基于异质边图注意力网络的电力系统振荡评估模型  

Power System Oscillation Evaluation Model Based on Heterogeneous Edge Graph Attention Network

作　　者：朱思婷 管霖[1,2] 郭梦轩 黄济宇 陈鎏凯 钟智 ZHU Siting;GUAN Lin;GUO Mengxuan;HUANG Jiyu;CHEN Liukai;ZHONG Zhi(School of Electric Power,South China University of Technology,Guangzhou 510641,Guangdong Province,China;Guangdong Provincial Key Laboratory of Intelligent Operation and Control for New Energy Power System,Guangzhou 510663,Guangdong Province,China)

机构地区：[1]华南理工大学电力学院,广东省广州市510641 [2]广东省新能源电力系统智能运行与控制企业重点实验室,广东省广州市510663

出　　处：《电网技术》2022年第7期2581-2592,共12页Power System Technology

基　　金：国家自然科学基金项目(52077080);南方电网公司科技项目(ZDKJXM20180084)。

摘　　要：新能源发电使电网潮流变化更加快速,跟踪潮流变化,在线预测电网关键振荡模式的阻尼比和机组参与因子对维护电网运行安全有重要意义。该文采用数据驱动的建模思路,设计了基于多任务学习和深度学习框架的电力系统小干扰稳定评估模型(small-signal stability assessment,SSA),可同时实现多振荡模式的阻尼比预测和机组参与因子预测任务。基于图注意力网络和异质图思想,设计了引入边信息的边图注意力机制和将节点和边分类处理的异质图处理方法,建立了能有效利用边信息的异质边图注意力网络模型(heterogeneous edge graph attention network,HEGAT)。以HEGAT的特征聚合为基础,通过多任务共享参数和基于联合误差函数的训练提高了特征提取能力。IEEE10机39节点算例的对比实验表明,HEGAT-SSA能快速准确的预测模式和模态变化,并具有对拓扑变化的良好适应能力。Renewable generation speeds the power flow variation.It is important for the security of the power system to on-line predict the damping ratios and the participation factors of key oscillation modes by following the power flow changes.Based on the idea of data driving modelling,a Small-Signal Stability Assessment(SSA)model is designed in the framework of deep learning and the multi-task learning,which may assesses the damping ratios and the participation factors of the multiple oscillation modes at the same time.On the basis of the graph attention network and the heterogeneous graphs,this paper designs a Heterogeneous Edge Graph Attention Network(HEGAT)model,a novel edge graph attention mechanism that integrates the edge information,as well as the heterogeneous graph processing method for different types of nodes and edges.Based on the feature aggregation of HEGAT,the ability of feature extraction is improved through the multi-task sharing of the parameters and the joint loss function training.The comparative experiments on the IEEE 39-bus system show that the HEGAT-SSA predicts the oscillation mode and modal changes accurately and have a good adaptability to the power system topological changes.

关 键 词：小干扰稳定评估 图深度学习 边注意力 异质图 多任务学习 

分 类 号：TM721[电气工程—电力系统及自动化]