Abstract:

Many human-based HAZOP (hazard and operability analysis) results were stored as paper documents. It is difficult to reuse and share the results. At the same time, the computer-aided HAZOP results can only be recognized by the special software. The results are also difficult to reuse and share. For the problem, the HAZOP information standardization framework was proposed. The HAZOP standard information model including the model structure, definitions of the elements and relationships of elements was proposed basing on knowledge ontology and HAZOP international standard IEC 61882. The HAZOP information standardization approach was proposed based on the model. Every HAZOP record was tagged, trained and identified using BiLSTM neural network in the approach. The human-based HAZOP results can be identified and standardized automatically by the approach. The framework was used for HAZOP of an oil synthesis equipment. It was proved that the HAZOP results can be transformed to the standard form automatically using the framework. The HAZOP information can also be reused and shared easily.

Key words: safety, knowledge ontology, hazard and operability analysis (HAZOP), standardization, model, neural networks

摘要：

大量的危险与可操作性分析（hazard and operability analysis，HAZOP）报告以纸质文档形式保存，难于复用、共享，同时基于计算机软件的分析结果也只有对应的分析软件才能识别，同样存在难于复用、共享的问题。针对此问题，本文提出了基于知识本体的HAZOP信息标准化框架。该框架以知识本体和HAZOP分析国际标准IEC 61882为基础，抽提归纳了HAZOP的标准化信息模型，给出了模型的整体结构、模型中各元素的定义与关系。并在此基础上，提出了HAZOP信息标准化方法，采用BiLSTM神经网络对每一条HAZOP分析的记录进行标注、训练与识别，实现了人工HAZOP分析结果的自动识别与标准化。以某油品合成装置为例，对HAZOP信息标准化方法进行了验证，结果表明基于知识本体的HAZOP信息标准化框架可以自动实现分析结果的标准化，便于分析知识的共享与复用。

关键词: 安全, 知识本体, 危险与可操作性分析, 标准化, 模型, 神经网络