Tackling control and modeling challenges in wind energy with data-driven tools

Driven by the abundance of data, new paradigms have lately emerged in science and engineering fields. Wind energy makes no exception and an increasing number of data-driven methods are used to tackle one of today’s biggest challenges in the wind community: reduce the levelized cost of energy. Many underlying themes are concerned, ranging from wind turbine and wind farm control for power maximization and fatigue reduction to wake modeling.
We present a couple of efforts into exploiting such data-driven paradigms to address important matters in wind energy. At the turbine scale, we focus on the detrimental effects of atmospheric boundary layer and turbulence on structural components. We propose an individual pitch controller based on a neural network trained with reinforcement learning. As for the wind farm scale, wake interaction is a big challenge as it generates major power losses. We investigate wake redirection strategies under the lens of reinforcement learning. The investigations reveal the need for low-cost yet accurate wake modeling, which leads us to bring data assimilation and physics together and develop an online dynamic wake meandering model.