The Research And Application Of Knowledge Graph In The Field Of High-speed Railway Turnout Fault Maintenance

As the key ground signal equipment in the process of high-speed train operation,the high-speed railway turnout will have a great impact on the efficiency and safety of high-speed railway operation once it breaks down in the actual process of high-speed train operation.Therefore,in the maintenance of high-speed railway turnout need to give full play to technical advantages,relying on big data research high-speed railway turnout fault maintenance methods,to assist high-speed railway turnout maintenance personnel to carry out fault maintenance,in order to improve maintenance efficiency.Many years of high-speed railway operation have produced a large number of text data of high-speed railway turnout faults recorded in natural language.However,the current research on the unstructured text data in the field of turnout fault maintenance fails to fully excavate the rich fault knowledge contained in the data.Only rough positioning based on the fault diagnosis results can not show the complex relationship between the various elements of the turnout fault.In order to solve the above problems,this paper applies the knowledge graph technology to the field of turnout fault maintenance.Based on the text data set of high-speed railway turnout fault,the knowledge graph technology is used to dig out the internal relations among the elements of turnout fault,so as to improve the maintenance efficiency.The main research contents of this paper are as follows:（1）A corpus for entity recognition of high-speed railway turnout faults is constructed.In the early stage of named entity recognition,marked data sets are needed,but currently there is no publicly marked data set in the field of high-speed turnout fault maintenance.Therefore,the primary task of this paper is to build a corpus of high-speed railway turnout fault.Firstly,eight types of fault entities are defined.Then,the"Genie Annotation Assistant"software is used to carry out text sequence annotation on the high-speed turnout fault data set.Finally,the data is processed into the BIO format that the entity recognition model allows to input,so as to build the corpus of high-speed railway turnout fault entity recognition.（2）The domain entity identification model of high-speed railway turnout fault is established.A BERT-Bi LSTM-CRF named entity recognition model is established by using the constructed corpus of high-speed railway turnout faults as the experimental data set.The model firstly obtains the high-quality quantifier vector representation of contextual semantic information through bidirectional encoder representations from transformers model（BERT）.Secondly,the word vector obtained by BERT is input into bi-direction long short term memory（Bi LSTM）neural network to learn contextual semantic features and score all kinds of tags.Finally,the scoring tag sequence is added with constraints through conditional random field（CRF）model to output the optimal tag sequence.The experimental results show that the accuracy rate,recall rate and F₁value of the proposed model are 93.41%,92.93%and 93.17%,respectively.Moreover,the comparison experiment shows that the proposed model is superior to other entity recognition models in the above three indexes.（3）A rule-based entity relation extraction method for high-speed rail turnout faults is proposed.In order to improve the quality of domain knowledge graph,a rule-based relationship extraction method is adopted.Under the guidance of experts in the field of turnout maintenance,a rule template for relationship extraction is developed according to the characteristics of the fault data set.Finally,different experts in the field are invited to judge the relationship between entities by confidence scoring according to the established rules.Although this method is manual relationship extraction,the extraction rules and relationship judgment are guided by domain experts,and the extracted relationships are highly correct,which can significantly improve the knowledge quality of knowledge graph.（4）Build the knowledge graph of high-speed railway turnout fault maintenance.The constructed domain knowledge graph is divided into four parts:named entity recognition,relationship extraction,knowledge fusion and knowledge storage,In the part of knowledge fusion,the key technology of knowledge fusion entity alignment is applied,by using the algorithm based on text semantic similarity and semantic similarity,the alignment of synonymous fault entities with similar structure and fault entities with similar structure but different semantics is achieved respectively.According to the results of entity recognition,relationship extraction and knowledge fusion,the knowledge graph of high-speed railway turnout fault maintenance field is constructed,and the knowledge storage and visualization display of the graph are carried out through the Neo4j diagram database.Finally,an automatic question answering system based on the knowledge graph of high-speed railway turnout fault maintenance field is designed.The system is developed under the web framework based on Fast API,and realizes the function of question analysis,query statement generation,database connection,answer conversion and interface display.Users can use this system to complete the identification of fault types,find the internal links and hidden knowledge between faults,and provide suggestions for maintenance measures.The system effectively assists the staff in the maintenance of high-speed rail turnout faults,and promotes the field work to operate more efficiently.