中国安全科学学报 ›› 2022, Vol. 32 ›› Issue (12): 10-18.doi: 10.16265/j.cnki.issn1003-3033.2022.12.2752

• 安全科学理论与安全系统科学 • 上一篇    下一篇

基于随机森林的矿工不安全行为预测预警模型

李红霞1,2,3(), 徐浩冉1(), 田水承2,3   

  1. 1 西安科技大学 管理学院,陕西 西安710054
    2 西安科技大学 安全科学与工程学院,陕西 西安710054
    3 西安科技大学 应急管理学院,陕西 西安710054
  • 收稿日期:2022-07-05 修回日期:2022-10-19 出版日期:2022-12-28 发布日期:2023-06-28
  • 作者简介:

    李红霞 (1965—),女,河北遵化人,博士,教授,博士生导师,主要从事安全与应急管理、矿工不安全行为等方面的研究。E-mail:

    徐浩冉 (1998—),男,河南鹤壁人,硕士研究生,研究方向为安全与应急管理。E-mail:

  • 基金资助:
    国家自然科学基金资助(51874237); 国家自然科学基金资助(U1904210); 国家自然科学基金资助(71273208)

A prediction and early warning model of miners' unsafe behavior based on random forest

LI Hongxia1,2,3(), XU Haoran1(), TIAN Shuicheng2,3   

  1. 1 Management School, Xi'an University of Science and Technology, Xi'an Shaanxi 710054, China
    2 School of Safety Science and Engineering, Xi'an University of Science and Technology, Xi'an Shaanxi 710054, China
    3 School of Emergency Management, Xi'an University of Science and Technology, Xi'an Shaanxi 710054, China
  • Received:2022-07-05 Revised:2022-10-19 Online:2022-12-28 Published:2023-06-28

摘要:

为减少煤矿工人不安全行为,科学研判其未来趋势和状态,通过事故统计和文献分析构建煤矿不安全事件与事故数据库和矿工不安全行为属性表,在此基础上,采用Apriori算法挖掘各不安全因素与不安全行为间的关联规则,将强相关规则作为随机森林预测模型的输入指标,不安全行为频次作为输出指标,对原始模型进行拟合训练,基于相关性分析与专家建议设置预警阈值,完善矿工不安全行为预警模型。结果表明:安全文化、现场监护、无效纠正、技术环境以及人员状态与矿工不安全行为构成强相关规则;对于矿工不安全行为的预测,预测值与实际值的偏差较小,预警模型准确度较高;且技术环境、工作环境、组织管理和隐患识别对预测结果影响最大。

关键词: 随机森林, 矿工, 不安全行为, 预测预警模型, 关联规则挖掘

Abstract:

In order to reduce the unsafe behaviors of coal miners and make scientific judgement of current status and future trend, this study constructed a database of accidents of coal mines and an attribute table of unsafe behaviors of miners with accident statistics and literatures. On this basis, association rules between unsafe factors and behaviors were explored by adopting Apriori algorithm. An original model was established to fit training, which was formed by taking strong correlation rule as input indicators to the random forest prediction model and the frequency of unsafe behaviors as output indicators. Early warning threshold based on correlation analysis and expert recommendations was also set to refine the early warning model of miners' unsafe behaviors. The results show that safety culture, on-site monitoring, ineffective correction, technical environment, and personnel status constitute strong correlation rules with miners' unsafe behaviors. For the prediction of miners' unsafe behavior, the predicted value has less deviations from the actual, and the early warning model has relatively high accuracy. Among them, technical environment, work environment, organizational management, and identification of hazards play the key role in predicting outcomes.

Key words: random forest, miner, unsafe behavior, prediction and early warning model, association rule mining