Data-driven $H_{\infty}$ predictive control for constrained systems: a Lagrange duality approach

TL;DR

Key closed-loop properties, including stability, disturbance attenuation, and constraint satisfaction, are examined by the proposed data-driven moving horizon predictive control algorithm, enabling more effective handling of external disturbances and time-domain constraints.

Abstract

This article proposes a data-driven $H_{\infty}$ control scheme for time-domain constrained systems based on model predictive control formulation. The scheme combines $H_{\infty}$ control and minimax model predictive control, enabling more effective handling of external disturbances and time-domain constraints. First, by leveraging input-output-disturbance data, the scheme ensures $H_{\infty}$ performance of the closed-loop system. Then, a minimax optimization problem is converted into a more manageable minimization problem employing Lagrange duality, which reduces conservatism typically associated with ellipsoidal evaluations of time-domain constraints. The study examines key closed-loop properties, including stability, disturbance attenuation, and constraint satisfaction, achieved by the proposed data-driven moving horizon predictive control algorithm. The effectiveness and advantages of the proposed method are demonstrated through numerical simulations involving a batch reactor system, confirming its robustness and feasibility under noisy conditions.

Figures (4)

• Figure 1: State trajectories under control input
• Figure 2: The comparison of control output between the unconstrained control and the constrained method
• Figure 3: The comparison of $r_t$ between moving horizion method and static method
• Figure 4: The comparison of $\gamma_t$ between moving horizion method and static method