基于K-DBSCAN的用电行为检测方法  

K-DBSCAN based method for detecting electricity behavior

作　　者：吴春鹏 郑敏 王岳 周飞 Wu Chunpeng;Zheng Min;Wang Yue;Zhou Fei(Artificial Intelligence Research Institute,China Electric Power Research Institute Co.,Ltd.,Beijing 102209,China)

机构地区：[1]中国电力科学研究院有限公司人工智能研究所,北京102209

出　　处：《国外电子测量技术》2024年第12期62-70,共9页Foreign Electronic Measurement Technology

基　　金：国家电网公司科技项目(5700-202358706A-3-3-JC)资助。

摘　　要：针对传统用电行为分析方法难以应对海量数据处理需求问题,提出了一种基于电力物联数据的用电行为检测方法。首先,融合业务和数据特征构建用电行为特征模型,提出了一种最大信息系数和方差融合的特征重要性评估方法。其次,提出了一种结合K均值聚类算法(K-means)辅助定参的密度聚类(DBSCAN)用电行为分类算法,同时通过递归特征消除确定特征集。最后,结合用户侧环境因素预测各类别用户用电负荷,并基于预测的负荷上下限深入分析各类用户的用电行为。实验结果表明,K-DBSCAN算法相较于基线算法,在用户行为聚类效果上实现了3.32%的显著提升。Aiming at the problem that traditional power consumption behavior analysis methods are difficult to cope with massive data processing demands,a power consumption behavior detection method based on power IOT data is proposed.First,the business and data features are fused to construct an electricity consumption behavior feature model,and a feature importance assessment method with maximum information coefficient and variance fusion is proposed.Second,a DBSCAN(K-DBSCAN)power usage behavior classification algorithm combined with K-means assisted parameterization is proposed,while the feature set is determined by recursive feature elimination.Finally,it predicts the electricity load of each type of user in combination with the environmental factors on the user's side,and analyzes the electricity behavior of each type of user in depth based on the upper and lower limits of the predicted load.The experimental results show that the K-DBSCAN algorithm achieves a significant improvement of 3.32%in the clustering effect of user behavior compared with the baseline algorithm.

关 键 词：用电行为检测 特征构造 DBSCAN 负荷预测 

分 类 号：TM715[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程] TN911.7[自动化与计算机技术—控制科学与工程]