一种基于深度神经网络的航天器追逃博弈方法  

A deep neural network-based method for spacecraft pursuit-evasion games

作　　者：王漱石 张新 孙继元 徐莹辉 WANG Shushi;ZHANG Xin;SUN Jiyuan;XU Yinghui(College of Aerospace,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China)

机构地区：[1]南京航空航天大学航天学院,南京210016

出　　处：《航天控制》2025年第2期49-55,共7页Aerospace Control

摘　　要：针对时间自由且考虑J2摄动的航天器追逃博弈问题,提出了一种基于深度神经网络的混合优化方法。首先,利用传统优化算法求解J2摄动下的两点边值问题,生成数据集;然后,在此基础上,构造深度神经网络来拟合初始状态和解之间的关系,并生成初始猜测解;最后,通过局部优化算法进一步优化求解。通过仿真校验,证明该方法具有良好的可行性和鲁棒性,而且与传统混合优化算法相比,显著提高了计算效率。A hybrid optimization method based on deep neural networks is proposed in this paper for the spacecraft pursuit-evasion game problem with free terminal time and J2 perturbation considerations.Firstly,the data set is generated by solving the two-point boundary value problem without J2 perturbation using traditional optimization algorithms.Then,on the basis of that,a deep neural network is established to fit the relationship between the initial state and the solution,and the initial guess solution is generated.Finally,the solution is further optimized by using a local optimization algorithm.Through simulation and verification,it is demonstrated that this method not only performs good feasibility and robustness but also significantly improves computational efficiency,which is compared with traditional hybrid optimization algorithms.

关 键 词：自由时间 追逃博弈 J2摄动 深度神经网络 

分 类 号：V249.1[航空宇航科学与技术—飞行器设计]