Link Prediction Method Based On Bayesian Recurrent Neural Network

Opportunistic Network(ON)is a new type of mobile Ad Hoc networks that establishes communication through the movement of network nodes without the need of complete communication path between the source node and the target node.The topology of the ON changes frequently and does not require full network connectivity,which is more in line with the actual networking requirements.Link prediction is a research difficulty in ONs.The goal of link prediction is to estimate the possibility of future links among nodes using known network information and the attributes of the nodes.Link prediction can dig out the potential relationships between nodes by analyzing the known link information,which is helpful to obtain the evolutional law of the network structure,and provide decisions for ON routing and forwarding,so as to provide better services for the upper application.According to the time-varying characteristics and the nodes' mobility of ON,this thesis considers the characteristics of link over time in network evolution,we proposed a novel link prediction method based on the Bayesian recurrent neural network(BRNN-LP)framework.This thesis analyzes the characteristics of the opportunistic network in different application scenarios,and obtains factors that affect the generation of links between nodes,including the location of nodes and the correlation between nodes.The time series data of the opportunistic network is divided into a series of network snapshots,each of which contains connection information and location information.Connection information and location information are used to define correlation and spatial location,the correlation and spatial position are used to construct the vectors in each snapshot,which characterize the links between the nodes.At the same time,considering the historical information of the link in the time dimension,spatiotemporal vector sequences are constitute by the vectors of multiple network snapshots,which are used as the input of the BRNN-LP model.Benefiting from the BRNN's ability of extracting the features of time series data,the correlation between spatiotemporal vector sequence and node connection states is learned,and the law of the link evolution is captured to predict future links.In this thesis,MIT reality(MIT),an ON real dataset,is used for experiments.MATLAB is used to process the data.The Accuracy,Area under the Receiver Operating Characteristic Curve and Precision,are adopted as evaluation indices,and zhusuan as an experimental platform.By comparing the accuracy and precision of multiple sets of BRNN-LP prediction models under different parameters,which to determine the number of iterations,time slice length,and input sequence length.Compared with prediction methods based on common neighbors,Adamic-Adar,resource allocation,Katz,local path and Bayesian networks,the BRNN-LP prediction model has better precision and accuracy.Compared with the link prediction model based on support vector classifier,recurrent neural network and graph neural network,the performance of BRNN-LP prediction model is more accurate and stable.