Hybrid Prediction Model for Passive Optical Network Traffic based on CEEMDAN and Machine Learning

This paper proposes a hybrid prediction model for passive optical network (PON) traffic, combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) with particle swarm optimization-support vector regression (PSO-SVR) and long short-term memory (LSTM) neural network. The model decomposes the original network traffic data by CEEMDAN, uses the approximate entropy value of the intrinsic mode function (IMF) component as the basis for selecting PSO-SVR or LSTM for individual prediction, and sums the forecast results of each component to obtain the final prediction result. Meanwhile, we selected four models based on PSO-SVR direct prediction, LSTM direct prediction, hybrid prediction based on local mean decomposition (LMD), and hybrid prediction based on empirical mode decomposition (EMD) to conduct experiments on network traffic data simulated by ON/OFF sources. The experimental results show that the model proposed in this paper has better prediction results and higher prediction accuracy.