Low-orbit giant constellation satellite switching method and device based on deep reinforcement learning

The invention discloses a low-orbit giant constellation satellite switching method and device based on deep reinforcement learning. The method comprises the following steps: acquiring satellite information in a visual range of a user terminal; determining the available channel capacity between the user terminal and the satellite, the remaining service time of the user terminal and the satellite, the lifting track type corresponding to the satellite and the state information of the minimum hop count from the satellite to the set gateway satellite; the state information is input into a first neural network model, a state-action value function output by the model is obtained, the model is trained through a deep reinforcement learning algorithm, actions are defined as satellites selected by a user terminal, and an action reward function is defined as a utility function constructed according to the available channel capacity, the remaining service time and the minimum hop count; and selecting the satellite corresp