Rare Event Prediction for Enhanced Control System Reliability of AWE Systems

Abstract

Reliable autonomous operation of Airborne Wind Energy (AWE) systems requires control algorithms that are able to attenuate the effect of stochastic disturbances on the control performance in continuously changing wind conditions. Assessing the stability and robustness of the control system is in general carried out using simplified system models where the real stochastic nature of the control problem is neglected. Therefore, a direct Monte Carlo approach is used in practice to increase the confidence in the control system’s reliability. However, this approach performs poorly if it is used to estimate the effect and the probability of rare events such as strong gusts. Statistically, these events are located at the tails of the underlying joint probability density function. Consequently, only a few samples leading to rare events can be identified in a reasonable amount of time which leads to a biased probability estimate. In addition, it is difficult to recognize and leverage patterns if only a small set of samples is available that lead to a violation of a critical control requirement