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 ›› 2026, Vol. 45 ›› Issue (9): 162-169.

• • 上一篇    下一篇

基于模糊神经网络和D-S证据理论的民机货舱火灾探测

何志祥, 王立纲, 董勤   

  1. （中国民用航空飞行学院广汉分院，四川 广汉 618307）
  • 收稿日期:2025-05-26 修回日期:2025-09-28 出版日期:2026-09-15 发布日期:2026-09-15
  • 作者简介:何志祥，中国民用航空飞行学院广汉分院工程师，主要研究方向为智能火灾技术应用，四川省广汉市南昌路四段 46 号，618307，2432166443@qq.com。
  • 基金资助:
    中央高校基本科研专项资金资助项目（25CAFUC05020）

Civil aircraft cargo hold fire detection based on fuzzy neural network and D-S evidence theory

He Zhixiang, Wang Ligang, Dong Qin   

  1. (Civil Aviation Flight University of China Guanghan Flight College, Guanghan Sichuan 618307, China)
  • Received:2025-05-26 Revised:2025-09-28 Online:2026-09-15 Published:2026-09-15

PDF (PC)

11

摘要： 针对火灾发生时传统监测装置容易出现的误报问题，提出了一种基于模糊神经网络和D-S证据理论的火灾探测算法。首先，将模糊神经网络和D-S证据理论进行融合构建智能火灾探测算法。其次，以温度、烟雾浓度、CO浓度等火灾特征参数作为智能火灾探测算法的输入，实现了对火灾的有效探测。最后，基于所提算法研制火灾探测系统，并在聚氨酯明火、乙醇明火、棉绳阴燃火和木材阴燃火等4种典型火下进行试验。结果表明, 4种火灾场景下，系统的探测准确率分别为92.8%、91.3%，97.8%和91.6%，其性能明显优于其他对比算法；所研制的火灾探测系统在火灾平均响应时间上比市售火灾探测器快2.1 s，火灾误报率约为0.63%。

关键词: 火灾探测, 火灾特征参数, 模糊神经网络, D-S证据理论

Abstract: Aiming at the false alarm defect prone to conventional monitoring devices during fire outbreaks, this paper proposes a fire detection algorithm combining fuzzy neural network with D-S evidence theory. Firstly, an intelligent fire detection algorithm is established by integrating the fuzzy neural network and D-S evidence theory. Secondly, fire characteristic parameters including temperature, smoke concentration and CO concentration are taken as the input variables of the proposed algorithm to realize reliable fire identification. Finally, a fire detection system based on the above algorithm is developed and verified via experiments under four typical fire conditions: polyurethane open flame, ethanol open flame, smoldering cotton rope fire and smoldering timber fire. Experimental results demonstrate that the detection accuracy of the developed fire detection system reaches 92.8%, 91.3%, 97.8% and 91.6% respectively for the four fire scenarios, delivering substantially superior performance compared with alternative contrast algorithms. In addition, the average fire response time of the developed system is 2.1 seconds shorter than that of commercially available fire detectors, with a false alarm rate limited to approximately 0.63%.

Key words: fire detection, fire characteristic parameter, fuzzy neural network, D-S evidence theory