Research On Opportunistic Network Link Prediction Based On CNN+LSTM

Opportunistic Networks is a kind of delay tolerant network(DTN).Because the nodes in the Opportunistic Networks can move,disconnect and connect frequently,there is no complete link between the end-to-end,so the Opportunistic Networks mainly relies on the way of 'store carry forward'.In order to improve the routing and forwarding rate of Opportunistic Networks and reduce the cost in the network,the historical information of Opportunistic Networks is used to predict the possible future node connections,and the data information can be forwarded to the maximum extent with the minimum cost,making link prediction become the key problem of Opportunistic Networks forwarding data,which is also one of the hot issues in academic researchIn recent years,deep learning has made great achievements in graphics and image,natural language processing,speech recognition and other fields.At the same time,deep learning technology is also widely used in the field of link prediction.It is also a common method to mine unknown node connections in the network structure by using the strong feature extraction ability of deep neural network.Because the link structure of Opportunistic Networks has the characteristic of time sequence,this paper first analyzes the research status of link prediction model at home and abroad,describes the characterization,communication characteristics and similarity index of Opportunistic Networks,and gives the data preprocessing method and evaluation index of this paper.On the basis of in-depth analysis of RNN and LSTM deep learning framework,using the feature extraction idea of RWM model for reference,this paper uses the feature extraction ability of CNN-Conv1 D extracts the features of time series,and proposes the fusion method of traditional features and historical information features of Opportunistic Networks;using the memory function of Long Short-Term Memory(LSTM)to time series,applying LSTM to the link prediction of Opportunistic Networks,a link prediction model of CNN + LSTM is proposed.The experimental results show that the algorithm is effective.Compared with LSTM and RNN,it improves the prediction accuracy and shortens the training period.The main work and innovation of the paper are as follows:(1)The preprocessing method of the Opportunistic Networks data set is proposed,and the two major data sets Infocom05 and MIT Reality commonly used in the Opportunistic Networks are used to perform trapping and deduplication operations,and the data set is cleaned without losing the original information as much as possible.After a clean data set is obtained,it is time-serialized and sliced again to finally obtain a usable data set.(2)Propose an Opportunistic Networks communication feature extraction and feature fusion method;this paper adds two features of the communication duration and communication times on the basis of the Opportunistic Networks communication feature TCL to form a new communication feature TCLC.And the experimental results show that the characteristics of TCLC proposed in this paper are better than TCLC.(3)For the first time in Opportunistic Networks link prediction,an Opportunistic Networks link prediction model and algorithm based on CNN+LSTM is proposed;on the basis of communication features TCLC,CNN's feature extraction and dimensionality reduction capabilities are used to extract features from the data.Then the data is input into LSTM for training and prediction.The final experimental results show that the CNN+LSTM prediction model has achieved good prediction accuracy on the Infocom05 and MIT Reality datasets.