Big Data and Safety Management Methods: The Reduction Model of Hot Work Number

The management of hot work is the key element of petroleum refinery safety management. A large amount of hot-work data has been accumulated, which is underexploited. Driving new insights using big data analytics is the trend. However, there have been few scientific studies on solving a specific problem using big data method in the field of safety management. Hence, the unstructured-data analysis of the hot-work permit-to-work text was investigated. The professional corpus in the hot-work field was constructed using word segmentation, stop list elimination, standardized process, and manual proofreading. All hot-work content collocates were manually grouped into semantic domains in the light of experts' experience. The data-driven reduction models of hot work number were proposed aiming at possible invalid hot work, long-time hot work, repetitive hot work and possible equipment defect. Based on the proposed reduction models, we mined the patterns of hot work and the unnecessary or high-risk hot work could be identified automatically. The result of the reduction model in training set indicated that the reduction models are reasonable. The reduction model 1, reduction model 2, reduction model 3, reduction model 4 could reduce the proportion of hot work number 5%, 3%, 4%, and 2%, respectively. Thus, the targeted measures could be put forward to optimize the safety management of hot work.