Machine Learning Substitute Modelling of CFD Simulations for Wind Turbine Wakes

Abstract

The growing need for sustainable energy sources has led to a focus on improving the power production capacity of wind turbines. The power optimization of wind turbines is chiefly hinged on the wind velocity which is mitigated by the wake and turbulence effects produced within turbine blades. Herein, computational fluid dynamics (CFD) techniques have been used for resolving wake effects nevertheless the computational method is expensive and tedious. In this context, a machine learning strategy named surrogate modeling was used to predict the reduced velocities inside the wake. These models were trained from a small data set attained from CFD simulations like input air velocities of 6 m/s , 9 m/s and 12 m/s. The machine learning surrogate models provided a data set that aided to find the wind velocity at any arbitrary point and helped in calculating required parameters in real-time without running costly CFD simulations with least error like mean absolute error of 0.000096 by GBR model. Additionally, contribute to power improvement and reliability of wind turbines and wind farm layout optimization. The findings are applicable for optimized performance of the wind turbine wakes at bench and commercial scales.