Stratospheric wind field feature extraction and energy management for hybrid electric solar airship with deep reinforcement learning

Sufficient energy is demonstrated overwhelming superiority in both vehicles and aircrafts. Limited by the energy density, energy storage represented by Lithium-ion battery cannot meet the increasing energy requirements of diverse payloads on solar-powered stratospheric airship for months or years. In this paper, the hybrid fuel cell/battery system for stratospheric airship is presented. The relationship between the real wind field and the demand power is illustrated. Based on the reanalysis of historical wind data, the probabilistic model of demand propulsion power is established and integrated with the training environment. The deep reinforcement learning method is adopted to solve the energy management problem. The prioritized experience replay with extra identifier, which encourages the utilization of high-value samples without identifier, is proposed. Comparative analysis shows the proposed method is effective in determining the management strategy with promising convergence speed. The results demonstrate that changing the SOC reference of the proposed method from 0.4 to 0.7 can result in 5.9% increment in energy consumption. Furthermore, the potential decline of regulation capability of the hybrid system and the corresponding influence on the nighttime energy balance is investigated. The proposed method has reference value for advance alarm of power supply failure during long term flight.