Efficient techniques for fast uncertainty propagation in an offshore wind turbine multi-physics simulation tool

E Fekhari (1) , B Iooss (1) , V Chabridon (1) , J Muré (1)

1 EDF R&D PRISME - Performance, Risque Industriel, Surveillance pour la Maintenance et l’Exploitation

Offshore wind turbines (OWT) are subject to uncertain environmental conditions (wind and tidal), making long-term investment decisions riskier. To study the impact of the environmental uncertainties on output variables of interest, we propagate them through costly multi-physics numerical models, simulating the OWT. Two main frameworks are possible to retrieve the random variables of interest. First, an efficient sampling method can improve the Monte Carlo reference convergence rate. Second, a regression model can be fitted over a few samples before using it as inexpensive approximation of the numerical model. Some advanced methods offer a sequential improvement of this strategy by iteratively adding samples enhancing the regression model. In this work, our aim is to perform a numerical comparison between various propagation methods to estimate the expected value of the mechanical loads of an OWT over environmental random variables. Additionally, theoretical equivalences between Bayesian quadrature and Kernel herding using maximum mean discrepancy shall be verified on an industrial use-case.

Domaines: Statistiques [math.ST]

E Fekhari, B Iooss, V Chabridon, J Muré. Efficient techniques for fast uncertainty propagation in an offshore wind turbine multi-physics simulation tool. 5th International Conference on Renewable Energies Offshore, Nov 2022, Lisbon, Portugal. ⟨hal-03757979⟩