Research On Key Technology Of Knowledge Graph Construction And Its Application For Intelligent High-speed Railway Safety Assurance

Safety is the lifeline of railway transportation production and the premise of promoting intelligent development of railway.Driven by booming technologies such as big data and artificial intelligence,the intelligent development of railway has become the direction and goal of the current railway industry.High-speed railway has been in operation for more than 10 years,and is now transforming from large-scale construction to comprehensive safety operation and maintenance.With the rapid expansion of the scale and volume of high-speed railway,the business information system has become more and more complex,and various safety problems have become diversified and complicated.As the service life of various high-speed railway equipment increases,the infrastructure gradually enters the maintenance cycle,and the mass implementation of advanced repair of EMU increases the pressure of safety guarantee and operation and maintenance of high-speed railway.In recent years,intelligent high-speed railway attaches great importance to the construction of security capability,and gradually builds a number of safety monitoring and testing systems such as comprehensive testing vehicles,works,electricity and power supply,accumulating massive structured,semi-structured and unstructured monitoring data.Each business unit based on the monitoring and testing data on the safety equipment related data analysis,relying on traditional knowledge engineering and expert system based on data driven to solve the key problems,but its application limits in the rules clear,clear boundary and closed application scenarios,strongly depends on the expert experience,it is difficult to deal with open era of big data is applied to the large-scale application of demand.In particular,the lack of processing ability for unstructured complex data restricts knowledge precipitation,application and sharing in the field of intelligent high-speed rail safety.Knowledge map technology relies on semantic friendly technical advantages such as rich,excellent quality and structure in the vertical layout,based on rich entities,concept,attribute and relation triples semantic knowledge to more complex modal data expressed by the closer to the world in the form of human cognition,this gives intelligent high-speed rail towards the development of more secure and reliable target provides a new solution.Therefore,in this paper,based on the current intelligent high-speed rail safety operation situation and existing problems,with the aid of knowledge atlas technology advantage,intelligent high-speed security field actual business needs as the guidance,to railway safety risk,safety concerns,such as failure safety data driven,knowledge map in the financial,medical,electricity,and other areas of the vertical ground application experience,Carry out in-depth research on the application of intelligent knowledge graph service for intelligent high-speed railway security assurance.The main work and contributions of this paper are as follows:(1)The construction and application method of knowledge graph oriented to intelligent high-speed railway security assurance is proposed,and the top-level architecture of knowledge graph oriented to intelligent high-speed railway security assurance is designed.Through the research on key technologies such as knowledge extraction and knowledge visualization for railway safety assurance,a method and algorithm chain for the construction and application of railway safety-oriented knowledge map is proposed,which provides theoretical and technical support for the construction of railway all-domain knowledge map.(2)A security "knowledge graph + event graph" bisspectral fusion method is proposed for the security guarantee of intelligent high-speed railway.In order to further explore the deeper correlation between security incidents and faults,based on the security domain entity graph and the strong dynamic characteristics of security data,a security event graph was constructed to reveal the evolution rules and logic of security incidents.By focusing on the causal,sequential and temporal relationships of safety incidents,the railway safety incident atlas analyzes the evolution law,mode and context of accidents and faults,accurately matching the application of intelligent service oriented to intelligent high-speed rail security,and providing decision-making assistance for the effective prevention and avoidance of safety incidents.(3)A railway safety entity recognition model based on ELMO-BI-LSTM +CRF is proposed.The data was vectorized by ELMO model,and the preprocessed results were input into BI-LSTM model.At the same time,CRF conditional random fields are used for global feature learning of data samples.By adding dynamic features and word character features,the entity extraction method makes up for the lack of character vector specificity of railway safety data,realizes the feature extraction of safety accident faults,and improves the level of complex multi-word recognition in railway safety field.The accuracy rate P,recall rate C and F-score are all above 80%,which verifies the effectiveness of the proposed method.(4)Propose an entity relation extraction method integrating Transformer multi-self-attention mechanism.In order to highlight the features of the key semantic information in the training corpus and fully understand and recognize it,Attention mechanism is introduced to label the tendentious features of the target sample data in the entity relation extraction task.In this method,the results are input into bi-LSTM neural network through feature screening of attention mechanism,which reduces irrelevant information existing in context features and achieves accurate identification of entity relationship of railway safety accident fault samples.(5)A method of extracting fault event causality based on stacked CNBiGRU-CRF combination model is proposed.The causality of railway safety was divided into explicit and implicit extraction tasks.CNN convolutional neural network and BI-LSTM bidirectional gated cyclic unit were used to extract global and local features of training sample data,and CRF random vector field model was used to complete sequence labeling tasks.The experimental results show that the model is feasible in complex event-oriented causality extraction.