Accident causes data-driven coal and gas outburst accidents prevention: Application of data mining and machine learning in accident path mining and accident case-based deduction

Analyzing the causes of accidents, excavating accident paths, and applying accident prevention are important tasks in safety management. Focusing on coal and gas outburst accidents, this study examined the primary accident path and conducted applied research on the reasoning of the accident case. First, combined with the obtained accident causes, a coupling analysis of the causes of coal and gas outburst accidents was conducted. Second, using the method of data mining coupled with Apriori algorithm, the coupling relationship between each cause module of the coal and gas outburst accident was obtained, and consequently, a path map of the coal and gas outburst accident was drawn. Third, a Bayesian network model for the causes of coal and gas outburst accidents was established based on the accident path map and the probability of occurrence of each cause. Finally, considering the safety concept element (SC1) as an example, the Bayesian network model was used to conduct a sensitivity analysis of accident causes. Thereafter, considering the coal and gas outburst accident of the Sanjia Coal Mine in Guizhou Province as an example, probabilistic reasoning research on the cause of the accident was conducted. The application results showed that (1) under normal conditions, there are approximately 797,280 accident paths for coal and gas outbursts. Following data mining, 188 main accident paths were found. (2) Sensitivity analysis determined 19 factors that were sensitive to safety concept elements (SC1), of which the three most sensitive factors were (i) resource management system procedures (SM7), (ii) safety policy (SM1), and (iii) safety training system procedure (SM8). 13 paths exhibited a sensitivity ≥0.5%, of which 7 exhibited strong sensitivity. (3) The absolute accuracy rate of accident cause reasoning in the Sanjia Coal Mine in Guizhou Province was 71.43%, while the relative accuracy rate was close to 100%. Thus, it was concluded that: (1) the accident path mining method proposed in this paper is feasible for main accident path mining. (2) The Bayesian network model for the causes of coal and gas outburst accidents established in this study can be practically applied for the sensitivity analysis of accident causes and exhibits high reliability in the probabilistic reasoning of accident causes. The results of this study is expected to aid in the prevention of coal and gas outburst accidents, and provide reference and help for the path mining of other accident causes and the probabilistic reasoning of accident causes.