Research And Application Of Spatial Situation Prediction Method Based On Graph Neural Networks

Nowadays,with the development of network technology,the research and application of Network Security Situational Awareness are getting more attention with the increasing importance of network security.Situational prediction and assessment,as a very important part of situational awareness,has long been an important research content in the field of network security.This thesis focuses on the research and implementation of the situation prediction algorithm,and proposes a spatial situation prediction method based on graph neural network.The method is divided into the following two algorithms:(1)Edge Connection Prediction Algorithm Based on Graph Convolutional Networks.Aimed at the problem is that in highly hidden networks,such as the dark web,it is difficult to obtain the topological relationship between nodes,even by injecting controlled nodes involved in the dark web packet forwarding process,also can get and controlled the node hop distance of 1 other network nodes,and to form the network topology range is very limited,even to inject more nodes,the correlation between nodes is also very difficult to find.In this application scenario,the core idea of the edge prediction algorithm is to use multiple subgraphs containing different number of edge links as a basis to predict the edges contained in the complete topological graph,and use local predictions to predict the overall,use known to perceive the unknown.In the dark web environment where it is difficult to obtain data,it is also possible to achieve complete supplementation from hidden data.(2)Situation Prediction Algorithm Based on Simplifying Graph Convolutional Network.This algorithm is the key research content of this thesis.Aiming at the problem that most of the current prediction algorithms only consider one single element or do not make full use of multiple elements,improvements and optimizations have been made in the algorithm.Process massive time-series data into graph structure,and use simplified graph convolutional network as a prediction tool,fuse the data in the two dimensions of time and space,and make full use of the correlation between the multiple elements of the node itself,and the mutual influence between nodes makes the model have the ability to extract the characteristics from temporal and spatial dimensions,and can quantify and calculate the node security situation value at the next moment,and then aggregate the situation value of each node to form a prediction of the overall network security situation,which improves the prediction accuracy and prediction efficiency,what is more,the prediction accuracy rate during the experiment can reach 95%.After the completion of the two spatial situation prediction algorithms,a Web-based network security situation warning system is designed and implemented.The system integrates the data acquisition module,data preprocessing module,data interaction module and deep learning module,and the deep learning module is mounted with the above two core algorithms.Real-time update and real-time prediction of network situation are realized,and the results of situation prediction are displayed by visualization technology.