Data Mining of Hidden Danger in Enterprise Production Safety and Research of Hidden Danger's Model Conversion

Abstract

As the adoption of the high-speed development of the information technology and continuous improvement of industrial technology, huge amount of statistical data and statistical documents are accumulated in environment-safe and environment-friendly field. The big data technology drives the attention points in the safety science field turn to the data and the exploration of the security management mode established on analysis of the data. When the big data oriented thinking and mode are used, the safety management knowledge underlines the hidden danger data in production safety is disclosed, and precision application of the safety management will be promoted. The hidden danger data of safe production is stored in text form, and data mining and machine learning model can’t deal with these non-structured (or half-structured) information directly, so it is necessary to deal with the text data through Natural language processing (NLP), and then use the data mining method to excavate the rules information. First of all, the words in the text are recombined through technologies such as Chinese word segmentation, part-of-speech labeling, named entity recognition and the like, the word of the combined vocabulary is marked, and the named entity is further identified. Secondly, the structured problem disclosing database is constructed through three steps including keyword mapping and extraction, data cleaning and integration, data selection and transformation. What’s more, by using data mining technology such as data stream sliding window model, association analysis and change mining algorithm, this paper constructs the related hidden danger analysis method. By means of mining association rules of the current hidden danger data, releasing the related type, the existence possibility and the change pattern of hidden danger and so on. Finally, the information is visualized and analyzed. In this paper, the data mining of 10,387 safety hidden danger samples of a refinery enterprise in 2012-2017, obtained 1091 association rules, constructed 5 kinds of associated hidden danger transformation mode: the emerging mode, attenuation mode, relation change mode, result transformation mode, new adding mode. These modes are summed up the following conclusions: (1) in view of the safety hidden danger data of petroleum and petrochemical industry, this paper analyzes the language characteristics of the domain corpus, and first forms a professional dictionary with industry characteristics; (2) using Eclat association analysis algorithm to excavate the association rules between hidden danger data, and classify the rules, which mainly focus on the pipeline, valve, safety valve and other parts; (3) after analysis of the hidden danger, the leakage of the bearing box, blind plate missing, the length of the coupling shield is too short, the discharge port setting height does not meet the requirements, the automatic valve failure, the export pressure gauge real fluctuation are six problems for the second half of 2017 new problems. By excavating the potential value of the hidden danger data, the hidden danger checking work is further optimized and improved, and the enterprise is guided to carry out targeted hidden danger checking, the centralized optimization of services is realized, the accurate application of safety management is promoted, and the safety and environmental protection management and risk control level are improved.