Approximation methods are applied to achieve computational efficiency, and to reduce conservativeness, a combination of approximation and data analysis techniques is proposed. Key results of this thesis include implementing an adaptive control horizon for the reduction of computational costs in nonlinear MPC under uncertainty. Moreover, an implementation of a data-based framework to improve uncertain model scenario selection is achieved. The limitations of the proposed methodologies are clearly highlighted through theoretical analysis, and their performances are demonstrated by simulation experiments on several benchmark numerical examples and on an application-based case study.