A new method of network traffic prediction based on combination model

Network traffic has time-varying and nonlinear characteristics, leading to relatively low accuracy of single linear and nonlinear prediction models. Therefore, a combination model prediction method of autoregressive integrated moving average (ARIMA) model and improved particle swarm optimization (IPSO) bi-directional long short-term memory (BiLSTM) is proposed. First, the ARIMA model is used for nonsmooth modeling to obtain linear variation characteristics. Then the ARIMA prediction values are subtracted from the actual data to obtain the residual series. The optimal parameters of the BiLSTM are also optimized by IPSO. Then the optimized BiLSTM is modeled on the residual series to obtain the residual variation characteristics. Finally, the ARIMA linear predictions and the IPSO optimized BiLSTM residual predictions are summed to obtain the final prediction results. The proposed combination model outperforms the comparison model, has smaller prediction errors, and can better characterize the complex features of network traffic.