Observation Method for Autonomous Maneuver of Spacecraft under Emergency Conditions

To deal with space threats with strong maneuverability such as kinetic energy interceptors, remote-sensing satellites need to perform autonomous avoidance while carrying out observation tasks. The traditional method is no longer suitable for such sudden and time-sensitive problems. A Reinforcement-learning method for remote-sensing satellite autonomous decision-making maneuver and completion of remote-sensing tasks is proposed.Assuming that remote sensing satellites only use velocity pulses for evasive maneuvering observations.By establishing the MDP model of orbital maneuver, the problem is abstracted as the order decision problem of impulse strategy in the process of maneuver observation. According to the discrete characteristics of impulse action, the classical Reinforcement-learning algorithm DDQN is used to solve it, so as to obtain the optimal multi impulse maneuver strategy under interception threat.Based on the DDQN algorithm, the parameter update mechanism of the target neural network has been improved.The experimental results show that the DDQN algorithm, which improves the parameter update mechanism of the target neural network, has a faster rate of convergence than the traditional DDQN algorithm in the training process;The remote sensing satellite adopts the Deep reinforcement learning algorithm for independent decision-making. Under the constraint of satisfying the observation conditions, the algorithm has lower fuel consumption than the current on orbit escape mode;When the interceptor has the same parameters but different maneuvering trajectories, the success rate of avoiding multiple trajectories within the arc segment it can hit is 97.69 % , of which 68.46 % meets the observation time requirements and 69.23 % meets the coverage requirements for the target area, effectively improving the ability to avoid autonomous decision-making under emergency conditions.