海量大气颗粒物成分分析系统-计算机系统应用.PDF

2015 年 第 24 卷 第 10 期 计 算 机 系 统 应 用 海量大气颗粒物成分分析系统① 1,2 2 2 2 张梦瑶 , 廉东本 , 赵 奎 , 马元婧 1( 中国科学院大学, 北京 100049) 2( 中国科学院沈阳计算技术研究所, 沈阳 110168) 摘 要: 2011 年以来, 我国多地出现了雾霾天气, 对大气颗粒成分分析有助于人们了解雾霾形成的原因, 制订有效 的应对措施. 本文的主要目的是对于大气颗粒物成分进行命名. 传统颗粒物的命名是在经验的基础上, 对颗粒进 行逐个的命名. 若将该过程自动化, 难点有两个: 数据规模太大、人工经验难以量化. 本文使用数据挖掘的工具, 首先进行了一次聚类分析, 降低了数据规模. 为了解决人工经验难以量化的问题, 使用逻辑回归分类算法, 并进行 了调优, 使正确率达到了业务处理的要求. 关键词: 单颗粒; 聚类; 神经网络; 分类; 逻辑回归 Massive Atmosphere Particulate Matter Analysis System 1,2 2 2 2 ZHANG Meng-Yao , LIAN Dong-Ben , ZHAO Kui , MA Yuan-Jing 1(University of Chinese Academy of Sciences, Beijing 100049, China) 2(Shenyang Institute of Computing Technology, Chinese Academy of Sciences, Shenyang 110168, China) Abstract: Since 2011, fog and haze has been appeared in many places of China. Atmosphere particulates components analysis could help people to know the reason of the fog and haze so that we can take measures to deal with it. The purpose of this system is to name the atmosphere particulates, which means to classify the atmosphere particulates into seven common types. In traditional practice, we need to name the particulates artificially and one by one. But, if we want to do this automatically, there are two difficulties we need to solve. First, the scale of data is large. Secondly, it is hard to make rules to summarize the human experience. In this system, we try to use data mining technology to solve the two problems. To decrease the scale of data, we use the Adaptive Resonance Neural Cluster Algorithm, and to summarize the human experience, we use the Logistic Regression Classification Algorithm. At last, we adjust the models to get a better accuracy and meet the needs of the actual application. Key words: