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 ›› 2020, Vol. 39 ›› Issue (12): 1735-1739.

• • 上一篇    下一篇

基于机器学习的火灾事故等级分类研究

王湛1,2,3,朱国庆1,2,3,柴国强1,2,3,姚斌1,2,3   

  1. 1. 城市地下空间火灾防护江苏省高校重点实验室,江苏徐州221116;2. 中国矿业大学煤矿瓦斯与火灾防治教育部重点实验室,江苏徐州221116;3. 中国矿业大学公共安全与消防研究所,江苏徐州221116
  • 出版日期:2020-12-15 发布日期:2020-12-15
  • 作者简介:王湛(1995-),男,河南开封人,中国矿业 大学安全工程学院硕士研究生,主要从事建筑火灾防护、 性能化防火设计和火灾数据分析等方面的研究,江苏省徐 州市泉山区大学路1 号,221116。
  • 基金资助:
    :中国矿业大学未来杰出人才助力计划资助项目(2020WLJCRCZL044);江苏省研究生科研与实践创新计划资助项目(KYCX20_2067)

Research on fire classification based on machine learning

WANG Zhan1,2,3,ZHU Guo-qing1,2,3, CHAI Guo-qiang1,2,3, YAO Bin1,2,3   

  1. 1. Jiangsu Key Laboratory of Fire Safety in Urban Underground Space, Jiangsu Xuzhou 221116, China; 2 Key Laboratory of Gas and Fire Control for Coal Mines, China University of Mining and Technology, Jiangsu Xuzhou 221116, China; 3 Public Safety and Fire Research Institute, China University of Mining and Technology, Jiangsu Xuzhou 221116, China
  • Online:2020-12-15 Published:2020-12-15

摘要:

为了探究火灾的严重程度与消防队响应时间、救援人数、火灾地点、火灾探测器以及自喷系统等之间的关系,采用随机森林、人工神经网络、支持向量机和极限学习机4 种机器学习算法对美国旧金山市的历史火灾数据进行了挖掘分析。运用模糊理论将消防队响应时间和救援人数转换为三角模糊数,提出了一种基于模糊推理的火灾因素分类方法。研究发现4 种算法的准确率均超过90%,并通过交叉验证的方法证明了这些算法的可靠性。在4 种算法中,随机森林算法的准确率和Kappa 值均高于其他算法,但是其计算得到的第三类火灾的AUC 值最小。

关键词: 数据挖掘, 机器学习, 火灾数据, 模糊理论

Abstract: In order to explore the relationship between the fire severity and the response time of fire brigades, the number of rescuers, the location of fires, the operation of fire detectors and the sprinkler system, etc., random forest, artificial neural network, support vector machine and extreme learning machine are used to classify and analyze the historical fire data of San Francisco, USA. Fuzzy theory is used to convert the response time of the fire brigade and the number of rescuers into triangular fuzzy numbers and a fire classification method based on fuzzy reasoning and data analysis is proposed. The research finds that the accuracy of the four algorithms exceeds 90% and the reli ability of these algorithms is proved by cross- validation. Among the four algorithms, the accuracy and kappa value of the random forest are higher than other algorithms, but the AUC value of the third fire type in the random forest algorithm is the smallest. 

Key words: data mining, machine learning, fire accident data, fuzzy theory