不确定性知识获取的粒计算方法及其应用研究

In recent years, lots of uncertainty measures based on the ideas of granular computing are proposed by national and international experts. However, most of them have more or less limitations or shortcomings. In particular, much less attention has been paid to the fusion of corresponding measure methods. Then, the main objective of this project is to solve the problems above. It follows that theories and methods of many novel uncertainty measures are presented, and the fusion formulas between these measures and the current corresponding measures are derived. Thus, a measurement system of uncertainty knowledge acquistion is proposed. In covering approximate space, the measures of roughness and fuzziness based on rough-fuzzy set model and rough-vague set model are investigated, and then their relatively unified measurement criterion is established. To overcome the limitations of dynamic data processing in rough set theory, the evolution model of granular decision based on time series and its predictive algorithm are proposed. Then, the approximate iterative formulas of decision rules are presented and the evolutionary tracks of decision rules are established. Furthermore, granular computing methods are introduced into gene expression data processing. Some extended rough set models, which are based on relative neighborhood relation, generalized binary relation, and covering granulation-based mutually exclusive relation, are presented, and then many feature gene selection algorithms are proposed. Meanwhile, granular computing methods are applied in image retrieval. The rough granular space model is proposed and some new image retrieval algorithms based on the proposed model are put forward. The image information retrieval models based on the probabilistic rough set and the image semantic annotation technology are constructed. These research results achieved in this project will provide some effective theories and methods to measure and process uncertain information for knowledge acquisition.

近年来，国内外专家基于粒计算思想提出了很多不确定性度量理论与方法，但这些理论与方法大都存在一定的局限性或不足，特别是对相关度量方法进行融合的研究较少。本项目提出一些新的不确定性度量理论与方法，研究这些度量方法与现有相关度量方法之间的融合形式，建立不确定性知识获取的度量系统。在覆盖近似空间下研究粗糙模糊集、粗糙Vague集的粗糙性和模糊性度量，建立其相对统一的度量标准；针对粗糙集理论在处理动态数据方面的局限性，从时间序列角度提出粒度决策演化模型及其预测算法，建立决策规则近似迭代公式和演化轨迹；将粒计算方法应用于基因数据处理，构建基于相对邻域关系、广义二元关系和覆盖粒相斥关系的扩展粗糙集模型，提出一些特征基因选择算法；将粒计算方法应用于图像检索，提出粗糙粒度空间模型及其图像检索算法，建立基于概率粗糙集的图像语义信息检索模型。该项目研究成果可为不确定性知识获取提供有效的理论和方法保证。

粒计算是当前人工智能领域中研究大规模复杂问题求解、大数据分析与挖掘、不确定性智能信息处理的有力工具。按照项目计划书的要求，项目组围绕不确定性知识获取的粒计算方法及其应用进行了系统而深入的研究工作，具体研究内容包括：给出了条件粗糙熵、不完备粗糙熵等不确定性度量方法，研究了这些度量方法之间的融合形式；在覆盖近似空间下研究了粗糙Vague集的粗糙性和模糊性度量；从时间序列角度，提出了粒度决策演化模型及其预测算法，建立了决策规则近似迭代公式和演化轨迹；基于相对邻域关系、广义二元关系和覆盖粒相斥关系构建了粗糙集扩展模型，提出了一些特征基因选择算法；提出了粗糙粒度空间模型及图像检索算法，建立了基于概率粗糙集的图像语义检索模型。该项目在理论和应用方面取得了有价值的研究成果，对传统的粒计算方法进行了扩展和丰富，对粒计算在模式识别、不确定性建模以及数据挖掘等相关领域的应用做了研究探索，为不确定性知识获取提供了重要的理论和技术支持。