燃气轮机故障知识图谱构建方法与应用研究  

Research on construction method and application of knowledge graph for gas turbine fault

作　　者：王明达[1] 吴志生 朱光辉 李云飞 张榜 WANG Mingda;WU Zhisheng;ZHU Guanghui;LI Yunfei;ZHANG Bang(College of Mechanical and Electrical Engineering,China University of Petroleum,Qingdao Shandong 266580,China;PipeChina Shandong Branch Co.,Jinan Shandong 250002,China)

机构地区：[1]中国石油大学(华东)机电工程学院,山东青岛266580 [2]国家石油天然气管网集团有限公司山东省分公司,山东济南250002

出　　处：《中国安全生产科学技术》2023年第11期121-128,共8页Journal of Safety Science and Technology

基　　金：国家自然科学基金项目(52075549)。

摘　　要：为更好地管理和利用燃气轮机故障知识,提高故障诊断的效率,提出燃气轮机故障知识构建方法。首先,根据故障文本知识特点,并结合专家知识设计燃气轮机故障文本知识本体概念模型。其次,采用BERT-BiLSTM-CRF、BERT-BiLSTM-Attention等深度学习模型实现燃气轮机故障命名实体识别及实体关系模型训练,在引入BERT模型获取动态字符后,相比BiLSTM-CRF模型,实体识别模型的综合评价指标F1提高7.98个百分点,相比Word2Vec特征表示方法提高0.89个百分点;在关系抽取中将BiLSTM-CRF抽取模型中的CRF模型替换为Attention模型并引入BERT模型后,综合评价指标F1提高8.49个百分点。最后,使用Neo4j图数据库完成知识的存储工作,并将知识图谱用于辅助故障分析。研究结果表明:知识图谱技术能够实现对燃气轮机组成部件故障先验知识的利用以及对故障原因的解释。研究结果可为燃气轮机故障诊断提供知识支持。In order to better manage and utilize the gas turbine fault knowledge and improve the efficiency of fault diagnosis,a construction method of gas turbine fault knowledge was proposed.Firstly,the ontology conceptual model of gas turbine fault text knowledge was designed according to the characteristics of fault text knowledge and combined with expert knowledge.Secondly,the deep learning models such as BERT-BiLSTM-CRF and BERT-BiLSTM-Attention were used to realize the named entity recognition and entity relationship model training of gas turbine fault,and after introducing into BERT model to obtain the dynamic characters,the comprehensive evaluation index F 1 of the entity recognition model was improved by 7.98 percentage points,and compared with the Word2Vec feature representation method,it was improved by 0.89 percentage points.After replacing the CRF model in the BiLSTM-CRF extraction model with the Attention model and introducing the BERT model in the relational extraction,the comprehensive evaluation index F 1 improved by 8.49 percentage points.Finally,the Neo4j graph database was used to complete the knowledge storage work,and the knowledge graph was used to assist fault analysis.The results show that the knowledge graph technology can realize the utilization of priori knowledge for gas turbine constituent component faults and the explanation of fault causes.The results can provide knowledge support for gas turbine fault diagnosis.

关 键 词：燃气轮机 故障诊断 知识图谱 实体识别 图数据库 

分 类 号：X913.4[环境科学与工程—安全科学] TK478[动力工程及工程热物理—动力机械及工程]