基于OPTICS聚类算法的流场结构特征分析方法  

A modified OPTICS clustering algorithm for analyzing flow characteristics

作　　者：王锐 辛大波 欧进萍[1] WANG Rui;XIN Dabo;OU Jinping(School of civil engineering,Harbin Institute of Technology,Harbin 150090,China;School of civil engineering,Northeast Forestry University,Harbin 150040,China)

机构地区：[1]哈尔滨工业大学土木工程学院,哈尔滨150090 [2]东北林业大学土木工程学院,哈尔滨150040

出　　处：《空气动力学学报》2021年第5期27-43,共17页Acta Aerodynamica Sinica

基　　金：国家重点研发计划重点专项项目(2018YFC0809600,2018YFC0809605);国家自然科学基金面上项目(51878131)。

摘　　要：为了改进现有的基于聚类分析的流场结构特征分析方法,使之更加适用于结构风工程领域的风场特征识别与分析,依托聚类分析的思想,结合一种基于密度的OPTICS聚类算法,并引入相关距离的概念替换了原算法中的欧氏距离,提出了采用一种基于相关距离的OPTICS聚类算法进行流场结构特征分析。实例分析利用基于大涡模拟的计算流体动力学数值模拟,对低雷诺数下经典圆柱绕流问题进行了瞬态求解,获取了2000个圆柱尾流中顺流向涡的瞬态涡量场样本。然后,以识别圆柱尾流中的顺流向涡旋涡脱落状态和顺流向涡A模式展向分布为目标,对比了k-means、原始的OPTICS算法和基于相关距离的OPTICS聚类算法等流场结构特征分析方法。分析实例的结果表明,基于相关距离的OPTICS算法能够在合适的初始参数设置下有效识别顺流向涡脱落状态和其A模式展向分布间距,相对k-means算法降低了对初始参数的敏感度,聚类结果稳定且唯一。Current clustering-based methods still need improvement for the flow features identifying and analysis in the field of wind engineering.Therefore,this paper proposed a modified OPTICS(Ordering Points to Identify the Clustering Structure)algorithm to extract flow features.It is a density-based clustering method with the Euclidean distance replaced by the correlation distance.It is employed to study the streamwise vortex shedding and the spanwise distribution of the A-mode in a low-Reynolds-number circular cylinder wake flow obtained by Large Eddy Simulation.Its performance is further compared with the k-means,and the original OPTICS method.The results indicated that the OPTICS algorithm based on the correlation distance can identify the streamwise vortex shedding and the spanwise distribution of the A-mode effectively with reasonable initial parameters.Compared with the k-means,this method is insensitive to the initialization parameters and its results are consequently more stable and well-determined.

关 键 词：流场结构特征 聚类分析 相关距离 OPTICS算法 顺流向涡 

分 类 号：O302[理学—力学]