Abstract：With the advancement of science and technology, an increasing number of industries and fields are striding towards the informationization. The utility tunnels have become important facilities for urban energy transmission. In the context of the centralized development of large-scale municipal infrastructure, the fire hazards of utility tunnels have gradually become prominent. The convolutional neural network established by YOLO V5 can perform high-precision fire recognition, and then is able to extract the important fire parameters, including real-time fire spread range, the fire spread speed and the flame width. By designing 12 sets of fire experiments in the utility tunnel with different fire development speeds, the convolutional neural network is trained and verified for the accuracy of its parameter extraction. The results show that the average relative error (ARE) of the extracted spread position of fire front, fire spread speed and flame width are around 5%～15%, 6%～20% and 10%～27%, respectively. Furthermore, it is verified that the method can guarantee good extraction accuracy. For informationization of building fire protection, this method can be applied to formulating the fire rescue measures at the fire scene, and make it possible to examine and judge the fire development trends, assess the severity of fire accidents, and estimate the accident losses in real time.
黄萍 陈铭 陈可欣 刘春祥 余龙星. 卷积神经网络在建筑消防信息化的应用——以城市综合管廊火灾监控为例[J]. 土木工程学报, 2022, 55(6): 112-120, 128.
Huang Ping Chen Ming Chen Kexin Liu Chunxiang Yu Longxing. Application of convolutional neural network in building fire informatization—taking urban utility tunnel fire monitoring as an example. 土木工程学报, 2022, 55(6): 112-120, 128.
