高精度在轨实时轨道机动决策

为保证在轨机动实时性和高精度的要求，提出了一种基于机器学习的在轨实时机动决策方法。通过优化算法离线获得摄动下的精确解，减去二体解得到速度增量差，将其投影到轨道坐标系获得速度增量摄动修正项，以此作为神经网络输出，设计网络参数并训练得到摄动修正网络、组合应用摄动修正网络和二体解实现高精度的在轨实时轨道机动决策。仿真结果表明:卫星按照该决策机动完成后的终端位置偏差与按照优化算法给出的决策机动完成后终端位置偏差精度一致，且前者决策耗时仅为后者决策耗时的0.01%左右。所提轨道机动决策方法兼顾了精度与实时性，适用于星上决策。 

High-precision on-orbit real-time orbital maneuver decision

In order to ensure the real-time maneuverability and high-precision requirements of orbital maneuver, a real-time maneuver decision-making method based on machine learning is proposed. The optimal solution under perturbation is obtained offline through the optimization algorithm. The two-body solution is subtracted to obtain the speed increment difference, which is projected onto the orbital system to obtain the speed increment perturbation correction term, which is used as the output of the neural network. The network parameters are designed and trained to obtain perturbation correction network. The combination of perturbation correction network and two-body solution is used to achieve high-precision real-time orbital maneuver decision. The simulation results show that the terminal position deviation after the completion of the maneuver according to the decision is consistent with the accuracy of the terminal position deviation after the completion of the decision maneuver according to the optimization algorithm, and the former decision time is only about 0.01% of the latter decision time. The orbital maneuver decision-making method proposed in this paper takes into account both accuracy and real-time performance, and is suitable for on-board decision-making.