Internet time-delay prediction based on EEMD-PACFElman-GWO combined prediction model

Aim at the multi-scale, self-similar, nonlinear and dynamic characteristics of internet time-delay, a prediction model combining ensemble empirical mode decomposition (EEMD), PACFElman neural network and gray wolf optimization (GWO) is proposed to predict the time-delay sequence of internet. Firstly, the model uses EEMD to decompose the internet time-delay sequence into multiple intrinsic mode functions (IMF) and residual components, the multi-scale mixed internet time-delay prediction problem is transformed into multiple fixed scale components prediction problem, which effectively reduces the complexity of the prediction target. Then, the partial autocorrelation function (PACF) was used to optimize the Elman neural network to obtain the PACFElman neural network. The optimized PACFElman neural network was used to construct sub-models and predict for each component. Finally, in order to make full use of the different scales prediction results of each sub-model and further improve the accuracy of the prediction model, GWO was introduced into the model to optimize the fusion weights, and the optimal fusion weights were used to weight the prediction results of each sub-model to obtain the final time-delay prediction results. The results show that the mean square error (MSE) of EEMD-PACFElman-GWO combined prediction model is 54.45% and 52.93% lower than that of Elman and ARMA single model respectively, and 15.62% and 15.53% lower than that of EEMD-Elman and EEMD-ARMA combined model respectively. Therefore, EEMD-PACFElman-GWO combined prediction model can effectively improve the accuracy of internet time-delay prediction.