Semi-parametric Regression Model Based Network End-to-End Delay Prediction

For the non-linearity and non-stationary characteristic of the network end-to-end delay, this paper establishes a semi-parametric regression model which is fit to describe the model uncertainty. In the modeling process, the Empirical Mode Decomposition (EMD) is employed to estimate the parametric ß and non-parametric g(-) separately using the trend vector and the intrinsic mode functions (IMFs). The estimation method can effectively suppress the phenomenon of partial regression. In order to achieve high accuracy network time-delay approximation based on the result of EMD, an improved method is proposed to mitigate the end effect which will affect the vibration and distortion errors. Due to the simple analytical expression, the semi-parametric regression model can be used as the network end-to-end delay real-time forecast. A large number of experimental results show that the proposed semi-parametric regression model can achieve high prediction accuracy of network end-to-end delay.