A novel method for satellite prediction of VDE-TER service traffic information based on multilayers bidirectional LSTM

Satellite prediction of very high frequency (VHF) Data Exchange (VDE)-Terrestrial (VDE-TER) ship-to-ship Self-Organized Networks (SON) service traffic information is crucial for wireless resource allocation of VHF Data Exchange System (VDES). Due to the VDE-TER, channel load observed by the satellite changes rapidly and the historical information cached on the satellite is limited; the prediction of VDE-TER service traffic is difficult. This paper proposes a multilayer bidirectional long short-term memory (LSTM) neural network model (MLB-LSTM), which controls the amount of forgetting of historical information and the amount of memory of current information through the LSTM unit, so that the model can better learn the nonlinear changes of service traffic. The bidirectional LSTM module combines forward and backward time series data, reducing the amount of data required for prediction and enhancing the prediction accuracy of the model. Numerical results show that the proposed prediction model significantly outperforms traditional methods and is able to adapt to rapidly changing VDE-TER data.