Construction of petrochemical knowledge graph based on deep learning

Hazard and operability analysis (HAZOP) is a safety evaluation method that is vital in the chemical process. HAZOP analysis uses the form of “brainstorming” and “counterfactual reasoning” to determine the potential events for danger in the chemical process. However, this method relies too much on expert experience and a large amount of existing HAZOP information has not been shared and reused. Herein we prompt a semi-automated HAZOP knowledge graph to overcome the problem of information reuse and sharing. In the implementation of the knowledge graph, the construction of the HAZOP ontology is completed by the seven-step rule in the top-down ontology construction. The named entity recognition task is one of the key tasks for the construction of knowledge graphs. The deep learning method is adopted in this paper and is based on the existing HAZOP data to recognize the named entity of HAZOP text. We adopt the deep neural network of BI-GAT-CRF, which has an accuracy of 90.75, a recall rate of 91.53, and an F1 score of 91.14 in the HAZOP Chinese text. BI-GAT-CRF aims to solve the problem of unclear entity boundary recognition and word ambiguity in named entity recognition tasks. The results show that the model performs well in HAZOP named entity recognition. BI-GAT-CRF can effectively determine entity boundaries and recognize named entity categories. BI-GAT-CRF also solves the problem of word ambiguity and improves the accuracy of named entity recognition tasks. Finally, a HAZOP analysis report of a certain oil equipment was used as an example to construct a HAZOP knowledge graph to verify the effectiveness of the construction method.