Collective prediction of power performance and fatigue loads in wave energy converters using a data-driven approach

Due to the complexity of the ocean environment and wave energy converter (WEC) system, it has been an effort-demanding work to assess either the power performance or fatigue loads of WECs. This work attempts to apply a data-driven approach to increase the efficiency of the collective prediction of the power and fatigue load of a point-absorber type WEC. Nonlinear time-domain modeling is first established to estimate the power and fatigue loads, which is considered the reference data in this work. To demonstrate the performance of the applied data-driven approach, two prevalent power take-off (PTO) mechanisms are implemented to represent different characteristics of WECs. A data-driven approach, active learning Kriging (AK), is adapted to predict power and fatigue loads collectively, and a new learning function is defined to select the enriched wave cases for the active learning process. Results show that the applied active learning approach can accurately and simultaneously predict power and fatigue loads in both PTO mechanisms. Compared to pure numerical simulation, the proposed method only requires 15 simulations of sea state, and the computational effort is reduced by more than 20 times. The maximum prediction error is less than 2%. The data-driven approach could be a powerful tool for WEC system optimization, considering both power performance and fatigue loads.