Research on Cognitive UAV Network Channel Prediction Algorithm Combining DQN and Bi- LSTM Mechanisms

With the increasing application of unmanned aerial vehicle (UAV) in communication tasks, dynamic and accurate channel prediction becomes particularly critical. Channel state prediction through cognitive technology is very important for optimizing the allocation and scheduling of communication resources and improving communication reliability. In this paper, an algorithm combining Deep Q-network (DQN) and bidirectional Long Short-Term memory network (Bi-LSTM) is proposed to improve the accuracy of channel state prediction in UAV networks. The algorithm uses the multi-layer perception mechanism of deep neural network to sense the unoccupied channel resources in the surrounding environment, and uses the bi-directional prediction information ability of Bi-LSTM to achieve accurate prediction of the channel state. The simulation results show that the proposed channel prediction algorithm not only reduces the Root Mean Square Error (RMSE) and Mean Absolute error (MAE), but also optimizes the value of the loss function, which verifies the effectiveness of the proposed algorithm in channel prediction.