Citation

Zhen GUO, Xiaoyan JIA, Fumin LI, Yan HU, Qiuyan YN. China Saf. Sci. J. 2023; 33(11): 117-125.

Copyright

(Copyright © 2023, China Occupational Safety and Health Association, Publisher Gai Xue bao)

DOI

10.16265/j.cnki.issn1003-3033.2023.11.2219

PMID

unavailable

Abstract

In order to quickly rehearse or reproduce the fire scene, the key influencing factors of building fire were studied by using fire dynamics software(FDS) and machine learning technology, so as to adjust the fire rescue strategy in real-time, and finally provide a space composition scheme conducive to fire engineering for architectural design. In this paper, a single-room fire smoke overflow was taken as a case, and 5 kinds of algorithm models were used to perform machine learning training and efficiency evaluation on 11 spatial composition parameters and fire conditions, a total of 7 776 sets of fire simulation result data. The experimental results show that the machine learning algorithm is suitable for parameter learning and evaluation prediction of discrete types such as building space. It can intuitively give the weight of each parameter, mine the key information in the fire dynamics system, and realize the visualization of fire data. The Random Forests(RF) algorithm has the highest prediction efficiency, and its best prediction accuracy can reach 91.82%.

Key words: building fires, rapid fire prediction, machine learning, weight analysis, spatial composition

Language: en