用自组织特征映射神经网络对飞行时间质谱采集的大气气溶胶单粒子进行分类

Classification of Atmospheric Individual Aerosol Particles Sampled by Time-of-flight Mass Spectrometry Using Self-Organizing Map

气溶胶飞行时间质谱仪(ATOFMS)在对气溶胶粒子的测量过程中,产生大量包含单粒子化学成分和粒径信息的数据。本研究采用具备矢量量化与数据降维能力的自组织特征映射网络(SOM),对自制的气溶胶飞行时间质谱仪24 h采集到的室内大气气溶胶质谱数据进行聚类分析。获得"含钙"、"盐类和二次气溶胶"、"二次颗粒"、"有机胺"、"富含钾有机物"、"无机盐"和"土壤"等20类颗粒。相比于其它聚类方法,SOM可进行可视化分析,对神经元进行再次聚类,聚类中心多。这些分类信息将有助于评估气溶胶粒子的反应和毒性,以及鉴别气溶胶粒子的起源。

Large amount of data including chemical composition and size information of individual particles would be generated in the measurement of aerosol particles using atmospheric aerosol time-of-flight mass spectrometry （ ATOFMS ） . Our home-made ATOFMS was used to measure the indoor individual aerosol particles in real-time for 24 h, and the obtained mass spectrometric data were clustering analysis by self-organizing map （ SOM ） because of its ability of vector quantization and data dimensionality reduction. 20 classification results were got which included＂Calcium-Containing＂,＂Salt＋Secondary particles＂,＂Secondary particles＂,＂Organic Amines＂,＂K＋-Rich Organics＂ and＂Soil＂ particles, etc. Compared with previous mass spectrometric methods, SOM is a natural visualization tool, more classification results can be obtained. This classification information would be useful to assess the response and toxicity of atmospheric aerosol particles and identify the origin of atmospheric aerosol particles.