Research on Pursuit-Evasion Strategies for GEO Satellites Using PD-DDPG

This paper addresses the problem of orbital pursuit-evasion game for geostationary orbit (GEO) satellites using the Deep Deterministic Policy Gradient (DDPG) algorithm to control impulsive orbital maneuvers, with both sides trained simultaneously. Firstly, the kinematics of GEO satellite orbital motion is modeled, and the Runge-Kutta method is employed for orbital propagation to provide an interactive environment for reinforcement learning. Secondly, a game-based training framework is established based on the DDPG algorithm. Finally, the state space and action space within the framework are designed, and a novel PD-DDPG algorithm is proposed to control the pursuit-evasion process. The results demonstrate that the proposed algorithm effectively guides the agents to complete the task and accelerates the convergence rate.