LSTM Based Network Flow Forecasting in High-Speed Railway Data Network

Abstract

Network flow forecasting is the basis of network performance analysis and bandwidth allocation. It is of great significance for routing path selection in network service establishment process. The accurate flow forecasting can effectively reduce the service blocking probability. Different forecasting models based on recurrent neural network (RNN) have got extensive attention recently. However, there is relatively less research focusing on network flow forecasting, which is an essential part of future network management. In this paper, we utilize long short term memory (LSTM) model to forecast network flow tendency. To avoid overfitting issue, dropout method is introduced into LSTM. The flow information is collected from High-Speed Railway data network and preprocessed to obtain time-flow series. Two time-flow series are used to verify the network flow forecasting accuracy. Experiments show that the LSTM model can provide accurate forecasting results. The median MAPE of two series forecasting are 0.014 and 0.043, respectively. A shuffled data is used to train the LSTM model too, and corresponding forecasting results indicate that the model has a good generalization ability and it can be applied in different networks.