Robust Data-Driven Predictive Control for Unknown Linear Systems With Bounded Disturbances

• 成果类型: Article
• 署名作者: Hu, Kaijian; Liu, Tao
• 署名单位: University of Hong Kong; University of Hong Kong; The University of Hong Kong Shenzhen Institute of Research & Innovation
• 刊物名称: IEEE TRANSACTIONS ON AUTOMATIC CONTROL
• ISSN/ISSBN: 0018-9286
• DOI: 10.1109/TAC.2025.3560697
• 发表日期: 2025
• 页码: 6529-6544
• 关键词: NOISE; Noise measurement; Linear systems; Predictive control; stability analysis; Upper bound; trajectory; Symmetric matrices; linear programming; ISO; Bounded disturbances and noise; linear time-invariant (LTI) systems; quadratic matrix inequalities (QMIs); robust data-driven predictive control (RDPC)

摘要

This article presents a robust data-driven predictive control (RDPC) framework for linear time-invariant (LTI) systems affected by bounded disturbances and measurement noise. Unlike traditional model-based approaches, the proposed method relies solely on input-state-output (ISO) data without requiring prior system identification. Given that multiple systems can be consistent with the collected data due to disturbances and noise, a set of all possible systems using quadratic matrix inequalities is constructed. The RDPC scheme is then formulated as an optimization problem that minimizes an upper bound on the control objective while ensuring robust constraint satisfaction for all systems in the set. Unlike the existing robust data-driven model predictive control methods based on behavioral system theory, the proposed method does not require the precollected data to satisfy the persistently exciting condition of a sufficiently high order. It only needs the stabilizability of the LTI system to be controlled. The proposed approach is further extended for systems where the state is unmeasurable. By reformulating the problem using an autoregressive exogenous model, a dynamic output feedback controller that leverages input-output data directly is designed. The effectiveness of the proposed methods is validated through a case study on an unstable batch reactor, demonstrating comparable performance to model-based robust MPC and data-driven MPC approaches while reducing conservatism and computational complexity.