Short-Term Network Traffic Prediction with ACD and Particle Filter

Abstract

Network traffic prediction is hot spot in recent years' research, which is of great significance in area such as congestion control, network management and diagnostic. Network traffic is non-linear, non-stationary, and uncertain, and its uncertainty increases rapidly when making short-term traffic flow prediction. After reviewing current network traffic prediction algorithms' merits and drawbacks based on Time-Series analysis, Artificial Neural Network here, a new network traffic prediction algorithms in short-term is proposed. The time interval when detecting that network data packet pass on certain section is treated as a stochastic process. In the ARCH (autoregressive conditional heteroskedasticity) framework, stochastic process is described by a marked point process that different point processes may generate different ACD (autoregressive conditional duration) model, then ACD model can be used to complete the description of time interval when network data packet passing. Based on this model, a particle filter is applied to non-stationary motion system for short-term network traffic prediction. At last, this algorithm is applied to real data for real-evidence analysis.