基于FSOM神经网络的电能质量数据缺失修复算法  

Power Quality Data Loss Repair Algorithm Based on FSOM Neural Network

作　　者：杨挺 何周泽 赵东艳 盆海波 姜含 蔡绍堂 原义栋 蔡玉朋 YANG Ting;HE Zhouze;ZHAO Dongyan;PEN Haibo;JIANG Han;CAI Shaotang;YUAN Yidong;CAI Yupeng(School of Electrical Automation and Information Engineering,Tianjin University,Nankai District,Tianjin 300072,China;Beijing Smartchip Microelectronics Technology Co.,Ltd.,Haidian District,Beijing 100080,China;State Grid Liaoning Electric Power Research Institute,Shenyang 110006,Liaoning Province,China)

机构地区：[1]天津大学电气自动化与信息工程学院,天津市南开区300072 [2]北京智芯微电子科技有限公司,北京市海淀区100080 [3]国网辽宁省电力有限公司电力科学研究院,辽宁省沈阳市110006

出　　处：《电网技术》2020年第5期1941-1949,共9页Power System Technology

基　　金：国家电网有限公司总部科技项目(2019GW-12)。

摘　　要：电力物联网承载的电气量测数据在采集、传输、转换等各个环节中受到干扰而导致数据出现缺失,影响状态估计精度和系统稳定运行。针对传统修复策略仅考虑一维量测数据横向分布规律造成数据修复精度较低的不足之处,充分考虑电力系统量测数据缺失点的邻域数据以及量测数据的周期性变化规律,提出一种基于模糊自组织(fuzzy self organizing map,FSOM)神经网络的电能质量量测缺失数据修复方法。首先,通过将电能质量一维测量数据映射为二维灰度图像,提升数据间的时-空相关性解析。进而采用人工智能FSOM神经网络算法对原始数据进行聚类,析构出数据的多层特征值,进行对聚类后数据的分层修复。最后,以劳伦斯伯克利国家实验室实测电能质量数据为基础实验验证FSOM算法性能。实验结果表明,无论是在随机缺失还是连续缺失情况下,提出的FSOM修复算法比现有算法在数据低丢失率和高丢失率下都有更低的修复误差和更高的信噪比。The ubiquitous electric measurement datacarried by the Internet of things are interfered in various linkssuch as collection, transmission and conversion, resulting in theloss of data and affecting the accuracy of state estimation andthe stable operation of the system. Due to the low data repairaccuracy caused by one dimension measurement datatransverse distribution of the traditional repair strategy, thispaper fully considers the neighborhood data of the powersystem measurement data missing points and the periodicvariation law of those data and puts forward a fuzzy selforganizing map(FSOM) neural network based power qualitymeasurement missing data repair method. Firstly, theone-dimensional measurement data of power quality aremapped into two-dimensional grayscale images to improve thetime-space correlation analysis between the data. Then, theoriginal data are clustered with artificial intelligence FSOMneural network algorithm to dissect their multi-layercharacteristic values so that they can be given a hierarchicalrepair. Finally, FSOM algorithm performance is verified by theexperiments of testing power quality data of LBNL(LawrenceBerkeley national laboratory). The experimental results showthat the FSOM repair algorithm proposed in this paper haslower repair error and higher signal-to-noise ratio than theexisting algorithm in the case of whether random or continuousloss.

关 键 词：FSOM神经网络 量测数据 二维映射 缺失数据修复 

分 类 号：TM721[电气工程—电力系统及自动化]