Research On The Construction Method Of Knowledge Graph For Urban Rail Train Fault

The urban rail transit system is huge and complex,and has a large number of subsystems and equipment.When the system is operating normally,the automation level is very high.In the event of a sudden fault during operation,the normal state is recovered manually in the traditional fault handling method,which has problems such as numerous abnormal disposal operations,lack of sufficient handling experience for fault handling personnel,and low degree of automated information processing.Therefore,the service capacity of urban rail transit is affected in this way to a large extent.In addition,urban rail transit operating companies have collected a large amount of valuable train fault records from the site,which described in detail the occurrence and development process of train fault,including some important information(e.g.,fault phenomenon,location,handling method and impact).However,these fault texts are recorded for the archive purpose instead of the analysis purpose,and there is still a lack of effective analytical methods for these data.Based on the above issues of fault handling,this thesis takes the unstructured texts of historical train faults in urban rail transit as the research object and studies the construction method of train fault knowledge graph.This thesis deeply excavates the key elements of fault events and the correlation of each element,which provides auxiliary support to fault handling personnel in urban rail transit,deals with the increasingly complex urban rail transit emergencies and improves the efficiency of fault handling.The research work of this thesis mainly includes:(1)According to the descriptive characteristics of the train fault text,combined with expert knowledge,the entity type and relationship type of the train fault are defined.Aiming at the current situation of lack of training corpus in the field of urban rail transit and referring to the standard dataset of the public domain knowledge graph,a train fault entity annotated dataset for entity extraction,a train fault knowledge representation dataset for knowledge representation,and a train fault reasoning dataset for knowledge reasoning are constructed.In addition,the train fault knowledge graph is stored and visualized in the Neo4 j graph database.(2)Aiming at the problem that the non-standard language description structure and the unclear entity boundary of the train fault entity,a word-enhanced train fault entity extraction method combining adversarial training and Lattice long short-term memory network is proposed.The method is verified based on the constructed dataset,which proves the effectiveness of the entity extraction method.(3)Aiming at the problem that train fault knowledge triples belong to the discrete symbolic representation and are difficult to apply neural network for further analysis,a train fault knowledge representation method based on knowledge graph attention network model is proposed.The method is verified based on the constructed dataset,which proves the effectiveness of the knowledge representation method.(4)Aiming at the problem that the train fault knowledge graph may be incomplete and include implicit regular knowledge and semantic connections,a train fault knowledge reasoning method based on the relational graph convolutional neural network model is proposed.The method is verified based on the constructed dataset.The train fault knowledge graph is also completed,which proves the effectiveness of the knowledge reasoning method.The research shows that a train fault knowledge graph that is relatively complete,intuitive,easy to store and mine can be obtained through the construction method of the train fault knowledge graph for urban rail transit proposed in this thesis.