摘要
电力系统数据采集、测量、传输和存储等过程均可能出现数据缺失问题,威胁电网安全。针对传统电力系统缺失数据重建方法仅考虑数据分布规律,忽略了数据时序与空间特性的问题,提出一种考虑时空特性的电力系统缺失数据重建模型(spatial-temporal seq2seq imputation model, ST-SSIM)。ST-SSIM具备编码-解码结构,编码器由基于图卷积层与长短时记忆单元构造的时空信息提取单元组成,用于提取数据高维时空特征,解码器由长短时记忆单元与全连接层组成,用于解码高维特征,生成电力系统数据。所提模型的输入包括电力系统数据时间序列与电网拓扑邻接矩阵,因此ST-SSIM可实现对电力系统数据复杂时空关系的自动学习。算例中,将所提方法与现有方法在不同规模电网下比较,ST-SSIM具有最高的重建精度,证明了ST-SSIM能有效地学习到电力系统数据的时空特性。通过讨论重建误差与数据缺失节点数以及缺失时间跨度的关系,验证了所提模型重建效果较稳定。
Power system data may miss during acquisition, measurement, transmission and storage, which threatens the security of power grid. Since traditional missing data reconstruction methods only consider the data distribution and ignore the spatio-temporal characteristics, a power system missing data reconstruction model called ST-SSIM(spatio-temporal seq2 seq imputation model) is proposed in this paper. ST-SSIM has an encoder-decoder structure. The encoder is composed of a spatio-temporal information extraction unit which is constructed by graph convolution layer and long short-term memory cell. The decoder is composed of long short-term memory cell and full connection layer. The input of the proposed model includes power system timeseries and adjacency matrix, so ST-SSIM can realize the automatic learning of complex time-space relationship of data. In experiment, compared the proposed method with the existing methods in power grids of different scales, ST-SSIM has the highest reconstruction accuracy, which proves that ST-SSIM can effectively learn the spatio-temporal characteristics of power system data. By discussing the relationship between reconstruction error and the number of missing nodes and time span, it is verified that the reconstruction effect of the proposed model is stable.
作者
宋铁维
施伟锋
毕宗
SONG Tiewei;SHI Weifeng;BI Zong(Department of Electrical Automation,Shanghai Maritime University,Shanghai 201306,China)
出处
《电力建设》
CSCD
北大核心
2022年第7期103-112,共10页
Electric Power Construction
基金
上海市科技计划项目资助(20040501200)。
关键词
电力系统
缺失数据重建
时空特性
图卷积
长短时记忆单元
power system
missing data reconstruction
spatio-temporal characteristics
graph convolution network
long short-term memory cell
