面向空气质量监测数据时空多维属性的可视分析方法  

Visual Analytics of the Spatio-temporal Multidimensional Air Monitoring Data

作　　者：周志光[1,2,3] 胡迪欣 刘亚楠[1] 陈伟锋[1] 陶煜波[2] 林海[2] 苏为华[4,3] Zhou Zhiguang;Hu Dixin;Liu Yanan;Chen Weifeng;Tao Yubo;Lin Hai;Su Weihua(School of Information, Zhejiang University of Finance and Economics, Hangzhou 310018;State Key Laboratory of CAD&CG, Zhejiang University, Hangzhou 310058;School of Statistics and Mathematics, Zhejiang Gongshang University, Hangzhou 310018;Postdoctoral Research Station of Applied Economics, Zhejiang University of Finance and Economics, Hangzhou 310018)

机构地区：[1]浙江财经大学信息学院,杭州310018 [2]浙江大学CAD&CG国家重点实验室,杭州310058 [3]浙江财经大学应用经济学博士后流动站,杭州310018 [4]浙江工商大学统计与数学学院,杭州310018

出　　处：《计算机辅助设计与图形学学报》2017年第8期1477-1487,共11页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61303133;61472354);全国统计科学研究项目(2015LD03);浙江省科技厅公益技术应用研究计划项目(2014C31057);教育部博士后科学基金(2015M571846);浙江省自然科学基金(LY15F020014);浙江省一流学科A类(浙江财经大学统计学)2016年度规划项目;浙江大学CAD&CG国家重点实验室开放课题(A1417)

摘　　要：空气质量监测数据具有显著的时空多维属性.传统的平行坐标技术虽然能够有效地展示数据的多维属性信息,却在分析与解读空气质量监测数据各个属性的时空变化规律方面表现出一定的局限性,在平行坐标展示空气质量监测数据多维属性的基础上,提出一种支持用户交互式探索大气污染时空特征的可视分析方法.首先利用平行坐标展示空气质量监测数据,支持用户交互改变时空维度以及指定坐标轴排列顺序;然后引入角度面积正负相关性等方式度量数据在平行坐标系中的布局差异,并且通过矩阵图和交互式柱状图分别展示不同时空维度下数据的布局差异;再综合考虑各个属性之间的数据布局差异,构建相似性矩阵,利用多维标度法对当前时空维度的数据进行降维,获得初始数据在低维空间的表示;最后利用层次聚类方法对低维空间的数据表达做聚类分析,并且分别设计时钟隐喻图和地域抽象图描述各个类别的时空节点组成.集成上述可视化算法设计便捷的用户交互模式,开发面向空气质量监测数据时空多维属性的可视分析原型系统,为用户快速分析和解读大气污染的时空特征及潜在规律提供有效手段.通过大量的可视化效果及用户反馈结果,进一步验证了文中所提可视分析方法的有效性和实用性.The air quality monitoring data presents significant spatio-temporal multidimensional attributes.Thetraditional parallel coordinates are able to reveal multiple attributes,but show some limitations in terms of spatio-temporal analysis of the air quality monitoring data.In this paper,we design a visual analysis system aiming atthe exploration of the spatio-temporal characteristics of atmospheric pollution.Firstly,the parallel coordinates areemployed to display multiple attributes of a section of air quality monitoring data,which also allow users to interactivelyexchange the spatio-temporal perspective and specify the axis order.Some metrics are used to measure the difference of data in the parallel coordinates,including angle,area,and correlation.The matrix graph andinteractive bar chart are applied to display the difference of data from different perspectives.We then weight thematrix graphs to construct a similarity matrix and use the MDS to reduce the original multidimensional data into2D coordinates.A hierarchical clustering scheme is used to classify the coordinates.Furthermore,a design of timemetaphor and an abstract of space are used to enable users achieve the spatio-temporal features in the air qualitymonitoring data.Convenient interactions are also integrated in this system to help users deeply investigate atmosphericpollution.A large number of experimental results as well as credible user studies further demonstratethe effectiveness and practicability of the system.

关 键 词：大气污染 可视分析 平行坐标 多维标度法 层次聚类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]