A research project at the University of Warwick in the UK is aiming to improve wind turbine efficiency using machine learning and computation fluid dynamics.
The Machine Learning in Wind Farm Modelling based on Computational Fluid Dynamics project arose through an industry study group on uncertainty quantification and management in 2017. Modelling wind farms is essential both during the planning and operation stages in order to maximise power output and predict maintenance levels, particularly for large arrays of turbines. However, current models of sufficient quality are computationally costly. This project is looking at ways to reduce the cost using machine learning (ML) techniques.
The research involves running high-fidelity computational fluid dynamics (CFD) simulations of wind farms at various scales. This data is used to build a more efficient ‘surrogate’ model based on an artificial neural network (ANN).
“We began with digging deep to figure out turbulence between two turbines for various spacing, wind angles and wind speeds. In particular, we defined the angle of independence (the angle of wind where two turbines will not affect each other after 0 degrees),” said researcher Muhammed Nedim Sogut.
The research uses the zCFD tool from Zenotech in Bristol to build a simulation of a wind farm with two turbines from three to seven diameters spacing for changing wind direction of 15 degrees on both sides at 10, 12 and 14 m/s wind speeds. “The angle of independence for changing wind direction, speed and distance between turbines has been investigated, the data required to train an ANN has been created and the individual effects of wind speed, wind direction and spacing between turbines have been examined,” said Sogut.
The results of the project will provide a better understanding of turbulence and its effect on the wind power output of a wind turbine n a large array, ultimately improving the output and reducing costs.
“The angle of independence is a useful measure for reducing the number of