Optimizing airborne wind energy with reinforcement learning.

Area of Science:

• Renewable Energy Systems
• Aerodynamics
• Artificial Intelligence

Background:

• Airborne wind energy (AWE) technology harnesses wind power using airborne devices like kites and gliders.
• Dynamic control of airfoil orientation is crucial for maximizing AWE performance.
• Turbulent aerodynamics presents significant challenges for conventional control methods.

Purpose of the Study:

• To address the control optimization problem in airborne wind energy systems.
• To explore the application of reinforcement learning for dynamic control of AWE devices.
• To develop an efficient control strategy for kite-based power generation.

Main Methods:

• Utilizing reinforcement learning (RL) for adaptive control without prior system knowledge.
• Simulating a controlled environment to train the RL agent.
• Developing an RL algorithm based on intuitive observations and transparent interpretation.

Main Results:

• The reinforcement learning agent successfully learned to control a kite in a simulated environment.
• The developed control strategy proved efficient for towing a vehicle over long distances.
• The optimal strategy was interpretable as a simple set of maneuvering instructions.

Conclusions:

• Reinforcement learning offers a viable solution for optimizing control in airborne wind energy systems.
• This approach bypasses the need for complex analytical models of turbulent aerodynamics.
• The RL-based strategy provides an efficient and understandable method for AWE device control.