Transfer Learning of Model-Based Neural Network for Transfer Function Inversion and Load Monitoring of Wind Turbines

Frédéric Delbos (1) , Fabien Caleyron (1) , Jean-Marc Leroy (1)

1 IFPEN - IFP Energies nouvelles

One main concern when using a generic model of a wind turbine is how can we apply it on real measurements for different turbines of the same type. This paper gives a first answer by proposing an inversion approach of a transfer function. Then that approach is tested on a real test case. Two transfer functions from two different turbines of the same farm are inverted on SCADA measurements with transfer learning of a model-base neural network. The inversion results show better predictions than ones with a transfer function classically trained with LIDAR data. An application to the prediction of damage equivalent load with a generic model is showed: the results of applying the inverted transfer function is compared to real DEL measurements and to the transfer function trained from LIDAR data.

Domaines: Physique des Hautes Energies - Théorie [hep-th]; Mathématiques [math]

Frédéric Delbos, Fabien Caleyron, Jean-Marc Leroy. Transfer Learning of Model-Based Neural Network for Transfer Function Inversion and Load Monitoring of Wind Turbines. Journal of Physics: Conference Series, 2022, 2265, pp.032081. ⟨10.1088/1742-6596/2265/3/032081⟩. ⟨hal-03910066⟩