【作者】 牛继强；

【导师】 刘耀林；

【作者基本信息】 武汉大学 ， 地图学与地理信息系统， 2010， 博士

【摘要】 近年来由于空间信息技术领域内对地观测技术、数据库技术、网络技术等的飞速发展,使得土地利用数据的获取与管理变得更为便利,我国已经实施的农用地分等定级、更新调查和“全国第二次土地大调查”等工程获得了大量的数据和资料,并建设了土地利用数据库。这些数据的复杂程度和数量远远超出人脑的分析能力,如何快速、定量地从这些大型时空数据库中挖掘有用的特征和知识已经成为土地利用数据库利用的瓶颈问题。空间数据挖掘可以从时空数据库中获取用户感兴趣的空间模式与特征、数据的关联关系以及其他一些隐含在空间数据中的规律和特征,目前已经成为国内外研究的热点。土地用途分区是土地利用规划的核心问题,但是目前还缺乏系统的深入研究,特别是在土地用途分区的智能化方面。因此,针对目前土地用途分区中存在的问题,发展面向领域的空间数据挖掘模型是时空数据不断积累过程中所提出的迫切要求。本文界定了面向土地用途分区的空间数据挖掘的研究内容和体系,并系统研究了该问题的理论方法和应用。基于土地利用分区问题研究的必要性,本文在分析国内外对土地用途分区和空间数据挖掘的研究进展的基础上,建立起面向土地信息的空间数据挖掘的基础理论和技术框架,进一步完善了空间数据挖掘的理论和方法。从土地用途分区、空间数据挖掘的定义出发,定义了面向土地利用分区数据挖掘的概念、特征和内容；提出了一种包括数据层、知识层、挖掘层和人机交互层的四层结构的空间数据挖掘体系结构；阐述领域空间数据挖掘的基本步骤和从土地利用数据库中能发现的知识类型；探讨了土地用途分区数据挖掘的基本方法,主要包括空间计算模型：空间关系度量的方法；空间数据关联规则的挖掘方法：模糊概念格；空间数据聚类分析的方法：人工免疫系统的聚类算法。在对土地用途分区的问题进行描述的基础上,分析了土地用途分区的知识体系,并构建了基于领域知识的土地用途分区模型。概念格是用数学的形式化的方法对从数据中产生概念的过程进行分析的有力工具。这与数据挖掘是从大量数据中产生知识的过程是一致的,因此,概念格理论经过改进是适于对空间数据库进行数据挖掘的。本文针对概念格难以表达空间概念的问题,研究了多值背景下概念格的构建方法,并对形式概念分析理论进行了扩展,研究了基于模糊概念格的土地利用数据空间关联知识的挖掘,构建了面向土地利用的模糊概念格渐进式算法和Hasse图绘制算法,针对土地利用空间数据海量的特征,引入了基于辞典序索引树算法,提出了土地利用空间关联规则的提取方法,以为土地用途分区提供指导。土地用途分区是综合考虑影响土地质量与土地利用方式的各类因素(包括自然、社会、经济方面的因素)的基础上,将研究区域划分为若干均质区片的方法。土地用途分区是一个非常复杂的多目标优化问题。而聚类分析是一种典型的解决组合优化问题的方法。在分析了传统的克隆选择算法的基础上,通过引入混沌理论对其进行了扩展,使用Logistic方程改进了克隆选择算法,并提出两种算法的三种结合方式,构建了混沌免疫克隆选择算法模型(CICSA)。传统聚类方法存在过分依赖数据集聚类原型的问题,为了解决这一问题,本文基于混沌免疫克隆选择算法提出了一种基刁知识的多目标优化聚类模型。该模型是用混免疫克隆选择算法进行聚类,借助混沌免疫克隆选择算子的优势,将进化搜索与随机搜索、全局搜索和局部搜索相结合,通过对候选解进行操作,能够快速得到全局最优解,而不受到样本集方差分布的影响。因此使用混沌免疫克隆选择算法能同时处理多类原型的数据聚类问题,并可以在聚类的过程中获得类数信息。本文在面向土地用途分区的空间数据挖掘的相关理论与技术研究的基础上,研究并开发了原型系统,该软件原型系统包括以下功能模块：土地利用数据管理模块、土地利用知识挖掘模块、土地用途分区挖掘模块、系统库管理模块和可视化表达模块。通过原型系统的开发,进一步明确了面向土地用途分区的空间数据挖掘的功能,解释了土地用途分区的具体过程。选择宜城市土地利用数据库和相关数据,进行数据整合,形成可用于挖掘的整合数据库,并以此数据库进行实验研究,使用模糊概念格获取了土地利用的空间关联规则,并将这些规则和其他领域知识用于混沌免疫克隆选择算法抗体的编码,使用混沌免疫克隆选择算法进行基于多目标的土地用途分区聚类实验,实验结果证明本文所研究的基于知识的土地用途分区聚类挖掘模型是一种智能、高效、准确的分区工具。更多还原

