基于自适应共振神经网络的单粒子激光电离质谱数据分析

气溶胶激光飞行时间质谱仪(ALUOFMS)可以在线地对气溶胶单粒子进行物理和化学特性分析，利用双束连续激光对单个粒子的空气动力学粒径进行测量，并通过飞行时间完成单粒子化学成分的检测。该仪器在运行过程中将产生海量的实验数据，对这些数据快速、自动处理并提取有价值的信息是整机系统的关键之一。文章介绍了基于神经网络的自适应共振算法(ART-2a)在随机混和的氯化钠、氯化钙、邻苯二甲酸二正辛酯(DOP)和2,5二羟基苯甲酸(DHB)气溶胶单粒子聚类分析中的成功运用。同以往的质谱分析方法相比，ART-2a可以实现对任意多和任意复杂的输入模式进行自组织，自适应和自稳定的快速识别，更有利于质谱数据的分析。实验结果表明，当警戒值为0.40，学习速率为0.05以及迭代次数为6时，ART-2a可以成功地对这四种物质进行分类，同时得到4类物质的聚类中心，每类的聚类中心都能很好的代表该类物质的特征。

Data Analysis of Laser Desorption/Ionization Mass Spectrum of Individual Particle Using Adaptive Resonance Theory Based Neural Network

On-line measurement of size and chemical composition of single particle using an aerosol laser time-of-flight mass spectrometer (ALTOFMS) was designed in our lab. Each particle’s aerodynamic diameter is determined by measuring the delay time between two continuous-wave lasers operating at 650 nm. A Nd∶YAG laser desorbs and ionizes molecules from the particle, and the time-of-flight mass spectrometer collects a mass spectrum of the generated ions. Then the composition of single particle is obtained. ALTOFMS generates large amount of data during the process period. How to process these data quickly and extract valuable information is one of the key problems for the ALTOFMS. In the present paper, an adaptive resonance theory-based neural network, ART-2a algorithm, was used to classify mixed mass spectra of aerosol particles of NaCl, CaCl2, dioctylphthalate (DOP), and 2,5-dihydroxybenzoic acid (DHB). Compared with the traditional methods, ART-2a can recognize input patterns self-organically, self-adaptively and self-steadily without considering the complexity and the number of the patterns, so it is more favorable for the analysis of the mass spectra data. Experimental results show that when vigilance parameter is 0.40, learning rate is 0.05 and iteration number is 6, ART-2a algorithm can successfully reveal these four particle categories. The weight vectors for these four particle classes were obtained, which can represent the characters of these four particle classes remarkably.