关键词: 空间飞行器, 控制软件, 动态自适应, 强化学习, 反馈

Abstract: Spacecraft needs to meet the requirements of multitask, multiworking mode and largescale maneuvering during space flight. Its control system has a lot of external interference and uncertain internal parameters under largescale maneuvering conditions. At the same time, the adaptive process of the aircraft is affected by limited resources, and manual intervention is difficult. The existing mature dynamic adaptive methods are not necessarily suitable for spacecraft autonomous control software, so the current dynamic adaptive adjustment methods of autonomous control system software cannot meet the higher requirements. Therefore, an adaptive framework based on the twolayer perceptionanalysisdecisionexecution (MAPE) control loop is proposed, which uses rule/strategybased decisionmaking methods and reinforcement learningbased decisionmaking methods to make decisions on local and global changes. In addition, a datadriven feedback method is used to periodically adjust and optimize the policy information in the rule library to ensure that the aircraft can dynamically complete adaptive adjustments and ensure the reliable execution of tasks when performing tasks in complex space environments.

Key words: space vehicle, controlling software, dynamic adaptation, reinforcement learning, feedback