Research On Knowledge Graph Completion Algorithm For High-Speed Railway Operation And Maintenance

At present,the Knowledge Graph(KG)in general field is relatively popular,but the Knowledge Graph in specific field still needs continuous development.For example,the Knowledge Graph in high-speed railway operation and maintenance field can provide rich information and technical support for prevention,emergency response and post-treatment,and has a broad prospect.Knowledge Graph Completion(KGC)is necessary for the automatic construction and verification of Knowledge maps in the field of high-speed railway operation and maintenance.It is an important means to discover new Knowledge and is widely used in advanced tasks of Knowledge maps.But the current study of Gao Tieyun d domain knowledge map completion content is less,the existing model used for the field of Chinese map completion work still exists many problems:(1)based on the translation of the translation distance model is strong interpretability,the advantages of simple operation and few parameters,but can lead to low quality of learning embedded vector;(2)The CNN-based model improves the performance of the translation model to a certain extent,but too many parameters increase the complexity of the model,and the usability is poor for large-scale knowledge maps.(3)Although GNN model based on graph neural network structure aggregates effective information from the entity and the relationship near the entity to update the embedded vector of the iterative central entity,it has a significant shortcoming,that is,it does not distinguish the importance of the contribution of the relationship,and uses fixed parameters to learn different relationship information.Based on the above analysis,this paper proposes a generalized attention completion model based on entity relation cross aggregation in view of the shortcomings of existing models.Taking traction motor,a key component of high-speed railway EMUS,as an example,this paper studies the knowledge completion problem in high-speed railway operation and maintenance in the Chinese field,and makes improvements and innovations in the following aspects:(1)Aiming at the lack of data information in the field of high-speed railway operation and maintenance,this paper constructed a new knowledge graph dataset TMFD(database with traction motor fault relationship types and repair strategies)by collecting traction motor fault information,which is suitable for the study of knowledge graph in the field of high-speed railway operation and maintenance.(2)This paper proposes a new relational messaging mechanism,which transfers the relationship aggregation information between entity nodes and relationship edges in turn on the knowledge graph,so as to reduce the number of storage entities embedded,reduce the storage efficiency,more in line with the nature of knowledge flow and enhance the interpretability under the same computational complexity as transmitting information along entity nodes.(3)In this paper,a generalized attention completion model based on cross aggregation of entity relations is proposed(Generalized attention entity relation cross aggregation completion model,GERCA).Two neighborhood topologies are added for a given central entity: 1)the entity itself is represented by the local subgraph of the entity,and more neighborhood information is captured to form relational context information;2)Determine the relative positions of predicted entities in the knowledge graph through relationship path information to solve the problem that relationship types are not evenly distributed on the graph but spatially correlated with each other.Experimental results show that GERCA is advanced in public data set and domain data set.