The Research Of Link Prediction Technology In Hybrid Opportunistic Networks

A large number of smart devices with short-range wireless communication capabilities have promoted the research and application of opportunistic networks. However, in many pr actical applications, because of nodes' high mobility and network's sparseness, signal atten uation and so on, links in opportunistic networks are generally highly dynamic, which cau se problems such like low message delivery success ratio and high delivery delay. These pr oblems restrict the practicability of opportunistic networks. We consider deploying infrastruct ures with large storage in opportunistic networks, which can be fixed or mobile to form a hybrid opportunistic network to reduce the dynamic nature of the network, and improve tra nsmission performance. Besides, networks like social networks on campus, on board self-org anizing networks with roadside nodes and fixed-route bus network etc. can actually be seen as a Hybrid structure. This thesis mainly researches how the deployment strategies of infr astructures will impact on the Network dynamics, the link dynamics and prediction, and ho w it will improe the transmission efficiency of the networks. The main contents are as foll ows:(1)Research the statistic characteristics in hybrid opportunistic networks. In order to res earch that how the quantity and density of infrastructures will impact on the networks, this thesis conducts a series of experiments by statistics and analysis and concludes the network dynamics and ink characters.The results indicate that:the the contact-inter CCDF curve ha s the "heavy tail" character, furthermore, the quantiy and density of infrastuctures can affec t the networks a lot:only a few ones which are deployed correctly can highly improve the performance, and after that, diminishing marginal utility dcreases.(2)Propose a combined link prediction algorithm in hybrid opportunistic networks. The algorithm applies periodic pattern mining based prediction to the nodes which contact frequ ently and periodically with each other; it applies machine learning based prediction to the nodes which contact frequently but not periodically with each other; it applies complex net work based prediction to nodes which don't contact frequently. The experiments'result indi cates that the combined algorithm has better performance than other single ones and it can predict more links.(3)Research message forwarding based on link prediction in hybrid opportunistic network s. Based on (1), propose "Prophet-F" and strategies of deploying infrastructures; baesd on (2), propose an opportunistic routing based on link prediction algorithm; conduct simulation experiments are on real dataset. The results indicate that it's best to deploy the infrastruct ures in hot areas, and in this case Prophet-F improve the performance by 15%-20%; besid es,the performance of the routing based on combined link prediction is better than others i ncluding Prophet-F.Results of this thesis have great significance for improving the availability of opportuni stic networks. Based on the results, transmission performance of community networks, vehic ular ad-hoc networks and so on can be improved.