Piecewise data-driven modelling of the RM3 wave energy converter for multi-condition prediction

Abstract

The modelling of wave energy converters (WECs) is essential for performance evaluation and optimisation. Existing methods often rely on assumptions that simplify system dynamics but may not fully capture the nonlinear effects introduced by wave-structure interactions and power take-off (PTO) mechanisms. High-fidelity numerical simulations and experimental testing provide detailed representations but are computationally demanding or constrained by practical limitations. In this context, data-driven models offer an alternative for describing system behaviour while balancing accuracy and computational feasibility. This study employs a piecewise Nonlinear Autoregressive Moving Average with Exogenous Inputs (NARMAX) approach to represent the dynamics of the Reference Model 3 (RM3) WEC, considering variations across eight sea states. Numerical simulations in WEC-Sim provide the dataset for system identification, using excitation force as input and PTO position and velocity as outputs. The root mean squared error (RMSE) values range from 0.02 to 0.53 for position, from 0.01 to 0.13 for velocity, and 0.058 for PTO force. The results indicate that the segmented models are able to describe the system response within the considered conditions while maintaining computational feasibility.