Prediction and Application of Network Business Traffic based on LSTM

Abstract Traffic prediction is the basis for guiding bandwidth adjustment in the process of network operation and maintenance. Most of the existing traffic prediction models focus on network traffic prediction based on a single model or single characteristic, and cannot accurately predict whether the current bandwidth can meet user requirements. This paper proposes a prediction method of network traffic based on the long short-term memory (LSTM) model. Based on LSTM model, the traffic interval deviation and the network traffic are predicted, and the two prediction results are combined, then the network traffic interval is predicted. First, this study uses the historical data from 2019 to 2021 to predict the historical deviation value in February based on the LSTM model, and then predicts the traffic interval on October 2021 based on the historical data in September 2021, and compares it with the bandwidth whether the forecast is out of bounds. Finally, compare with other models, including Auto Regression Model (AR), Moving Average Model (MA), Autoregressive Integrated Moving Average Model (ARIMA) and linear regression model, the precision rate(p), recall rate(r) and comprehensive value(f) are all higher than those of the other two models, which proves the reliability of the LSTM model.