Efficient data-driven predictive control of nonlinear systems: A review and perspectives

Abstract

Model predictive control (MPC) has become a key tool for optimizing real-time operations in industrial systems and processes, particularly to enhance performance, safety, and resilience. However, the growing complexity and nonlinearity of modern industrial systems present significant challenges for both first-principles modeling and real-time implementation of typical non-convex optimization associated with conventional MPC designs based on nonlinear models. In this review, we aim to provide an overview of current data-driven predictive control methods that have attributes of being computationally efficient as well as having the distinctive potential to address the above two challenges simultaneously. We focus particularly on two promising frameworks: (1) Koopman-based model predictive control, and (2) data-enabled predictive control, both of which are capable of formulating the optimization problem into a convex form even in the presence of strong nonlinearity in the underlying system. Additionally, we provide an outlook on the potential applications of these methods and briefly discuss their future directions across various industrial sectors.