基于GCN-GRU的短期时空负荷预测方法  

Spatio-temporal load forecasting method based on GCN-GRU

作　　者：朱力 李成 郭龙 刘云鹏 史炯 Zhu Li;Li Cheng;Guo Long;Liu Yunpeng;Shi Jiong(State Grid Hubei Electric Power Company Xiangyang Power Supply Company,Xiangyang 441002,China;Wuhan Flyminer Technology Co.,Ltd.,Wuhan 430070,China)

机构地区：[1]国网湖北省电力公司襄阳供电公司,湖北襄阳441002 [2]武汉飞脉科技有限责任公司,湖北武汉430070

出　　处：《能源与环保》2022年第4期211-215,221,共6页CHINA ENERGY AND ENVIRONMENTAL PROTECTION

基　　金：国网湖北省电力有限公司科技项目(5215D020001X)。

摘　　要：随着城市的快速发展,城市配电系统也进行了快速的扩展。空间负荷预测研究可以指导电力系统的管理与调度,并且其准确性会影响到方案的合理性。首先分析并总结了常用空间负荷预测方法的特点,然后提出了GCN-GRU时空负荷预测模型。GCN-GRU模型充分利用图神经网络在网络拓扑数据方面的优势以及GRU在时间序列建模方面的优势,对电网进行建模,考虑了负荷的空间特性和时间特性,并将影响负荷的因素转换为特征向量进行算法训练,提高了负荷预测的准确度。最后以湖北省某市区电网为研究对象,证明了该方法的有效性。With the rapid development of city,the urban power distribution system has also undergone rapid expansion.Space load forecasting research can guide management and dispatch of power systems,and its accuracy will affect rationality of plan.This article first analyzes and summarizes the characteristics of commonly used spatial load forecasting methods,and then proposes GCN-GRU spatiotemporal load forecasting model.The GCN-GRU model makes full use of advantages of graph neural network in network topology data and the advantages of GRU in time series modeling,modeling the power grid,considering the spatial and temporal characteristics of the load,and transforming factors that affect load algorithm training for feature vectors improves the accuracy of load forecasting.Finally,the power grid of a certain urban area in Hubei Province is used as research object to prove effectiveness of method.

关 键 词：配电网 空间负荷预测 图神经网络 GRU 网络拓扑 

分 类 号：TV213.4[水利工程—水文学及水资源]