列控车载设备故障知识图谱的重叠三元组联合抽取方法OA

[Objective]As the core of the Chinese Train Control System(CTCS),on-board equipment utilizes the knowledge graph to analyze fault cases and realize rapid and accurate identification of fault causes and treatment solutions,which is crucial for railway safety and transportation efficiency.[Methods]To address issues such as error accumulation and poor interactivity caused by traditional pipeline methods in knowledge extraction,a joint extraction method for overlapping triples based on a sequence labeling strategy was proposed.Firstly,based on the defined entity and relationship types,a"position-entity-relationship-role"sequence labeling strategy was employed to achieve the joint labeling of overlapping triples.Secondly,the FMBCR(FaultRoBERTa-MHSA-BiLSTM-CRF incorporating regularization methods)model was constructed to accurately extract the global semantic information from fault texts of CTCS on-board equipment,and its predictive performance was optimized by incorporating a KL(Kullback-Leibler)divergence loss algorithm based on the regularization module READ(REgularization method with adversarial training and dropout).Finally,experimental analysis and knowledge graph visualization construction were conducted,using fault data of CTCS on-board equipment as an example.[Results]Through sufficient comparative experiments and ablation experiments,the results showed that the proposed method realized the joint extraction of overlapping triples under small-sample conditions,and the extraction performance was significantly improved compared with traditional methods.Compared with the widely used CASREL joint extraction model,the FMBCR model improved the precision,recall,and F1 score by 3.0%,4.3%,and 3.7%,respectively.[Conclusion]The research findings provide support for advanced applications such as intelligent information retrieval and decision-making assistance for fault maintenance of CTCS on-board equipment.