基于粒计算的动态知识发现中若干关键问题研究

项目名称： 基于粒计算的动态知识发现中若干关键问题研究

项目编号： No.60873108

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 李天瑞

作者单位： 西南交通大学

项目金额： 30万元

中文摘要： 粒计算研究的核心问题之一是在粒度变换中如何利用已有的信息来实现快速动态知识发现。粗糙集是粒计算的一个主要分支，其中的近似集可直接用来导出决策规则。因此，粒度变换下如何设计增量更新近似集和知识维护方法已成为人们研究的一个热点问题。针对信息系统中属性集粗化细化，提出了变精度粗糙集等模型下动态更新近似集方法，建立了知识增量更新的数学模型，设计了动态维护知识算法。刻画了对象集粗化细化时集值粗糙集等模型中近似集的变化机理，给出了其动态维护方法。刻画了多粒度多层次属性值粗化细化的原理，建立了属性值粗化细化时增量计算变精度粗糙集等模型中近似集的一些理论与方法。设计了并行计算经典粗糙集中近似集的方法，并提出了对象、属性集同时粗化细化时动态更新其近似集的增量方法。设计了基于矩阵的计算概率粗糙集和邻域粗糙集等模型中近似集的方法。对所提出的这些方法利用公共数据集等进行了性能评测，验证了其具有高效性。并提出限制容差关系变精度粗糙集、累计变精度粗糙集、偏好关系变精度粗糙集和集值有序变精度粗糙集等模型以适应复杂现实问题。为系统建立处理动态变化及大规模复杂现实信息的粗糙集和粒计算理论与方法提供了学术思想与技术路线。

中文关键词： 知识发现；粒计算；数据挖掘；粗糙集

英文摘要： One of the core issues of granular computing is how to use the previous information to achieve dynamic knowledge discovery under the transformation of granularities. Rough set theory is a major branch of granular computing. Its approximations can be directly used to induce decision rules. Therefore, how to design approaches to incrementally update approximations and dynamically maintain knowledge under the transformation of granularities has been a hot research topic. Aiming to the coarsening and refinement of the attribute set of information systems, approaches for dynamically updating approximations in the Variable Precision Rough Set model (VPRS) etc. were presented, a mathematical model for dynamic maintenance of knowledge was established and algorithms for incrementally learning knowledge were developed. The variation mechanism of approximations in the set-valued rough set model etc. under the coarsening and refinement of the object set was characterized. Furthermore, approaches for dynamic maintenance of their approximations were proposed. In addition, principles for coarsening and refinement of multi-granularity and multi-level attribute values were outlined. Then, several properties and methods for incrementally updating approximations in VPRS etc. were given when attribute values are coarsened and refined. A parallel method for computing approximations in classic rough sets was designed. Accordingly, an incremental method for calculating its approximations was developed when the attribute and object sets are coarsened and refined simultaneously. Matrix-based approaches for computing approximations in the probabilistic rough sets and neighborhood rough sets, etc were presented. The effectiveness of all the proposed methods was validated by performance evaluations on public data sets. Moreover, a limited tolerance relation based VPRS, a cumulative VPRS, a preference relation based VPRS and VPRS on set-valued ordered information systems, et al. were proposed to adapt to complex practical problems. All these outcomes may contribute to provide feasible theoretical support and technical routes for establishing the theoretical system of rough set and granular computing aiming to deal with dynamic, massive and complex data.

英文关键词： Knowledge Discovery; Granular Computing; Data Mining; Rough Sets