【作者】 甄宇峰；

【导师】 施化吉；

【作者基本信息】 江苏大学 ， 计算机应用技术， 2010， 硕士

【摘要】 研究如何从大量的数据中智能地抽取出有价值的知识和信息,已成为当前人工智能研究中非常活跃的研究领域。目前,知识发现面临着不能有效地处理不完备、不确定性数据以及知识的可解释性比较差的问题。而作为集合论的扩展,粗糙集理论是一种新的软计算方法,可以有效处理模糊的、不确定性知识。它不需要先验知识和外界参数,近年来已经被成功地应用于人工智能、数据挖掘、模式识别等诸多领域。因此,研究基于粗糙集的知识发现方法具有十分重要的意义。本文对粗糙集的基本理论和概念进行了分析和研究,在这些基本理论的框架下,主要做了以下几个方面的研究:(1)粗糙集属性离散化粗糙集的属性离散化要求:离散化结果要保持决策系统的不可分辨关系,以此来确保系统的分类能力不变;求得最小数目的断点集。针对这两点要求,本文首先介绍目前文献中已有的一些连续属性离散化的算法,并对其进行研究与分析,得出这些算法在上述两方面或其他方面的不足;然后,针对这些不足,提出了基于改进遗传算法的数据离散化算法;最后通过实例验证,该算法具有较好的离散化效果。(2)粗糙集属性约简针对传统的粗糙集属性约简算法效率不高,速度不快的问题,本文提出基于条件信息熵和相关系数的属性约简算法,把决策表的非核属性约简过程转化为相关系数的运算,能减少对决策表的扫描次数,降低算法时间复杂度,降低算法冗余,提高属性约简的效率。并利用k-fold轮换对比方法计算相关系数,较大幅度的减少了计算量,同时能得到次优属性约简结果。文中结合实验对算法的性能进行了验证。(3)基于粗糙集的柴油机燃油喷射智能故障诊断系统本文最后将对粗糙集的相关研究应用在故障诊断方面。在对柴油机及其燃油喷射系统进行了介绍之后,基于本文提出的算法构造了柴油机燃油喷射智能故障诊断系统,以帮助工作人员更好地完成故障诊断工作。更多还原

【Abstract】 The study on how to automatically extract valuable information and knowledge from large scale of data has become very active research area in current artificial intelligence research. Nowadays, knowledge discovery is facing the problems that incomplete and uncertain data is not processed effectively and interpretability of knowledge is weak. As a new soft computing method, rough set theory is the extension of set theory, and it is efficient in processing incomplete and uncertain data without knowing prior knowledge and external parameters. And it has been successfully used in areas of artificial intelligence, data mining, pattern recognition, and so on. Therefore, the research of knowledge discovery technology based on rough set theory is of great practical significance.In this dissertation, basic theories and conceptions of rough set are analyzed and studied. And in the framework of them, these researches are done:(1) Discretization of continuous attributes in rough setDiscretization of continuous attributes in rough set requires: the indiscernibility of decision system can not be changed by results of discretization so as to make sure that classification capacity of the decision system is not going to change; and the number of breakpoints in breakpoints set is as small as possible. Aiming at these two requests, firstly some discretization algorithms of continuous attributes are introduced, and they are studied and analyzed to expose their deficiencies on the above aspects or other ones; after that, aiming at these deficiencies, a mew data discretization algorithm based on advanced genetic algorithm is proposed; at last, experiments are carried out to prove its performance.(2) Attribute reduction in rough setAiming at the problem of inefficiency and low velocity with the traditional attribute reduction algorithm, an attribute reduction algorithm based on conditional information entropy and correlation coefficient is proposed, which changes attribute reduction process of non core attributes in the decision table into calculation of correlation coefficient, and reduces the number of scanning decision table, algorithmic time complexity and redundancy of the algorithm, and improves the efficiency of attribute reduction. Then the k-fold rotation comparison method is used to calculate correlation coefficient, which largely reduces calculation amount, and attains sub optimal attribute reduction result. The algorithm details are given and an experiment is carried out, the result of which verifies the efficiency of the algorithm.(3) Intelligent fault diagnosis system of fuel injection system in diesel enginesbased on rough set theoryRelative researches on rough set theory in this dissertation are used in fault diagnosis. After the introduction of diesel engine and its fuel injection system, theoretic basis of fault diagnosis based on rough set theory is analyzed, and an intelligent fault diagnosis system of fuel injection system in diesel engines based on rough set theory is established to help the staff finish fault diagnosis job better.更多还原