模式识别在火灾调查中的汽油分类问题的应用研究(英文)  

The application of pattern recognition to the classification of regular gasoline in fire investigation

作　　者：支有冉[1,2] 宗若雯[1,2] 王荣辉 李松阳[1,2] 

机构地区：[1]中国科学技术大学火灾科学国家重点实验室,安徽合肥230026 [2]中国科学技术大学苏州研究院苏州市城市公共安全重点实验室,江苏苏州215123 [3]北京市消防局,北京102308

出　　处：《火灾科学》2009年第2期108-114,共7页Fire Safety Science

摘　　要：在火灾调查中,检测汽油成分并对其进行正确分类尤为重要。运用GC-MS对90#和93#两种普通汽油的共50个样本进行检测,所得的GC-MS原始数据通过PCA方法进行处理,以提取有用信息,避免冗余变量进入后续计算。在此基础上应用KNN方法对这两种汽油助燃剂进行分类。结果表明,KNN方法对这两种汽油的分类准确率达到100%,且当初始数据未经标准化预处理时也能达到同样准确的分类效果。研究表明:将模式识别方法正确地运用到助燃剂鉴定和分类工作中有助于火灾调查。Detection and accurate classification of gasoline is very important in fire investigation. In this paper, a total of 50 samples of regular gasoline, covering two different grades （90# and 93#） , were examined by gas chromatography - mass spectrometry （GC -MS）. The GC -MS data were treated by Principal Component Analysis （PCA） to distill the in- formation from the original dataset in order to avoid the redundant variables to be calculated. And k - nearest neighbors algorithm （KNN） was further applied to classify the two types of accelerant. The results showed that KNN could classify the two types of gasoline effectively, with the 100% probability （no prediction error） , whether the data were normalized or not. The results indicated that the proper application of pattern recognition to the identification and classification of accelerant provided positive help in fire investigation.

关 键 词：助燃剂 主成分分析 KNN GC-MS 模式识别 

分 类 号：X928.1[环境科学与工程—安全科学] X928.9