Autonomous Planning Of Spacecraft Pursuit-Evasion Maneuver Trajectory Based On Artificial Intelligence Method

Spacecraft chasing game is a research hotspot in the field of orbital mechanics.Traditionally,differential countermeasures are used to obtain the optimal control strategy of both sides.However,this method has the disadvantages of complicated solving process and complicated calculation,which is difficult to meet the strong real-time requirements of the confrontation task.The development of artificial intelligence technology makes it possible to use artificial intelligence to realize all or part of online decision making.This dissertation studies the problem of generating optimal control strategy of space target pursuit game based on deep neural network and reinforcement learning algorithm to realize intelligence,autonomy,and fast spacecraft pursuit game online orbital independent planning.The main research contents of the thesis are as follows:Firstly,the optimal maneuver strategy generation algorithm for space target pursuit game based on differential game theory is studied.Based on the CW equation,the relative motion model of space target pursuit is established.Then the differential countermeasure theory is used to analyze and solve the three types of differential game models of fixed stay,infinite time domain and survival type,which lays a foundation for the generation of training and test data sets in the subsequent artificial intelligence algorithm research.Secondly,the self-planning method of maneuvering trajectory of space target pursuit game based on deep neural network is studied.Different neural network models were established for three different types of space target pursuit game.Both the fixed stay period and the infinite time domain pursuit game model can directly obtain the maneuvering strategy of the spacecraft via the deep neural network.The numerical simulation results show that the neural network method is effective,and the generated flight trajectory is basically consistent with the optimal trajectory.The survival-type chasing game model can only fit the four intermediate quantities needed to solve the maneuver strategy by the neural network.It can be seen from the numerical simulation results that the neural network method is much faster in terms of calculation speed than the traditional optimization algorithm.Finally,the self-planning method of maneuvering trajectory of space target pursuit game based on reinforcement learning algorithm is studied.For the problem of solving three-dimensional continuous space trajectories such as space target pursuit,the learning efficiency of training directly using reinforcement learning is relatively low,and the training takes a long time.In response to this problem,this dissertation conducts supervised learning before intensive learning,and uses the results of supervised learning to reinforce the initialization of the learning strategy network,thereby accelerating the learning process of intensive learning.The simulation results show that the reinforcement learning algorithm can adjust the network parameters online when the environment of the spacecraft is deviated from the dynamic model used for supervised learning,so that the network model can be gradually applied to the current environment.