基于Spark的电能计量数据异常辨识方法设计  

Design of Spark-Based Anomaly Recognition Method for Electricity Metering Data

作　　者：杨劲锋 郑楷洪 刘玉仙 张伟 曾璐琨 YANG Jinfeng;ZHENG Kaihong;LIU Yuxian;ZHANG Wei;ZENG Lukun(China Southern Power Grid Company Limited,Guangzhou 510663,China;Southern Power Grid Digital Grid Research Institute Co.,Ltd.,Guangzhou 510663,China;School of Computer Science,Guangdong Polytechnic Normal University,Guangzhou 510665,China)

机构地区：[1]中国南方电网有限责任公司,广东广州510663 [2]南方电网数字电网研究院有限公司,广东广州510663 [3]广东技术师范大学计算机科学学院,广东广州510665

出　　处：《自动化仪表》2025年第4期92-96,共5页Process Automation Instrumentation

基　　金：中国南方电网有限责任公司营销技改基金资助项目(JY-KF-02-QT-21-031-TQ)。

摘　　要：为了准确辨识电能计量数据中的异常数据、提高电能计量数据质量,提出一种基于Spark的电能计量数据异常辨识方法。以Spark框架为核心支撑,基于张量核范数约束的低秩张量补全模型补全电能计量数据后,采用最小生成树并行聚类方法进行聚类。依据Spark并行优势改进K-means算法,形成基于Spark改进的K-means并行算法,以辨识聚类后电能计量数据中的异常数据。采用蝙蝠算法优化辨识方法的关键参数,优化电能计量数据异常辨识结果。测试结果表明,该方法可以全面、完整地实现电能计量数据补全,标准互信息(NMI)和调整兰德指数(ARI)的最大值分别为0.984和0.988,因而聚类效果好。该方法能够有效辨识出不同类型的异常数据,为电力的综合运行管理提供可靠依据。To accurately identify abnormal data in electricity metering data and improve the quality of electricity metering data,a Spark-based anomaly recognition method for electricity metering data is proposed.With the Spark framework as the core support,the electricity metering data are complemented by a low rank tensor completion model based on tensor kernel paradigm constraints and clustered by the minimum spanning tree parallel clustering method.The K-means algorithm is improved based on the parallel advantage of Spark,and the K-means parallel algorithm is formed based on the improvement of Spark,to recognize the abnormal data in the clustered electricity metering data.The bat algorithm is used to optimize the key parameters of the identification method and optimize the results of abnormal identification of electricity metering data.The test results show that the method can comprehensively and completely realize the electricity metering data complement,and the maximum values of the normalized mutual information(NMI)and the adjusted Rand index(ARI)are 0.984 and 0.988,respectively,thus the clustering effect is good.The method can effectively identify different types of abnormal data and provide a reliable basis for the comprehensive operation and management of electric power.

关 键 词：电能计量 Spark框架 异常辨识 数据补全 数据聚类 K-MEANS算法 蝙蝠算法 

分 类 号：TH-86[机械工程]