The minimum flight time of spacecraft rendezvous is one of the fundamental indexes for mission design. This paper proposes a rapid trajectory planning method based on convex optimisation and deep neural network (DNN). The time-optimal trajectory planning problem is reconstructed into a double-layer optimisation framework, with the inner being a convex optimisation problem and the outer being a root-finding problem. The thrust properties corresponding to time-optimal control are analysed theoretically. A DNN-based rapid planning method (DNN-RPM) is put forward to improve computational efficiency, in which the trained DNN provides a high-quality initial guess for Newton’s method. The DNN-RPM is extended to search for the optimal entering angle of natural-motion circumnavigation orbit injection problem and the minimum reconfiguration time of spacecraft swarm. Numerical simulations show that the proposed method can improve the computational efficiency while ensuring the calculation accuracy.

中文翻译：

---

基于凸优化和DNN的时间最优交会轨迹快速生成

航天器交会最短飞行时间是任务设计的基本指标之一。本文提出一种基于凸优化和深度神经网络（DNN）的快速轨迹规划方法。将时间最优轨迹规划问题重构为双层优化框架，内部为凸优化问题，外部为寻根问题。从理论上分析了时间最优控制对应的推力特性。为了提高计算效率，提出了一种基于DNN的快速规划方法（DNN-RPM），其中经过训练的DNN为牛顿法提供了高质量的初始猜测。将DNN-RPM扩展用于搜索自然运动环绕轨道注入问题的最佳进入角和航天器群的最小重构时间。数值模拟表明，该方法能够在保证计算精度的同时提高计算效率。