考虑数据缺失的图注意力网络暂态稳定评估
Transient Stability Assessment of Graph Attention Networks Considering Data Missing
周生存 1罗毅 1易煊承 1吴亚宁 1李丁 1熊逸1
作者信息
- 1. 强电磁工程与新技术国家重点实验室(华中科技大学电气与电子工程学院),湖北武汉 430074
- 折叠
摘要
基于人工智能的暂态稳定评估模型的性能高度依赖于系统的可观测性,而通信延迟和相量测量单元(phasor measurement units,PMU)故障等因素易导致数据缺失,使模型的评估性能下降.针对该问题,提出了一种基于图注意力网络(graph attention network,GAT)的暂态稳定评估模型.首先,根据原始网络拓扑及PMU配置方案获得表征系统可观测性的掩码矩阵,在任意PMU缺失的条件下,利用掩码矩阵训练模型;其次,通过GAT网络的多头注意力机制提取输入节点的时空信息,利用不同的权重聚合目标节点的邻域特征,实现对可观测数据的充分利用;最后,利用焦点损失函数加强模型对失稳样本的学习能力.仿真结果表明,所提方法可以最大限度地利用可观测数据,具有高精度和强鲁棒性,并且不受网络拓扑的限制,易于迁移.
Abstract
The performance of the transient stability assessment model based on artificial intelligence is highly dependent on the observability of the system,while the factors such as communication delays and PMU faults can easily lead to data missing,which degrades the model's assessment performance.To address this problem,this paper proposes a transient stability assessment model based on graph attention network(GAT).Firstly,a mask matrix representing the system observability is obtained based on the original network topology and PMU configuration scheme,and the mask matrix is used to train the model under the condition of any PMU missing.Secondly,the spatio-temporal information of the input node is extracted through the multi-head attention mechanism of the GAT network,and different weights are used to aggregate the neighborhood characteristics of the target node to make full use of observable data.Finally,the focus loss function is used to enhance the model's learning ability for unstable samples.The simulation results show that the proposed method can maximize the use of observable data with high precision and strong robustness,and is not limited by the network topology and easy to migrate.
关键词
暂态稳定评估/数据缺失/图注意力网络/掩码矩阵/PMU故障Key words
transient stability assessment/data missing/graph attention network/mask matrix/PMU fault引用本文复制引用
基金项目
中国南方电网有限责任公司科技项目(EDRI-GH-KJXM-2021-101)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据