Comprehensive tradeoff and utilization of airborne renewable energy and uncertain stratospheric wind potential based on reinforcement learning

Solar-powered airships are demonstrated overwhelming superiority in plentiful application scenario. Adaptation to the flight environment and efficient energy management are essential during the mission. To improve the operating efficiency of airborne energy system, the tradeoff and integration of airborne renewable energy and uncertain stratospheric wind potential is studied. To complete the station keeping mission utilizing external and internal energy which has complex decision support parameters in different scales and continuous control action spaces with different characteristics, a Noisy Heterogeneous Policy Network Proximal Policy Optimization method is proposed. The state standardization, piecewise reward function, output action with noise, and heterogeneous policy network are designed. The results show that the proposed method has better convergence speed under different degrees of uncertainty of wind field and at different starting points. When the prediction error of the wind velocity is less than 2 m/s, the effective time within the region of the airship starting at specific positions is more than 80 %. When the error reaches 5 m/s, the time percentage is reduced to 50 %. The research results of this paper can provide some valuable reference for improving the performance of renewable energy system on stratospheric airship during the long-time flight in uncertain wind fields.