文献类型：期刊文献

中文题名：基于极限学习机的密度聚类离群点检测研究

英文题名：Research on Density Clustering Outlier Detection Based on Extreme Learning Machine

作者：邱华[1];乔涵哲[2];虞董平[1];葛泉波[2];李小凯[1];姜淏予[3]

机构：[1]国家电网浙江省电力有限公司丽水供电公司,浙江丽水323000;[2]杭州电子科技大学自动化学院,浙江杭州310018;[3]广东海洋大学电子与信息工程学院,广东湛江524088

年份：2021

卷号：28

期号：8

起止页码：1676

中文期刊名：控制工程

外文期刊名：Control Engineering of China

收录：CSTPCD、、北大核心、CSCD、北大核心2020、CSCD_E2021_2022

基金：国家自然科学基金资助项目(61803136)。

语种：中文

中文关键词：极限学习机;密度聚类;离群点检测;报文数据;数据挖掘;

外文关键词：Extreme learning machine;density clustering;outlier detection;message data;data mining;

中文摘要：针对智能电网大数据背景下传统密度聚类离群点检测方法在适应性和异常点样本获取成本上的不足,研究一种新的基于权值的密度聚类离群点检测算法,并用极限学习机来预测离群点判断的阈值。对海量历史报文数据进行数据预处理后,将其放入极限学习机进行训练,并预测得到基于权值的局部离群因子(weight-based local outlier factor,WLOF)的阈值。应用预测的WLOF阈值对实时数据进行密度聚类,以实现高性能的离群点检测。最后,采用某电网公司的实际数据进行实验。实验结果表明,所提算法具有较高的检测准确率,尤其适用于阈值未知情况下的离群点检测。

外文摘要：Aiming at the shortcomings of traditional density clustering outlier detection methods in the high cost of adaptability and outlier sample acquisition in the context of smart grid big data,a new density clustering outlier detection algorithm based on weight is studied,and the threshold of outlier judgment is predicted by using extreme learning machine.A large number of historical message data are pre-processed,and then put into the extreme learning machine for training.The threshold of weight-based local outlier factor(WLOF)is predicted.The predicted WLOF threshold is applied to density clustering of real-time data to achieve high-performance outlier detection.Finally,the actual data of a power grid company are used to carry out the experiment.The experimental results show that the proposed algorithm has a high detection accuracy,and is especially suitable for the outlier detection in the case that the threshold is unknown.