基于粒计算的信息系统知识发现研究

发布时间：2018-02-08 11:42

本文关键词： 信息系统知识发现粒计算规则提取真值表　出处：《太原理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：信息系统是数据的一种重要的表现形式,从信息系统中通过算法搜索隐藏信息的过程是知识发现的主要内容。真值表是一种特殊形式的信息系统,在数字电路的组合逻辑的应用中占有重要的地位。粒计算是近年发展起来的用来解决复杂问题、处理智能信息的一种新的计算方式。粗糙集是粒计算中重要的理论工具,可以对数据进行分析和推理,从中发现隐含的知识,揭示潜在的规律。规则提取是粗糙集中知识发现的重要研究内容之一,是一种获得信息系统隐含知识的理论方法。本文从研究粒计算和粗糙集理论出发,研究信息系统的知识发现,重点讨论了现有的规则提取算法及所存在的缺陷,基于粒计算提出了新的信息系统的规则提取算法,并针对真值表提出了新的并行约简算法。具体工作如下:首先,针对信息系统的主要形式——决策表,利用粒计算中粒化的思想,从多粒度角度出发,定义判别向量,在由粗到细的粒度空间下分别对决策表进行分析,根据得出的判别向量的元素值提取出信息系统中的规则;而针对不一致决策表,需要将不一致决策表转换为一致决策表,然后进行规则提取。本文通过定理证明和实例分析说明了新算法的有效性,并用UCI数据集与现有的规则提取算法进行了对比试验,实验结果显示了新算法的有效性和快速性。然后,针对信息系统的特殊形式——真值表,首先分析了传统约简算法所存在的缺陷,并基于粒计算知识定义了判别矩阵,在多粒度空间下,根据得出的判别矩阵的元素值提取每个输出的最简规则,实现了真值表的约简,并通过并行计算加快了算法的效率。本文以发光二极管的真值表为例,阐述了新算法计算的具体过程,并比较了公式法、卡诺图法、Q-M算法等传统的真值表约简算法,通过数据集的测试表明新算法具有准确性和快速性。最后,在本文的基础上设计了一个简易的信息系统知识发现系统,该系统集成了现有的一些决策表规则提取算法,并且针对真值表设计了一个对真值表进行约简的子系统,便于用户操作。本文提出的3种信息系统知识发现算法,克服了现有算法的一些弊端,通过算法得到的决策规则在准确性和简易性方面得到了提升,实现了数据的快速规则提取过程。
[Abstract]:Information system is an important representation of data. The process of searching and hiding information through algorithm in information system is the main content of knowledge discovery. Granular computing is a new computing method developed in recent years to solve complex problems and deal with intelligent information. Rough set is an important theoretical tool in granular computing. The data can be analyzed and inferred, and hidden knowledge can be found from it. Rule extraction is one of the important research contents of knowledge discovery in rough sets. This paper studies the knowledge discovery of information system from the perspective of granular computing and rough set theory, and discusses the existing rules extraction algorithms and their defects. A new rule extraction algorithm for information system based on granular computing is proposed, and a new parallel reduction algorithm for truth table is proposed. The main work is as follows: firstly, the decision table, the main form of information system, is proposed. The discriminant vector is defined from the point of view of multi-granularity, and the decision table is analyzed separately in the coarse to fine granularity space, and the rules in the information system are extracted according to the element value of the discriminant vector. For the inconsistent decision table, it is necessary to convert the inconsistent decision table into the consistent decision table, and then to extract the rules. This paper proves the validity of the new algorithm by theorem proof and example analysis. The UCI data set is compared with the existing rule extraction algorithm. The experimental results show the effectiveness and rapidity of the new algorithm. Then, aiming at the special form of information system-truth table, Firstly, the defects of the traditional reduction algorithm are analyzed, and the discriminant matrix is defined based on the granular computing knowledge. In the multi-granularity space, the minimum rule of each output is extracted according to the element value of the discriminant matrix, and the reduction of the truth table is realized. The efficiency of the algorithm is accelerated by parallel computation. Taking the truth table of LED as an example, this paper expounds the concrete process of the new algorithm, and compares the traditional truth-table reduction algorithms such as formula method, Carnot diagram method and Q-M algorithm. The test of data sets shows that the new algorithm is accurate and fast. Finally, a simple information system knowledge discovery system is designed based on this paper. The system integrates some existing decision table rule extraction algorithms. A sub-system is designed to reduce the truth table, which is easy for users to operate. The three knowledge discovery algorithms of information system proposed in this paper overcomes some disadvantages of the existing algorithms. The decision rules obtained by the algorithm are improved in accuracy and simplicity, and the fast rule extraction process is realized.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN79;TP18

【参考文献】