Modeling of a hinged-raft wave energy converter via deep operator learning and wave tank experiments

Model identification for a hinged-raft wave energy converter (WEC) is investigated in this paper, based on wave tank experiments and deep operator learning. Different from previous works which all formulated this issue as a function approximation task, this work, for the first time, formulates it as an operator approximation task (which learns the mapping from a function space to another function space). As such, a continuous-time WEC model is identified from data, greatly expanding the horizon of data-based WEC modeling because previous works were limited to discrete-time model identification. The error accumulation for multi-step predictions in the discrete-time formulation is thus also addressed. The model is developed by first carrying out a set of wave tank experiments to generate the training data, and then the deep operator learning model, i.e. the DeepONet, is constructed and trained based on the experimental data. The validation study shows that the model captures the WEC dynamics accurately. A new set of experimental runs are further carried out and the results show that after training, the model can be used as a digital wave tank, an alternative to the expensive numerical and physical wave tanks, for accurate and real-time simulations of the WEC dynamics.