核动力装置报警信号知识图谱构建方法研究  

Research on construction method of nuclear power plant alarm domain knowledge graph

作　　者：王迪 余刃[1] 彭俏[1] 王天舒 宋霄森 WANG Di;YU Ren;PENG Qiao;WANG Tianshu;SONG Xiaosen(College of Nuclear Science and Technology, Naval University of Engineering, Wuhan 430033, China)

机构地区：[1]海军工程大学核科学技术学院,武汉430033

出　　处：《兵器装备工程学报》2022年第S01期310-315,共6页Journal of Ordnance Equipment Engineering

基　　金：院校“双重”建设项目“核动力+人工智能”;国防科工局核能开发科研项目。

摘　　要：研究了核动力装置报警领域知识图谱的构建方法,以便能够从大量核动力装置技术文档中自动提取报警领域知识,结合专家知识,形成可用于报警信号抑制和确定根报警的知识图谱。首先针对核动力装置报警信号的特点,确定其知识图谱的构建流程,完成其知识图谱模式层本体概念的定义,以及本体模型的构建。然后利用基于BILSTM-CRF算法实现核动力装置报警领域知识的命名实体识别,利用GLU Dilated CNN-Attention关系三元组抽取方法,实现对于技术文档中所包含的报警领域实体间关系的自动抽取,完成数据层构建,最终形成报警领域知识图谱,为实现基于知识图谱的报警信号抑制和根报警判定奠定基础。The alarm signal inhibition and root alarm determination to the nuclear power plant(NPP)under accident condition is of great significance to ensuring the plant operation safety.The construction method of the NPP alarm domain knowledge graph is studied in the paper,so as to automatically extract the alarm domain knowledge from a large number of technical documents,and combine the expert knowledge to form a knowledge graph that can be used to determine the root alarm signal.Firstly,according to the characteristics of the NPP alarm signal,the construction process of its knowledge graph is determined,the ontology concepts of the knowledge graph model layer are defined,and the ontology model is formed.Then,the BILSTM-CRF algorithm is utilized to realize the recognition of named entities in the alarm domain knowledge of NPP,and the automatic extraction of the relations between the alarm domain entities contained in the technical documents is realized by a relation triple extraction method,that is,the GLU Dilated CNN-Attention method.The data layer is then constructed,and the knowledge graph of the alarm domain is completed.The result can be used for the alarm signal inhibition and root alarm determination based on knowledge graph.

关 键 词：核动力装置 知识图谱 报警信号 关系抽取 实体抽取 图数据库 

分 类 号：TL361[核科学技术—核技术及应用]