【作者】 刘蓉；

【导师】 陈晓红；

【作者基本信息】 中南大学 ， 管理科学与工程， 2006， 博士

【摘要】 多属性复杂大群体决策(Multi-attribute Complex Huge Group-decision，MCHGD)作为一种决策领域新的发展趋势，越来越被理论界和实业界所重视，它迫切需要相应决策方法来提供支持。因此，本文将从新的角度并基于聚类算法来研究多属性复杂大群体决策方法。本文研究的群体复杂性主要体现在：复杂群体行为对群决策的影响、多方案群决策评判准则权重的确定、个体意见发散的一致性修正等，针对考虑以上群体复杂性的大群体高效集结，本文设计了一种改进的聚类算法(Minimum Fuzzy C-means，MFCM)，并提出了基于此算法的具有复杂群体行为的大群体决策方法、基于此算法及优化硬C-均值聚类算法(Weighted Hard C-means，WHCM)的群体一致性修正方法，同时还设计了基于多智能体(Multi-agent)和数据仓库(Data Warehouse，DW)的多属性复杂大群体决策支持系统(Multi-attribute Complex Huge Group-decision Support System，MCHGDSS)框架结构，实现了其中基于聚类算法的MCHGDSS，最后对中国网络消费行为群体决策进行了实证研究。本文主要的研究成果如下：(1)提出了一种改进的聚类算法和基于此算法的具有复杂群体行为的大群体决策方法。本文提出一种将多种复杂群体行为对群决策的影响和大群体高效集结综合考虑的两阶段群体决策理论和方法。第一阶段为群体思维交互决策，通过设计群体思维的五阶段发展过程和基于民主型领导控制的多维空间冲突协调机制来控制复杂群体行为对群决策结果的影响；同时，为更好描述群体成员之间的关系，将群体成员视为一张图，用邻接矩阵表示成员之间的关系，应用MFCM中的全部最小连通支配集算法(Minimum Connected Donating Set Algorithm，MCDSA)得到控制整个群体的民主型领导成员集。第二阶段是群体集结，借助群体思维交互结果，将MFCM继续应用于多属性复杂大群体决策中，能有效解决600个以上成员的集结。接着，本文还定义了群体偏好矢量和群体一致性指标等概念，利用熵权法得到多方案群决策评判准则权重，给出多属性复杂大群体决策结果，最后通过计算机仿真实验和与其它方法的对比分析验证了该方法的有效性和正确性。作为以上新方法的技术基础，本文针对传统模糊C-均值聚类算法(Puzzy C-means，甽中存在的大数据量算法耗时和局部极值等问题，结合图论中最小连通支配集理论给出了一种新的聚类算法MFCM。改进的MFCM算法一方面可以辅助大规模群体集结，另方面由于采用了图论理论，可从关联、控制的角度较好描述复杂群体行为对群决策的影响。(2)提出了基于聚类算法的群体一致性修正方法。针对某个决策方案，本文从新的角度提出了一种基于聚类算法、且能面向较大规模群体、考虑成员学习进化能力并有效收敛群体意见的群体一致性修正方法。本文认为群体一致性修正是一个包含多轮决策过程的不断进化过程，并设计了群体成员学习进化决策程序；接着，利用群体成员相似性和群体一致性的紧密关系，借助一种能够处理大数据量聚类的C-均值类型聚类算法，通过梯度下降法以逐步优化属性权重，并逐渐修正群体一致性来避免因个别成员意见偏离太大而引起的群决策失误；最后通过计算机仿真和实验对比分析验证了该方法的正确、有效性。经过群体一致性修正后，可继续利用基于聚类算法的具有复杂群体行为的大群体决策方法进行决策。(3)提出了一种新的数据挖掘关联算子模型。分布、异构式的多属性复杂大群体决策支持系统需要利用数据挖掘手段获取面向主题的、概括和聚集的信息，本文基于Anindya Datta等提出的一种实现数据挖掘分析的代数模型，通过增加新的关联算子在跨多个多维立方体的分析方面拓展了这个模型。更多还原

【Abstract】 As a new development trend in decision field, the MCHGD (Multi-attribute Complex Huge Group-decision) is accounted by Theory and Business Career, it urgently demands proper decision methods to support the decision processes. Therefore, this thesis will study the methods of MCHGD basing on the clustering algorithm in a different light.According to the problems of group-complexity such as the influence of decision results imposed by complex group-behaviors, the verifying of the weight of alternative group-decision evaluation criteria, and the group-consistency amendment of individual’ s divergence of their owns, in addition, the problem of huge group’s high efficiency aggregation considering complex group characteristics, this thesis outlines the improved MFCM (Minimum Fuzzy c-means) basing on the Graph Theory, and proposes the huge group-decision method with the group-complexity basing on MFCM, the method of group-consistency amendment basing on MFCM and WHCM (Weighted Hard C-means). In the mean time, the thesis presents the framework of the MCHGDSS (Multi-attribute Complex Huge Group-decision Support System) basing on Multi-agent and DW (Data Warehouse), develops the MCHGDSS basing on the clustering algorithm among it. At last, a case research of the group-consume behavior decision of Chinese consumer in network economic environment is performed.Main contents of this thesis are outlined as follows:(1) An improved clustering algorithm and the efficient huge group-decision theory and method with complex group-behavior basing on the algorithm. To combine the group-thinking interaction decision with the efficient huge group-integration decision, the thesis points out a kind of 2-phase group-decision theory and method. First, to deal with the problems of decision quality caused by many kinds of complex group-behaviors, the thesis designs 5-phase development model of group-thinking, and the coordinating mechanism of multi-dimension supervised by the group-leaders which is introduced to control the process of group-thinking interaction. In the mean time, to describe the relations among the group-member, the thesis views the group-member as a picture, representing their relations by matrix adjacency, the democratic leaders’ set is also produced by MFCM. And then the thesis aggregates the huge group efficiently using MFCM continuously, defines group-preference vector and group-consistency index. The result of MCHGD is put forward after getting the weight of alternative group-decision criteria using entropy method. The validity and accuracy of the method is verified by the computer simulation and comparing analysis.As the technical basis of the above method, an improved MFCM is presented to support the MCHGD. To answer the questions of the existing fuzzy clustering algorithms involving local limit value, bad scalability and only for the statistic data, this thesis puts forward a new kind of MFCM basing on MCDSA (Minimum Connected Donating Set Algorithm) which clusters only the dominating points to improve the traditional FCM from the view of overall.The improved MFCM could be used to aggregate the group, on the other hand, it also could be used to particular describe the influence of the complex group-behaviors from the point of correlation and control.(2) A huge group-consistency amendment method with learning ability basing on clustering. According to a certain decision project, the thesis puts forward a group-consistency amendment method basing on a kind of optimized c-means clustering algorithm, aiming at lager scale group, and considering the ability of evolution by learning. First, the thesis regards the group-consistency amendment is an evolutionary process and designs the evolutionary program of the group-member. Then, by using the tightness relations of the group-comparability and the group-consistency, a gradual decision method of amendment group-consistency by optimizing attributes’ weight is presented to avoid the fault caused by individual idea divergence. Lastly, the method is also validated by the computer simulation and comparing analysis. The above efficient huge group-decision theory and method with complex group-behavior could be used to make decision after the consistency amendment.(3) A new DW associational operator model is brought forward. Because the MCHGDSS needs the general and assemble information to support its distribution, the thesis extends the DW analysis algebra model advanced by Datta, realizes the operations among multiple multi-dimensional cubes by adding a kind of new associational operator.更多还原