期刊
EXPERT SYSTEMS WITH APPLICATIONS
卷 239, 期 -, 页码 -出版社
PERGAMON-ELSEVIER SCIENCE LTD
DOI: 10.1016/j.eswa.2023.122251
关键词
CNN; Bidirectional long short-term memory; Smart firefighting; Fire detection; Extrawide immersed tunnel; Computational fluid dynamics
The confined spaces of extrawide immersed tunnels pose challenges to firefighting strategies. This study proposes a deep-learning-based fire location detection model and creates a proprietary database for fire location detection tasks in extrawide immersed tunnels. The proposed model achieves exceptional performance in fire source location detection tasks.
The confined spaces of extrawide immersed tunnels presents a significant challenge to the effective implementation of firefighting strategies in the event of a fire. In such an event, it is crucial to accurately and promptly identify the fire source location, choose the optimal smoke exhaust scheme, and closely monitor the smoke exhaust conditions. Existing methods rely on human experience to predict the location of a fire source, and few utilize neural network models with self-learning capabilities. In response, we propose a deep-learning-based fire location detection model that provides intelligent and timely smoke extraction schemes for application in extrawide immersed tunnels. Numerical simulation tools were used to create a proprietary database of immersed tunnel fires. We have compiled a comprehensive dataset for fire location detection tasks in extrawide immersed tunnels, integrating soot concentration and temperature data through expert knowledge. This dataset has been publicly released at (https://github.com/zhang-zhen-project/immersed-tunnel-fire-location-detection-data). To the best of our knowledge, this is the first dataset of its kind in the field of extrawide immersed tunnel fires. We compared six existing fire-location detection models: backpropagation neural network (BPNN) using multilayer perceptron (MLP), convolutional neural network (CNN), long short-term memory (LSTM), CNN-LSTM, bidirectional long short-term memory (BiLSTM), and CNN-BiLSTM. Our proposed CNN-BiLSTM-based approach achieved an exceptional detection performance and a remarkable robustness, with an F1 score of over 99% in the fire source location detection task, optimal exhaust damper combination task, and the detection task with different heat release rates (HRRs), even with sensors spaced at over 40-m intervals inside the tunnel.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据