基于边缘计算的台区短期负荷预测方法  

Short-term substation load forecasting method based on edge computing

作　　者：张明泽 栾文鹏 艾欣[1] 刘博 ZHANG Mingze;LUAN Wenpeng;AI Xin;LIU Bo(School of Electrical and Electronic Engineering,North China Electric Power University,Beijing 102206,China.;School of Electrical Automation and Information Engineering,Tianjin University,Tianjin 300072,China)

机构地区：[1]华北电力大学电气与电子工程学院,北京102206 [2]天津大学电气自动化与信息工程学院,天津300072

出　　处：《电测与仪表》2024年第4期93-99,共7页Electrical Measurement & Instrumentation

基　　金：国家电网有限公司科技项目(520600230011)。

摘　　要：配变台区是配电物联网与用户交互的重要纽带,台区短期负荷预测对实现配电物联网的精益化管理具有重要意义。为缓解全部台区上传负荷数据所带来的通信压力,提出一种基于边缘计算的台区短期负荷预测方法,将台区智能配变终端存储30天的历史负荷数据作为样本数据,通过核平滑法对样本数据进行清洗,因样本数据较少考虑将样本归一化后,拆分为标幺曲线与基值分别计算提高预测结果精度。然后通过相关系数法构建历史负荷数据的相关系数矩阵,用相关系数矩阵替换仿射传播相似度矩阵后聚类求得相似日的标幺曲线,再通过加权求和求得待测日的标幺曲线。同时,按照相似日原理预测待测日基值,最终通过待测日标幺曲线和基值反归一化后得到待测日负荷曲线完成预测工作。通过山东某配变30天的历史负荷数据计算验证,表明所提方法可以实现台区负荷量级小、样本少、波动大情形下的合理预测,占用主站计算资源较少,对配网精益化运维具有积极意义。Distribution substation is an important link between distribution IoT and user interaction,and the substation short-term load forecasting is of great significance for achieving the lean management of the distribution Internet of Things.In order to alleviate the communication pressure caused by uploading load data from all substations,this paper proposes a short-term substation load forecasting method based on edge computing.The 30-day historical load data stored by the intelligent distribution terminal is used as sample data,and the sample data is cleaned using Nadaraya-Watson method.Due to the small amount of sample data,it is considered to normalize the sample and split it into standard unit curves and base value.Then,PCC matrix of historical load data is constructed,and the unit curve of similar days is obtained through replacing the affine propagation(AP)similarity matrix with the correlation coefficient matrix,the unit curve of the similar day is obtained by clustering,and the unit curve of the test day is obtained through weighted summation.At the same time,forecast the base value of the test day and ultimately obtain the load curve of the test day.The historical load data of a distribution substation in Shandong Province for 30 days show that the proposed method can achieve reasonable prediction under the condition of small load magnitude,small sample and large fluctuation,and it occupies less computing resources in the main station.It has a positive significance for the lean operation and maintenance of the distribution network.

关 键 词：配电物联网 智能终端 短期负荷预测 仿射传播聚类 数据挖掘 

分 类 号：TM715[电气工程—电力系统及自动化]