Abstract:In order to explore the evolution law of pocket switched network (PSN), the link prediction which is a part of network behavior prediction problem in PSN was studied in this paper. PSN has lots of features such as node mobility, intermittent connection and high latency. The challenges of link prediction are the opportunistic connection and the time-varying topology. The key to obtain high-quality link prediction is how to obtain the attributes of nodes comprehensively. A link prediction method was proposed in the paper, which is based on learning automaton and firefly algorithm (FA-LP). The learning automaton was employed to cluster nodes adaptively so as to complete the community division of the network. The node community attribute influence coefficient and mobile behavior influence coefficient were defined to construct the similarity index which reflects the community attributes, node mobility and intermittent connection between nodes. After fusing the index with CN, RA, AA, etc., a similarity vector of node pairs was achieved. Taking the advantages of analyzing time series of autoregressive integrated moving average model, the evolution law of the vector sequence was extracted. The binary classifier optimized by firefly algorithm was constructed to predict the connection of node pairs at the next moment. The experimental results on INFOCOM2006 and MIT datasets show that the proposed method has better accuracy and better stability than the ones of RBM and week estimators.