Online learning robust MPC: an exploration-exploitation approach. Issue 2 (2020)

• Record Type: Journal Article
• Authors: Manzano, J.M.; Calliess, J.; de la Peña, D. Muñoz; Limon, D.
• Is Part Of: IFAC-PapersOnLine. Volume 53:Issue 2(2020)
• Page Start: 5292
• Page End: 5297
• Publication Date: 2020
• Subjects: Learning control -- Nonlinear control -- Target tracking -- Robust stability -- Predictive control -- Sampled-data systems -- Output feedback; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2020.12.1210
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: This paper presents a predictive controller whose model is based on input-output data of the nonlinear system to be controlled. It uses a Lipschitz interpolation technique in which new data may be included in the database in real time, so the controller improves the system model online. An exploration and exploitation policy is proposed, allowing the controller to robustly and cautiously steer the system to the best reachable reference, even if the model lacks data in such region. The conditions needed to ensure recursive feasibility in the presence of output and input constraints and in spite of the uncertainties are given. The results are illustrated in a simulated case study.