Abstract: The conventional knowledge graph can only deal with binary relations,while knowledge of bridge crane faults contains a large number of multi-nary relations of “multiple phenomena,multiple causes,and multiple methods”.If forced to transform,the integrity of the relations will be destroyed,causing serious information distortion.To deal with such complex multi-nary relational knowledge to ensure integrity,the Knowledge hypergraph was proposed and a hypergraph embedding-based representation method for multi-nary relational knowledge of bridge crane faults was designed.Through sorting the correlation among phenomena,causes,methods and other data in the driving fault sheet,a driving fault ontology model suitable for characterizing the multi-nary relation  was constructed,which was taken as the schema for establishing the knowledge hypergraph of bridge crane faults.Based on the BERT model in natural language processing and the hypergraph convolutional network,the embedding representation of fault knowledge was obtained,hence similar fault retrieval could be carried out.By exploiting the fault sheets of bridge cranes collected from a steel factory,the effectiveness of the proposed method was verified.

Key words: knowledge hypergraph, multi-nary relation, knowledge representation, graph embedding algorithm, bridge crane fault

摘要： 鉴于常规知识图谱仅能处理二元关系,而故障知识包含大量“多现象—多原因—多方法”的多元耦合关系,强制转化将会破坏关系的完整性,造成严重的信息失真,为采用知识超图处理此类复杂多元关系以保证数据的完整性,设计了一种基于超图嵌入的行车故障多元关系知识表示方法。通过梳理行车故障单中现象、原因、方法等数据之间的多元关联,构建适用于表征多元耦合关系的行车故障本体模型,以该本体模型为知识超图的模式层建立行车故障知识超图;基于BERT模型和超图卷积网络获取故障知识的嵌入向量表示,并实现了相似故障检索。最后,以上海某钢铁公司收集的行车故障调查单为实例,验证了所提方法的有效性。

关键词: 知识超图, 多元耦合关系, 知识表示, 图嵌入算法, 行车故障