基于非洲秃鹫优化算法改进的密度峰值聚类  

Improved density peak clustering approach based on African vultures optimization algorithm

作　　者：罗兴隆 贺兴时[1] 周洁[1] 杨新社[2] LUO Xinglong;HE Xingshi;ZHOU Jie;YANG Xinshe(School of Science,Xi'an Polytechnic University,Xi'an 710600,Shannxi,China;School of Science and Technology,Middlesex University,Cambridge CB21TN,UK)

机构地区：[1]西安工程大学理学院,陕西西安710600 [2]密德萨斯大学科学与技术学院,英国剑桥CB21TN

出　　处：《山东大学学报（理学版）》2024年第1期46-55,71,共11页Journal of Shandong University(Natural Science)

基　　金：国家自然科学基金资助项目(12101477);陕西省教育厅自然科学基金资助项目(21JK0654)。

摘　　要：密度峰值聚类算法是一种自动寻找簇中心的新型快速搜索算法。针对其截断距离的不确定和一步式分配策略不稳健等缺陷,本文提出一种基于非洲秃鹫优化算法改进的密度峰值聚类算法。通过准确率(accuracy,Acc)这一评价指标建立优化问题的目标函数,利用非洲秃鹫优化算法强大的寻优能力对不确定的截断距离dc进行优化,降低了人为取值的不准确性;其次,根据数据集密度均值将其划分为高低不同的密度区域,对不同区域采用不同的分配策略,针对高密度区域内的数据点采用与原密度峰值聚类相同的分配方法,对低密度区域内数据点则根据其k近邻数量进行聚类;最后,将该算法在合成和真实数据集上进行实验验证,算法的聚类性能有了很大的提升,且对密度差异性较大的数据集划分也更加精确。Density peak clustering algorithm is a new fast search algorithm for automatically finding cluster centers.Aiming at the uncertainty of its cut-off distance and the instability of the one-step allocation strategy,an improved density peak clustering approach based on African vultures optimization algorithm is proposed.The objective function of the optimization problem is established through evaluating accuracy(Acc),and the uncertain cut-off distance dc is optimized by the powerful optimization ability of the African vultures optimization algorithm,which reduces the inaccuracy of artificial values.Secondly,according to the average density of the data set,it is divided into different density areas,and different allocation strategies are used for different areas.For data points in the high-density area,the same allocation method as the original density peak clustering is used,and for data points in the low-density area,the k-nearest neighbor method is used for clustering.Finally,the algorithm is experimentally verified on synthetic and real data sets,the clustering performance of the algorithm has been greatly improved,and the division of data sets with large density differences is also more accurate.

关 键 词：非洲秃鹫优化算法 密度峰值 截断距离 分配策略 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]