Research On Representation Learning Method Of Electromagnetic Situation Knowledge Graph

With the rapid development of wireless communication technology,the use of a large number of wireless devices not only facilitates people's life,but also leads to more and more complex electromagnetic environment.In response to the massive and complex electromagnetic situation data,how to store and manage the data and design effective analysis methods,quickly extract useful data information from the complex data,it is very important to accurately evaluate the electromagnetic situation information.In recent years,the emerging knowledge graph technology can quickly build a structured knowledge base,clearly describe the relationship and conceptual attributes between different entities,and provide very rich feature information for the management,analysis and situation awareness of electromagnetic data.This thesis studies the representation learning of knowledge graph,puts forward some innovative improvements combined with the structural characteristics of electromagnetic situation data,and applies them to the electromagnetic situation knowledge graph system.The research contents and main contributions of this thesis are as follows:(1)To solve the problem that the Conv E model is not interactive enough in the knowledge graph representation learning process and that the feature information extraction effect is not ideal,a knowledge representation learning model based on Inception module is proposed,called Inception E.Firstly,an input layer processing module is defined.The input entities and relations are stitched,arranged,remolded and merged to improve the interaction times of entities and relations.Then,using the idea of Inception network for reference,the convolution kernel of different sizes is used for calculation to increase the receptive field of the network and extract feature information more efficiently.To prevent the loss of original feature information,this thesis also introduces residual network to further optimize the model.The Inception E model has been tested on multiple public datasets and electromagnetic data,and the experimental results are superior to the current mainstream algorithms.(2)To solve the sparse problem of electromagnetic situation knowledge graph,a representation learning algorithm for sparse knowledge graph is proposed in this thesis.Firstly,the Inception E model was used to calculate the entity similarity of the sparse knowledge graph,and the sparse knowledge graph was normalized by adding a confidence edge.Then,the graph attention network and the cyclic neural network are used to encode the entities and relations of the normalized knowledge graph respectively.Finally,the Dist Mult network is used to decode the entities and relations to obtain the learning result of knowledge representation.The effectiveness and practicability of the proposed algorithm are proved by comparing the relevant data sets with the current mainstream algorithms.(3)Based on the application of electromagnetic situation,a complete electromagnetic situation knowledge graph system is developed by collecting and sorting electromagnetic situation data.The system provides the function of electromagnetic knowledge graph management,and completes the link prediction,triple classification application,and electromagnetic situation analysis combined with representation learning algorithm.The system designed in this thesis has passed the relevant experimental tests and can be used in the future electromagnetic situation analysis.