If we wish to prevent rendering the planet we live on irreversibly uninhabitable, we need to change the way we do many things, generating energy among them. Wind energy, produced by wind farms, can support this, but only if we optimize their operation to maximize the power they can harvest from the wind without incurring excessive damage to the wind turbines. This calls for active control of the individual turbines. In particular, we can adjust their yaw angles (the direction they face about a vertical axis), such that the turbines on the boundaries first hit by the wind (the ‘upstream’ turbines) of the farm don’t shield the ‘downstream’ turbines that lie in their wake from the wind. To optimally adjust these yaw angles, we need to predict what the wind speed at the downstream turbines will be for different wind speeds, wind directions, and turbine yaw angles. This project presents a collection of recurrent neural networks to predict the wind speeds at downstream wind turbines over a future time horizon, that could be used to better control the yaw angles of wind farms, and thereby save the planet from harrowing environmental destruction.
