基于GCN-LSTM的电力系统暂态电压稳定评估  

Transient Voltage Stability Assessment of Power System Based on GCN⁃LSTM

作　　者：徐焕 夏凡 陈铈 赵青尧 魏晓燕 梅子薇 XU Huan;XIA Fan;CHEN Shi;ZHAO Qingyao;WEI Xiaoyan;MEI Ziwei(State Grid Hubei Electric Power Company Information and Communication,Wuhan 430000,China)

机构地区：[1]国网湖北省电力公司信息通信公司,武汉430000

出　　处：《高压电器》2025年第1期104-111,144,共9页High Voltage Apparatus

摘　　要：为了提高电力系统暂态电压在系统拓扑结构发生变化时能够稳定评估,以及提高在时空方面的特征提取能力,提出一种图卷积网络与循环神经网络相融合的方法。首先,引入图卷积网络对电力数据进行图表示,将电力系统建模为网络结构,自动学习电压节点的特征表示。其次,提出使用循环神经网络来处理暂态电压数据的时间依赖关系,捕捉暂态电压数据的时序特征。然后,提出自适应增强模块,用于将两个输出特征表示相互融合,提高模型在系统拓扑结构上的时空特征提取能力。最后,通过算例验证表明,相比于传统的评估模型,所提方法具有更高的预测精度和有效性。In order to improve the stability assessment of transient voltage of power system in case of topology varia⁃tion and enhance the feature extraction ability in spatiotemporal aspects,a method combining graph convolutional network and recurrent neural network is proposed.Firstly,a graph convolutional network is introduced to represent power data,modeling the power system as a network structure and automatically learning the feature representations of voltage nodes.Then,the recurrent neural network is proposed to handle the temporal dependencies of transient voltage data and capture the temporal characteristics of transient voltage data.After that,an adaptive enhancement module is proposed to fuse the two output feature representations with each other and improve the spatiotemporal fea⁃ture extraction ability of the model on the system topology structure.Finally,it is shown by numerical examples that the proposed method,compared to traditional assessment models,has higher prediction accuracy and effectiveness.

关 键 词：电力系统 暂态电压稳定性 循环神经网络 图卷积网络 

分 类 号：TM712[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]