Tuesday, January 29, 2019, 11:00 - 12:30
Room 01-012, Georges-Köhler-Allee 102, Freiburg 79110, Germany
Airborne Wind Energy (AWE) systems are a new and promising way to satisfy
the growing demand on energy with renewable technologies that are competitive to
classical non-renewable power plants. To produce energy a kite is connected with a
tether to the ground and harvests wind energy by flying in a crosswind motion.
Flying a kite autonomously is a complex task which needs research in advanced
control techniques. To accomplish the task this thesis combines techniques from
the field of control theory and machine learning to control the behavior of a model
airplane attached to a rotating carousel. A data driven approach is used to train
a neural network as a model of the airplane. The learned model is passed to a
Non-linear Model Predictive Control (NMPC) scheme to let the airplane follow a
reference trajectory. To improve the control performance the model is adapted online
during the flight. The controller is tested in real-time on an experimental setup.
The results show that it is possible to train a neural network as a model and use it in
a model-based controller to achieve that an airplane follows a reference trajectory.
Adapting the model during the flight reduces the steady-state error and increases the
With this thesis hopefully another step is done towards a working AWE system and
setting a base for further research of a promising control technique that is able to
accomplish the autonomous flight of tethered kites to harvest wind energy.