Model Predictive Control of WECs: A Comparative Study of Single-Step and Multi-Step Dynamic Prediction Models

Ocean waves represent a vast and promising renewable energy resource. Wave Energy Converters aim to harness this potential by converting wave-induced mechanical motion into electrical power. Despite their promise, WECs face challenges related to efficiency, reliability, and high capital and operational costs. Among the most promising solutions for improving energy extraction is Model Predictive Control, a control strategy that explicitly handles system constraints and optimises a performance objective over a finite time horizon. This thesis investigates the design of MPC strategies tailored for WECs, with a focus on understanding and improving the models used for dynamic prediction within the control loop. We begin by analysing the effect of different energy-maximising objective function formulations commonly used in the literature. Although all aim to approximate the integral of absorbed power, their numerical properties may vary significantly and influence both the quality of the resulting control action. The core contribution of this work addresses a key challenge in WEC-MPC design: the need for accurate and efficient dynamic models that predict the system’s velocity output over the entire prediction horizon. Standard Model Predictive Control implementations rely on a single-step model, iteratively applied to compute the system trajectory. However, this approach may suffer from compounding model inaccuracies and limited flexibility. To overcome these limitations, we propose a multi-step-ahead model, trained via system identification on simulated input-output data. This model directly predicts the velocity profile over the MPC horizon as a function of the initial state, control and wave excitation inputs. It enables the construction of a quadratic cost function, simplifying the optimisation process. The model is validated under various conditions, including measurement noise, nonlinear effects, and inaccurate wave forecasts. Finally, a full MPC loop is implemented to compare the proposed multi-step model against the classical single-step approach in a realistic simulation environment.