Safe reinforcement learning control for adaptive station-keeping of a stratospheric airship in complex environments

High-precision, long-duration station-keeping is an enabling technology for stratospheric airships to perform critical missions such as Earth observation and communication relay. However, achieving adaptive and safe station-keeping control under both persistent wind disturbances and critical safety constraints presents a primary challenge. To address this issue, this paper proposes an intelligent control framework that integrates weather optimal control (WOC) with safe reinforcement learning (SRL). The framework leverages WOC to generate guidance commands that dynamically adapt to wind disturbances, which are precisely tracked by a soft actor–critic (SAC) based reinforcement learning controller. Crucially, high-order control barrier function (HOCBF) is innovatively introduced to construct a safety filter. This approach provides a formal safety guarantee for the airship’s autonomous obstacle avoidance in complex environments, without requiring precise information on the wind field. Simulation results validate the effectiveness of the proposed framework, demonstrating its ability to achieve autonomous, safe, and adaptive station-keeping while satisfying multiple constraints.