【作者】 刘盾；

【导师】 胡培；

【作者基本信息】 西南交通大学 ， 管理科学与工程， 2011， 博士

【摘要】 随着社会的不断进步和科技的迅猛发展,多属性决策理论及其方法已在管理决策领域得到了广泛的应用。然而,面对信息系统“日益复杂”和“动态变化”的今天,如何调节决策过程中单目标与多目标、静态决策与动态决策间的矛盾；如何制定合理有效的权重及规则挖掘机制；如何从外界环境变更中获取动态决策规则,实现知识的高效更新,已成为现代决策理论所面临的重点和难点。本文在充分考虑管理决策语义环境下,将粗糙集理论引入多属性决策过程,回答了如何从众多粗集模型选取适用的决策模型,以及怎样将这些模型应用到管理决策系统中的问题。进而,分别从属性权重设定和规则获取两个方面由浅入深、循序渐进地探讨粗集多属性决策的理论和方法。首先,通过综合分析现有权重设定的不足,将粗集理论与信息熵引入多属性决策权重确立中,提出一种基于粗集和信息增益的属性权重获取方法来解决权重的主观性和冗余性的不足,并与其他决策方法(AHP等)进行对比研究。其次,考虑到经济管理系统的数据大多具有偏好特征,将偏好关系引入概率粗集模型中,提出一种新的基于偏好下的概率度量关系,并分别建立在完全信息和不完全信息条件下的基于偏好关系的概率粗集模型,提出一种适应于管理信息系统的规则获取方法。由于该方法充分考虑到决策尺度的弹性问题和决策属性中“序”的特征,这使得决策结果更具说服力。再者,考虑当外界环境发生变化后,如何寻找合理有效的动态决策学习模型及策略的角度出发,分别从属性集不变,对象集变化、对象集不变,属性集变化,以及对象集和属性集不变时,属性值变化三个方面来探讨动态粗集多属性决策问题,并提出一系列新的解决动态决策问题的权变方法。实验结果表明,新的方法不仅在决策效率上有较大提高,而且使得决策过程更为直观、简单,这为人们提供了一种新的解决动态决策难题的思路。本论文基于管理信息系统的视角,结合数学中的粗集理论及计算机相关数据挖掘技术,从简单到复杂、从静态到动态,初步而又系统地建立一套解决多属性决策问题的体系,并在某种程度上弥补了现有多属性决策方法的缺陷,具有一定的理论意义和应用价值。更多还原

【Abstract】 With the advancement progress of society and the fast development of technology, the researches on the theories and approaches of multiple-attribute decision making have receive great achievements in management science. However, as the result of the information system is becoming more complex and constantly changing all the time in nowadays, the approaches of how to adjust the conflict between single target and multiple targets (criteria), static environment and dynamic environment; how to acquire the reasonable and efficiency weight and rule mining mechanism, how to obtain the dynamic decision rules and updating strategies for generating knowledge, have become the new keystone and difficult in decision making problems. By considering of the semantic environment in management decisions, rough set theory is induced into multiple-attribute decision procedure. The problems of how to choose the property rough set models and how to use them in management decision system is clearly discussed in our paper. In addition, the two problems about weight setting and rules acquirement are proposed step by step to illuminate the theories and approaches of the multiple-decision making, respectively.Firstly, observed by the lack the weight setting, rough set theory as well as information entropy are induced into multiple-attribute decision making, an approach for attribute weights acquisition based on rough sets theory and information gain is propose to overcome the subjectivity and redundancy, and our approach is also compare with other methods (i.e. AHP).Secondly, with the insightful gain from the ordinality and inconsistency in real management information system simultaneously, the preference-orders relation is induced into Probabilistic rough sets model (PRS), and a Probabilistic model of strict-dominance-based rough set approaches (P-SDRSA) based on complete information system and incomplete information system are proposed respectively. Due to these approaches consider the flexibility. problem and the "order" character in decision making process, it make the decision result more reasonable and suitable for us to acquire the decision rules in management decision environment.In addition, considering the changes of environment, three different models and strategies are proposed for the knowledge incremental learning in dynamic decision system, including the following three cases:(1) The object set in the information system evolves over time while the attribute set remains constant; (2) The attribute set in the information system evolves over time while the object set remains constant; (3) The attribute value in the information system evolves over time while the object set and attribute set remain constant. These models give a series of new approaches to deal with the dynamic decision problems. Furthermore, the experiments results not only provide the efficiency of our approaches, but also make the decision process simpler and more clearly, which gives us a new viewpoint to solve the dynamic decision puzzles.Overall, rough set theory in mathematic and data mining technologies in computer science are integrated into multiple-attribute decision making in this paper, a new approach based on the view of management information system is set up to solve the multiple-attribute decision problems from simple to complex, from static to dynamic, which remedy the defects of classical multiple-attribute decision methods somehow. In a short, these researches have some theoretical significance and clinical value in multiple-attribute decision making studies.更多还原