《火灾科学（中英文）》编辑部

王力,李泊宁,张曦,梅志斌.基于门控循环单元网络的层次化多参量高大空间火灾探测模型[J].火灾科学,2025,34(3):182-192.

基于门控循环单元网络的层次化多参量高大空间火灾探测模型

Hierarchical mult-parameter gated recurrent unit model for fire detection in large space buildings

DOI：10.3969/j.issn.1004-5309.2025.03.03

基金项目:国家重点研发计划项目(2021YFC3001603);辽宁省自然基金项目(2024010753-JH3/107)

作者	单位
王力	1.应急管理部沈阳消防研究所,沈阳,110034 2.消防与应急救援国家工程研究中心,沈阳,110034 3.辽宁省火灾防治技术重点实验室,沈阳,110034
李泊宁^^	1.应急管理部沈阳消防研究所,沈阳,110034 2.消防与应急救援国家工程研究中心,沈阳,110034 3.辽宁省火灾防治技术重点实验室,沈阳,110034
张曦	1.应急管理部沈阳消防研究所,沈阳,110034 2.消防与应急救援国家工程研究中心,沈阳,110034 3.辽宁省火灾防治技术重点实验室,沈阳,110034
梅志斌	1.应急管理部沈阳消防研究所,沈阳,110034 2.消防与应急救援国家工程研究中心,沈阳,110034 3.辽宁省火灾防治技术重点实验室,沈阳,110034

中文关键词: 火灾探测门控循环单元时序特征多参量模型火灾数据采集

英文关键词:Fire detection GRU Temporal feature Hierarchical Multi-Parameter Model Fire data collection

摘要点击次数: 37

全文下载次数: 27

中文摘要:

在当前高大空间建筑火灾监测场景中,由于火灾探测器安装高度高,能够到达探测器的火灾信号强度有限,易造成火灾报警发生延误,而单纯通过降低报警阈值来提升灵敏度的方法则易造成较高的误报率。本研究以温度、羽流速度和烟雾浓度为传感对象,基于门控循环单元网络提出了层次化多参量火灾探测网络模型(HMPM)。该模型具有三层结构,可聚合多类传感器在低响应强度下的时序特征,通过层次化的多参量融合判决实现高大空间火灾信号的识别。通过设计高度为18.5 m的数据采集塔,采集明火和阴燃火火灾参量,验证了算法模型的有效性。实验表明,HMPM在误报率和漏报率方面优于传统火灾探测方法和常规机器学习方法,在解决高大空间场所火灾探测的早期性和准确性方面具有显著潜力。

英文摘要:

In high-rise building scenarios, traditional fire detection methods have certain limitations in perceiving environmental conditions due to factors such as high detector installation height and a single sensing type, resulting in high detection delay. Moreover, such methods neglect temporal changes in sensor data, leading to high false alarm rates. To overcome the limitations of traditional fire detection methods in large space buildings, a Hierarchical Multi Parameter Model (HMPM) is proposed based on the Gate Recurrent Unit, which uses temperature, plume velocity, and smoke concentration as sensing objects and can aggregate time-series features from multiple sensor types. To verify the effectiveness of the model, a data acquisition tower with a height of 18.5 meters is designed. The early fire time series feature data set is collected and constructed under open-fire and smouldering conditions for experimental comparison. The experimental results demonstrate that the proposed HMPM outperforms not only traditional fire detection methods but also common machine learning methods in large-scale building scenes, with high accuracy and low false alarm rates, proving that HMPM has significant advantages for early and accurate fire detection in high-rise spaces.

关闭