地球信息科学学报 ›› 2013, Vol. 15 ›› Issue (6): 793-800.doi: 10.3724/SP.J.1047.2013.00793

• 本期要文(可全文下载) • 上一篇    下一篇

时空点过程:一种新的地学数据模型、分析方法和观察视角

裴韬, 李婷, 周成虎   

  1. 中国科学院地理科学与资源研究所 资源与环境信息系统国家重点实验室, 北京 100101
  • 收稿日期:2013-11-14 修回日期:2013-12-03 出版日期:2013-12-25 发布日期:2013-12-25
  • 作者简介:裴韬(1972-),男,江苏扬州人,博士,研究员,研究方向为时空数据挖掘。E-mail:peit@lreis.ac.cn
  • 基金资助:

    国家自然科学基金面上项目(41171345);国家高技术研究发展计划项目(2012AA12A403)。

Spatiotemporal Point Process:A New Data Model, Analysis Methodology and Viewpoint for Geoscientific Problem

PEI Tao, LI Ting, ZHOU Chenghu   

  1. State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China
  • Received:2013-11-14 Revised:2013-12-03 Online:2013-12-25 Published:2013-12-25

摘要:

栅格计算因其具有简单的构架成为目前地学分析的主流模型,然而,由于栅格计算平均分配计算和存储资源的弱点,不仅容易产生冗余,更重要的是难以凸显研究对象的突变部分,从而使研究者有可能忽略地学现象的变化特征。为此,本文提出将时空点过程模型应用于地学研究。时空点过程不仅适用于模拟以点事件为基本单元的地学现象,而且由于大多数地学过程可以转化为时空点过程,故其具有更广泛的应用范围。因此,时空点过程不仅是一种数据模型,同时也是地学问题的分析方法,更是观察和理解地学问题的一种新视角。为了实现从点过程数据中提取模式,作者经过多年研究提出了时空点过程层次分解理论框架,该理论与信号处理理论中的谱分析思路类似,首先,假设任意点集为有限多个均匀点过程的叠加,然后,通过点局部密度表达工具K阶邻近距离,将空间点转换为混合概率密度函数,再应用优化方法将混合密度函数进行分解得到丛集点和噪声,最终利用密度相连原理从丛集点中提取模式。该理论框架可适用于绝大多数点集数据,初步实现了点集数据的“傅里叶变换”。

关键词: 泊松过程, 数据挖掘, 非均匀点过程, 聚类, K阶邻近距离

Abstract:

The gridding computation is a major model in current geoscientific research due to its simplicity in organizing data resources. However, because the gridding computation equally distributes computational resources, it brings redundancy to the computational process and neglects catastrophe points in geoscientific phenomena, which might overlook the important patterns and bring more uncertainties to the research result. To overcome this weakness, this paper proposes to use the spatial point process model in geoscientific research. The spatial point process model is used to model spatial point based geoscientific phenomenon, also is applied to most of the other geoscientific processes (because they can be transformed into spatial point processes). In this regard, the spatial point process is not only a data model, but also an analysis tool for geoscientific problems. Moreover, it provided a new angle of view for observing geoscientific problems. To extract patterns from point process data, the authors propose the frame of multilevel decomposition of spatiotemporal point process. This frame is similar to the basic idea of signal decomposition. We first assume that any point data set is the overlay of an unknown number of homogeneous point processes. Then, the points are transformed into a mixture probability density function of the Kth nearest distance of each point. After that, the optimization method is used to separate clustering points from noise. Finally, the patterns are extracted using the density connectivity mechanism. The theory can be used to any type of point process data. It can be considered as the "Fourier transform" of point process data.

Key words: Poisson process, data mining, nonhomogeneous point process, clustering, Kth nearest distance