Research On Reasoning Method Of Aviation Safety Events Based On Knowledge Graph

With the rapid development of the civil aviation industry,ensuring flight safety,improving flight capabilities,and reducing the incidence of aviation safety incidents have always been important aspects of aviation safety management.At present,the causes of aviation safety incidents are complicated and diversified.The prediction of development trends also faces new challenges.Knowledge graph is a large-scale fact database with graph structure.It can learn the characteristics of historical events by multiple factors and attributes to infer the rules of event development.This paper focuses on the method and application of knowledge graph reasoning for aviation safety events in the following aspects:Aiming at the problem that the current traditional convolutional neural network model processes each triplet information separately in the process of knowledge graph link prediction,and cannot effectively learn the rich semantics and potential relationship information existing in the entity neighborhood.A knowledge graph link prediction method(AttGCN)based on attention mechanism and convolutional neural network is proposed.This method first obtains all the triples information associated with each entity and uses the convolutional neural network to learn the characteristics of the triples.Then use the attention mechanism to measure the degree of influence of different neighborhood triples on the target entity,and use the attention weight to update the feature vector of the entity,so that the entity feature vector fuses all neighborhood entity features and their corresponding relationship features.Finally,the convolutional neural network is used to interactively learn the head entities and the relationship features using filters based on specific relations,and perform link prediction on all the tail entities at the same time.The experimental results show that the AttGCN method can effectively learn the neighborhood triples of each entity in the knowledge map,and the MRR and H @ 10 indicators on the WN18 RR dataset are increased by 29.1% and 9% compared to TransE,and in FB15k-237 The MRR and H @ 10 indicators on the dataset have increased by 15.2% and 18.1% compared to TransE.Take the 2117 accidents in the aviation accident investigation and tracking report and the voluntary aviation safety report as the basic data,and generate a knowledge map,and use the TransE algorithm to vectorize the entities and relationships;then use AttGCN method to learnthe basic information and The information of all triples of risk factors,because the potential relationship between multiple risk factors and event results in historical events is comprehensively considered,and link prediction is performed through multi-event knowledge,the characteristics and evolution rules of similar events can be inferred.Thus,serious consequences may be found,which provides a basis for the prediction and early warning of aviation safety events.Compared with TransE,the MRR and H @ 3 indicators of AttGCN method on the knowledge map of aviation safety incidents are improved by 13.4%and 13.2%.