基于粒化机理的粗糙计算模型研究

All kinds of computing models have been proposed in the area of intelligent information processing. Noting that people always granulation the data during the problem sloving, this project is planned to study the rough computing models under the framework of granular computing. It aims to illustrate the granulation mechanism, the uncertainty in granulation, granulation semantics and the superiority of the corresponding models and may provided new thoughts and methods for the deep study of the rough computing models. The main work of this project are included as follows: (1) The systemically study for granulation mechanism and the corresponding mathematical structures. The uncertainty analyses of concepts, knowledge and rules of the approximation space will also be considered. (2) By employing new granulation metheod, basic semantics granules and the corresponding data granules will be discussed. (3) A system of region granules is explored, the hierachy structures is studied to explore both the quantitative semantics extraction and optimal computing of the models. (4) New rough computing models based on granulation mechanism will be studied, a system of some concrete models and the expansion structures are explored. These studies will give us a further understaning of granulation mechanism and also be helpful in solving the real life complex problems where hierarchy and structure is involved.

针对复杂数据的处理，人们已提出许多计算模型。根据人类在问题处理时会对数据进行粒化这一行为特征，本项目拟运用粒计算方法，进行粗糙计算模型的相关研究，阐明数据粒化过程中的不确定性、粒化的语义及相应模型的形成机理和应用功效，为系统研究基于粒化机理的粗糙计算模型提供研究思路和方法。研究内容主要有：（1）提取一般粒化机制，构建相应近似空间的数学结构，从概念、知识和规则三个层面分析其不确定性；（2）研究粒化后的基本语义粒和基本数据粒；（3）研究基于粒化机理的粗糙计算模型的区域粒体系，构建整体粒层次结构，探讨模型的粒化语义提取与优化计算；（4）研究基于粒化机理的粗糙计算模型的扩张理论，构建具体的粗糙计算模型系统并研究其扩张结构，挖掘良性粒化机制。通过以上研究，本项目将为探讨粒化机制与不确定性的关系奠定基础，并推进基于粒化机理的粗糙计算模型的知识发现方法的发展。

当人们在面对复杂的、难于准确把握的、涉及高维和海量数据的问题时，是通过逐步尝试的办法达到有限的合理的目标，也就是采用逐步求精的多粒度分析法，从而获得足够满意的解。粒计算正是研究将人类这种智能形式化的理论体系，而且有望成为复杂系统中智能信息处理的一种有效的理论框架。本课题组根据粒化机理，分别从粒的不确定性度量，基于粒的模型构造以及基于粒计算的规则提取等三个研究方向对粒计算理论进行了系统、深入、有效的研究工作，并在理论和应用方面取得了有价值的研究成果。课题对传统的粒表示和粒计算的方法进一步扩展与丰富，提升其处理实际数据的能力，并且对粒计算在模式识别以及数据挖掘等相关领域的应用做了积极的研究探索，对于智能信息处理的研究具有重要的理论和应用价值。