中国空间科学技术

中国空间科学技术 ›› 2025, Vol. 45 ›› Issue (1): 34-45.doi: 10.16708/j.cnki.1000-758X.2025.0004

• 智能航天器专栏 • 上一篇    

高轨航天器集群在轨服务智能任务规划方法

郑鑫宇1，曹栋栋1，唐佩佳1，张轶1，彭升人1，*，周杰1，党朝辉2   

  1. 1.中国空间技术研究院 钱学森空间技术实验室，北京100094
    2.西北工业大学 航天学院，西安710072
  • 收稿日期:2023-10-23 修回日期:2024-02-19 录用日期:2024-03-01 发布日期:2025-01-23 出版日期:2025-02-01

Intelligent mission planning method for on-orbit service of high-orbit spacecraft cluster

ZHENG Xinyu1，CAO Dongdong1，TANG Peijia1，ZHANG Yi1，PENG Shengren1,*，ZHOU Jie1，DANG Zhaohui2   

  1. 1.Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, Beijing 100094, China
    2.School of Astronautics，Northwestern Polytechnical University，Xi’an 710072，China
  • Received:2023-10-23 Revision received:2024-02-19 Accepted:2024-03-01 Online:2025-01-23 Published:2025-02-01

PDF (PC)

66

摘要： 针对高轨航天器多对多在轨服务任务规划问题，建立了考虑燃料消耗和任务时长两个优化目标的高轨航天器在轨服务任务规划模型，并提出了一种Q学习改进的多目标遗传算法（Q-learning-based Multi-objective Genetic Algorithm , QMGA）。首先，建立了基于四脉冲Lambert转移的多对多目标分配模型，并同时以速度脉冲和任务用时为目标函数，通过将问题解耦为轨道转移优化问题和目标分配优化问题实现了优化变量的降维，简化了计算过程。然后，结合Q学习提出了QMGA算法，采用Q学习在线更新多目标遗传算法的交叉概率与变异概率，提高了算法的寻优能力。最后采用QMGA算法求解模型，并将其计算结果与传统多目标遗传算法计算结果进行对比，发现QMGA算法可以得到更优的任务规划结果，计算得到的总速度增量消耗和总任务时间分别平均比MGA计算得到的结果减少了6.2%和19.7%。这一结果证明强化学习方法可进一步赋能传统智能优化方法，从而提升航天器集群任务能力。

关键词: Q学习, 多目标遗传算法, 多目标分配任务规划, 多脉冲Lambert转移, 集群任务规划

Abstract: A mission planning model for on-orbit service of high-orbit spacecraft with two optimization objectives, fuel consumption and time consumption, is developed for the high-orbit spacecraft multi-to-multi on-orbit service mission planning. And the Q-learning-based Multi-objective Genetic Algorithm(QMGA) is proposed to solve the model. Firstly, a multi-to-multi objective assignment model based on four-impulse Lambert transfer is established. The velocity impulse consumption and time consumption are taken as the objective functions. By decoupling the problem into the orbit transfer optimization problem and the target assignment optimization problem, the dimension of the optimization variables is reduced, and the calculation process is simplified. Then, combined with Q-learning, the QMGA algorithm is proposed. The Q-learning is used to update the crossover probability and mutation probability of the multi-objective genetic algorithm, which improves the optimization ability of the algorithm. Finally, the QMGA algorithm is adopted to solve the model, and the calculation results are compared with that of the traditional multi-objective genetic algorithm. It is found that the QMGA algorithm can obtain better results and complete multi-to-multi on-orbit service tasks with less fuel consumption in a shorter time. The fuel consumption and the time consumption computed with the QMGA algorithm were 6.2% and 19.7% lower than those computed with MGA algorithm on average, respectively. This proves that the reinforcement learning method can further empower the traditional intelligent optimization method, thereby improving the mission capability of the spacecraft cluster.

Key words: Q-learning, multi-objective genetic algorithm, multi-objective assignment mission planning, multi-pulse Lambert transfer, cluster task planning