Abstract:

Hazard and operability analysis (HAZOP) takes the form of brainstorming, and it relies too much on expert experience. It is hard to share and reuse for the discussion results which were recorded in paper documents. Under the guidance of international standard IEC 61882, the ontology rules were developed through the perspective of dangerous events, and the model layer construction was completed after that. Then, based on the existing HAZOP analysis data, the attention mechanism was introduced on the BiLSTM-CRF model, which realized the automatic extraction of key HAZOP information and graph database was used for data storage to complete the construction of the data layer. Taking the HAZOP analysis report of oil equipment as an example, the HAZOP knowledge graph was constructed to verify the effectiveness of the construction method. It was proved that HAZOP knowledge graph can clearly demonstrate the relationship between HAZOP knowledge, quickly provide existing HAZOP information, assist the development of manual HAZOP, reduce labor and time costs, provide knowledge support for the production and maintenance of later projects, and realize the sharing and reuse of HAZOP information.

Key words: safety, hazard and operability analysis (HAZOP), knowledge graph, model, named entity recognition, neural networks

摘要：

危险与可操作分析方法（hazard and operability analysis，HAZOP）主要采用头脑风暴的形式，将讨论结果记录在纸质文档中，该方法过于依赖专家经验且大量现有的HAZOP信息并没有得到共享与复用。针对此问题，本文提出了结合自顶向下和自底向上的HAZOP知识图谱半自动构建方法。在解析HAZOP信息基础上，结合HAZOP分析国际标准IEC 61882，以危险事件角度设计了HAZOP本体规则，完成了模式层的构建；以现有HAZOP数据为基础，在BiLSTM-CRF命名实体识别模型基础上引入了Attention机制，实现了关键HAZOP信息的自动提取并利用图数据库进行数据存储，完成了数据层的构建。以某油品装置HAZOP分析报告为例构建HAZOP知识图谱，验证了构建方法的有效性。结果表明HAZOP知识图谱能够清晰地展示HAZOP知识之间的联系，快速提供现有HAZOP分析信息，辅助人工HAZOP的开展，降低人工成本与时间成本，也为后期项目的生产与维护提供知识支撑，实现了HAZOP信息的共享与复用。

关键词: 安全, 危险与可操作分析方法, 知识图谱, 模型, 命名实体识别, 神经网络