气溶胶单粒子化学成分在线聚类分析  

Real-time Classification of Aerosol Single Particles Chemistry Composition

作　　者：夏玮玮[1] 顾学军[1] 郑海洋[1] 方黎[1] 

机构地区：[1]中国科学院安徽光学精密机械研究所环境光谱学研究室,合肥230031

出　　处：《分析化学》2010年第6期795-799,共5页Chinese Journal of Analytical Chemistry

基　　金：国家自然科学基金(No.20477043);中国科学院研究生科技创新与社会实践资助专项(科技创新类)资助

摘　　要：为了在线快速、自动处理气溶胶激光飞行时间质谱仪在运行过程中产生的海量数据并提取有价值的信息,将基于神经网络的自适应共振算法运用于气溶胶单粒子化学成分在线聚类分析。利用该算法对NaCl粒子、NaCl和CaCl2的气溶胶混合物、三聚氰胺气溶胶单粒子以及大气气溶胶单粒子进行在线分类,被成功分类的粒子数占命中粒子总数的100%,当警戒阈值为0.1,学习速率为0.05时,获得的每类聚类中心都能很好地代表该类物质的特征;深入研究了警戒阈值对NaCl粒子在线聚类数的影响。结果表明:当警戒阈值增大到0.8时,在线聚类数增多,分类更加精细。本算法的质谱分析软件基本满足大气气溶胶单粒子在线聚类分析的要求。To quickly process the laser desorption/ionization mass spectra which were generated by aerosol laser time-of-flight mass spectrometer （ALTOFMS） during its working and extract the valuable information, an adaptive resonance theory-based neural network, ART-2a algorithm, was successfully applied to the real-time classification of single particles of NaCl, the mixture of NaCl and CaCl2 aerosol, melamine and the atmospheric aerosol. Experimental results showed that when vigilance factor was 0.1 and learning rate was 0.05, ART-2a algorithm could successfully reveal the aerosol particles categories. Besides the successful probability was near 100%, the centroid mass spectra for the single particle classes were obtained, which could represent the characteristic of single particle classes remarkably. The number of NaCl single particle classes as a function of vigilance factor was also discussed. The result showed that the number of NaCl particle classes was larger when vigilance factor reached 0.8; furthermore the classification was much more precise. The mass spectra acquisition and control software using ART-2a can basically meet the requirements of real-time classification of atmospheric aerosol single particles.

关 键 词：大气气溶胶 气溶胶单粒子 化学成分 气溶胶激光飞行时间质谱仪 自适应共振神经网络 

分 类 号：X513[环境科学与工程—环境工程]