Neural network topology for wind turbine analysis

Audrey  Gaymann; Giorgio  Schiaffini; Michela  Massini; Francesco  Montomoli; Alessandro  Corsini

doi:10.29008/ETC2019-174

13^th European Conference on Turbomachinery Fluid dynamics & Thermodynamics

Paper ID:

ETC2019-174

Main Topic:

Wind Turbines

https://doi.org/10.29008/ETC2019-174

Authors

Audrey Gaymann - Imperial College of London
Giorgio Schiaffini - Imperial College of London/Universita' di Roma la Sapienza
Michela Massini - UQuant
Francesco Montomoli - Imperial College of London
Alessandro Corsini - Università di Roma La Sapienza

Abstract

In this work Artificial Neural Networks (ANN) are used for a multi-target optimization of the aerodynamics of a wind turbine blade. The Artificial Neural Network is used to build a meta-model of the blade, which is then optimized according to the imposed criteria. The neural networks are trained with a data set built by a series of CFD simulations and their configuration (number of neurons and layers) selected to improve performances and avoid over-fitting. The basic configuration of the airfoil is the profile S809, which is commonly used in horizontal axis wind turbines (HAWT), equipped with a Coanda jet. The design position and momentum of the jet are optimized to maximize aerodynamic efficiency and minimize the power required to activate the Coanda Jet.