主管：中华人民共和国应急管理部
主办：应急管理部天津消防研究所
ISSN 1009-0029  CN 12-1311/TU
消防科学与技术

消防科学与技术 ›› 2023, Vol. 42 ›› Issue (9): 1265-1269.

• • 上一篇    下一篇

飞行时间质谱结合机器学习识别香烟烟灰研究

王靖童1, 刘术军2, 杨明3, 徐芷芊2   

  1. （1. 沈阳航空航天大学 安全工程学院，辽宁 沈阳 110136；2. 应急管理部沈阳消防研究所，辽宁 沈阳 110000；3. 中国科学院大连化学物理研究所，辽宁 大连 116000）
  • 出版日期:2023-09-15 发布日期:2023-09-15
  • 作者简介:王靖童(1999- )，女，辽宁锦州人，沈阳航空航天大学硕士研究生，主要从事火灾科学与飞机火爆防控方面的研究，辽宁省沈阳市沈北新区道义南大街37号，110136。
  • 基金资助:
    国家重点研发计划项目（2022YFC3006304）

Identification of cigarette ashes by time-of-light mass spectrometry combined with machine learning

Wang Jingtong1, Liu Shujun2, Yang Ming3, Xu Zhiqian2   

  1. (1. College of Safety Engineering, Shenyang Aerospace University, Liaoning Shenyang 110136, China; 2. Shenyang Fire Science and Technology Research Institute of MEM, Liaoning Shenyang 110000, China; 3. Dalian Institute of Chemical Physics, Chinese Academy of Sciences, Liaoning Dalian 116000, China)
  • Online:2023-09-15 Published:2023-09-15

摘要： 为了对不同品牌、不同厂家的香烟烟灰进行准确、快速的识别，收集了58种香烟、26种干扰物以及干扰物和香烟的混合烟灰样品，通过飞行时间质谱仪得到相应的飞行时间质谱数据，再通过系统聚类对飞行时间质谱数据进行分类，针对质谱图特征峰进行比对，最后运用机器学习的主成分分析、偏最小二乘判别分析方法，建立不同方法的判别分析。主成分分析结果表明，此模型具有良好的可靠性，并且具有较好的预测能力；偏最小二乘判别分析结果表明，模型建立可靠且有很好的预测香烟和干扰物烟灰的能力。此外，对该模型进行了200次置换验证，结果表明偏最小二乘判别模型在建立时未发生过拟合。因此，利用飞行时间质谱谱图并结合2种机器学习算法能够帮助查勘人员对香烟烟灰样本进行精确迅速辨别和检测。

关键词: 香烟烟灰, 飞行时间质谱, 机器学习, 主成分分析, 偏最小二乘判别分析

Abstract: In order to accurately and quickly identify cigarette ash from different brands and manufacturers, 58 types of cigarettes, 26 different interfering substances, and mixed cigarette ash samples from different brands and manufacturers were collected. The corresponding time-of-flight mass spectrometry data was obtained through a time-of-flight mass spectrometer, and then the time-of-flight mass spectrometry data was classified through system clustering. The characteristic peaks of the mass spectrometry were compared. Finally, principal component analysis and partial least squares discriminant analysis methods of machine learning are used to establish discriminant analysis for different methods. The results of principal component analysis indicate that this model has good reliability and predictive ability; partial least squares discriminant analysis indicates that the model is reliable and has good ability to predict cigarettes and interfering substance smoke ash. In addition, the model was verified for 200 times, and the results showed that the partial least squares discriminant model did not have overfitting when it was established. Therefore, combining time-of-flight mass spectrometry with two machine learning algorithms can help surveyors accurately and quickly identify and detect cigarette ash samples.

Key words:  cigarette ash, time-of-flight mass spectrometry, machine learning, principal component analysis, partial least squares discriminant analysis