【作者】 杨恶恶；

【导师】 王坚强；

【作者基本信息】 中南大学 ， 管理科学与工程， 2013， 博士

【摘要】 语言多准则决策是现代多准则决策研究的一个重要方向,在诸多领域中有着广泛的的应用背景。尽管针对语言多准则决策的研究己取得了丰硕的成果,但无论理论还是应用方面,现有研究均还存在许多尚待解决的问题,尤其是在处理语言信息的不确定性特征时还很不成熟。为此,本文提出了双语言集的概念,并针对双语言环境下的不确定多准则决策问题加以系统研究,本文的主要工作包括：(1)提出了双语言集的概念。可以利用它同时表达对备选方案在准则下表现的评价和决策者对该评价的信心水平。这一工具与传统的语言变量相比,能够更好的包容决策过程中来自不同源头的不确定性。(2)定义了双语言集间的优势关系,并在此基础上建立了方案间的级别高关系。利用级别高关系,综合方案间两两比较的结果,可以得到方案间的不完全序。(3)在上述级别高关系的基础上,进一步扩展得到了一种双语言多准则分组评级决策方法。该方法利用一组标志性的虚拟参考方案,通过备选方案与这些虚拟参考方案相比较而将方案划分进合适的分组中。适用于备选方案较多、不便于方案间两两比较时对方案进行初步评价的决策环境。本文将其用以处理城市绿化树种选择问题,取得了较好的效果。(4)提出了语义占优技术,并证明了五种典型的语义结构下所对应的语义占优规则,研究了不同语义占优的性质。利用这一组规则,提出了基于语义占优的双语言多准则决策步骤,从而挑选出“非劣”方案的集合。(5)为了尽可能多获得的方案间偏好关系,并减少对语言变量的语义设定施加过多人为影响,通过建立语义规划模型对语言变量的语义加以设定,语言变量中包含的不完全偏好通过模型的限制条件表示,并提出了两种利用语义规划技术的双语言多准则决策方法。(6)针对权重信息不完全的双语言多准则决策问题进行了研究,提出了基于扩展语言运算的两种决策方法。其中一种方法将决策过程看作是决策者与自然间的博弈,利用矩阵博弈理论对准则权重加以设定；而另外一种方法则通过最大化模型离差来设定权重。(7)提出了直觉正态云模型来处理双语言多准则决策问题,决策者给出的决策信息被看作是方案的综合评价云中部分云滴的集合,通过这些云滴可估计出该评价云的相关参数,随后,利用所设计的云发生算法运用蒙特卡洛技术产生云滴并对云滴计分加以统计,进而对不同方案的综合评价云加以比较排序。(8)将所提出的双语言多准则决策技术应用于新能源公交车选型决策案例中,通过综合运用上述方法,可以揭示蕴含在双语言决策信息之内的决策者偏好,帮助决策者更好的理解问题和影响方案间偏好的具体因素,从而帮助其得到更加理性的决策结果,提高决策质量。更多还原

【Abstract】 Linguistic multi-criteria decision-making (MCDM) problem is an important research topic in the nowadays MCDM theory. It has widely application background in many fields. Although the linguistic MCDM research has achieved fruitful results, the existing methods still have many unsolved problems in both theory and application aspects. Especially, they are very immature in handling the uncertainty in linguistic information. Thereby, the duplex linguistic (DL) sets are introduced in the thesis. Moreover, the uncertain MCDM problems under duplex linguistic environment are studied systematically. It includes:(1) Defined the DL set, which can be used to express the evaluation for an alternative with respect to a criterion and the confidence on such evaluation simultaneously. Comparing to the classical linguistic variable, the DL set can comprise the uncertainty from different sources.(2) Defined the dominance relation between DL sets, and established the outranking relation between alternatives based on such dominances. Integrating the outranking relations after pairwise comparing the alternatives, the partial order of alternatives is reached.(3) Based on the outranking relation above, a DL multi-criteria classifying and rating method is proposed. By comparing the alternatives to a group of virtual reference alternatives, the evaluated alternatives are classified into the proper groups. This method is suitable for solving the problem that involves too many alternatives to compare them pairwise. An urban tree species selection was conducted by using this method.(4) The semantic dominance (SD) technique is proposed. The SD rules about five typical semantics structures were proved, and the properties of SD were studied. The DL MCDM procedure based on SD was introduced to find all the "non-inferior" alternatives.(5) For obtaining the preferences about the alternatives as much as possible, and avoiding assigning the semantics to the linguistic variables artificially, a semantic programming model was introduced to set the semantics to the linguistic variables. The incomplete preference involving in the linguistic variables are expressed by the constraints of the model. Two DL MCDM methods based on semantic programming are introduced.(6) For handling the DL MCDM problems with incomplete weights, two methods based on the computing with expanded linguistic variables are proposed. One of them regards the decision as a game between decision makers and the nature, and set the weights with a matrix game. The other method sets the weights to maximize the deviation of the model.(7) The intuitionistic normal cloud model is proposed for solving the DL MCDM problems. The decision information is regarded as the drop sets of the clouds that evaluate the alternatives. Parameters of these clouds then can be estimated from such drops. Further, the drops of these clouds are generated by using the cloud-generating algorithm. The statistical results of the drop score can be used to rank the clouds.(8) The methods proposed are used to select the alternative fuel bus species. These methods reveal more detail of the preference that involves in the DL decision information. They help the decision maker to understand the problem and the factors that affect the preferences, thus make decision more reasonable.更多还原