基于粒子群优化聚类算法的研究.docx

摘要　聚类分析是数据预处理的一种重要工具，是一种无监督分类方法．计算机及网络技术的发展，为聚类分析的应用提供了广阔的舞台；特别是在模式识别、图像分割、计算机视觉和模糊控制等诸多领域，聚类分析都扮演着重要的角色．　随着其应用的拓广，传统的聚类分析方法正面临着越来越大的挑战，特别是在正确的分类和大数据集及高维数据的处理上，传统聚类方法有时候根本就不能完成任务．　本文通过对粒子群算法和聚类方法的探索，针对上述问题做出如下工作：　　１、对基于传统距离的聚类算法在高维数据聚类中所面临的困难及传统距离度量高维数据对比性下降的问题，文中定义了一种面向高维数据的距离度量，来改善聚类算法在处理高维数据时的性能．为克服划分方法容易陷入局部最优的缺陷，将之与智能算法中粒子群算法相结合，以种群寻优的方式，寻找全局最优解．　２、将粒子群优化算法善于全局寻优的特点和聚类算法中的逼近思想结合提出了基于粒子群的动态聚类算法，其主要思想是：首先，将整个的聚类过程看作一个复杂的优化问题，然后通过效用函数的判定逐层进化、逐步寻找最优的聚类及聚类类别数目，即在不断的进化和逼近过程中得到最优解．在该算法中，我们使用效用函数来评价寻优结果的优劣，故效用函数选择尤为重要．　　３、在高维数据的聚类及类别的寻优上，文中将定义的距离应用于基于双粒子群的动态聚类算法中，并改写了效用函数．最后，对高维数据的聚类寻优做了有益的尝试．　　关键词：粒子群优化；聚类分析；模糊聚类；效用指标　IＡｂｓｔｒａｃｔ　Cluster analysis is an important tool for data preprocessing and an unsupervised classificationmethod. The application of cluster analysis is becoming broader as the development of computer andnetwork technology. Cluster analysis is playing an important role, especially in pattern recognition,image segmentation, computer vision and fuzzy control, and many other fields.Along with the extension of the application of the cluster analysis, the traditional method of clusteranalysis is facing increasing challenges, particularly in proper classification, large data sets andhigh-dimensional data processing; sometimes the traditional clustering method can not complete thetask.In this paper, we study the particle swarm optimization and clustering method. To address theseissues, we do some work as follows:1. For the difficulties of the clustering high dimensional date by the traditional distance-basedclustering algorithm and the problem of declining contrast in the high dimensional data by traditionaldistance.In this paper, we define a new distance to measure the high dimensional data, and to improvethe performance of clustering algorithm addressing the high dimensional data. To overcome theshortcomings of partitional methods which are easy to fall into local optimization, we combined theparticle swarm optimization with partional methods to find the globa