尺度驱动的时空集聚模式挖掘模型与方法

In spatio-temporal databases, there are usually spatio-temporal clusters formed by objects with similar attribute values that occur together across space and time. Discovery of spatio-temporal clusters plays a key role in understanding spatio-temporal patterns by allowing us to learn from the past and to prepare better for the future. It has been known that most geographical phenomena are scale-dependent, and modeling scale-dependency of spatio-temporal clusters is vital for understanding the evolving patterns of geographical phenomena. However, the scale-dependency of spatio-temporal clusters is still not well handled by existing methods, thus the characteristics of geographical phenomena cannot be fully revealed by using the discovered spatio-temporal clusters. On that account, this project aims to develop a scale-driven model for detecting spatio-temporal dynamic clustering patterns through mimicking the physiological model of multi-scale human visual perception. In this project, scale is quantitatively modeled as parameter in clustering, and scale-driven spatio-temporal clustering model will be developed. Specially, the project mainly consists of three parts: (i) quantitatively modeling of the multi-scale features of spatio-temporal clustering patterns; (ii) construction of multi-scale spatio-temporal clustering model; (iii) assessment of validity of the multi-scale spatio-temporal clustering patterns. The method developed in this project will enrich the methodology of spatio-temporal data mining, and the scale-driven clustering model will be helpful for mining the spatio-temporal evolving rules of geographical phenomena. The scale-driven clustering model will also be applied in geographic national condition monitoring, and it will be used to discover spatio-temporal clustering patterns of air particulate matter.

地理时空数据中广泛存在着由时空邻近且专题属性相似的实体所构成的时空集聚模式，这类模式对于研究地理现象的演化规律与形成机理具有重要价值。时空多尺度特性是地理现象的内蕴特征，亦是时空集聚模式挖掘所需要关注的核心内容。现有研究尚没有在挖掘模型中充分考虑时空集聚模式的尺度依赖特征，导致挖掘结果辅助理解地理现象的能力不足。本项目受到人类视觉多尺度认知生物机制的启发，拟将尺度特征定量建模并融入于时空集聚模式挖掘模型中，研究一套尺度驱动的时空集聚模式挖掘模型与方法，主要包括：（1）时空集聚模式多尺度特征定量描述；（2）时空集聚模式多尺度挖掘模型构建；（3）多尺度时空集聚模式有效性评价。本项目研究将有利于从多尺度的视角对地理现象的时空变化规律进行深层次探究，丰富时空数据挖掘的理论方法体系。本项目提出的模型与方法将应用于地理国情监测中的空气颗粒物多尺度时空分布模式挖掘，提高地理国情监测辅助决策的能力。

地理时空数据中广泛存在着一系列由专题属性相似的、时空邻近的实体组成的潜在结构模式，即时空集聚模式。时空集聚模式挖掘对于深入研究地理现象的发展变化规律与形成机理具有重要的价值。本项目针对当前时空集聚模式挖掘模型缺乏尺度依赖性建模能力的缺陷开展研究，受到人类视觉系统多尺度模式识别能力启发，建立了时空集聚模式多尺度挖掘的理论与方法，主要成果包括：1）创新性地采用简洁的统计学方法构建了仿人类视觉系统的时空集聚模式多尺度挖掘理论框架，将时空集聚模式多尺度挖掘建模为以尺度为参数的假设检验问题，解决了尺度参数化描述、尺度依赖性形式化描述的难题；2）建立了数据尺度与分析尺度联合效应统计建模方法，克服了挖掘结果有效性判断难、可解释性差的弊端；3）提出了时空集聚模式多尺度挖掘模型尺度参数自适应控制方法，解决了尺度改变与最优尺度筛选的难题。本项目研究成果提升了对时空集聚模式尺度依赖机制的建模能力，为从多尺度的视角对地理现象进行深入剖析提供理论方法支撑。