Machine learning can help kites and gliders capture wind energy

Aerial wind energy (AWE) is a lightweight technology that uses flying devices including kites and gliders to capture energy from the atmosphere. To maximize the energy they harness, these devices need to precisely control their orientation to account for turbulence in Earth’s atmosphere.

Through new research published in EJ EAntonio Celani and colleagues at Abdus Salam International Center for Theoretical Physics, Italy, demonstrate how Reinforcement Learning algorithms can significantly enhance the explanatory power of AWE devices distortion.

With a much lower construction cost than traditional wind turbines, AWE could prove of great value in extending wind energy’s reach to poorer, more remote communities. To solve the problem wind energyflying devices or tethered to a ground stationwhere electricity is converted into electricity, or used to pull cars.

The main challenge facing this technology is maintaining its performance in wind and weather conditions. To do this, the researchers are now using computer models to predict the future state of the atmosphere, allowing kites and gliders to automatically adjust their direction. However, because perturbation requires a large amount of computing power for accurate estimation, it is often ignored in existing models, resulting in suboptimal performance in AWE systems.

In their study, Celani’s team solved the problem using Reinforcement Learning: a machine learning algorithm that uses trial and error interactions with its surroundings to calculate the direction a kite or glider will take. obtain the maximum possible energy from the atmosphere. As a proof of concept, the researchers applied the algorithm to a simulated ship being towed by a kite.

When given a set of simple maneuvering instructions, the kite used Reinforcement Learning to tow the ship long distances, even without knowing in advance of the ups and downs it would encounter. Given the initial success of their method, Celani and colleagues now hope that the use of Reinforcement Learning may soon allow the reach of AWE to expand further in the future.