To solve the problem of spectrum resource shortage, a key technology of cooperative spectrum prediction in cognitive radio network is adopted to improve the accuracy of spectrum prediction in this paper. This technology can effectively avoid the problem that the single user spectrum prediction is easy to be affected by environmental interference. In this paper, we first model the authorized channel state using queuing theory, predict the future time slot state of the channel by LSTM, and then summarize the final prediction results. Finally, by comparing LSTM with RNN and MLP methods in Python simulation, this paper takes prediction accuracy and F1 value as performance indicators to verify, and the results show that LSTM is superior to the other two algorithms.