Time Slicing Method Based on Network Variation Indicator in Pocket Switched Network

摘要

The size of time slicing affects link prediction accuracy in pocket switched network (PSN), this paper proposes a network variation indicator slicing algorithm(NVISA) to determine it. The local network variation indicator is defined to characterize the degree of dynamic change of the network between adjacent time slices. In order to determine whether the sequence of local network variation indicator corresponding to different slice sizes is stable, stationarity test is used as filter. After that, the global network variation indicator is defined to characterize the topology evolution of the whole network under different slice sizes. Meanwhile, by determining the global network variation indicator interval (0.1, 0.3) and the value of 0.2, appropriate slice size can be determined. And the Encoder-LSTM-Decoder(E-LSTM-D) link prediction model is used to verify the slice effect. The experimental results show that, compared with some slice size commonly used on Infocom05, Infocom06 and Cambridge datasets, the one determined by NVISA improves the prediction effect 2.7%, 6.3% and 5.86% of the link prediction model. So the algorithm by this paper can effectively select the appropriate slice size.