基于花朵授粉算法的软子空间聚类算法优化研究

发布时间：2018-10-05 19:36

【摘要】：随着信息技术、数据收集和存储技术的不断发展,数据规模逐渐扩大、维度逐渐增高,传统聚类算法受高维数据稀疏性和维度灾难的影响无法进行有效聚类,为解决高维数据聚类问题,软子空间聚类分析技术应运而生并得到广泛关注。软子空间聚类通过描述各样本隶属于不同簇的不确定性进行聚类,具有更好的适应性和灵活性、更接近于客观世界。但现有软子空间聚类算法主要存在以下两方面不足:采用随机选取样本点的方法初始化聚类中心,导致算法的聚类精度和稳定性依赖于初始簇心;采用的局部搜索策略,导致算法在聚类过程中易陷入局部最优。本文针对上述问题进行深入研究,具体研究内容如下:(1)针对聚类结果依赖于初始簇心的问题,本文对快速搜索算法(CFSFDP)进行优化,通过引入投影分区和类合并技术,提出一种基于投影分区及类合并技术优化算法(PM-CFSFDP),可以获得更加精准的类中心点。将PM-CFSFDP作为初始化步骤应用于软子空间聚类中,为其选择最佳聚类中心,降低算法对初始簇心的依赖。(2)针对在聚类过程中易陷入局部最优的问题,本文对花朵授粉算法(FPA)进行优化,通过引入混合蛙跳思想和自适应高斯变异策略,提出一种基于自适应高斯变异的混合蛙跳花朵授粉算法(AGM-SFLFPA),可有效避免陷入局部最优且收敛速度较快。将AGM-SFLFPA作为全局优化搜索策略应用于软子空间聚类中,为其搜索最优权值,有效避免陷入局部最优。(3)将两个改进的算法PM-CFSFDP和AGM-SFLFPA引入软子空间中,提出一种基于花朵授粉算法的软子空间聚类算法(FPASC)。在UCI标准数据集上的实验结果表明,在处理高维数据时,FPASC算法可降低对初始簇心的依赖,避免在搜索过程中陷入局部最优,有效提高了软子空间算法的聚类精度和稳定性。
[Abstract]:With the development of information technology, data collection and storage technology, the scale of data is gradually expanding and the dimension is gradually increasing. The traditional clustering algorithm is unable to cluster effectively because of the sparsity of high-dimensional data and the disaster of dimensionality. In order to solve the problem of high dimensional data clustering, soft subspace clustering analysis technology emerged as the times require and received wide attention. Soft subspace clustering by describing the uncertainty of samples belonging to different clusters has better adaptability and flexibility and is closer to the objective world. However, the existing soft subspace clustering algorithms mainly have the following two shortcomings: the clustering center is initialized by randomly selecting sample points, which results in the clustering accuracy and stability of the algorithm depend on the initial cluster center, and the local search strategy is adopted. As a result, the algorithm is prone to fall into local optimum in the process of clustering. The main contents of this paper are as follows: (1) aiming at the problem that the clustering results depend on the initial cluster center, this paper optimizes the fast search algorithm (CFSFDP), and introduces the projection partition and class merging techniques. An optimization algorithm based on projection partitioning and class merging (PM-CFSFDP) is proposed to obtain more accurate center points of classes. PM-CFSFDP is applied to soft subspace clustering as an initialization step to select the best clustering center to reduce the dependence of the algorithm on the initial cluster center. (2) aiming at the problem that the clustering process is prone to fall into local optimum. In this paper, the flower pollination algorithm (FPA) is optimized by introducing the mixed leapfrog idea and adaptive Gao Si mutation strategy. A hybrid leapfrog flower pollination algorithm (AGM-SFLFPA) based on adaptive Gao Si mutation is proposed, which can effectively avoid falling into local optimum and converge quickly. AGM-SFLFPA is applied to soft subspace clustering as a global optimal search strategy to search the optimal weights. (3) two improved algorithms, PM-CFSFDP and AGM-SFLFPA, are introduced into soft subspace. A soft subspace clustering algorithm (FPASC). Based on flower pollination algorithm is proposed. The experimental results on the UCI standard data set show that the algorithm can reduce the dependence on the initial cluster center, avoid falling into local optimum in the search process, and effectively improve the clustering accuracy and stability of the soft subspace algorithm.
【学位授予单位】：中国矿业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;TP311.13

【参考文献】