Unsteady data-driven aerodynamics for floating offshore wind turbines

With the growing availability of high quality data, data-driven modelling is an increasingly attractive
option for wind turbine and wind farm aerodynamics, in particular for floating offshore wind turbines
(FOWTs).
Physical models of FOWTs are computationally expensive, and difficult to integrate into control or
optimisation algorithms. Data-driven models are in principle faster and readily integratable in control
and optimisation. The inherent unsteadiness and nonlinearity of the underlying physics combined
with the high-dimensionality of turbine wakes, put data-driven modelling of FOWTs at the edge of
what is possible today.
A system identification-based framework is proposed for aerodynamics of FOWTs, leveraging recent
developments in system identification as well as work on data-driven unsteady aerodynamics of
simpler, but still nonlinear, systems. The models are trained on mid-fidelity physical models that
allow good coverage of the parameter space. The same framework can be used to train models on
high-fidelity data. This offers a flexible model structure where the fidelity of the predicted outputs
and training data grow together.
Finally, the data-driven models will be tailored towards application in optimisation and control. The
resulting models would have the fidelity of the physical models used to generate the training data, at
lower cost. This makes the proposed data-driven models promising to ultimately improve
performance and reduce structural fatigue.