【Abstract】 A rapid development trend emerges in the domain of spatial information technology that contains the earth observation technology, database technology and network technology. Because Spatial information technology provide the convenient ways, in recent years, the land use database was established through actualize the agricultural land classification and gradation, land use investigation and other projects that get a lot of data and information. The complexity and volume of data overstep the analytical capacity of the human brain. How to mining the useful features and knowledge from the land use database become a bottleneck by frequent and quantitative method. From the spatial database, spatial data mining can extract the spatial patterns and characteristics, general relations of spatial and non spatial data, and other data features in common that hidden in the spatial database. As a part of data mining, spatial data mining is a hot issue for the scholars in China or abroad. Land use zoning is one of the core issues, as well as the popular of land use planning. But there is no systematic and intelligent method in land use zoning. It is urgent to develop a domain model of spatial data mining for the problems of land use zoning currently in the course of the accumulation of spatial data. This dissertation defines the research content and system of spatial data mining for land use zoning. Its theory and application is studied systematically.Based on the essentially of land use zoning research and the related research of land use zoning and spatial data mining, the theory and technology framework of spatial data mining for land information is established, and the theory and method of spatial data mining is put forward. Concretely, this dissertation defines the concept, features and content of spatial data mining for land use mining, from the aspect of the concept of land use zoning, spatial data mining. A whole architecture of spatial data mining is bring forward, including the data layer, knowledge level, mining layer and human-computer interaction layer. It is described that is basic steps and how to find the type of knowledge from land use database. Spatial relationship measurement methods that can be considered as space calculation model is a basic method of land use zoning. And then a domain knowledge-based model of land use zoning is designed through describe the issue and analysis the knowledge system of land use zoning.Formal concept analysis theory, also known as concept lattice theory is a powerful tool to analysis the course of from data to concept through the formal method of mathematics.This method is same to the course of data mining that can get the knowledge from large amounts of data. Therefore, the formal concept analysis theory is very suitable for data mining research. Because concept lattice is difficult to express the spatial problem, this dissertation studied the construction algorithm of multi-value context concept lattices. Based on the extention of formal concept analysis theory, a fuzzy concept lattice is proposed to mine the spatial association of knowledge. The incremental algorithm and drawing algorithm of Hasse is established. But for the characteristics of mass spatial data, this algorithm cannot be applicated efficiently. So the index tree is applied in this algorithm to solve the problem of complex spatialsystem. Beside, this dissertation presents a method of acquired for spatial association rules.Land use zoning is a method on divide the study area into a number of homogeneous areas, considering the impact factor of land quality and land use patterns including the physical, social, and economic factors comprehensively. Therefore, land use zoning is a very complex multi-objective optimization problem. The cluster analysis, a typical combinatorial optimization problem, can solve multi-objective optimization problem. Based on the traditional algorithm that over-reliance on data clustering prototype, this dissertation proposes a clustering model of chaos immune clonal selection algorithm (CICSA) based on the knowledge through the Logistic equation of chaos theory. This algorithm can integrate evolution search and random search, global search and local search. Through clonal selection operation on candidate solution, the global optimal solution is acquired quickly, rather than by the variance distribution of sample set. This model can transact data clustering problem with multi-prototype, and the information of classes can be gained automatically.A prototype system was developed based on the study in theory and technology of spatial data mining for land use zoning. This prototype system includes the following modules:data management module of land use, land-use knowledge mining module, land use zoning mining module, the system database management module and visualization modules. By means of prototype system development, the function of spatial data mining for land use zoning is defined, and specific process of land use zoning is explained furthermore. Yicheng, located in Hubei province of China, is an agriculture-based small city. This dissertation selects the land use database and other data, which is integrated by existing mathematical models. These data compose a new data set that can be mining by proposed algorithm. The fuzzy concept lattice is applied to acquire the spatial association rules of land use. This rules and other knowledge that come from domain is used to coding for antibody of CICSA, which is applied to the experiment of multi-objective land use zoning clustering. Experimental results show that spatial clustering for land use zoning based on the knowledge is an intelligent, efficient, accurate zoning tool.更多还原