主管：中华人民共和国应急管理部
主办：应急管理部天津消防研究所
ISSN 1009-0029  CN 12-1311/TU
消防科学与技术

消防科学与技术 ›› 2022, Vol. 41 ›› Issue (10): 1384-1389.

• • 上一篇    下一篇

机器学习在煤粉着火敏感性预测中的应用

雷友学   

  1. （陕西省消防救援总队，陕西 西安 710016）
  • 出版日期:2022-10-15 发布日期:2022-10-15
  • 作者简介:作者简介：雷友学（1982- ），男，陕西商洛人，陕西省消防救援总队防火监督处副处长，主要从事消防监督工作，陕西省西安市凤城二路15号，710016。

Application of machine learning in prediction of coal dust explosion sensitivity

Lei Youxue   

  1. （Shaanxi Fire and Rescue Brigade, Shaanxi Xi'an 710016, China）
  • Online:2022-10-15 Published:2022-10-15

摘要： 摘 要：为探究机器学习模型预测煤粉云最低着火温度（Minimum Ignition Temperature,MITc）的可行性，收集Godbert-Greenwal炉测试得到的煤粉云最小着火温度和影响因子数据，并分析了影响因子的关联性。利用AUC/ROC、Kappa系数、敏感性、特异性、MAE、RMSE等指标对三种机器学习模型在煤粉云最小着火温度和着火概率两方面的预测效果进行评价。结果发现：RSM模型的预测效果最差；RF模型在预测煤粉云MITc和着火概率时具有较好的精度和稳定性；Bagging模型在预测着火概率时AUC值均大于0.85，但预测MITc时效果较差。结果为煤粉云着火敏感性预测提供了一种新的研究思路。

关键词: 关键词：煤粉云, 着火敏感性, 随机森林, Bagging模型

Abstract: Abstract: This article presented a full-scale UHVDC converter transformer fire model, which was developed on account of the characteristics of special fire accidents such as high-temperature oils and explosion shock of converter transformers, and the performance of the compressed air foam spraying system in firefighting was validated by carrying out the spill fire test. Due to the absence of fire branch pipes and sprinklers, the compressed air foam sprinkler system will not fail in blast shock. Under the application rate of 12.4 L/(min·m2), the fire was controlled at 28 s and extinguished at 181 s. During the fire test, re-ignition occurred first on the top fuel tank due to the higher temperature of the fuel in the top fuel tank. In addition, the fire plume tended to the direction of the valve hall firewall, and the "fireproof foam curtain" had the effect of thermal radiation protection, resulting in higher temperature of the valve hall firewall.

Key words: Key words: compressed air foam, fire extinguishing system, UHVDC converter transformer, high-temperature oils, spill fire