基于网格的并行聚类算法及数据流聚类算法研究-计算机软件与理论专业论文.docxVIP

兰州大学硕士学位论文 兰州大学硕士学位论文 基于网格的并行聚类算法及数据流聚类算法研究 Abstract Clustering analysis，as an important task of data mimng，has wide application fields．These different applications raise some novel requirements for clustering analysis algorithm． This thesis proposes a novel grid-based parallel clustering algorithm for multi—density datasets，called PGMCLU．The innovative works of it are as follows． Define the concepts，including grd compactness，grid density-connected，grid feature， cluster density and cluster similarity．Propose the method for data partition based on grd partition，the method for local clustering based on grid density-connected concept， and the method for merging local clusters based on cluster similarity measure．Realize the adaptive set for parameter minPts．PGMCLU algorithm can better handle high—dimensional and massive datasets，and Can be capable of identifying clusters with distinguished shape and density． Data stream is a sequence composed of a series of infinite，successive， high—speed，and time-ordered data objects．Data stream has the characteristics of real—time and infinity,which determines that clustering algorithm for data stream compared with traditional clustering algorithm for static dataset has some distinguished properties． This thesis proposes the grid-based clustering algorithm for data stream，shorten for GC-Stream．The innovative works of it are as follows．Propose the concept of grid feature vector for describing the grid summary information．Improve the SP—Tree structure，and propose the novel spatial index structure LSP—Tree based on List data structure．Propose the exponential damped strategy for grid information，and the pruning strategy for noisy grid and outdated grid．GC—Stream algorithm con better satisfy the real-time requirement of data stream clustering,and can be adaptive for memory size． ．Detailed and complete experiments have proved the correctness and effectiveness of PGMCLU and GC—Stream algorithm，therefore，these novel algorithms will have s