Satellite dynamic spectrum access prediction method and device considering incomplete spectrum sensing mode

The invention discloses a satellite dynamic spectrum access prediction method and device considering an incomplete spectrum sensing mode, and the method comprises the steps: constructing a deep reinforcement learning frame based on an underlying satellite communication system, enabling the deep reinforcement learning frame to comprise a training network, a target network, an agent, an experience pool and an environment, enabling a secondary satellite to serve as the agent, and enabling the target network to serve as a target network; a transmission channel is used as an environment; establishing a target optimization problem according to the deep reinforcement learning framework, defining parameters in the deep reinforcement learning framework, and converting the target optimization problem into a sequential decision problem; performing reinforcement learning training by using the experience pool based on the sequential decision problem to obtain a trained training network; and obtaining an observation value