《火灾科学（中英文）》编辑部

莫力科,张璐,范成,周海俊,赵杨平.深度学习代理模型的消防传感系统优化方法[J].火灾科学,2024,33(1):50-60.

深度学习代理模型的消防传感系统优化方法

Fire sensing system optimization method based on deep learning surrogate model

DOI：10.3969/j.issn.1004-5309.2024.01.07

基金项目:国家重点研发计划项目(2020YFB2103503)

作者	单位
莫力科	1.深圳大学土木与交通工程学院,深圳,518060 2.滨海城市韧性基础设施教育部重点实验室,深圳,518060
张璐	1.深圳大学土木与交通工程学院,深圳,518060 2.滨海城市韧性基础设施教育部重点实验室,深圳,518060
范成^*	1.深圳大学土木与交通工程学院,深圳,518060 2.滨海城市韧性基础设施教育部重点实验室,深圳,518060
周海俊0	1.深圳大学土木与交通工程学院,深圳,518060
赵杨平0	1.深圳大学土木与交通工程学院,深圳,518060

中文关键词: FDS 深度学习 LSTM 系统优化智慧消防

英文关键词:FDS Deep learning LSTM System optimization Smart fire

摘要点击次数: 1530

全文下载次数: 115

中文摘要:

传感器合理布置显著影响消防监测系统的实际效用。为确保火灾预警及特性识别效率,建立了一套基于深度学习代理模型的消防监测传感系统优化方法。通过建立火灾仿真模型获取多场景模拟数据,利用深度学习技术学习火场演绎规律,结合不同数据设定下的实验量化模型的识别准确率、查准率、查全率和F1-score,从空间布局和采集频率、探测时间角度对整体监测系统进行了优化。结果表明,优化后的传感系统可在满足火灾特征识别准确率的前提下,显著减少传感器数量及数据采集成本。该解决思路可推广应用于不同建筑场景,为消防监测系统设计提供决策支持,研究成果有利于智慧消防和智慧建筑发展。

英文摘要:

The reasonable arrangement of sensors significantly affects the actual effectiveness of the fire monitoring system. A fire sensor system optimization approach based on a deep learning surrogate model is developed to ensure the effectiveness of early warning and characteristic identification of fires. The multi-scene simulation data is obtained using the fire simulation model, and the deduction law of the fire field is learned using deep learning technology. After studying and quantifying the model identification accuracy, precision, recall and F1 score in various data settings, the monitoring system is optimized in terms of spatial layout, acquisition frequency, and detection time length. The results show that the optimized sensor system can significantly reduce both the number of sensors and the data acquisition cost while maintaining fire detection's effectiveness. This solution provides decision support for designing a fire monitoring system and can be used in different building scenarios. The study results are helpful for the advancement of intelligent buildings and fire protection.

关闭