Intelligent HAZOP analysis method based on data mining
Feng Wang,Wunan Gu
DOI: https://doi.org/10.1016/j.jlp.2022.104911
IF: 3.916
2022-12-01
Journal of Loss Prevention in the Process Industries
Abstract:Accidents in production equipment are occurring with increasing frequency, which has caused serious negative impacts on society and petrochemical enterprises. It is important to learn prevention and control rules and lessons from accident cases to prevent accidents and improve intrinsic safety levels of enterprises. The hazard and operability (HAZOP) analysis method has been widely used for risk identification and accident prevention in petrochemical enterprises. However, the rules contained in many completed HAZOP analysis reports have not been deeply studied, inherited, or extensively applied. The analysis of a new process, however, relies on expert experience, which is usually subjective and time-consuming. This paper proposes an intelligent HAZOP analysis method based on data mining to explore the law of accident cause mechanisms, help promote intelligent analysis, and improve the accuracy of analysis results. A HAZOP analysis data table structure is proposed. The frequencies of all words in the process parameter, guide word, cause, and consequence were calculated, and the words were sorted based on the word frequency statistics method. Words with high word frequencies were considered key in formulating the safety inspection checklists. The latent Dirichlet allocation (LDA) model and contingency table were used to explain the correlation between the process parameters, guide words, causes, and consequences. The co-occurrence probability sequence of the elements can then be acquired. Using the naive Bayes algorithm, a method for calculating the likelihoods, severities, and risk levels of accidents is proposed. The accident risk ranks can be intelligently predicted to realize an intelligent HAZOP analysis. This study developed a software to help execute the method. The intelligent HAZOP analysis method was applied based on data from 14 petrochemical units; the raw oil buffer tank in the wax oil hydrocracking unit was taken as an example to illustrate the application of the method. The method provides a technical basis and basic guarantee for risk identification, accident prevention, and rescue of petrochemical plants.
engineering, chemical