基于GCN和HGP-SL的电力系统暂态稳定评估  

Power System Transient Stability Assessment Based on Graph Convolutional Network and Hierarchical Graph Pooling with Structure Learning

作　　者：周宇 肖健梅[1] 王锡淮[1] ZHOU Yu;XIAO Jianmei;WANG Xihuai(Logistics Engineering College,Shanghai Maritime University,Shanghai 201306)

机构地区：[1]上海海事大学物流工程学院,上海201306

出　　处：《电气工程学报》2024年第4期246-254,共9页Journal of Electrical Engineering

基　　金：国家自然科学基金资助项目(71771143)。

摘　　要：当前基于人工智能的电力系统暂态稳定评估研究多以欧式结构数据为输入,为了考虑系统拓扑结构对电力系统暂态稳定的影响,提出一种基于图卷积神经网络(Graph convolutional network,GCN)和具有结构学习的层次图池化(Hierarchical graph pooling with structure learning,HGP-SL)的电力系统暂态稳定评估模型。首先,解构电力系统,以母线为节点,输电线路为边,创建图这一典型非欧式结构数据;然后,结合图深度学习思想,通过提出的GCN+HGP-SL模型对解构后形成的电力系统潮流数据进行特征提取,建立其与电力系统暂态稳定之间的映射关系,其中HGP-SL包含对节点降采样和学习节点间结构两个步骤,其目的是捕捉重要节点的同时不破坏结构本身;最后,建立性能评价指标体系,选取对照神经网络组,对所提模型进行评估,结合算例分析各因素对模型的影响。算例分析表明,所提模型具有更好的综合性能表现。At present,the research of power system transient stability assessment based on artificial intelligence mostly takes euclidean structure data as input.In order to consider the influence of system topology on power system transient stability,a model based on graph convolutional neural network and hierarchical graph pooling with structure learning for power system transient stability assessment is proposed.Firstly,the power system is deconstructed.Taking the buses as the nodes and the transmission lines as the edges,and then a graph which is a typical non-euclidean structure data is created.Secondly,combined with the idea of graph deep learning,the proposed GCN+HGP-SL model is used to extract the features of power flow data formed after the deconstruction and establish the mapping relationship between the features and the power system transient stability.HGP-SL includes two steps:dropping the sampling and learning the structure between nodes,which aims to catch the important nodes without destroying the structure itself.Finally,to evaluate the proposed model,the performance evaluation index system is established,and the control group is selected.The influence of various factors on the model is analyzed with the experiment.The experiment shows that the proposed model has better comprehensive performance.

关 键 词：电力系统暂态稳定评估 非欧式结构数据 图深度学习 图卷积神经网络 具有结构学习的层次图池化 

分 类 号：TM712[电气工程—电力系统及自动化]