Path Following or Tracking Model Predictive Control for Towing Kites – A Question of Formulation or Learning?
Towing kites bear large potential for carbon neutral energy generation using high altitude wind. However, to realize this potential reliable, safe, and efficient control is essential. This task is often tackled using a trajectory tracking model predictive control formulation. We show that such a formulation limits the achievable performance due to a fixed and predefined trajectory speed. This can be bypassed by a tailored path following predictive control formulation, enhancing the controller’s robustness in different scenarios. Additionally, we show that learning an accurate model using input-output data improves the performance of both path following and trajectory tracking. In case of the trajectory tracking controller the improvements allow to mask its inherent drawbacks, raising the question whether learning can be used to compensate an unfavorable problem formulation. Our results show that the path following controller performs well for a wide range of parameters and exhibits a trade-off between generated thrust and tracking accuracy. Furthermore, the path following controller significantly outperforms the trajectory tracking controller in terms of robustness.