面向航天器远距离协同交会的超参数 自主调优-同伦方法

In response to the challenges of slow convergence and difficulty in obtaining high-quality solution for the cooperative rendezvous fuel optimization problem of far-distance spacecraft under J₂ perturbation，a method combining hyperparameter autonomous tuning global search and homotopy method is proposed. To quickly obtain high-quality optimal initial co-states in the energy-optimal problem，the deep deterministic policy gradient（DDPG）neural network is combined with the hybrid cluster particle swarm optimization（HCPSO）to form a reinforcement learning enhanced particle swarm optimization（RLEPSO）. The algorithm parameters are autonomously and dynamically tuned according to the particle search status by RLEPSO，which improves the searchability and convergence speed of the optimization algorithm significantly. The homotopy method is adopted to realize the smooth transition from energy optimization to fuel optimization， thereby effectively solving the non-smooth integration problem of bang-bang control. A series of simulation verification results show that compared with particle swarm optimization（PSO）and hybrid cluster particle swarm optimization （HCPSO），RLEPSO can obtain higher-quality initial co-states，which provide a high-quality foundation for the homotopy process and enable it to achieve a better control strategy. In addition，the proposed method has a faster convergence speed，which can better meet the timeliness requirements for solving the spacecraft cooperative rendezvous problem and provide reliable technical support for fuel-optimal cooperative rendezvous under complex perturbations.