维度映射下动态非平衡流数据DBSCAN离群点检测  

DBSCAN outlier detection in dynamic imbalanced flow data under dimension mapping

作　　者：秦康平 陈新仪 QIN Kangping;CHEN Xinyi(Dispatching and Control Center of State Grid East China Branch,Shanghai 200120,China)

机构地区：[1]国家电网华东分部调度控制中心,上海200120

出　　处：《电子设计工程》2025年第6期136-139,144,共5页Electronic Design Engineering

基　　金：上海横向科技项目(SGHD0000DKJS2310158)。

摘　　要：现有检测方法在处理高维动态非平衡流数据时,难以捕获数据非线性结构,导致检测结果不精准。为了解决该问题,提出了维度映射下动态非平衡流数据DBSCAN离群点检测方法。分析高维数据维度之间相似性,将高维数据集映射在簇特征上,按照类圆簇两维度映射,计算映射点与维度锚点距离,确定映射点位置,构建数据聚类中心。根据聚类参考点间距离与参考点所代表数据点到聚类中心的距离关系,判断离群对象。采用动态非平衡数据流的滑动窗口,更新窗口中数据近邻数。构建类似簇矩阵,检测离群点。由实验结果可知,研究方法正常点均在象限一,离群点分别在象限二、三、四,与数据集检测理想结果一致,能够精准检测离群点。Existing detection methods find it difficult to capture the nonlinear structure of high-dimensional dynamic unbalanced flow data,resulting in inaccurate detection results.To address this issue,a DBSCAN outlier detection method for dynamic unbalanced flow data under dimension mapping is proposed.Analyze the similarity between high-dimensional data dimensions,reflect the high-dimensional dataset on cluster features,map according to the two dimensions of circular clusters,calculate the distance between mapping points and dimension anchor points,determine the location of mapping points,and construct a data clustering center.Determine outliers based on the distance between cluster reference points and the distance between the data points represented by the reference points and the cluster center.Using a sliding window with dynamic unbalanced data flow to update the number of nearest neighbors in the window.Construct a cluster like matrix to detect outliers.From the experimental results,it can be seen that the normal points of the studied method are all in quadrant one,and the outliers are in quadrant two,three,and four,which are consistent with the ideal results of dataset detection and can accurately detect outliers.

关 键 词：维度映射 动态非平衡流 DBSCAN聚类 离群点检测 

分 类 号：TN919.5[电子电信—通信与信息系统]