Spacecraft attitude weakening and takeover control method based on deep reinforcement learning

Abstract

The application specifically relates to a spacecraft attitude weakening-takeover control method based on deep reinforcement learning, which comprises the following steps: in the weakening phase, an interference strategy is learned by a double-delay deep deterministic policy gradient algorithm agent to quickly consume the fuel of a target spacecraft; in the takeover phase, a takeover torque output by a service spacecraft is determined by a linear quadratic regulator; when the takeover torque output by the service spacecraft makes the attitude and angular velocity of a combination converge to a target attitude angle and a target angular velocity respectively, it is determined that the service spacecraft completes the attitude takeover of the target spacecraft; and the method has high robustness and can effectively cope with control challenges in a complex space environment by combining the adaptability of the double-delay deep deterministic policy gradient algorithm with the stability of the linear quadratic regulator.

Images