Research Of Link Prediction And Routing Strategy In Opportunistic Network

Recent years, with the development of wireless communication technology and the emergence of large amount of smart mobile devices, research in OppNets(Opportunity Networks) has raised a wide concern in the academic. Studies in this thesis will expand on link prediction and routing methods in OppNets. The delivery rate of packets is an important indicator to meature the network performance, and the key to improve the delivery rate is how to conduct a proper routing decision which selects an appropriate forwarding opportunity or node. Most of the existing routing methods in OppNets follow a utility-based delivery strategy, which leads to a lack of consideration in choosing an optimal forwarding option and reflecting the dynamic and regularity of the OppNets. This thesis will introduce the research of routing in OppNets from two main aspects:the definition of delivery utility and the conduct of the optimal forwarding option.Firstly, OSR, an optimal stopping theory based routing method in OppNets is proposed, which aims to offer best options for the routing decision. In OSR, delivery utility of a node is defined as the average encounter interval to the destination node. A node with a series of observed delivery utility of the encounters will duplicate and forward a copy of the packet to an appropriate intermediate node according to the optimal stopping rule. OSR aims to minimize the expected average delivery delay by weighing the delivery utility benefits and observing cost. This optimal stopping model will receive an excellent performance in making the optimal decision, decrease blind forwarding behaviors and maxmize forwarding benefits with limited forwarding times. And in simulation, OSR has also achieved a better performance in delivery rate, average delay and cost, compared with other routing protocols.In OppNets, limited knowledge of the future links leads to blind and unpredictable packet forwarding behavior in routing decisions. To this end, this thesis proposed KRLP, a kernel regression link prediction method in OppNets. In KRLP, we first extract features of node pairs to capture the evolution of the local network topology over time. Then, we use kernel regression estimation method to model the hitorical evolution of the topology and ouput the probability of a future link. In the comparsion experiments, we eventually proved KRLP outperforms on prediction accuracy. With this link prediction method, we can well predict changes in netwok links and provide better guidance for opportunistic routing.Finally, LP-OSR, a link prediction based optimal stopping routing method in OppNets is proposed, as a combination of the two parts of the work. Using the result of link prediction as the delivery utility instead of the average encounter interval, the LP-OSR method can perform better than OSR. The new utility makes up for lack of the original one, better reflects the dynamic and regularity of links between nodes. And LP-OSR method overcomes two major problems in routing in OppNets:accuracy of the delivery utility and optimality of the forwarding option. Experimental results also demonstrate the performance of LP-OSR, show that this method can effectively improve the packet delivery rate, with no defects on delays and costs.