摘要随着现实生活中决策问题复杂度的提升，决策中涉及的人员对象也越来越复杂，由单个或几个决策者变成复杂的大规模群体。大规模的对象群体在信息、能力、目标和行为方式等方面具有多样性和分布性，表现为成员来源的异质性、利益的冲突性、评价形式的差异性、协调的复杂性。要做出正确的决策，实现群体的有效协调，就要对对象的分类特征有所了解，因此首要做的就是对大规模的服务对象进行聚类研究，即将大规模群体对象按成员评价信息的相似度划分成若干个数量相对较少的聚集。

    目前关于聚类的研究有很多，但关于大规模群体的聚类研究却很少，一般的聚类算法普遍存在算法伸缩性差、易陷入局部极值等问题，对大规模群体聚类不适用，本文针对此情况，重点研究了FCM聚类算法的伸缩性差与易陷入局部极值等问题，并针对原有算法的缺点做了改进。本文首先对研究大群体决策聚类问题的相关文献做了阐述，并再次基础上较为详细的叙述了聚类算法的定义及分类，接着本文针对FCM聚类算法的缺陷做了改进，较好的解决了原有算法的缺陷，并通过仿真实验证明了此算法的有效性。66358

    本文主要的研究成果如下：针对FCM聚类算法的初始聚类中心的随机性和易陷入局部最优解的缺点，提出了一种改进的FCM聚类算法，新算法运用精简数据的基本思想，提出了一种选择初始聚类中心点的方法，获得了一种改进的FCM聚类算法。该算法首先利用精简数据的基本思想，将数据集精简并得到初始聚类中心点集合，将产生的聚类中心提供给FCM 算法进行再次聚类。经过通过计算机仿真实验和与其他方法的对比分析验证了该方法的有效性和正确性。

毕业论文关键词：群体决策方法、精简数据、FCM聚类算法、复杂大群体

毕 业 论 文 外 文 摘 要

   With the real-life decision-making to enhance the complexity of the problem, the personnel involved in decision-making and more complex objects, by a single or a few large groups of decision-makers become complicated. Large target groups in the information, capabilities, goals and behavior, and so has the persity and distribution, the performance of members of the sources of heterogeneity, conflict of interest, evaluation form differences, coordination complexity. To make the right decisions, to achieve effective coordination groups, it is necessary for an understanding of object classification feature, so the first thing to do is to conduct a large-scale clustering service objects, large group objects upcoming evaluation information by member similarity pided into a number of relatively small number of the congregation.

    Currently there are many studies about clustering, but clustering on large groups rarely, the general algorithm clustering algorithm widespread poor scalability, easy to fall into local and other issues, large-scale group clustering NA, in this paper, this case focuses on the FCM clustering algorithm is poor scalability and easy to fall into local  and other issues, and for the shortcomings of the original algorithm has been improved. Firstly, the study of large group decision clustering problem related literature are described in detail, and again on the basis of a more detailed description of the definition and classification of clustering algorithm, then this defect for FCM clustering algorithm has been improved, better algorithm to solve the original defect, and through the simulation results show that this algorithm.

In this paper, the research results are as follows: For the FCM clustering algorithm randomness of the initial cluster centers and easy to fall into local optimal solution shortcomings, an improved FCM clustering algorithm, the new algorithm uses the basic idea of streamlining the data, presents a selection of the initial cluster center, we obtain an improved FCM clustering algorithm. The algorithm uses the basic idea of streamlining the data, the data set and get the initial clustering centers to streamline the collection, the resulting cluster centers to FCM clustering algorithm again. After the computer simulation experiments and comparative analysis with other methods to verify the effectiveness and correctness.