Robust adaptive control for kite wind energy using evolutionary robotics

Recently designed wind energy systems use large traction kites to drive electricity generation equipment at ground level; this exploits stronger and more consistent winds available at higher altitudes than used by traditional wind turbine systems. These kites require active control; in this study we build upon past work demonstrating the use of evolutionary robotics techniques to build neural network controllers that maximize energy recoverable from wind in a simulated kite system using only information available at ground level from the line angles and forces. Neurocontrollers are evolved under selective pressure to fly the kite in order to maximise forces through the lines, resulting in optimal figure-eight trajectories. We allow evolution to converge completely and compare the flight trajectory with analytically derived solutions. We consider the robustness of the neurocontrollers to large gust deviations in speed and direction. Finally we address the problem of controlling the kite with different line lengths, which dramatically alters the response properties of the kite.