基于集成图卷积变分变换器的电力负荷数据补全方法  

Power Load Data Completion Method Based on Integrated Graph Convolutional Variational Transformer

作　　者：严莉 呼海林 史磊 吴钦政 吕天光[2] 徐英东 张闻彬 王高洲 YAN Li;HU Hailin;SHI Lei;WU Qinzheng;LÜTianguang;XU Yingdong;ZHANG Wenbin;WANG Gaozhou(Information and Telecommunications Company,State Grid Shandong Electric Power Company,Jinan 250001,China;College of Electrical Engineering,Shandong University,Jinan 250061,China)

机构地区：[1]国网山东省电力公司信息通信公司,济南市250001 [2]山东大学电气工程学院,济南市250061

出　　处：《电力建设》2025年第4期49-57,共9页Electric Power Construction

基　　金：国网山东省电力公司科技项目(52062723000A)。

摘　　要：【目的】随着电力系统的发展和能源系统规模的不断扩大,产生了海量的负荷功率数据,但在数据采集、传输中不可避免会发生数据缺失现象,这极大制约了系统协调优化和高级数据应用的发展。【方法】为此,提出了一种基于集成图卷积变分变换器(integrated graph convolutional variational transformer,IGCVT)网络的新型电力负荷缺失数据补全模型。IGCVT模型将改进的图卷积网络(graph convolutional network,GCN)和Transformer模型利用变分自编码器(variational auto-encoder,VAE)的架构进行聚合。通过GCN对原始数据进行处理,以学习空间特征并深度挖掘空间依赖关系;利用VAE对隐藏层数据进行重构,更为有效地还原数据的分布特性;基于Transformer模型对序列时间自相关信息进行挖掘。此外,引入了改进的鲸鱼优化算法(whale optimization algorithm,WOA)以优化网络模型超参数,以提高补全精度和模型的适用性。同时,针对电力负荷数据极端变化点补全误差较大的问题,采用了数据双向补全方法,充分利用缺失点前后的数据信息。【结果】实验结果表明,与基准模型相比,均方根误差(root mean square error,RMSE)指标分别提升了24.3%、44.0%和47.9%,验证了所提方法的优越性。【结论】文章所提方法为解决电力负荷数据缺失问题提供了可行的解决方案,并有望进一步扩展该模型的应用范围。[Objective]With the development of power systems and continuous expansion of energy systems,massive load power data have been generated.However,missing data are inevitable in the collection and transmission of power data,which greatly restricts the development of system-coordination optimization and advanced data applications.[Methods]To this end,this paper proposes a new power load missing data completion model based on an integrated graph convolutional variational transformer(IGCVT)network.The IGCVT model aggregates an improved graph convolutional network(GCN)and Transformer model using the variational auto-encoder(VAE)architecture.The raw data are processed by the GCN to learn spatial features and deeply mine spatial dependencies;the hidden layer data are reconstructed by the VAE to more effectively restore data distribution characteristics;and the temporal autocorrelation information of the sequence is mined based on the Transformer model.In addition,an improved whale optimization algorithm(WOA)is introduced to optimize the network model hyperparameters and improve the completion accuracy and applicability of the model.Simultaneously,to solve the problem of large errors in the completion of extreme change points of power load data,a two-way data completion method is adopted to make full use of the data information before and after the missing points.[Results]Experimental results show that,compared with the baseline model,the RMSE index is improved by 24.3%,44.0%,and 47.9%,which verifies the superiority of the proposed method.[Conclusions]The results show that the proposed method provides a feasible solution to the problem of missing power load data and is expected to further expand the application scope of the model.

关 键 词：数据补全 图卷积网络 Transformer模型 电力负荷数据 

分 类 号：TM714[电气工程—电力系统及自动化] TP274[自动化与计算机技术—检测技术与自动化装置]