A Vector Auto-Regression Based Forecast of Wind Speeds in Airborne Wind Energy Systems

This paper presents two wind energy forecast methods for control of an airborne wind energy system (AWE). The primary objective is to maximize the energy production of the AWE system under a spatio-temporally varying environment with uncertainty in the future wind speed. The controller for the AWE system is formulated as a model predictive controller (MPC). We employ data-driven models to generate probabilistic forecasts of the wind using vector auto-regression (VAR) and time-of-day forecasts. Bayesian optimization is employed to find the optimum of an unknown and expensive to evaluate function. Specifically, the objective function is modelled via wind speed forecasts and then Bayesian optimization optimizes the altitude trajectory while balancing exploitation and exploration of the available altitudes. The performance of the AWE system under the VAR forecast model significantly improves energy production by incorporating wind speed correlations for nearby altitudes.