Robust Learning-based Model Predictive Control for Wave Energy Converters

This paper proposes a robust learning-based model predictive control (MPC) strategy tailored for sea wave energy converters (WECs). The control algorithm aims to maximize power extraction efficiency and maintain the WECs' operational safety over a wide range of sea conditions, subject to system constraints and plant-model mismatches. The theoretical basis is the robust tube-based MPC (RTMPC), enabling WEC system state trajectories to evolve around the noise-free nominal WEC model state trajectories. The disturbances can be bounded by pre-computed uncertainty sets for tightening the WEC's physical constraints to guarantee the constraint satisfaction of an uncertain WEC system. Typically, RTMPC constructs a tube with constant sets of uncertainties, which is likely to be overly conservative and hence potentially degrades energy conversion performance. In this work, a machine learning-based uncertainty set is introduced to dynamically predict and quantify the model uncertainties at each sampling instant, which can effectively enlarge the feasible region of the WEC TMPC control problem. The proposed RTMPC not only ensures improved energy conversion efficiency but also guarantees the operational safety of WECs under uncertain conditions. Numerical simulations demonstrate the efficacy of the proposed controller.