没有合适的资源?快使用搜索试试~ 我知道了~
计算机研究 -基于量子粒子群优化的自动聚类算法研究.pdf
1.该资源内容由用户上传,如若侵权请联系客服进行举报
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
版权申诉
0 下载量 6 浏览量
2022-06-28
03:54:34
上传
评论
收藏 668KB PDF 举报
温馨提示
试读
52页
计算机研究 -基于量子粒子群优化的自动聚类算法研究.pdf
资源推荐
资源详情
资源评论
摘 要
摘 要
现如今,科技领域多学科交叉,学科间交互渗透已然成为众多学科的所共同
具有的特征。由信息科学与生命科学共同结合而成的智能计算即是这类新的交叉
学科。在已存在的智能计算领域中的所有群体智能算法,都是起源于科学家模拟、
分析自然界中生物群体行为的群智能算法(Swarm Intelligence Algorithm)。该类算
法在近几年中成为研究的热点,其中作为群体智能算法的典型代表,粒子群算法
(Particle Swarm Optimization, PSO),由于其具有易于实现、高效等特点受到人们
的重视,而孙军等人提出的量子粒子群算法(Quantum-Behaved Particle Swarm
Optimization, QPSO),不但具有传统粒子群算法的优点,更克服了其非全局收敛
的缺点。聚类(Clustering)这一概念,一直存在于人们生活之中,聚类方法也为解
决实际问题而服务。多年来,聚类已成为数据挖掘领域的一个重点。然而,随着
社会的发展,实际问题越来越复杂,以往那种需要预先确定类别数的算法已不能
满足现代复杂的实际问题。而实现数据的自动聚类近年来引起学术界的广泛关
注,成为当今科研的热点。核(Kernel)方法亦是模式识别方向的一种方法。当前
核方法已在诸多方面如支持矢量积、高斯过程等方面获得应用,而其由低维向高
维转变问题的特点决定了核方法在数据聚类领域能够使得原来的那些聚类方法
性能有所提高。本论文综合上述三个方面,提出了基于量子粒子群算法的自动核
聚类方法。不但能够很好地解决球形数据自动聚类问题,而且能够实现非球形数
据的自动聚类。
本文主要工作如下:
1) 提出了基于 QPSO 算法的自动聚类算法。该方法充分利用了 QPSO 算法
的全局收敛性,能克服早收敛的特点,从而精确找寻到待分类数据集的准确分类
个数。
2)针对上一算法无法对非簇状数据进行处理的问题加以改进,加入核的思
想。在保证对簇状数据集分类准确度不变的情况下,实现对非簇状数据集的分割。
3) 利用引入权重的量子粒子群算法(WQPSO 算法)在原算法的基础上做进行
进一步的改进。通过引入带权重值,使得算法更贴近实际应用,并使用 UCI 数据
对有权重和无权中的两种自动聚类算法做了对比实验。
本文得到如下基金资助:国家自然科学基金:61272279 和 61001202;中国
博士后科学基金特别资助:200801426;中国博士后科学基金:20080431228 以
及中央高校基本科研业务费专项资金资助:JY10000902040。
关键词: 自动聚类 群体智能 量子理论 核方法 量子粒子群算法
Abstract
Abstract
Today, technology is developing rapidly, and multidisciplinary field of science and
technology has become a common feature of many disciplines of the interdisciplinary
interaction penetration. The combination of intelligent computing information science
and life science is a kind of new interdisciplinary.
All swarm intelligence algorithms
already exists in the field of intelligent computing, are originated in the scientist
simulation. What’s more, analysis of the behavior of groups of organisms in nature
swarm intelligence algorithms (swarm intelligence algorithm), such algorithms in recent
years become a hot research field which as a typical representative of the swarm
intelligence algorithms. PSO (particle swarm optimization PSO) algorithm, due to it is
easy to implement, efficient attention has been paid, and the quantum particle swarm
algorithm Sun Jun et al (quantum-behaved particle swarm optimization, QPSO) not
only has the advantage of traditional particle swarm algorithm but overcome the
shortcomings of its non-global convergence as well. The concept of clustering ,which
has come into people's lives, clustering methods to solve practical problems and
services, over and over the years, clustering has become a focus of the field of data
mining, with the development of society, however, the actual of the problem is more
complex, and the past that need to pre-determine the number of categories algorithm
already can not meet the practical problems of modern complex, and automatic data
clustering cause academic attention in recent years, become the hotspot of the research
today. Kernel method is also a method of pattern recognition direction. Recent years,
Kernel methods are applied in many aspects, such as support vector product, Gaussian
process, which is determined by the characteristics of the low-dimensional to high
dimensional changes Kernel who makes the original clustering method even more
powerful in the field of data clustering. In this thesis the three proposed nuclear
clustering methods based on quantum particle swarm algorithm are not only a good
solution to the spherical automatic data clustering problem but non-spherical data
automatic clustering as well.
This thesis is organized as follows:
1) Based the QPSO algorithm automatically clustering algorithm, the method
takes full advantage the QPSO global convergence of the algorithm,
overcomes the premature convergence characteristics to accurately find
Abstract
accurate classification number to be disaggregated data set.
2) For the previous algorithms can not be non-clustered data improved, thought to
join the nuclear guarantee clustered data set classification accuracy of the same
case, the non-clustered data set segmentation.
3) The use of the quantum particle swarm algorithms improved further
improvements to the original algorithms; quantum particle swarm algorithm
through the introduction of a weighted value, the algorithm is closer to the
actual application.
This research is supported by the National Natural Science Foundation of
China(Grant Nos.61272279 and 61001202),the China Postdoctoral Science
Foundation Special funded project (No. 200801426), the China Postdoctoral
Science Foundation funded project (No. 20080431228) and the Fundamental
Research Funds for the Central Universities (No.JY10000902040).
Keywords: Automatic Clustering Swarm Intelligence QPSO
Quantum Theory Kernel Method
剩余51页未读,继续阅读
资源评论
programyp
- 粉丝: 86
- 资源: 1万+
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功