基于多维特征融合的全局图注意力网络的配电网脆弱线路辨识

范文礼; 王雨婕; 赵宇川; 张叶清; 曾怡达

doi:10.19783/j.cnki.pspc.260259

欢迎来到能源电力期刊网！

English Version 期刊登录

您当前的位置：

首页 >

文章列表页 >

基于多维特征融合的全局图注意力网络的配电网脆弱线路辨识

工程应用 | 更新时间：2026-09-23

- 基于多维特征融合的全局图注意力网络的配电网脆弱线路辨识
- Identification of vulnerable lines in distribution networks based on global graph attention network with multi-dimensional feature fusion
- 电力系统保护与控制 2026年54卷第18期
- 作者机构：
  
  西南交通大学电气工程学院,四川,成都,611756
- 作者简介：
- 基金信息：
- DOI：10.19783/j.cnki.pspc.260259
  中图分类号：
- 纸质出版：2026
- 稿件说明：
移动端阅览
范文礼, 王雨婕, 赵宇川, 等. 基于多维特征融合的全局图注意力网络的配电网脆弱线路辨识[J]. 电力系统保护与控制, 2026,54(18).

FAN Wenli, WANG Yujie, ZHAO Yuchuan, et al. Identification of vulnerable lines in distribution networks based on global graph attention network with multi-dimensional feature fusion[J]. Power System Protection and Control, 2026, 54(18).
范文礼, 王雨婕, 赵宇川, 等. 基于多维特征融合的全局图注意力网络的配电网脆弱线路辨识[J]. 电力系统保护与控制, 2026,54(18). DOI： 10.19783/j.cnki.pspc.260259.

FAN Wenli, WANG Yujie, ZHAO Yuchuan, et al. Identification of vulnerable lines in distribution networks based on global graph attention network with multi-dimensional feature fusion[J]. Power System Protection and Control, 2026, 54(18). DOI： 10.19783/j.cnki.pspc.260259.

摘要

为解决电动汽车充电集群自主接入会显著加剧电网结构脆弱性及局部停电风险的问题，提出一种基于全局图注意力网络(global graph attention network

GGAT)的配电网脆弱线路辨识方法。首先，针对图注意力网络(graph attention network

GAT)采用默认边权重而无法区分其边重要性的问题，引入边特征加权修正注意力计算分数，以体现边的重要性差异。其次，考虑到GAT模型只依赖局部邻域信息的问题，增加全局特征汇总模块，充分利用多维特征以增强模型的表达能力。然后，构建线路状态偏斜度指标，并利用GGAT挖掘电网运行状态与脆弱性之间的映射关系，实现脆弱线路的准确辨识。最后，算例仿真表明，所提方法相较于传统方法表现出最优辨识效果，从而验证了所提方法的有效性和准确性。

Abstract

To address the issue that autonomous integration of electric vehicle charging clusters significantly exacerbates grid structural vulnerability and the risk of local power outages

a method for identifying vulnerable lines in distribution networks using a global graph attention network (GGAT) is proposed. First

to address the inability of conventional graph attention network (GAT) to distinguish the importance of edges due to the use of default edge weights

edge feature-weighting is introduced to modify the attention scores

thereby capturing differences in edge importance. Second

considering that GAT models rely solely on local neighborhood information

a graph-level feature aggregation module is incorporated to fully utilize multidimensional features and enhance the model’s expressive power. Then

a line state deviation metric is developed

and GGAT is utilized to uncover the mapping relationship between the power grid operating states and vulnerability

thereby enabling accurate identification of vulnerable lines. Finally

case simulations demonstrate that the proposed model achieves superior identification performance compared to traditional methods

thereby validating its effectiveness and accuracy.

关键词

Keywords

references

ROOSENDAAL E M W. Electricity demand of passenger cars on the grid in the Netherlands in case all vehicles are electric[D]. Enschede: University of Twente, 2024.

LIU Nian, HU Xuejun, MA Li, et al. Vulnerability assessment for coupled network consisting of power grid and EV traffic network[J]. IEEE Transactions on Smart Grid, 2021, 13(1): 589-598.

LI Zhendong, YAN Jiongcheng, LIU Yutian, et al. Power system transient voltage vulnerability assessment based on knowledge visualization of CNN[J]. International Journal of Electrical Power Energy Systems, 2024, 155.

ZHU Yuhong, ZHOU Yongzhi, WEI Wei, et al. Real-time cascading failure risk evaluation with high penetration of renewable energy based on a graph convolutional network[J]. IEEE Transactions on Power Systems, 2023, 38(5): 4122-4133.

GONG Liyu, CHENG Qiang. Exploiting edge features for graph neural networks[C] // 32nd IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 15-20, 2019, Long Beach, CA, USA: 9203-9211.

BHATTI U A, TANG Hao, WU Guilu, et al. Deep learning with graph convolutional networks: an overview and latest applications in computational intelligence[J]. International Journal of Intelligent Systems, 2023(1).

ZHOU Jin, ZHANG Heng, CHENG Haozhong, et al. Resilience-oriented transmission expansion planning under hurricane impact considering vulnerable line identification and hardening[J]. Protection and Control of Modern Power Systems, 2025, 10(3): 98-113.

WU Yun, PAN Xingyu, YANG Jieming. VMD-informer- DCC for photovoltaic power prediction[J]. IEICE Transactions on Communications, 2024, 107(7): 487-494.

QU Haoao, KUANG Haoxuan, WANG Qiuxuan, et al. A physics-informed and attention-based graph learning approach for regional electric vehicle charging demand prediction[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(10): 14284-14297.

JIANG Enyu, ZHANG Wentao, XUE Ang, et al. Vulnerable node identification method for distribution networks based on complex networks and improved TOPSIS theory[J]. IET Generation, Transmission Distribution, 2023, 17(22): 4991-5002.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

信息-物理-社会耦合视角下电动汽车集群调频控制综述

计及电池损耗的电动汽车 V2G 多目标优化与功率控制

面向电压三相不平衡抑制的电动汽车聚合商优化调度

考虑温度影响的电动汽车集群市场化双型需求响应运营策略

考虑截获交通流量与充电行驶距离的电动汽车充电网络规划