基于图卷积神经网络与K-means聚类的居民用户集群短期负荷预测  

Short-term Load Forecasting of Residential User Groups Based on Graph Convolutional Neural Network and K-means Clustering

作　　者：董雷 陈振平 韩富佳 王晓辉 蒲天骄 DONG Lei;CHEN Zhenping;HAN Fujia;WANG Xiaohui;PU Tianjiao(State Key Laboratory of Alternate Electrical Power System With Renewable Energy Sources(North China Electric Power University),Changping District,Beijing 102206,China;China Electric Power Research Institute,Haidian District,Beijing 100192,China)

机构地区：[1]新能源电力系统全国重点实验室(华北电力大学),北京市昌平区102206 [2]中国电力科学研究院有限公司,北京市海淀区100192

出　　处：《电网技术》2023年第10期4291-4301,共11页Power System Technology

基　　金：国家电网公司科技项目(5400-202112507A-0-5-ZN)。

摘　　要：随着智能电表等高级量测装置在用户侧的广泛部署与使用,海量多源异构的居民用户数据得以采集与存储,为用户级负荷预测提供良好的数据基础。精准的居民用户集群负荷预测是促进智能配电网需求侧管理、辅助电网公司实现削峰填谷的重要基础。然而,现有的用户级负荷预测方法大多利用历史负荷序列的时间相关性构建数据驱动模型,却忽视相邻用户用电行为之间存在的潜在空间相关性。因此,提出一种基于K-means聚类和自适应时空同步图卷积神经网络的居民用户集群负荷预测方法。首先,采用K-means聚类将居民用户集群按照用电行为相似性划分成不同组;然后,基于居民用户集群的分组数量、各组居民用户的历史负荷数据以及各组居民用户负荷序列之间的相关性,构建面向居民用户集群负荷预测的时空图数据;最后,使用自适应时空同步图卷积神经网络实现居民用户集群短期负荷预测。文章通过真实的爱尔兰居民用户负荷公开数据集测试并验证所提方法的准确性和有效性,实验结果表明,相较于各个基准预测方法,所提方法能够充分挖掘并利用不同居民用户用电负荷之间的时空相关性,进而提高居民用户集群负荷预测精度。With the widespread deployment and use of advanced measurement devices such as smart meters on the user side,massive multi-source heterogeneous residential user data can be collected and saved,providing a good data foundation for user-level load forecasting.Accurate residential user cluster load forecasting is an important basis for promoting demand-side management of smart distribution networks and assisting power grid companies to achieve peak shaving and valley filling.However,most of the existing user-level short-term load forecasting methods use the temporal correlation of historical load sequences to build data-driven models,but ignore the potential spatial correlation between adjacent users'electricity consumption behaviors.Therefore,this paper proposes a cluster load forecasting method for residential users based on K-means clustering and adaptive spatiotemporal synchronous graph convolutional neural network.First,K-means clustering is used to divide the residential user clusters into different groups according to the similarity of electricity consumption behavior;then,based on the number of groups of residential user clusters,the historical load data of each group of residential users,and the sum of the load sequences of each group of residential users The spatial-temporal graph data for the load forecasting of residential user clusters is constructed based on the correlation between them.Finally,the adaptive spatiotemporal synchronous graph convolutional neural network is used to realize the short-term load forecasting of residential user clusters.This paper tests and verifies the accuracy and effectiveness of the proposed method on a real public dataset of Irish resident user load,and the experimental results show that,compared with each benchmark traditional prediction method,the proposed method can fully mine and utilize different resident users.The spatial and temporal correlation between electricity loads can be used to improve the prediction accuracy of residential user cluster load.

关 键 词：智能配电网 用户级负荷预测 居民用户集群 图数据 时空同步图卷积神经网络 

分 类 号：TM721[电气工程—电力系统及自动化]