Fast convergence dynamic induction spectrum access method based on LSTM and Q-Learning fusion

The invention provides a fast convergence dynamic spectrum access method based on LSTM (Long Short Term Memory) and Q-Learning fusion, which comprises the following steps of: firstly, constructing an online learning model of a cognitive user by adopting an LSTM (Long Short Term Memory) network, feeding back a real-time ACK (Acknowledgement Character) message according to channel access, taking the real-time ACK message as model input, obtaining predicted occupancy probabilities of all accessible channels, and finally obtaining the predicted occupancy probabilities of all the accessible channels; secondly, combining a channel occupancy prediction probability learned by the online learning model with a state action Q value table of a Q-Learning reinforcement learning algorithm to obtain a new Q value strategy matrix, performing dynamic spectrum access by a cognitive user according to the combined strategy matrix, and after multiple access iterations, obtaining an optimal spectrum access strategy under each spec