【摘要】：聚類分析為非監(jiān)督分類的一種,是將一組數(shù)據(jù)或?qū)ο笸ㄟ^某種特定規(guī)則分成不同類的過程。跟監(jiān)督分類相比,雖然沒有較高的分類精度,但是不需要先驗知識,是數(shù)據(jù)或?qū)ο髢?nèi)部之間的聚合,在實際應(yīng)用中能夠得到更好的應(yīng)用。但是,由于聚類分析問題初始聚類中心敏感,且容易陷入局部最優(yōu),針對此問題,許多改進方法被提出,群體智能優(yōu)化算法的聚類分析是其中重要的研究方向,群體智能優(yōu)化算法的聚類分析是通過將聚類問題歸結(jié)為一個優(yōu)化問題,然后通過全局并行搜索方式進行啟發(fā)式搜索。本文主要的研究重點是群體智能優(yōu)化算法的聚類分析問題,通過對常用的聚類分析方法和經(jīng)典的群體智能優(yōu)化算法聚類分析方法進行分析,研究其過程以及存在問題,針對聚類分析初始聚類中心敏感的問題,提出基于兩種新型群體智能優(yōu)化算法的聚類分析,即基于煙花算法的聚類分析和基于混合編碼方式的聚類分析,并通過其與傳統(tǒng)聚類分析方法和經(jīng)典群體智能優(yōu)化算法聚類分析方法的比較,評價其性能。基于煙花算法的聚類分析,是將煙花算法這種新型的智能優(yōu)化算法應(yīng)用到聚類分析中,該算法將兩種搜索策略不同的煙花算法進行結(jié)合,分別采用實數(shù)編碼和二進制編碼的方式,提出了基于兩種編碼方式的煙花聚類算法,并通過仿真實驗,對兩種不同編碼方式的算法性能進行分析,通過實驗得出二進制煙花算法的聚類分析聚類效果好、穩(wěn)定性高,并且分類精度高于經(jīng)典的群體智能優(yōu)化算法的聚類分析。基于混合編碼方式的聚類分析,是將基于聚類中心的編碼方式和基于樣本編號的編碼方式混合,并且在不同編碼方式下分別采用QPSO和改進的雨林算法進行聚類分析。通過仿真實驗得出,使用聚類中心的編碼方式,在搜索過程中,容易產(chǎn)生超出搜索空間的解,從而使搜索陷入局部最優(yōu)。而用樣本編號的編碼方式,搜索空間范圍固定,雖然便于控制搜索范圍,但是限制了搜索空間的范圍,不利于進一步提高最優(yōu)解的質(zhì)量�；诨旌暇幋a方式的聚類分析算法,既解決了超出搜索空間問題,同時能夠保持種群多樣性,并且通過實驗比較,分類精度優(yōu)于傳統(tǒng)的聚類分析方法。
[Abstract]:Cluster analysis is a kind of unsupervised classification, which is the process of dividing a group of data or objects into different classes through a specific rule. Compared with supervised classification, although there is no higher classification accuracy, but there is no need for prior knowledge, it is an aggregation between data or objects, and can be better applied in practical applications. However, because the initial clustering center of clustering analysis problem is sensitive and easy to fall into local optimization, many improved methods are proposed, and clustering analysis of swarm intelligence optimization algorithm is one of the important research directions. The clustering analysis of swarm intelligence optimization algorithm is based on the clustering problem, which is reduced to an optimization problem, and then heuristic search is carried out by global parallel search. The main research focus of this paper is the clustering analysis of swarm intelligence optimization algorithm. Through the analysis of the commonly used clustering analysis methods and the classical swarm intelligence optimization algorithm clustering analysis method, the process and existing problems are studied. In order to solve the problem of sensitivity of initial clustering center in clustering analysis, clustering analysis based on two new swarm intelligence optimization algorithms is proposed, that is, clustering analysis based on fireworks algorithm and clustering analysis based on hybrid coding. The performance of the method is evaluated by comparing it with the traditional clustering analysis method and the classical swarm intelligence optimization algorithm clustering analysis method. Based on the clustering analysis of fireworks algorithm, a new intelligent optimization algorithm, is applied to clustering analysis. The algorithm combines two fireworks algorithms with different search strategies. The fireworks clustering algorithm based on two coding methods is proposed by using real number coding and binary coding respectively, and the performance of the two different coding methods is analyzed through simulation experiments. The experimental results show that the clustering effect of binary fireworks algorithm is good, the stability is high, and the classification accuracy is higher than that of classical swarm intelligence optimization algorithm. The clustering analysis based on hybrid coding is to mix the coding method based on clustering center and the coding method based on sample number, and QPSO and improved rainforest algorithm are used for clustering analysis under different coding methods. The simulation results show that using the coding method of clustering center, it is easy to produce solutions beyond the search space in the search process, which makes the search fall into local optimization. However, using the coding method of sample number, the search space range is fixed, although it is convenient to control the search space, but it limits the scope of the search space, which is not conducive to further improving the quality of the optimal solution. The clustering analysis algorithm based on hybrid coding not only solves the problem of beyond search space, but also maintains the diversity of population. Through experimental comparison, the classification accuracy is better than the traditional clustering analysis method.
【學位授予單位】：哈爾濱工程大學
【學位級別】：碩士
【學位授予年份】：2016
【分類號】：TP18;TP311.13

【參考文獻】

相關(guān)期刊論文 前10條

1 饒U，

本文編號：2485850