【作者】 唐槐璐；

【导师】 须文波；

【作者基本信息】 江南大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 聚类算法在数据分析,数据挖掘等许多地方有广泛的应用,该文探索了基于量子行为的微粒群优化算法(QPSO)及FCM的数据聚类。首先,在分析PSO聚类、QPSO算法聚类的基础上,使用一种新的距离度量方法进行聚类,实验证明了新的度量方法比Euclidean标准更具有健壮性,聚类的结果更精确。在此基础上使用QPSO算法进行数据聚类,实验结果证明了QPSO算法优于PSO算法。QPSO算法不仅参数个数少,随机性强,并且能覆盖所有解空间,保证算法的全局收敛。其次,在QPSO算法中,收缩-扩张系数对于QPSO中的单个粒子的收敛来说是一个至关重要的参数,提出了一种新的聚类算法——适应性的基于量子行为的微粒群优化算法的数据聚类(AQPSO)。AQPSO在全局搜索能力和局部搜索能力上优于PSO和QPSO算法,它的适应性方法比较接近于高水平智能群体的社会有机体的学习过程,并且能保证种群不断地进化。最后,本文针对模糊C均值(FCM)聚类算法存在的缺点,利用量子粒子群优化(QPSO)算法的全局搜索能力,提出了一种新的聚类算法——基于量子粒子群优化的FCM聚类算法(QPSO-FCM)。QPSO-FCM算法先对随机初始点利用QPSO进行优化,然后利用产生的中心点进行聚类。新算法可以降低FCM算法对初始点的敏感度,一定程度上避免了FCM算法易陷入局部极优的缺陷。几组数据实验结果表明,与FCM和PSO-FCM算法相比,本文提出的QPSO-FCM算法聚类结果更可靠。继续在QPSO中使用新的距离公式与FCM相结合,数据表明新的算法能得到更优的结果。更多还原

【Abstract】 Clustering algorithm has a wide application in many fields, for example data analysis and data excavation .In this paper we explore data clustering and the application with Quantum-behaved Particle Swarm Optimization (QPSO) and FCM.Firstly, advancing QPSO algorithm to cluster data based on the PSO clustering and QPSO clustering, I propose a new distance metric in clustering procedures. Experiment results show that this new metric is more robust and accuracy than common-used Euclidean norm, I use QPSO algorithm to data clustering based on the new metric .The experiment results show that QPSO is superior to PSO. Not only the parameters of QPSO are few and randomicity of QPSO is strong, but also QPSO cover with all solution space and guarantee global convergence of algorithms.Secondly, In QPSO, Contraction-Expansion Coefficient is a vital parameter to the convergence of the individual particle in QPSO. In this paper we use adaptive mechanism, therefore we use Adaptive Quantum-behaved Particle Swarm Optimization (AQPSO) to cluster date. The AQPSO outperforms PSO and QPSO in global search ability and local search ability, because the adaptive method is more approximate to the learning process of social organism with high-level swarm intelligence and can make the population evolve persistently.Finally, After analyzing the disadvantages of the Fuzzy C-means (FCM) clustering algorithm,this paper proposes a novel Fuzzy C-means clustering based on Quantum-behave Particle Swarm Optimization algorithm. Not only parameters of QPSO are few and randomicity of QPSO is strong, but also QPSO covers with all solution space and guarantees global convergence. And it avoids the local minimum problems of FCM.At the same time,using QPSO to optimize initial centers first, FCM is no longer a large degree dependent on the initialization values.Numerical experiments show that the proposed algorithm is more robust and accuracy than FCM and PSO-FCM..Using new distance metric in QPSO with FCM, The new QPSO-FCM can get more robust result.更多还原