The Result and Performance Analysis of Long Short-Term Memory Algorithm in Mobile Ad Hoc Network for Transmission Channel Prediction

Link prediction aims to identify key network properties to forecast topology changes. Most studies on temporal channel prediction focus on networks with moderate topological variations” often overlooking topological intervals and link duration. This study proposes an approach using Long Short-Term Memory (LSTM) networks for link prediction. LSTMs are integrated into the prediction system due to their effectiveness in handling sequential input” outperforming multilayer perceptron networks and other artificial neural structures. The problem is framed as a multipurpose mobile network traffic prediction task” leveraging real downlink control channel data to minimize prediction errors. To optimize the LSTM network's memory length and determine the necessary data volume for accurate traffic forecasting” both one-step and long-term prediction errors are evaluated across different time intervals. Extensive testing on various datasets demonstrates that our LSTM model surpasses existing baseline methods in prediction accuracy.