Speaker
Abstract
Photoelectric fire smoke detectors are sensitive to false alarms caused by nuisance aerosols, which causes massive losses. To address this, techniques such as multiple optical channels and wavelengths have been developed to capture more particle scattering information. This paper presents an innovative method using broadband light (400–800 nm) to obtain multi-dimensional scattering information in a single measurement, enhancing particle discrimination. The approach was proved to be effective through numerical simulations using Mie scattering theory, which generated scattering spectrum for five types of fire smoke and five types of nuisance aerosols across various angles. These spectrum were input into five machine learning models for particle classification, with random measurement noise introduced to test robustness. These results indicate that forward scattering angles (60°–90°) combined with nonlinear machine learning models like Random Forest and XGBoost achieved almost 100% precision and recall in discriminating fire smoke from nuisance aerosols. This study highlights the potential of broadband visible light sources in fire detection, offering a robust solution to reduce false alarms and improve detection accuracy.
摘要
光电火灾烟雾探测器对干扰气溶胶非常敏感，会因为干扰源颗粒物和光源进行散射而引发误报，造成损失。为解决这一问题，本文提出了一种创新方法，即利用宽带光源（400-800 纳米）而非激光对颗粒物进行散射，获取散射光谱的高维信息，增强颗粒识别能力。该方法通过米氏散射理论进行数值模拟验证，生成了五种火灾烟雾和五种干扰气溶胶在不同角度下的散射光谱。将这些光谱输入五种机器学习模型进行颗粒分类，并引入随机测量噪声测试鲁棒性。结果表明，前向散射角（60°-90°）结合随机森林和 XGBoost 等非线性机器学习模型，在区分火灾烟雾与干扰气溶胶时实现了 接近100% 的精确率和召回率。本研究彰显了宽带可见光源在火灾探测中的潜力，为减少误报和提升检测精度提供了可靠解决方案。
| 关键词 | 火灾探测、宽光谱、数值模拟、颗粒散射、干扰源气溶胶 |
|---|---|
| Keywords | Fire detection, Nuisance Aerosols, Broadband Light, Particle Scattering, Numerical Study |
