Towards Robust and High-performance Operations of Wave Energy Converters: an Adaptive Tube-based Model Predictive Control Approach

Abstract

Model predictive control (MPC) is an effective control method to improve the energy conversion efficiency of wave energy converters (WECs). However, the current developed WEC MPC has not reached commercial viability since the control performance is significantly dependent on the WEC model fidelity. To overcome the plant-model mismatch issue in the WEC MPC control problem, this paper proposes a robust tube-based MPC method to bound plant states within disturbance invariant sets centered around the noise-free model trajectory. The invariant sets are also utilized for tightening the nominal model's constraints that robustly enable constraint satisfaction. Yet overly conservative invariant sets can narrow the feasible region of the states and control inputs, and hence a data-driven quantile recurrent neural network (QRNN) is proposed in this work to form a learning-based adaptive tube with reduced conservatism by quantifying WEC model uncertainties. The theoretical root is that time-dependent historical data can offer valuable insight into the future behaviour of uncertainties. Numerical simulations have validated that the proposed method can improve the energy capture rate compared to the TMPC approach, by synthesizing the QRNN-based tube with MPC.