航天器在轨加注技术可以显著提升在轨卫星的使用寿命，将推动载人登月以及深空探测任务进一步发展。然而由加注运载器和受注航天器组成的航天器组合体质量大、加注流量高，在轨加注过程面临着刚-液耦合、质量转移所引起的参数时变及非线性扰动问题。本文建立了航天器在轨加注系统的刚-液-控耦合动力学模型，将液体推进剂等效为运动脉动球模型；并设计了一种基于长短期记忆神经网络的自适应控制方法，利用神经网络在线学习并预测由液体晃动等未知动态引起的非线性时变扰动。数值仿真表明，所提出的控制方法可有效处理复杂航天器系统的姿态控制问题，并在不同的加注速度下表现出良好的鲁棒性。本文考虑的微重力环境下多贮箱推进剂晃动与控制问题将为航天运输系统的在轨加注任务提供总体设计参考。

On-orbit refueling technology for spacecraft can significantly extend the operational lifetime of satellites and further advance manned lunar landing and deep space exploration missions. However, the spacecraft combination consisting of a refueling carrier and a receiving spacecraft has a large mass and high propellant transfer rates, which introduces significant challenges during the refueling process, including rigid-liquid coupling, time-varying parameters, and nonlinear disturbances. This paper establishes a coupled rigid-liquid-control dynamic model that incorporates mass transfer effects, in which the liquid propellant is equivalent to moving pulsating ball model. Furthermore, an adaptive control strategy based on Long Short-Term Memory neural network is developed to online learn and predict nonlinear time-varying disturbances caused by unknown dynamics such as liquid sloshing. Numerical simulations demonstrate that the proposed control method can effectively handle the attitude control problem of complex spacecraft systems and exhibits good robustness at different refueling speeds. The investigation of multi-tank propellant sloshing and control under microgravity environment considered in this paper will provide overall design reference for on-orbit refueling tasks of space transportation systems.