Topology prediction mechanism for pocket switched network based on deep belief network

摘要

Pocket switched network is a new kind of delay tolerance network. The topology control algorithms for MANET and the link prediction methods for social network are not suitable for pocket switched network. A novel PSN topology prediction mechanism is proposed, named DBN-LS-SVR, which uses DBN to build the feature extractor and predicate the regression by LS-SVR. It uses the common neighbor strength between nodes as training samples, and the number of nodes of hidden layer of restricted boltzmann machine (RBM) is computed in terms of information entropy theory. Through tuning learning rate self-adaptively, the reconstruction error of RBM goes to stable rapidly, so that the convergence time is shortened. Least squares support vector regression machine (LS-SVR) is taken to predict common neighbor strength between the nodes, so as to judge whether there is a link between nodes. The mechanism is verified by real data from INFOCOM set and MIT set. The result shows that the mechanism can predict topology of PSN effectively.